Field Guide: Research

Across 1,000 children followed from birth to age 32, childhood self-control predicted adult physical health, substance dependence, personal finances and criminal offending, separately from intelligence and social class. In 500 sibling pairs, the sibling with less self-control had worse outcomes despite the same household.

Measured in: The Dunedin birth cohort, New Zealand, 1,000 children followed to age 32

It measures self-control in childhood, so it establishes how much the capacity matters rather than proving that building it later produces the same result. It runs as a gradient across the whole population, not a threshold you clear or fail.

What could explain it instead: Observational. The sibling comparison is the strongest part: siblings share household, income and much of their genetics, so it removes most of what would otherwise explain the gap.

Moffitt et al., a gradient of childhood self-control predicts health, wealth and public safety · PNAS 2011

In a conceptual replication of the marshmallow test, each additional minute a 4-year-old managed to wait for a treat predicted about one tenth of a standard deviation (0.1 SD) higher academic achievement at age 15. That link was about half the size found in the original 1990 study, and it shrank by roughly two thirds once family background, early cognitive ability and the home environment were accounted for. Most of the gain came from being able to wait at least 20 seconds; associations with behavior at 15 were much smaller and rarely reached statistical significance.

Measured in: United States children followed from preschool to age 15, focusing on those whose mothers had not completed college

It measures a single behavior at age 4, not adult discipline, and the effect that survives the controls is small. It supports the idea that early self-control counts for a little while showing that circumstances explain most of the raw gap.

What could explain it instead: Observational. The replication was built around the confounds: once family background, early cognitive ability and the home environment were controlled, about two thirds of the association disappeared, so much of what looks like early willpower reflects the setting a child grows up in.

Watts, Duncan and Quan, revisiting the marshmallow test, a conceptual replication of early delay of gratification and later outcomes · Psychological Science 2018

Benefit peaks around 30 to 60 minutes a week, at roughly 10 to 20% lower risk, and the estimated benefit fades to nothing by about 130 to 140 minutes. The curve was J-shaped for mortality, cardiovascular disease and cancer, and L-shaped for diabetes, where it did not turn back up.

Measured in: 16 prospective cohort studies, between roughly 167,000 and 540,000 people depending on the outcome

Whether higher volumes actively harm is unclear, and the authors say so directly rather than claiming reversal. The upturn in the curve rests on sparse data at high volumes, exposure was self-reported, and the shape did not hold for every outcome. Treat this as a suggestive hormetic curve rather than a demonstrated one.

What could explain it instead: Self-reported training minutes in observational cohorts, which are poorly measured at the high end. People who train very large volumes may also differ systematically in ways adjustment misses.

Momma et al., muscle-strengthening activities and risk and mortality in major non-communicable diseases · Br J Sports Med 2022;56(13):755-763

In a double-blind randomized trial, 54 adults took 1000 mg of vitamin C and 235 mg of vitamin E or a placebo daily through 11 weeks of endurance training. The mitochondrial-building markers COX4 and PGC-1 alpha rose 59% and 19% in the placebo group but did not rise in the supplement group (COX4 -13%, PGC-1 alpha -13%; P 0.03 or lower between groups). VO2 max rose about 8% in both groups, so the blunting showed up in the muscle markers while the performance tests were unchanged.

Measured in: 54 young men and women, 11 weeks of supervised endurance training

This is one 11-week trial, and the blunting was measured in muscle markers, while the performance tests improved about equally in both groups. It concerns large supplemental doses taken around training, not vitamin C and E from food. It supports the idea that the training stress is itself the signal the adaptation reads.

Paulsen et al., vitamin C and E supplementation hampers cellular adaptation to endurance training in humans, a double-blind randomised controlled trial · J Physiol 2014;592(8):1887-1901

In animals, heat acclimation raises tolerance to other stressors including low oxygen and restricted blood flow, apparently because they share protective machinery. Small human studies show improved cellular tolerance and exercise performance in acute low oxygen after heat acclimation.

Measured in: Largely rodent work, with small human studies in exercise performance

This is the best available case that heat trains a general defense rather than only warming tissue, and most of it is in animals. The human evidence is thinner and concerns exercise performance rather than clinical outcomes. No health benefit has been demonstrated in people.

Horowitz, heat acclimation-mediated cross-tolerance, origins in within-life epigenetics · Front Physiol 2017;8:548 Lee et al., cross acclimation between heat and hypoxia, heat acclimation improves cellular tolerance and exercise performance in acute normobaric hypoxia · Front Physiol 2016

Cells generally, not only nerve and muscle, hold a voltage across their membrane, maintained by ion pumps and channels. Values run from around minus 10 millivolts in rapidly dividing and cancerous cells to about minus 90 in mature, fully differentiated ones.

Measured in: Universal in biology

The spread matters more than the numbers. Cells that have not yet differentiated sit depolarized, and cells that have finished sit polarized, which is the observation the whole bioelectric research program is built on.

Levin, bioelectric signaling: reprogrammable circuits underlying embryogenesis, regeneration and cancer · Cell 2021;184(8):1971-1989

A cut generates a direct-current field at the wound edge, measured at about 107 to 148 millivolts per millimeter in the skin of young adults, and weakening as it heals. In cell culture and in mice, an applied field of similar strength acts as the dominant directional cue and overrides others.

Measured in: Human skin measurements; mechanism from cell culture and genetically modified mice

The field measured directly in human skin is about 107 to 148 millivolts per millimeter in young adults, roughly 122 in fresh mouse wounds, and about a third weaker in people over 65 than in those under 25. The mechanism work is in cells and knockout mice rather than in intact people.

Nuccitelli et al., the electric field near human skin wounds declines with age and provides a noninvasive indicator of wound healing · Wound Repair Regen 2011;19(5):645-655 Zhao et al., electrical signals control wound healing through PI3K-gamma and PTEN · Nature 2006;442(7101):457-460

Langevin and colleagues proposed that rotating a needle winds connective tissue around it and transmits a mechanical signal to fibroblasts, and that this coupling produces the resistance practitioners call needle grasp. Separately, they mapped an 80% correspondence between acupuncture point locations and connective tissue planes in serial sections of the human arm.

Measured in: A hypothesis paper drawing on the authors' human and rat work, plus a cadaveric mapping study of the arm

The mechanism paper is framed by its authors as a hypothesis rather than a demonstration, and the 80% correspondence is an anatomical correlation in cadaver sections, not evidence of a functional channel. It is a testable physical account of needling that does not require meridians to be conduits.

Langevin et al., mechanical signaling through connective tissue, a mechanism for the therapeutic effect of acupuncture · FASEB J 2001;15(12):2275-2282 Langevin & Yandow, relationship of acupuncture points and meridians to connective tissue planes · Anat Rec 2002;269(6):257-265

Loading bone generates electric potentials at the strained tissue, an effect tied to collagen's piezoelectricity and to fluid movement through bone. These stress-generated potentials alter cell proliferation and matrix production, which is one proposed route by which bone builds where it is loaded. Electrical stimulation devices grew out of this and have been used to help stubborn fractures knit.

Measured in: Review of laboratory, animal and device studies of bone bioelectricity

The piezoelectric effect of loaded bone is well established, but its exact role in everyday remodeling and the clinical value of electrical stimulation are still debated, with the review noting uneven trial design and dosing. This card grades the mechanism, not any stimulation product.

Isaacson & Bloebaum, bone bioelectricity, what have we learned in the past 160 years · J Biomed Mater Res A 2010;95(4):1270-1279

The exclusion zone near strongly hydrophilic surfaces, a layer that repels microspheres, is real and has been independently demonstrated by several groups. What is contested is the explanation: Schurr's diffusiophoresis account competes with Pollack's idea that the water there is structurally different from bulk water, and the reviewers judge the diffusiophoresis explanation the more compelling one.

Measured in: Physical chemistry, laboratory measurement

The interface phenomenon is well documented and still needs a theoretical explanation, which is a different statement from the one printed on a bottle. That no structuring device has shown a health benefit is our own reading of the absence of trials, not a finding from this review.

Elton et al., exclusion zone phenomena in water, a critical review of experimental findings and theories · Int J Mol Sci 2020;21(14):5041

Rising core temperature activates HSF1, which produces chaperone proteins that refold damaged proteins and dampen inflammation

Measured in: Humans, repeatedly demonstrated

A narrative review, not a demonstration, so it is a fair source for the mechanism and a weak one for any specific number. That heat induces these proteins in people is solid. That this specific induction produces the epidemiological signal is an inference.

Nagai & Kaji, thermal effect on the heat shock protein 70 family to prevent atherosclerotic cardiovascular disease · Biomolecules 2023;13(5):867

Growth hormone rose sixteen-fold in one 1986 Finnish study, measured in ten men under an unusually heavy protocol: an hour in a 176°F (80 °C) sauna, twice a day, for a week. The rise faded after the third day.

Measured in: 10 men (the women in the study took part only in the prolactin measurements)

Two hours of sauna a day for a week is not normal use, so this is not what a typical session does. The study's own authors attribute the hormone rise partly to dehydration, not to heat itself. It has not been shown to change body composition or any aging marker.

Leppaluoto et al., endocrine effects of repeated sauna bathing · Acta Physiol Scand 1986;128(3):467-470

Eight weeks of hot-water immersion, four to five times a week, took flow-mediated dilation from 5.6% to 10.9%, aortic pulse-wave velocity from 7.1 to 6.1 m/s, carotid wall thickness from 0.43 to 0.37 mm and mean arterial pressure from 83 to 78 mmHg. Nothing moved in the sham arm.

Measured in: 20 young sedentary adults, 10 heated and 10 in a thermoneutral sham

The protocol was demanding: a 104.9°F (40.5 °C) bath held long enough to keep core temperature above 101.3°F (38.5 °C) for 60 minutes, four to five times a week. The authors report the gains as similar to what exercise training produces in the same population, and larger for dilation. Twenty people is a small trial.

Brunt et al., passive heat therapy improves endothelial function, arterial stiffness and blood pressure · J Physiol 2016;594(18):5329-42

One session against a sham matched for duration. On the 17-item Hamilton depression scale (0 to 52), scores fell 6.5 points versus sham at week one and were still 4.3 points lower at week six, the last measurement taken.

Measured in: 34 adults with major depressive disorder randomized, 29 in the efficacy analysis

The benefit had not worn off when the trial stopped looking at six weeks. It is one small double-blind trial, and a later trial pairing whole-body hyperthermia with therapy saw the sham arm improve about as much, which is why this sits at emerging.

Janssen et al., whole-body hyperthermia for major depressive disorder · JAMA Psychiatry 2016

At 4 to 7 sessions a week versus one, cardiovascular deaths ran 12.0% against 22.3%, and all-cause deaths 30.8% against 49.1%, over a median 20.7 years. Sudden cardiac death was the sharpest single figure, at a hazard ratio of 0.37.

Measured in: 2,315 middle-aged men, aged 42 to 60, in Eastern Finland (the KIHD cohort)

The dose-response is consistent. This and the dementia finding below come from the same 2,315 Finnish men, so they are two results from one cohort, not two independent lines of evidence.

What could explain it instead: Healthy-user bias: people well enough to sauna four to seven times a week are healthier to begin with. Early illness also reduces sauna use, which can make sauna look more protective than it is.

Laukkanen et al., sauna bathing and fatal cardiovascular and all-cause mortality · JAMA Intern Med 2015;175(4):542-548

Dementia at a hazard ratio of 0.34 and Alzheimer's at 0.35, for 4 to 7 sessions a week against one

Measured in: The same 2,315 Finnish men, followed a median 20.7 years

Adjusted for age, alcohol, body mass, blood pressure, smoking, diabetes, prior heart attack, heart rate and cholesterol. It is the same cohort as the mortality finding, and no randomized trial of heat and dementia exists.

What could explain it instead: Healthy-user bias and reverse causation. Early cognitive decline itself reduces sauna use.

Laukkanen et al., sauna bathing and risk of dementia and Alzheimer's disease · Age Ageing 2017

The American College of Physicians recommends that all adults receive cognitive behavioral therapy for insomnia as the initial treatment for chronic insomnia. It is a strong recommendation resting on moderate-quality evidence.

Measured in: Adults with chronic insomnia, across many trials

A strong recommendation on moderate-quality evidence means the direction is clear while the size of the effect is less settled. It takes several weeks, and the component people abandon first, restricting time in bed, is the one doing the work.

Qaseem et al., management of chronic insomnia disorder in adults, ACP clinical practice guideline · Ann Intern Med 2016

At six months, mean ISI (Insomnia Severity Index, 0-28) was 10.9 (SD 5.5) with four sessions of sleep restriction therapy versus 13.9 (SD 5.2) with a sleep hygiene booklet, an adjusted mean difference of -3.05 (95% CI -3.83 to -2.28; p<0.0001), Cohen's d -0.74. 642 adults randomized, 90% followed up.

Measured in: 642 adults with insomnia disorder recruited from 35 general practices across England

Sleep restriction is the component people drop first because it means spending less time in bed at the start, and it is the one doing most of the work. The trial was open-label, so participants knew their arm, which can lift self-reported scores, and the comparator was a booklet, not active attention. A gain of about 3 points on a 0-28 scale is clinically meaningful and moved many people below the clinical-insomnia threshold, at a cost per quality-adjusted life-year of about GBP 2,076.

Kyle et al., clinical and cost-effectiveness of nurse-delivered sleep restriction therapy for insomnia in primary care (HABIT), pragmatic randomised controlled trial · Lancet 2023;402(10406):975-987

In a component network meta-analysis of 241 trials, remission odds rose with cognitive restructuring (incremental odds ratio 1.68, 95% CI 1.28-2.20), third-wave methods (1.49, 95% CI 1.10-2.03), sleep restriction (1.49, 95% CI 1.04-2.13) and stimulus control (1.43, 95% CI 1.00-2.05). Sleep hygiene alone did nothing (1.01, 95% CI 0.77-1.32) and relaxation procedures trended against remission (0.81, 95% CI 0.64-1.02). The best combination reached a risk difference of 0.33 (95% CI 0.23-0.43), a number needed to treat of 3.0 versus education. In-person delivery outperformed other formats (incremental odds ratio 1.83, 95% CI 1.19-2.81).

Measured in: 31,452 adults with chronic insomnia across 241 trials, mean age 45, 67% women, mostly North America and Europe

This isolates the working parts of CBT-I. The behavioral core, restricting time in bed and reserving the bed for sleep, plus the cognitive work carry the effect, while sleep hygiene advice and progressive relaxation do not, and relaxation may slightly work against remission. A number needed to treat of 3 for the best package is strong for any treatment. Component network meta-analysis models the pieces statistically, so the estimate for any single component is less certain than a head-to-head trial of that component.

Furukawa et al., components and delivery formats of cognitive behavioral therapy for chronic insomnia in adults, systematic review and component network meta-analysis · JAMA Psychiatry 2024;81(4):357-365

Versus sleep hygiene education, digital CBT-I delivered by the Sleepio app improved sleep-related quality of life at week 8 (adjusted difference -17.60 on the Glasgow Sleep Impact Index, where lower is better, 95% CI -20.81 to -14.39), psychological well-being (+2.68 on the WEMWBS 14-70 scale, 95% CI 1.89-3.47) and functional health (+1.76, 95% CI 1.24-2.28). Improvement in insomnia itself mediated 45.5% to 84.0% of these gains. 1,711 adults, intention-to-treat.

Measured in: 1,711 adults with insomnia symptoms, mean age 48, 78% women, treatment delivered entirely online

This is the scalable form of the first-line treatment: an automated app with no therapist reached over 1,700 people. The comparator was sleep hygiene education, not face-to-face CBT-I, and the effects on general health were small even where the sleep-specific gains were large. The value is reach and access, since the same behavioral and cognitive core can be delivered digitally when a therapist is not available.

Espie et al., effect of digital cognitive behavioral therapy for insomnia on health, psychological well-being, and sleep-related quality of life, randomized clinical trial · JAMA Psychiatry 2019;76(1):21-30

In a network meta-analysis of 154 double-blind trials (44,089 adults), for acute treatment benzodiazepines, eszopiclone, zolpidem, zopiclone, doxylamine, lemborexant and seltorexant beat placebo (standardized mean differences roughly 0.36 to 0.83). Over the longer term only eszopiclone (SMD 0.63, 95% CI 0.36-0.90) and lemborexant (0.41, 95% CI 0.04-0.78) beat placebo. Zolpidem and zopiclone had higher dropout for side effects (odds ratios about 1.8 to 2.0 versus placebo); melatonin and ramelteon showed no material benefit.

Measured in: 44,089 adults with insomnia disorder across 154 double-blind trials

These come after the behavioral first line because the guideline-preferred treatment, CBT-I, works without side effects or dependence risk and holds up after it stops. Among the drugs, the newer orexin blocker lemborexant and eszopiclone had the most durable benefit; the older z-drugs zolpidem and zopiclone work but more people stop them for side effects, and long-term safety data are thin for most agents. Any of these is a shared decision with a prescriber, and none removes what is disrupting sleep.

De Crescenzo et al., comparative effects of pharmacological interventions for the acute and long-term management of insomnia disorder in adults, systematic review and network meta-analysis · Lancet 2022;400(10347):170-184

A taught self-acupressure routine beat sleep-hygiene education on insomnia severity at both four and eight weeks, with a moderate effect size rising from 0.51 to 0.67.

Measured in: 200 adults with insomnia disorder, both arms given matched training time

Both groups received the same amount of teaching, which makes this a fairer comparison than most in this area. The gain of roughly 2 to 3 points on the insomnia severity scale (ISI, 0-28) is statistically clear and modest in size. Blinding a hands-on practice is close to impossible.

Yeung et al., self-administered acupressure for insomnia disorder, randomized controlled trial · Phytomedicine 2022;99:153993

Across 22 RCTs (1,806 adults), exercise improved subjective sleep quality by a pooled mean difference of -2.19 on the PSQI (Pittsburgh Sleep Quality Index, 0-21; 95% CI -2.96 to -1.41) versus controls, with similar gains from physical exercise (-2.22) and mind-body exercise such as yoga and tai chi (-2.17). Insomnia severity fell -1.52 on the ISI (0-28; 95% CI -2.63 to -0.41) and daytime sleepiness fell -2.55 on the Epworth scale. Sleep-onset latency did not change.

Measured in: 1,806 adults across 22 randomized trials, poor sleepers and insomnia populations, studies from 1995-2019

A 2-point PSQI gain is real and about the size you would expect from a change you can start on your own. Shorter programs of three months or less helped more than longer ones in this pooled data, which most likely reflects adherence fading over time. Exercise cannot be blinded, so some of the effect is expectation, and the pooled trials varied in intensity and format.

Xie et al., effects of exercise on sleep quality and insomnia in adults, systematic review and meta-analysis of randomized controlled trials · Front Psychiatry 2021;12:664499

Across 18 RCTs (1,654 adults), mindfulness meditation improved sleep quality against nonspecific active controls with an effect size of 0.33 (95% CI 0.17-0.48) at post-treatment and 0.54 (95% CI 0.24-0.84) at follow-up. Measured head-to-head against evidence-based sleep treatments, the difference was 0.03 (95% CI -0.43 to 0.49) at post-treatment and -0.14 (95% CI -0.62 to 0.34) at follow-up, essentially level.

Measured in: 1,654 adults across 18 randomized trials, mixed clinical populations

Mindfulness beats spending an equal amount of time on some other activity, and the effect grows in the months after training, which fits a skill you keep practicing. In the head-to-head trials it sat level with established treatments such as CBT-I, so it belongs alongside the behavioral first line, not ahead of it. It is a reasonable choice for people who prefer a meditative practice or want to combine it with sleep restriction and stimulus control.

Rusch et al., the effect of mindfulness meditation on sleep quality, systematic review and meta-analysis of randomized controlled trials · Ann N Y Acad Sci 2019;1445(1):5-16

Across 19 RCTs (1,683 people) with primary sleep disorders, melatonin reduced sleep-onset latency by a weighted mean of 7.06 minutes (95% CI 4.37-9.75), increased total sleep time by 8.25 minutes (95% CI 1.74-14.75) and improved overall sleep quality by a standardized mean difference of 0.22 (95% CI 0.12-0.32) versus placebo. Effects were larger at higher doses and longer durations.

Measured in: 1,683 people with primary sleep disorders across 19 randomized placebo-controlled trials

Seven minutes is small, and it suits sleep-onset and body-clock problems, where timing the dose a few hours before target sleep matters more than the size of the effect. A separate 154-trial network analysis of insomnia disorder specifically found melatonin gave no material benefit, so its best fit is circadian and older-adult sleep-onset use, not long-standing insomnia. It is one of the better-tolerated options and useful for shift work and jet lag.

Ferracioli-Oda et al., meta-analysis, melatonin for the treatment of primary sleep disorders · PLoS One 2013;8(5):e63773 De Crescenzo et al., comparative effects of pharmacological interventions for insomnia disorder in adults, systematic review and network meta-analysis (context: melatonin no material benefit for insomnia disorder) · Lancet 2022;400(10347):170-184

Across 15 RCTs (1,108 patients), real acupuncture beat sham or placebo acupuncture on the PSQI (Pittsburgh Sleep Quality Index, 0-21): standard acupuncture -7.34 (95% CI -8.02 to -6.66; 3 trials, I-squared 86%), minimal acupuncture -3.29 (95% CI -3.95 to -2.63; 5 trials, I-squared 53%), auricular acupressure -4.16 (95% CI -6.57 to -1.75; 1 trial). Insomnia severity, total sleep time, sleep-onset latency and sleep efficiency also favored acupuncture.

Measured in: 1,108 patients with insomnia across 15 randomized trials, largely from East Asia

The comparison against sham needling is the demanding test, and acupuncture passed it, which is more than many hands-on therapies manage. Heterogeneity was high (I-squared up to 86%) and the largest effect came from only three trials, so the true size is less settled than the direction. Fully blinding a needle is difficult, and trials from a single region can run larger than average, so treat the -7.34 figure as an upper bound and the minimal-acupuncture -3.29 as the steadier estimate.

Zhang et al., the effects of acupuncture versus sham or placebo acupuncture for insomnia, systematic review and meta-analysis of randomized controlled trials · Complement Ther Clin Pract 2020;41:101253

Light therapy cut time spent awake after falling asleep by about 11 minutes on actigraphy. Sleep latency, total sleep time and sleep efficiency did not improve.

Measured in: 685 people across 22 studies, 13 pooled, only 5 rated high quality

One outcome of four moved, which is narrower than the usual summary of this literature. The authors flag heterogeneity and publication bias. Note the gap between 11 minutes measured by actigraphy and 36 minutes reported by the sleepers themselves. The pooled result is not specific to morning light; the morning finding in the same review is about shifting the body clock, not sleep quality.

Chambe et al., light therapy in insomnia disorder, systematic review and meta-analysis · J Sleep Res 2023;32(6):e13895

In laboratory conditions a bright light pulse after the core body temperature minimum, which is early morning for most people, advanced the clock by up to 2.3 hours. Light before that point delayed it.

Measured in: 23 adults aged 19 to 44 under laboratory constant-routine conditions, 16 men and 7 women, with phase data analyzable in 21

A small, young, male-skewed laboratory study, which is what the human phase-response literature largely consists of. It measures the clock moving, not sleep improving; the step to falling asleep earlier is an inference from the mechanism.

Khalsa et al., a phase response curve to single bright light pulses in human subjects · J Physiol 2003;549(Pt 3):945-952

Across randomized trials, bright light treatment reduced seasonal affective disorder symptom severity with an effect size of 0.84 (95% CI 0.60 to 1.08, eight studies), which the reviewers place in the same range as most antidepressant drug trials; dawn simulation gave 0.73 (0.37 to 1.08, five studies). A newer meta-analysis of 19 trials found a smaller but still significant benefit, a standardized mean difference of -0.37 (95% CI -0.63 to -0.12, 610 patients) and a 42% higher response rate (risk ratio 1.42, 95% CI 1.08 to 1.85, 559 patients).

Measured in: Randomized controlled trials of light therapy for seasonal affective disorder; the older review included only 13% of the studies it screened, the newer pooled 19 trials

Blinding is the structural weakness, since a person sitting at a light box knows it, and the sham conditions used are not obviously inert. The older review is from 2005 and predates the LED devices most people now buy; the newer 19-trial analysis reported moderate heterogeneity and a moderate-to-high risk of bias, with no sign of publication bias.

Golden et al., the efficacy of light therapy in the treatment of mood disorders, review and meta-analysis · Am J Psychiatry 2005;162(4):656-662 Pjrek et al., the efficacy of light therapy in the treatment of seasonal affective disorder, meta-analysis of randomized controlled trials · Psychother Psychosom 2020;89(1):17-24

Across 40 controlled studies (49 comparisons), light interventions advanced a delayed sleep timing by an effect size of 0.34 (p = 0.010), improved sleep continuity by 0.23 (p < 0.001), and reduced self-reported sleep disturbance by 0.32 (p = 0.014). Adding evening-light avoidance was linked to a greater gain in total sleep time.

Measured in: 40 controlled studies, 49 intervention comparisons, in people with intrinsic circadian rhythm sleep disorders such as delayed sleep phase and in neuropsychiatric conditions

The effects are small, and the trials ran mostly in people with a diagnosed circadian rhythm disorder or a psychiatric condition rather than in healthy sleepers, so the size in an ordinary good sleeper is likely smaller. Light doses and schedules varied across the trials.

Faulkner et al., light therapies to improve sleep in intrinsic circadian rhythm sleep disorders and neuro-psychiatric illness, systematic review and meta-analysis · Sleep Med Rev 2019;46:108-123

Over about eight years, people in the brightest daytime-light group had roughly 17 to 34% lower all-cause mortality than the darkest, and those with the brightest nights had 21 to 34% higher.

Measured in: 88,905 UK Biobank participants, mean age 62, 57% women, one week of wrist light sensing

Light was sensor-measured rather than recalled, which makes this better than most observational work in this area. These are relative hazards; absolute mortality in the cohort was about 4% over eight years. Measured in the same UK Biobank light-sensor substudy as our other light finding, by the same research group, so the two are separate analyzes of one dataset rather than independent corroboration.

What could explain it instead: Reverse causation is live: people who are already unwell go outside less. Daytime light also correlates with physical activity, which has its own effect.

Windred et al., brighter nights and darker days predict higher mortality risk · PNAS 2024;121(43):e2405924121

More daytime light was associated with lower odds of major depression, PTSD, psychosis and self-harm. More nighttime light went the other way, taking in generalized anxiety and bipolar disorder as well.

Measured in: 86,772 UK Biobank participants, mean age 62, 57% women

A single snapshot in time, so it cannot establish direction. Measured in the same UK Biobank light-sensor substudy as our other light finding, by the same research group, so the two are separate analyzes of one dataset rather than independent corroboration.

What could explain it instead: People who are depressed go outside less, so the arrow plausibly runs the other way. Being one moment in time, this design cannot separate the two.

Burns et al., day and night light exposure are associated with psychiatric disorders · Nature Mental Health 2023;1:853-862

Cold water triggers an involuntary gasp and uncontrollable hyperventilation in the first minute. Cardiac arrhythmias are common when cold-water immersion is combined with breath-holding, the pairing that drives the autonomic conflict, and some drownings attributed to other causes likely began this way.

Measured in: Human experimental work, characterized by the authors as young, fit and healthy participants

The higher figures in circulation come from submersion protocols and could not be traced to a readable source; both papers are paywalled. The readable evidence shows the breath hold, not the depth, raises the danger. That strengthens the instruction: never hold your breath, never go in alone, and never combine breathing exercises with cold water.

Datta & Tipton, respiratory responses to cold water immersion, neural pathways, interactions and clinical consequences · J Appl Physiol 2006;100(6):2057-2064 Shattock & Tipton, autonomic conflict, a different way to die during cold water immersion · J Physiol 2012;590(14):3219-3230

One hour of head-out immersion at 57.2°F (14°C) raised metabolic rate about 350% and noradrenaline about 530% against thermoneutral 89.6°F (32°C) water. Dopamine rose 250% and diuresis 163%. At 68°F (20°C) the cardiovascular responses looked like thermoneutral water despite metabolic rate rising 93%.

Measured in: Young healthy men, one-hour immersions at three water temperatures

This measures the size of the response, not a health outcome. A large catecholamine surge is the most plausible mechanism behind the alertness people report, and no trial has connected the magnitude of that surge to any downstream benefit. The 57.2°F (14°C) condition also drove rectal temperature down, so part of what was measured is early cooling rather than the cold itself.

Šrámek et al., human physiological responses to immersion into water of different temperatures · Eur J Appl Physiol 2000;81(5):436-42

Brown adipose tissue activity was detected in 23 of 24 healthy men during exposure to 60.8°F (16°C) air and in none of them at 71.6°F (22°C). Activity was significantly lower in the overweight and obese participants than the lean ones, and correlated negatively with BMI and body fat percentage.

Measured in: 24 healthy men, 10 lean and 14 overweight or obese, imaged by FDG-PET/CT

This is mild cold in air over hours, not immersion, and it establishes that adult humans have functioning brown fat rather than that activating it changes body weight. The inverse correlation with body fat is cross-sectional within the study, so it cannot say whether low brown fat contributes to obesity or follows from it.

van Marken Lichtenbelt et al., cold-activated brown adipose tissue in healthy men · N Engl J Med 2009;360(15):1500-8

Cold-water immersion reduced delayed-onset soreness and creatine kinase, and modestly improved jump height. Ten to fifteen minutes at 41–59°F (5–15°C) was the best-supported dose. Strength was not measured.

Measured in: 1,139 participants across 55 randomized trials, roughly 21 people per trial, almost all men

Our earlier wording said this showed no benefit for jump recovery. That was the reverse of what the paper found. This is a dose-comparison analysis, so its finding is that particular dose windows work rather than that cold works generally. The separate and better-known finding that routine cold after lifting blunts training adaptation comes from a different body of work.

Wang et al., dose of cold water immersion for recovery from exercise-induced muscle damage, network meta-analysis · Front Physiol 2025;16:1525726

Six three-minute immersions in 59°F (15°C) water cut the gasping response on a later cold plunge by about half: breathing rate fell 47 to 24 breaths a minute and air moved per minute fell 57%. Heart rate fell much less, from 128 to 109. A 2024 meta-analysis puts the onset of habituation at around four immersions.

Measured in: 12 men in the original trial, and 17 groups in the later meta-analysis

The halving applies to the breathing components, and hyperventilation is the part that drowns people. Heart rate habituates far less, so the cardiac risk does not fall as much as the panic does. Some protection was still measurable seven months later; by fourteen months only the heart-rate change remained.

Tipton et al., permanence of the habituation of the initial responses to cold-water immersion in humans · Eur J Appl Physiol 2000;83(1):17-21 Barwood et al., habituation of the cold shock response, systematic review and meta-analysis · J Therm Biol 2024;119:103775

In eight people cooled in 46.4°F (8°C) water to a core temperature of about 96.6°F (35.9°C), core temperature continued to fall by 0.4 °C for roughly 21 minutes after the immersion ended. When shivering was suppressed with meperidine, the afterdrop grew to 1.1 °C and lasted 89 minutes, and the rewarming rate fell 37%.

Measured in: Eight adults cooled in a laboratory to mild hypothermia, in a crossover against their own control trial

This measured deliberate cooling to mild hypothermia, which is colder than a two-minute recreational plunge, so the size of the afterdrop after a short immersion is smaller than these numbers. The mechanism is the same, and the practical point stands: anything that suppresses shivering, including alcohol, sedatives and rushing into a hot shower, makes the afterdrop deeper and longer.

Giesbrecht et al., inhibition of shivering increases core temperature afterdrop and attenuates rewarming in hypothermic humans · J Appl Physiol 1997;83(5):1630-4 Giesbrecht, cold stress, near drowning and accidental hypothermia, a review · Aviat Space Environ Med 2000;71(7):733-52 Tipton et al., cold water immersion, kill or cure? · Exp Physiol 2017;102(11):1335-1355

Across 8 randomized studies, resistance training alone produced hypertrophic adaptations likely to be at least small in magnitude, while resistance training plus post-session cold water immersion produced adaptations that were small to negligible. In one 7-week trial, 15 minutes at 50°F (10°C) after each session attenuated the growth of type II fiber cross-sectional area while 1-RM leg press gains were similar between groups.

Measured in: Roughly 116 participants across 8 studies, aged 20 to 26, trained and untrained

Size and strength come apart here, and reporting them together overstates the finding: fiber hypertrophy was attenuated while dynamic strength gains were not. Eight small studies is a thin base, and the effect applies to cold taken immediately after the session. Cold on a separate day or several hours later has not been tested.

Piñero et al., throwing cold water on muscle growth, systematic review with meta-analysis of postexercise cold water immersion and resistance training-induced hypertrophy · Eur J Sport Sci 2024;24(2):177-189 Fyfe et al., cold water immersion attenuates anabolic signaling and skeletal muscle fiber hypertrophy, but not strength gain, following whole-body resistance training · J Appl Physiol 2019;127(5):1403-1418

11 women and 14 men were immersed in 64.4°F (18°C) water for up to 90 minutes. Rectal temperature in the women fell at about half the rate it did in the men, while metabolic rate roughly tripled in both. When six women and five men were matched for body fatness, no significant sex difference remained in cooling rate, energy metabolism or fat oxidation. Cooling rate tracked the ratio of body surface area to size, and shivering metabolism tracked body fatness inversely.

Measured in: 25 healthy adults, 11 women with mean body fat 22.4% and 14 men, single 64.4°F (18°C) immersion

The direction reverses with the state the body is in. In hyperthermic people being treated for heat illness, women appear to cool faster than men, and the cooling guidelines were written from male data. So the popular claim that women cool faster is right in one setting and backwards in the other, and body composition, not sex, is what the measurements track.

Tikuisis et al., comparison of thermoregulatory responses between men and women immersed in cold water · J Appl Physiol 2000;89(4):1403-11 Hutchins et al., treating exertional heat stroke, limited understanding of the female response to cold water immersion · Front Physiol 2022;13:1055810

Stress fell sharply twelve hours after immersion, and not at any other timepoint measured: not immediately, not at one hour, not at 24 or 48 hours. Sleep quality and quality of life improved in single studies. Mood itself did not change.

Measured in: 3,177 people across 11 randomized trials, ten of which were entirely male

A benefit absent at four timepoints and large at a fifth is a fragile pattern rather than a durable effect. The same review found inflammation significantly increased immediately after immersion and at one hour, which belongs beside the stress finding rather than after it.

Cain et al., effects of cold-water immersion on health and wellbeing, systematic review and meta-analysis · PLoS One 2025;20(1):e0317615

3,018 adults were randomized to finish their daily shower with 30, 60 or 90 seconds of cold, or not, for 30 consecutive days. Sickness absence fell 29% against control (incidence rate ratio 0.71, P = 0.003). Self-reported illness days did not differ between groups. Mental health scores were slightly better at 30 days and no different at 90.

Measured in: 3,018 adults aged 18 to 65 without severe comorbidity and no prior cold-shower habit; 79% completed the 30-day protocol

Absence and illness moving in different directions is the whole result, and it is usually reported as though only the first number existed: people took less time off work while reporting the same number of days ill. Nothing was blinded, which for an outcome as expectation-sensitive as deciding whether to go to work is a real limit. Duration of cold made no difference between 30, 60 and 90 seconds.

Buijze et al., the effect of cold showering on health and work, a randomized controlled trial · PLoS One 2016;11(9):e0161749

Ten days of daily exposure to air at 57.2–59°F (14–15°C) raised peripheral insulin sensitivity by about 43% in eight people with type 2 diabetes. Skeletal muscle GLUT4 translocation increased markedly at baseline, with no effect on insulin signaling or AMPK activation, and only minor increases in brown adipose tissue glucose uptake.

Measured in: Eight patients with type 2 diabetes, measured before and after acclimation

Eight people, no control group, and a 43% change measured against their own baseline rather than against anyone else. The exposure was prolonged mild cold in air across ten days, which is a different stimulus from a short immersion, and the authors themselves note the effect ran through muscle glucose transport rather than through the brown fat mechanism that made the field interesting.

Hanssen et al., short-term cold acclimation improves insulin sensitivity in patients with type 2 diabetes mellitus · Nat Med 2015;21(8):863-5

The often-quoted eleven minutes a week is a baseline characteristics row describing seven winter-swimming men, whose inclusion criterion was having started a second season, a mean of 1.8 years of practice.

Measured in: 7 winter-swimming men and 8 male controls, young and healthy

No dose was assigned to anyone. This is what a small group of habitual swimmers happened to be doing when they were measured, reported in a table of who they were, and nearly every swimmer also used a sauna. Reading it as a prescription reverses the direction of the evidence.

What could explain it instead: The swimmers chose this habit and had sustained it for years, so they differ from controls in every trait that predicts sticking with cold water. Nearly every swimmer also used a sauna (all but one, 7 of 8), which makes heat acclimation inseparable from cold acclimation in this group, and the authors say the pattern is consistent with both.

Søberg et al., altered brown fat thermoregulation and enhanced cold-induced thermogenesis in young, healthy, winter-swimming men · Cell Rep Med 2021;2(10):100408

A single set of 6 to 12 repetitions at roughly 70 to 85% of a one-rep maximum, performed 2 to 3 times a week with high effort for 8 to 12 weeks, produced a pooled 1RM increase of 26.7 lb (12.09 kg) (95% CI 8.16 to 16.03), made up of 38.5 lb (17.48 kg) on the squat and 18.2 lb (8.25 kg) on the bench press.

Measured in: 6 studies in resistance-trained men, meta-analysis performed on 5 of them

Six studies is a small base, the participants already trained, and the authors describe these gains as suboptimal rather than as a target. There is no data at all for the deadlift, and none in trained women or in highly trained strength athletes.

Androulakis-Korakakis, Fisher and Steele, the minimum effective training dose required to increase 1RM strength in resistance-trained men · Sports Med 2020;50(4):751-765

Resistance training lowered HbA1c by 0.39 percentage points versus control (weighted mean difference -0.39, 95% CI -0.60 to -0.18, p<0.001), on a scale where 6.5% and above defines diabetes and where a drop of about 0.5 points is the size clinicians expect from adding a standard glucose-lowering medication. Pooled across 20 randomized controlled trials and 1,172 adults with type 2 diabetes; larger gains in muscular strength predicted larger HbA1c reductions.

Measured in: 1,172 adults with type 2 diabetes across 20 randomized controlled trials of supervised or structured resistance training versus non-exercising controls.

Trial durations and training protocols varied and most programs ran 8 to 16 weeks, so the durability of the effect beyond a few months is less well characterized. The pooled estimate showed statistical heterogeneity, and diet and medication were not uniformly controlled across trials.

Jansson AK, et al. Resistance training and HbA1c in type 2 diabetes: systematic review and meta-analysis · BMJ Open Diabetes Res Care 2022;10(2):e002595

Resistance training reduced depressive symptoms with a moderate standardized effect of 0.66 (95% CI 0.48 to 0.83, p<0.001), on the standardized effect-size scale where 0.2 counts as small, 0.5 as moderate and 0.8 as large. The number needed to treat was 4, meaning about 1 in every 4 people who trained gained a clinically meaningful improvement over control. Pooled from 54 effects across 33 randomized controlled trials and 1,877 participants.

Measured in: 1,877 participants across 33 randomized controlled trials (947 assigned to resistance training, 930 to control), spanning healthy adults and people with physical or mental illness.

Effects were statistically heterogeneous across trials (I-squared 76%), and many studies used self-reported symptom scales rather than clinician diagnosis. Blinding participants to an exercise assignment is not possible, so some of the benefit may reflect expectation.

Gordon BR, et al. Resistance exercise training and depressive symptoms: meta-analysis and meta-regression of RCTs · JAMA Psychiatry 2018;75(6):566-576

Resistance training was associated with 21% lower all-cause mortality compared with no exercise (HR 0.79), and 40% lower when combined with aerobic activity (HR 0.60), again against no exercise.

Measured in: 370,256 people across 11 studies, ten of them prospective cohorts and one a randomized trial

Both figures are measured against the same no-exercise comparison, so the 40% is not an increment on top of the 21%. Cardiovascular mortality did not reach significance and there was no cancer signal, so this applies to all-cause death only. Ten of eleven studies are observational.

What could explain it instead: Healthy-adherer bias. People who lift weights differ from people who do not in many ways at once, and observational designs cannot separate the training from the person doing it.

Saeidifard et al., the association of resistance training with mortality, systematic review and meta-analysis · Eur J Prev Cardiol 2019;26(15):1647-1665

The fittest group had about 80% lower mortality than the least fit (HR 0.20), and still lower mortality than the next-fittest group (HR 0.77). No upper limit of benefit appeared.

Measured in: 122,007 patients referred for exercise treadmill testing, 41% women, mean age 53, median follow-up 8.4 years

Fitness here was measured directly on a treadmill rather than self-reported, which is what separates this from most activity research and is the reason it carries weight. These were clinically referred patients rather than a general-population sample.

What could explain it instead: Observational. Undetected early illness lowers measured fitness and independently raises mortality, which explains part of the gradient.

Mandsager et al., cardiorespiratory fitness and long-term mortality among adults undergoing exercise treadmill testing · JAMA Netw Open 2018;1(6):e183605

Pooling 38 cohort studies, higher handgrip strength carried a hazard ratio of 0.69 for all-cause mortality, stronger in women (0.60) than in men (0.69), and higher knee extension strength carried a 14% lower risk. In the PURE cohort, every 11 lb (5 kg) of grip strength lost carried a 16% higher rate of death (HR 1.16, 95% CI 1.13 to 1.20), and grip predicted mortality better than systolic blood pressure did.

Measured in: Approximately 1.9 million participants and 63,087 deaths across 38 studies in the pooled analysis; 139,691 people in 17 countries followed a median of 4 years in PURE, with 3,379 deaths

Grip strength is a proxy for whole-body condition rather than an outcome of training, and neither dataset asked whether the strong people had trained. It establishes that being strong tracks with living longer, not that becoming stronger extends life.

What could explain it instead: Reverse causation dominates here. Undiagnosed illness, low-grade inflammation, cancer and neurological disease all reduce grip strength for months to years before diagnosis, and they independently raise mortality. Social and economic position also tracks both grip strength and death rates.

García-Hermoso et al., muscular strength as a predictor of all-cause mortality in an apparently healthy population · Arch Phys Med Rehabil 2018;99(10):2100-2113.e5 Leong et al., prognostic value of grip strength, findings from the PURE study · Lancet 2015;386(9990):266-73

At matched weekly volume, training a muscle group twice a week produced larger hypertrophy effect sizes than once a week (0.49 versus 0.30). A Bayesian network meta-analysis of 178 strength trials and 119 hypertrophy trials ranked higher-load multiset thrice-weekly training best for strength and higher-load multiset twice-weekly training best for size.

Measured in: 10 studies directly comparing weekly frequencies; the network meta-analysis pooled 5,097 participants for strength and 3,364 for hypertrophy

The reviews report their samples as 45% women for strength and 47% for hypertrophy, which makes this one of the better sex-represented findings on the page rather than one of the gaps.

Schoenfeld, Ogborn and Krieger, effects of resistance training frequency on measures of muscle hypertrophy · Sports Med 2016;46(11):1689-1697 Currier et al., resistance training prescription for muscle strength and hypertrophy in healthy adults, a Bayesian network meta-analysis · Br J Sports Med 2023;57(18):1211-1220

Each additional weekly set per muscle group was associated with an effect size increase of 0.023, equal to about 0.37% more growth. Comparing the higher and lower volume arms within each study gave an effect size difference of 0.241, about 3.9% more growth.

Measured in: 34 treatment groups from 15 studies

The slope is small, and it was fitted over the range the studies happened to use, so it says nothing about where the curve turns down. The three-band analysis (under 5, 5 to 9, and 10 or more weekly sets) reached only a trend (p = 0.074), which is weaker than the way this finding is usually quoted.

Schoenfeld, Ogborn and Krieger, dose-response relationship between weekly resistance training volume and increases in muscle mass · J Sports Sci 2017;35(11):1073-1082

When every set went to momentary failure, changes in muscle size were similar between loads at or below 60% of a one-rep maximum and loads above it. One-rep-max strength gains were significantly greater with heavy loads; isometric strength showed no significant difference between conditions.

Measured in: 21 studies comparing low-load and high-load training with sets taken to failure

The equivalence holds only because effort was matched to failure. A light set stopped well short of failure is not equivalent to a heavy one, and light-load training to failure is considerably more unpleasant, which is a compliance problem rather than a physiological one.

Schoenfeld, Grgic, Ogborn and Krieger, strength and hypertrophy adaptations between low- and high-load resistance training · J Strength Cond Res 2017;31(12):3508-3523

Training to momentary muscular failure showed no advantage over non-failure training for muscle growth (effect size 0.12, p = 0.343). Training to the looser definition of set failure gave a trivial edge (0.19, p = 0.045). Velocity-loss thresholds showed no significant advantage (0.08, p = 0.529). The authors describe the relationship between proximity to failure and growth as potentially non-linear.

Measured in: 15 studies in healthy adults, predominantly resistance-trained

The pooled trials are close to evenly split between untrained and trained lifters, with 7 studies and 252 people in the untrained arm against 8 studies of trained lifters, so this is not a trained-lifter finding. That split matters here, because judging reps in reserve is a skill and untrained lifters are systematically bad at it.

Refalo et al., influence of resistance training proximity-to-failure on skeletal muscle hypertrophy · Sports Med 2023;53(3):649-665

In 100 frail nursing home residents of mean age 87, ten weeks of progressive resistance training raised muscle strength 113% against 3% in the non-exercising groups, raised gait velocity 11.8% against a 1.0% decline, and raised stair-climbing power 28.4% against 3.6%. Multinutrient supplementation alone changed no primary outcome. An earlier uncontrolled study in ten nonagenarians of mean age 90 found 174% strength gains and a 9.0% increase in mid-thigh muscle area over eight weeks.

Measured in: 100 frail nursing home residents (63 women, 37 men), mean age 87.1, range 72 to 98, in the randomized trial; 10 institutionalized nonagenarians in the earlier study

Ten weeks is short, the setting was supervised institutional care, and the earlier nonagenarian study had no control group and nine completers. Thigh muscle area in the randomized trial rose only 2.7% and did not reach significance, so most of the strength gain over this window is neural rather than new tissue.

Fiatarone et al., exercise training and nutritional supplementation for physical frailty in very elderly people · N Engl J Med 1994;330(25):1769-75 Fiatarone et al., high-intensity strength training in nonagenarians · JAMA 1990;263(22):3029-34

Eight months of twice-weekly, 30-minute supervised training at five sets of five repetitions above 85% of a one-rep maximum raised lumbar spine bone mineral density 2.9% against a 1.2% fall in the low-intensity control group, raised femoral neck density 0.3% against a 1.9% fall, increased femoral neck cortical thickness, and improved every functional measure. One adverse event was reported across the trial, a minor back spasm.

Measured in: 101 postmenopausal women with osteopenia or osteoporosis, mean age 65, screened to exclude conditions and medications affecting bone

Every session was supervised, participants were otherwise healthy and screened, and compliance was 92%. The safety record belongs to that setting rather than to unsupervised heavy lifting, and the authors are explicit about it. The lead author directs a clinic and licenses the commercial program derived from this trial. There is also an erratum on the paper that we could not read: the journal record carries no text and the publisher refused access, and nothing suggests it touches the bone results, but it is unresolved on the most load-bearing women's trial on this page.

Watson et al., high-intensity resistance and impact training improves bone mineral density and physical function in postmenopausal women, the LIFTMOR randomized controlled trial · J Bone Miner Res 2018;33(2):211-220

Protein supplementation during prolonged resistance training added about 0.7 lb (0.30 kg) of fat-free mass and increased one-rep-max strength beyond training alone. The dose-response plateaued at approximately 1.62 g of protein per kilogram of bodyweight per day, and the size of the fat-free mass benefit fell as participant age rose.

Measured in: 49 studies with 1,863 participants, all combining prolonged resistance training with a protein supplementation arm against a control arm

0.7 lb (0.30 kg) over a training block is a small effect, and it is measured against control diets that were often already adequate. The 1.62 g/kg figure is the point at which the fitted curve stops rising in this dataset, with wide confidence intervals at the top end, so it is a plateau estimate rather than a requirement. The senior author has run trials funded by a dairy industry body that are themselves inside this meta-analysis.

Morton et al., systematic review, meta-analysis and meta-regression of the effect of protein supplementation on resistance training-induced gains in muscle mass and strength · Br J Sports Med 2018;52(6):376-384 Bauer et al., evidence-based recommendations for optimal dietary protein intake in older people, the PROT-AGE position paper · J Am Med Dir Assoc 2013;14(8):542-59

In adults over 50 doing identical programs, women gained more relative lower-body strength (g = -0.21, 95% CI -0.33 to -0.10). Relative upper-body strength showed no significant difference (g = -0.29, p = 0.08) and relative muscle size showed no difference (g = 0.10, p = 0.16). Men gained more in absolute terms for upper-body strength (g = 0.48), lower-body strength (g = 0.33) and muscle size (g = 0.45).

Measured in: 30 studies with 41 comparison groups, 1,410 participants (651 men, 759 women), all older than 50

The direction of this finding depends entirely on whether results are expressed in absolute or relative terms, and the authors say so. Mean study quality was 14.7 out of 29 on a modified Downs and Black checklist. It covers adults over 50 only, so it does not answer the same question for younger trainees.

Jones et al., sex differences in adaptations in muscle strength and size following resistance training in older adults · Sports Med 2021;51(3):503-517

Across 1,382 original research articles in three major sports and exercise medicine journals over three years, covering 6,076,580 participants, 39% of participants were female and 61% male. The average share of female participants per article ran 35% to 37%, and the under-representation was significant in all three journals.

Measured in: 1,382 articles in Medicine and Science in Sports and Exercise, the British Journal of Sports Medicine and the American Journal of Sports Medicine

A three-year window in three journals, and the participant total is dominated by a handful of very large epidemiological datasets, so the per-article figure of 35 to 37% describes the typical study better than the pooled 39% does. The survey ends in the early 2010s and representation has improved since, without closing. Women remain underrepresented in this literature, and combined resistance-training-plus-protein trials in particular have rarely been analyzed by sex at all, which is why `measuredIn` appears on every row of this page.

What could explain it instead: The counting method mixes study types: one large cohort with balanced enrollment can offset dozens of all-male laboratory trials, which is exactly the pattern in exercise physiology. The journals sampled skew toward sports medicine and orthopedics, where the athlete populations studied are themselves sex-imbalanced.

Costello, Bieuzen and Bleakley, where are all the female participants in sports and exercise medicine research? · Eur J Sport Sci 2014;14(8):847-51 Noh and Park, effects of resistance training and protein supplementation interventions on muscle volume and muscle function, sex differences in humans · Phys Act Nutr 2023;27(4):15-25

Across 12 weeks of twice-weekly strength training, the group doing active recovery gained more strength and more muscle mass than the group doing 10 minutes of cold water immersion afterwards. A parallel mechanistic study found suppressed anabolic signaling and fewer NCAM-positive and Pax7-positive satellite cells after cold immersion than after active recovery.

Measured in: 21 physically active men in the 12-week training study; 9 active men in the single-leg mechanistic study

Twenty-one men over 12 weeks is a small trial for a training outcome, and it compares cold immersion against active recovery rather than against doing nothing. The finding applies to cold taken immediately after resistance training; it says nothing about cold on a separate day or several hours later.

Roberts et al., post-exercise cold water immersion attenuates acute anabolic signalling and long-term adaptations in muscle to strength training · J Physiol 2015;593(18):4285-301

In a population-based medical records cohort, the age- and sex-adjusted incidence of exertional rhabdomyolysis was 1.06 per 100,000 person-years (95% CI 0.59 to 1.52). Of 431 rhabdomyolysis cases, 4.9% were exertional. The common triggers were endurance activity (7 cases), manual labor (5) and weight lifting (4). Five patients had kidney injury, 16 were hospitalized for a median of 2 days, there were no deaths and one recurrence.

Measured in: Olmsted County, Minnesota, 2003 to 2015, via the Rochester Epidemiology Project record linkage system; 20 exertional cases, 18 of them male

Twenty cases in one American county over 13 years, ascertained from medical records, so mild cases that never reached care are invisible and the true incidence of muscle damage after an unaccustomed session is far higher than this. The civilian rate is much lower than the military one, where forced unfamiliar exertion is routine.

What could explain it instead: Ascertainment. Only people who sought medical care and had creatine kinase measured appear in the count, which selects for severity, for insurance coverage and for proximity to a major medical center. The activity attribution comes from clinical notes rather than from any structured exposure record.

Luetmer et al., exertional rhabdomyolysis, a retrospective population-based study · Med Sci Sports Exerc 2020;52(3):608-615

Dynamic resistance training lowered resting blood pressure by about 1.8 mmHg systolic and 3.2 mmHg diastolic overall, with the largest reductions in prehypertensive people, about 4.0/3.8 mmHg. The effect was explicitly smaller in people who were already hypertensive.

Measured in: 93 trials totalling 5,223 participants, of which 29 groups performed dynamic resistance training and 5 isometric resistance training

The direction of that gradient matters: this is better at holding a borderline reading down than at bringing an established one back, so it belongs alongside treatment rather than instead of it.

Cornelissen and Smart, exercise training for blood pressure, a systematic review and meta-analysis · J Am Heart Assoc 2013;2(1):e004473

Resistance training reduced anxiety symptoms with a small-to-moderate standardized effect of 0.31 (95% CI 0.17 to 0.44, p<0.001), on the standardized effect-size scale where 0.2 counts as small and 0.5 as moderate. The improvement was larger in healthy participants (effect 0.50) than in those with a physical or mental illness (effect 0.19). Pooled from 31 effects across 16 randomized controlled trials and 922 participants.

Measured in: 922 participants across 16 randomized controlled trials (486 assigned to resistance training, 436 to non-active control); mean age 43 years, about 68% female.

Fewer trials and participants inform this estimate than the depression evidence, and the overall effect is smaller. Anxiety was measured with self-report scales, and participants cannot be blinded to an exercise assignment.

Gordon BR, et al. Resistance exercise training and anxiety: meta-analysis and meta-regression of RCTs · Sports Med 2017;47(12):2521-2532

Across 16 prospective cohorts, muscle-strengthening activity carried 10 to 17% lower risk of all-cause mortality, cardiovascular disease, total cancer, diabetes and lung cancer. For all-cause mortality, cardiovascular disease and cancer the dose-response was J-shaped, with maximum risk reduction of roughly 10 to 20% at about 30 to 60 minutes a week and the benefit flattening or bending back above that. Diabetes was the exception, falling steadily up to 60 minutes a week.

Measured in: 16 prospective cohort studies of adults aged 18 and over without severe health conditions

The J-shape rests on the subset of cohorts that reported enough dose detail, all using self-reported weekly minutes. The authors themselves conclude that the influence of higher volumes is unclear rather than harmful. Treat this as an absence of evidence for a large marginal gain above an hour a week, not as evidence that more training shortens life.

What could explain it instead: Reverse causation at the low-volume end, where illness reduces training long before diagnosis. At the high-volume end, self-reported minutes are least reliable in the people reporting the most, and very high self-reported strength training clusters with occupational physical strain and with people managing existing conditions, both of which carry their own mortality.

Momma et al., muscle-strengthening activities are associated with lower risk and mortality in major non-communicable diseases · Br J Sports Med 2022;56(13):755-763

After resistance exercise with 25 g of whey protein, myofibrillar protein synthesis rose 2.3-fold in men and 2.7-fold in women over the 1 to 5 hour window, with no meaningful difference at 24 to 28 hours either. Men showed a 45-fold greater testosterone response and greater acute mTOR phosphorylation, with no corresponding difference in synthesis rate.

Measured in: 8 young men (mean age 20) and 8 young women (mean age 22), stable isotope infusion with serial vastus lateralis biopsies over two consecutive days

Sixteen participants, single-session, acute synthesis rates rather than measured muscle growth. Acute protein synthesis is a signal, not an outcome, and short-term synthesis rates have repeatedly failed to predict long-term hypertrophy. The women's menstrual cycle phase is not a variable this design controlled for across the group.

West et al., sex-based comparisons of myofibrillar protein synthesis after resistance exercise in the fed state · J Appl Physiol 2012;112(11):1805-13

Basal muscle protein synthesis was about 30% higher in older women than in younger women (0.060 versus 0.046 %/h), while in men the basal rate did not differ with age. Men showed roughly a 60% smaller synthesis response to hormonal and nutrient stimulation when older; older women showed no significant increase to the clamp, from an already higher baseline.

Measured in: 8 young men, 10 young women (25 to 45), 10 older men and 10 older women (65 to 85)

The trial excluded users of hormone therapy and hormonal contraceptives by protocol, so the sex difference here is not an artifact of who was taking what. It is a small mechanistic study measuring synthesis rates rather than muscle gained, and a synthesis rate is not an outcome by itself.

Smith et al., sexually dimorphic effect of aging on skeletal muscle protein synthesis · Biol Sex Differ 2012;3(1):11

Over 8 weeks of twice-weekly calf raises with one leg doing a 20-second loaded inter-set stretch, the soleus showed a possible benefit of 0.7 mm of thickness (90% CI 0 to 1.6 mm), the lateral gastrocnemius was ambiguous (0.4 mm, -0.4 to 1.3) and the medial gastrocnemius equivocal (0 mm, -0.6 to 0.7). In a separate 5-week within-participant trial adding inter-set static stretching to flywheel squats, quadriceps thickness rose 7.3% without stretching and 8.0% with it, a non-significant difference, alongside clear flexibility gains and greater isometric knee extension strength.

Measured in: 21 untrained young men (calf study, contralateral limb design) and 16 untrained men aged 21 (flywheel squat study, contralateral limb design)

Two small contralateral-limb studies, 8 and 5 weeks, both in untrained men, with confidence intervals that include zero on the primary measures. A published bench press study found the same pectoral thickness with and without a loaded inter-set stretch, and the largest reported effect in this area was a conference abstract that was never published in a peer-reviewed journal.

Van Every et al., loaded inter-set stretch may selectively enhance muscular adaptations of the plantar flexors · PLoS One 2022;17(9):e0273451 Nakamura et al., effects of adding inter-set static stretching to flywheel resistance training on flexibility, muscular strength and regional hypertrophy in young men · Int J Environ Res Public Health 2021;18(7):3770

Static stretching between sets reduced performance in subsequent sets. The size could not be confirmed: the figures appear only in a chart, and the formula printed in the paper makes their stated direction self-contradictory.

Measured in: 12 resistance-trained men, mean age 29, isokinetic knee extension in three counterbalanced conditions

We give the direction and not the magnitude, because the magnitude is not readable from the published record. Measured in resistance-trained men only (mean age 29).

Padilha et al., could inter-set stretching increase acute neuromuscular and metabolic responses during resistance exercise? · Eur J Transl Myol 2019;29(4):8579 Chaabene et al., acute effects of static stretching on muscle strength and power, an attempt to clarify previous caveats · Front Physiol 2019;10:1468

Direct intra-arterial recording during heavy lifting to failure gave group mean peak pressures of 320/250 mmHg on the double-leg press, with one participant exceeding 480/350 mmHg. Pressures rose during the concentric phase of each lift and fell during the lowering phase.

Measured in: 5 experienced bodybuilders with a brachial artery catheter, lifting at 80, 90, 95 and 100% of maximum

Five people, all experienced heavy lifters using a deliberate Valsalva at maximal loads, which is the extreme end of the exposure. The pressures last seconds. What this establishes is the mechanism behind the dizziness and graying vision that beginners experience, not a measured rate of harm.

MacDougall et al., arterial blood pressure response to heavy resistance exercise · J Appl Physiol 1985;58(3):785-90

During the fourth repetition of a bench press, intraocular pressure rose an average of 4.3 mmHg when the breath was held and 2.2 mmHg when the participant exhaled normally. Pressure rose in 90% of participants with the breath held and 62% without, and exceeded a 5 mmHg rise in 30% and 21% respectively.

Measured in: 30 men aged 18 to 40, mean age 26, without glaucoma; right eye measured during the breath-held repetition, left eye during the normally-breathing repetition

Thirty men without glaucoma, measured during single repetitions, all starting from normal baseline pressure. The authors raise glaucoma risk as a hypothesis rather than a finding, and no study has followed lifters to see whether repeated transient rises cause optic nerve damage.

Vieira et al., intraocular pressure variation during weight lifting · Arch Ophthalmol 2006;124(9):1251-4

Vastus lateralis cross-sectional area rose about 2.7% by week 3 and about 10.4% by week 10 of training. Ultrasound echo intensity, normalized to the change in area, was significantly raised only at week 3, alongside elevated myoglobin and interleukin-6, indicating that a large share of the week-3 increase was swelling from muscle damage rather than new tissue. Maximal voluntary contraction rose only by week 10.

Measured in: 10 untrained young men across a 10-week resistance training program

Ten men, one muscle, ultrasound rather than biopsy or MRI. It establishes that early cross-sectional area measurements overstate growth in untrained people, without quantifying by how much in any individual.

Damas et al., early resistance training-induced increases in muscle cross-sectional area are concomitant with edema-induced muscle swelling · Eur J Appl Physiol 2016;116(1):49-56

Exercise reduced the rate of falls by 23% (rate ratio 0.77, 95% CI 0.71 to 0.83; 59 studies, 12,981 participants) and the number of people who fell at least once by 15% (risk ratio 0.85; 63 studies, 13,518 participants). Both high-certainty. Fall-related fractures fell about 27%, at low certainty.

Measured in: 108 randomized trials across 25 countries, 23,407 community-dwelling participants, average age 76, 77% women

This review covers community-dwelling older adults and excludes residential care by design, so it says nothing either way about care homes. Adherence in the trials was supported and supervised in ways an unsupervised reader at home is not.

Sherrington et al., exercise for preventing falls in older people living in the community · Cochrane Database Syst Rev 2019;1(1):CD012424

Balance and functional exercise reduced the rate of falls by 24% (rate ratio 0.76, 95% CI 0.70 to 0.81; 39 studies, 7,920 participants), on high-certainty evidence. It is the only exercise type here with high-certainty evidence behind it.

Measured in: 39 trials of balance and functional training in community-dwelling older adults, plus 88 trials and 19,478 participants in the 2017 dose analysis

The dose finding is a subgroup comparison across trials rather than a randomized comparison of doses, so the people who did three hours a week were also the people well enough to do three hours a week.

Sherrington et al., exercise for preventing falls in older people living in the community · Cochrane Database Syst Rev 2019;1(1):CD012424 Sherrington et al., exercise to prevent falls in older adults, updated systematic review and meta-analysis · Br J Sports Med 2017;51(24):1750-1758

The rate of a first adjudicated serious fall injury did not differ: 4.9 per 100 person-years in the intervention group against 5.3 in the control group (hazard ratio 0.92, 95% CI 0.80 to 1.06, P = 0.25). Participant-reported fall injuries were slightly lower (25.6 against 28.6 per 100 person-years, hazard ratio 0.90, 95% CI 0.83 to 0.99).

Measured in: 5,451 community-dwelling adults aged 70 and over at increased risk of fall injury, in 86 primary care practices across 10 US health systems

Randomized by practice rather than by person, which reduces the effective sample below its headline size. The endpoint was time to first ADJUDICATED SERIOUS fall injury, which is a much higher bar than "did you fall", so this is not evidence that nothing changed at all. It is evidence that the serious-injury endpoint did not move.

Bhasin et al., a randomized trial of a multifactorial strategy to prevent serious fall injuries (STRIDE) · N Engl J Med 2020;383(2):129-140

Home fall-hazard interventions reduced the overall rate of falls by 26% (rate ratio 0.74, 95% CI 0.61 to 0.91; 12 studies, 5,293 participants) at moderate certainty. In people already at high risk of falling the reduction was 38% (rate ratio 0.62, 95% CI 0.56 to 0.70; 9 studies) at high certainty. In unselected community groups there was no reduction at all (rate ratio 1.05, 95% CI 0.96 to 1.16).

Measured in: 22 randomized trials in 10 countries, 8,463 community-dwelling older adults, average age 78, 65% women

Almost all of the benefit is concentrated in people who have already fallen or are otherwise at high risk, and the effective versions were delivered by an occupational therapist rather than by a checklist. Doing this to a low-risk household changed nothing.

Clemson et al., environmental interventions for preventing falls in older people living in the community · Cochrane Database Syst Rev 2023;3(3):CD013258

Pooled odds of falling: antidepressants 1.57 (95% CI 1.43 to 1.74), antipsychotics 1.54 (1.28 to 1.85), benzodiazepines 1.42 (1.22 to 1.65), tricyclic antidepressants 1.41 (1.07 to 1.86), SSRIs 2.02 (1.85 to 2.20), long-acting benzodiazepines 1.81 (1.05 to 3.16), short-acting benzodiazepines 1.27 (1.04 to 1.56).

Measured in: Adults aged 60 and over, or study populations with a mean age of 70 and over, across a large body of observational studies

An association, not a trial. The trials that actually removed the drugs found no fall reduction, so this cannot be read as "stopping them will help". Our note about SSRIs specifically is an inference from the drug classes involved rather than something the paper separates out.

What could explain it instead: Confounding by indication runs through all of it: depression, anxiety, insomnia and psychosis are themselves fall risk factors, and the people prescribed these drugs are frailer and more cognitively impaired than those who are not. No observational design separates the drug from the reason for the drug.

Seppala et al., fall-risk-increasing drugs, systematic review and meta-analysis II: psychotropics · J Am Med Dir Assoc 2018;19(4):371.e11-371.e17

The repositioning maneuver resolved vertigo in about 56% of people against 21% with a sham or no treatment. It treats posterior canal benign paroxysmal positional vertigo only.

Measured in: Eleven mostly small randomized trials in adults with posterior canal benign paroxysmal positional vertigo

This does not treat dizziness in general. It does not treat the anterior or horizontal canal forms, and it does nothing for light-headedness on standing, inner-ear disease, or dizziness from medication or heart rhythm. If the room does not spin, and spin specifically when you move your head, this is not the problem it solves.

Hilton and Pinder, the Epley (canalith repositioning) manoeuvre for benign paroxysmal positional vertigo · Cochrane Database Syst Rev 2014;(12):CD003162

In the first three months after a hip fracture, mortality ran 5.75 times higher in women and 7.95 times higher in men than in age- and sex-matched people without a fracture. The excess narrows afterwards but does not disappear.

Measured in: Prospective cohort studies of older adults with hip fracture, published 1957 to May 2009

These are excess hazards relative to matched controls, not crude death rates, and the age-80 figures are reported for white women and men specifically. A hip fracture also selects for people who were already frailer, so this is not the fracture acting alone.

Haentjens et al., meta-analysis: excess mortality after hip fracture among older women and men · Ann Intern Med 2010;152(6):380-390

Programs combining several exercise types reduced the rate of falls by 34% (rate ratio 0.66, 95% CI 0.50 to 0.88; 11 studies, 1,374 participants), on moderate-certainty evidence.

Measured in: Community-dwelling older adults within the same Cochrane review of 108 trials

This is the largest point estimate in the review and it rests on the fewest trials, so the confidence interval is wide and the certainty is moderate rather than high. It should not be read as proof that adding weights beats balance work alone.

Sherrington et al., exercise for preventing falls in older people living in the community · Cochrane Database Syst Rev 2019;1(1):CD012424

Tai Chi reduced the rate of falls by 19% (rate ratio 0.81, 95% CI 0.67 to 0.99; 7 studies, 2,655 participants) on low-certainty evidence. A later review of 24 randomized trials found a 24% lower risk of being a faller (risk ratio 0.76, 95% CI 0.71 to 0.82), with more effect at three or more sessions a week (risk ratio 0.67, 95% CI 0.58 to 0.79) than at two (0.78, 0.73 to 0.84).

Measured in: 7 trials, 2,655 community-dwelling older adults in the Cochrane analysis; 24 trials in the 2023 review, predominantly older women

The Cochrane certainty grade for Tai Chi is low, and the pooled point estimate sits slightly behind general balance and functional training rather than ahead of it. Trials vary widely in style, form length and teacher, and the comparator is often stretching or usual care rather than another balance program.

Sherrington et al., exercise for preventing falls in older people living in the community · Cochrane Database Syst Rev 2019;1(1):CD012424 Chen et al., Tai Chi for fall prevention and balance improvement in older adults · Front Public Health 2023;11:1236050

Over 24 weeks, a therapeutic Tai Ji Quan program cut falls 58% against stretching (incidence rate ratio 0.42, 95% CI 0.31 to 0.56) and 31% against a multimodal exercise program of balance, aerobics, strength and flexibility (IRR 0.69, 95% CI 0.52 to 0.94). Multimodal exercise itself cut falls 40% against stretching (IRR 0.60, 95% CI 0.45 to 0.80).

Measured in: 670 community-dwelling adults aged 70 and over in seven Oregon cities who had fallen in the past year or had impaired mobility. Mean age 77.7, 436 women (65%), 92% White

The trial that produced this is not inside the Cochrane pool, which searched only to May 2018 and lists it as awaiting classification, so those two really are separate. The other review cited alongside it does contain this trial and re-pools the Cochrane studies, so it is not a third independent confirmation. The trial author holds the license to the intervention being tested.

Li et al., therapeutic Tai Ji Quan vs multimodal exercise to prevent falls among older adults at high risk of falling · JAMA Intern Med 2018;178(10):1301-1310

Deprescribing as a standalone intervention did not change the rate of falls (rate ratio 0.98, 95% CI 0.63 to 1.51), the number of people who fell (risk ratio 1.04, 95% CI 0.86 to 1.26) or fall-related injuries (rate ratio 0.89, 95% CI 0.57 to 1.39) over 6 to 12 months.

Measured in: Five randomized trials, 1,305 participants

Two independent reviews reach the same null, which is why this sits at moderate rather than preliminary. Withdrawal is hard to achieve and hard to sustain in these trials, so this measures the whole strategy of trying to stop the drugs rather than the effect of successfully being off them.

Lee et al., deprescribing fall-risk increasing drugs (FRIDs) for the prevention of falls and fall-related complications · BMJ Open 2021;11(2):e035978

Orthostatic hypotension was associated with falls at an odds ratio of 1.73 (95% CI 1.50 to 1.99). The association held regardless of study population, design, quality, which definition of orthostatic hypotension was used, or how blood pressure was measured.

Measured in: 63 studies and 51,800 individuals reviewed, 50 studies and 49,164 individuals pooled; 39 of the 63 were cross-sectional and 24 longitudinal

The pooled odds ratio is unadjusted, so some of it is the illness and medication that cause orthostatic hypotension rather than the blood-pressure drop itself.

What could explain it instead: Orthostatic hypotension travels with the things that cause falls independently: dehydration, Parkinson's disease, diabetic autonomic neuropathy, frailty, and the same psychotropic and antihypertensive drugs listed above. It may be a marker of that whole cluster rather than the mechanism.

Mol et al., orthostatic hypotension and falls in older adults, systematic review and meta-analysis · J Am Med Dir Assoc 2019;20(5):589-597

Higher doses did worse. In a daily-dosing trial (STURDY) in older adults at elevated fall risk, 2,000 IU/day carried more falls than 1,000 IU/day (HR 1.54, 95% CI 1.01 to 2.34), and 4,000 IU/day trended the same way without reaching significance (HR 1.41, 95% CI 0.92 to 2.16). A separate trial giving 500,000 IU once a year for three to five years raised falls (IRR 1.15) and fractures (IRR 1.26). No regimen in any of these trials reduced falls.

Measured in: Community-dwelling older adults across four trials and reviews, including a daily-dose trial that was 44% women and 56% men, an annual-megadose trial in women only, and a monthly-dose trial 67% women

The harm is not confined to exotic megadoses. The clearest dose-response evidence here comes from a trial giving vitamin D DAILY, where 2,000 and 4,000 IU a day did worse than 1,000. The annual 500,000 IU result is the most dramatic but it is also the least like anything a reader takes. The three-month timing detail inside that trial was a post hoc finding. None of this speaks to treating a diagnosed deficiency, which is a different question with a different answer.

Sanders et al., annual high-dose oral vitamin D and falls and fractures in older women · JAMA 2010;303(18):1815-1822 Bischoff-Ferrari et al., monthly high-dose vitamin D treatment for the prevention of functional decline · JAMA Intern Med 2016;176(2):175-183 Appel et al., the effects of four doses of vitamin D supplements on falls in older adults (STURDY) · Ann Intern Med 2021;174(2):145-156 US Preventive Services Task Force, interventions to prevent falls in community-dwelling older adults · JAMA 2018;319(16):1696-1704

Bringing first-eye cataract surgery forward to about four weeks rather than a twelve-month wait reduced the rate of falling by 34%, with fewer fractures and better visual function and general health status.

Measured in: 306 women over 70 with cataract, randomized to expedited or routine surgery

The fall RATE fell while the proportion of people who fell at least once did not (49% of operated participants against 45% of controls), so this shifts repeat fallers rather than preventing first falls. The second-eye trial found a reduction of the same size that did not reach significance, so calling the second eye’s benefit smaller overstates what was shown.

Harwood et al., falls and health status in elderly women following first eye cataract surgery · Br J Ophthalmol 2005;89(1):53-59

Switching habitual multifocal wearers to single-lens glasses for outdoor use cut falls in the most active participants. Among the least active it may have done harm: outside falls rose (IRR 1.56, 95% CI 1.11 to 2.19), while all falls (1.29, 0.95 to 1.75) and injurious falls (1.34, 0.96 to 1.88) both crossed one.

Measured in: 606 regular multifocal wearers, mean age 80, all at increased fall risk and using multifocals outdoors at least three times a week

Only the outside-falls endpoint reached significance in the least active group, and the trialists themselves put it as "may be harmful". The direction is consistent across all three endpoints, which makes it a signal rather than a settled finding.

Haran et al., effect on falls of providing single lens distance vision glasses to multifocal glasses wearers (VISIBLE) · BMJ 2010;340:c2265

Concern about falling at baseline independently predicted falls at follow-up. A single-item measure of high versus low concern carried an odds ratio of 1.60 (95% CI 1.36 to 1.89). The 16-item Falls Efficacy Scale International carried 1.03 per point (1.02 to 1.05) and the 7-item short form 1.08 per point (1.05 to 1.11). Balance confidence measured by the ABC scale did not predict falls (0.97, 0.93 to 1.01).

Measured in: 53 prospective studies, 75,076 participants, minimum six months of follow-up. Female representation ranged from 30% to 100%, median 60.3%

Twenty-six of the 53 studies were rated poor quality, though the associations held across quality strata. The measure that performs best is the crude single question, and the more elaborate scales perform worst, which is the opposite of what you would expect if the construct were cleanly defined.

What could explain it instead: Concern about falling rises after a fall and rises with the impairments that cause falls, so some of this is the body reporting a real deficit rather than a fear causing one. The analyzes adjust for previous falls, which helps and does not settle it.

Ellmers et al., does concern about falling predict future falls in older adults? A systematic review and meta-analysis · Age Ageing 2025;54(4):afaf089

Nocturia was associated with about a 1.2-fold increase in falls (risk ratio 1.20, 95% CI 1.05 to 1.37, moderate quality) and a possible 1.3-fold increase in fractures (risk ratio 1.32, 95% CI 0.99 to 1.76, low quality).

Measured in: Nine longitudinal observational studies of adults with and without nocturia

The reviewers graded the evidence moderate for nocturia as a PREDICTOR of falls and very low for it as a CAUSE, which is the published answer to the obvious question. Gender was tested as an effect modifier and did not modify the association.

What could explain it instead: Nocturia is a symptom of heart failure, diabetes, prostate disease, sleep apnea and diuretic use, all of which raise fall risk on their own. It also tracks poor sleep, and sleeping badly impairs balance the next day.

Pesonen et al., the impact of nocturia on falls and fractures, systematic review and meta-analysis · J Urol 2020;203(4):674-683

20.4% of participants could not hold a 10-second one-legged stance. Over a median seven years, 17.5% of them died against 4.6% of those who could. Adjusted for age, sex, body mass index and comorbidities, the hazard ratio for all-cause mortality was 1.84 (95% CI 1.23 to 2.78).

Measured in: 1,702 individuals aged 51 to 75, 68% men, assessed between 2008 and 2020 at a Brazilian exercise medicine clinic

The outcome is death, not falls, and the authors say plainly that they could not control for recent falls or physical activity. A test failed today is a summary of everything that has already happened to the legs, the inner ear and the nervous system, so it forecasts well without implying that training the test changes the forecast.

What could explain it instead: People who cannot balance for ten seconds are more likely to have undiagnosed neurological disease, sarcopenia, obesity and inactivity, and the cohort comes from a self-selected clinic population rather than a community sample. Reverse causation is live: early illness degrades balance before it is diagnosed.

Araujo et al., successful 10-second one-legged stance performance predicts survival in middle-aged and older individuals · Br J Sports Med 2022;56(17):975-980

Eight months of twice-weekly, 30-minute supervised high-intensity resistance and impact training beat home-based low-intensity exercise for lumbar spine bone mineral density (+2.9% against -1.2%), femoral neck density (+0.3% against -1.9%), cortical thickness and every functional performance measure. Compliance was 92%, with one adverse event, a minor lower back spasm.

Measured in: 101 postmenopausal women with osteopenia or osteoporosis, screened to exclude conditions and drugs affecting bone

Every session was supervised by staff trained to coach heavy lifting, and the authors are explicit that the safety finding applies under those conditions. Bone density is a surrogate: this trial was not sized to measure fractures.

Watson et al., high-intensity resistance and impact training improves bone mineral density and physical function in postmenopausal women (LIFTMOR) · J Bone Miner Res 2018;33(2):211-220

Multifactorial programs, an individual risk assessment followed by a tailored set of actions, reduced the rate of falls by 23% (rate ratio 0.77, 95% CI 0.67 to 0.87; 19 studies, 5,853 participants) on low-certainty evidence, while the number of people who fell at least once was unchanged (risk ratio 0.96, 95% CI 0.90 to 1.03; 29 studies, 9,637 participants). Multiple-component programs, a fixed set of the same components given to everyone, reduced the rate of falls by 26% (rate ratio 0.74, 95% CI 0.60 to 0.91; 6 studies, 1,085 participants) and the number who fell by 18% (risk ratio 0.82, 95% CI 0.74 to 0.90; 11 studies, 1,980 participants), both on moderate-certainty evidence.

Measured in: 62 randomized trials of community-dwelling older adults in the Cochrane review, spanning multifactorial and multiple-component designs

The two designs differ in a way that matters. Multifactorial means each person is assessed and then given their own mix of actions, and its fall-rate benefit is low-certainty while the number of people who fell did not move. Multiple-component means the same bundle of a few proven parts is delivered to everyone, and it carries moderate-certainty benefit on both measures. This sits alongside the STRIDE trial elsewhere on this page, a large newer multifactorial trial that assessed people and referred them onward and found no drop in serious fall injuries, which points to how much depends on the actions actually being carried out rather than recommended.

Hopewell et al., multifactorial and multiple component interventions for preventing falls in older people living in the community · Cochrane Database Syst Rev 2018;7(7):CD012221

Training that provokes fast stepping, either voluntary (stepping to a cue) or reactive (recovering from a shove or a treadmill slip), reduced the rate of falls by 52% (rate ratio 0.48, 95% CI 0.36 to 0.65) and the proportion of fallers by 49% (risk ratio 0.51, 95% CI 0.38 to 0.68), with no heterogeneity between trials.

Measured in: Seven randomized trials, 660 older adults

660 people across seven small trials is a thin base for an effect this large, and the reactive versions need a harness, a treadmill or a person to push you. What transfers to a kitchen counter is the voluntary stepping half.

Okubo, Schoene, Lord, step training improves reaction time, gait and balance and reduces falls in older people · Br J Sports Med 2017;51(7):586-593

Vision improvement interventions made little or no difference to the rate of falls (rate ratio 1.12, 95% CI 0.84 to 1.50) or to the number of people who fell (risk ratio 1.09, 95% CI 0.79 to 1.50), on low-certainty evidence.

Measured in: Three trials, 1,489 community-dwelling older adults within the Cochrane environmental review

Three trials and a wide interval, so this rules out very little. It pools different things under one label, including new spectacle prescriptions, which have their own adaptation risk, and it is not a verdict on treating a specific eye disease.

Clemson et al., environmental interventions for preventing falls in older people living in the community · Cochrane Database Syst Rev 2023;3(3):CD013258

A twelve-month podiatry package cut the rate of falls by 36% (incidence rate ratio 0.64, 95% CI 0.45 to 0.91, P=0.01): 0.67 falls per person-year against 1.06 in the control group. The proportion who fell at least once did not differ (42% against 49%, P=0.19). At six months the package also improved ankle eversion strength, ankle range of motion and standing balance.

Measured in: 305 community-dwelling older adults with disabling foot pain and an increased risk of falling, mean age 74, 69% women, in Melbourne, Australia

This is one trial in people who already had disabling foot pain, so it does not speak to feet that are comfortable. The package combined several parts, prefabricated orthoses, footwear advice with a purchase subsidy, a home foot-and-ankle exercise program three times a week, and a falls-education booklet, so it cannot separate which part carried the effect. As with several other levers here, the rate of falls dropped while the share of people who fell at least once did not, so it reduces repeat falls more than it prevents a first one.

Spink et al., effectiveness of a multifaceted podiatry intervention to prevent falls in community dwelling older people with disabling foot pain · BMJ 2011;342:d3411

Pooled separately, none of these showed a fall reduction: resistance training rate ratio 1.14 (95% CI 0.67 to 1.97; 5 studies, 327 participants), dance 1.34 (0.98 to 1.83; 1 study), walking 1.14 (0.66 to 1.97; 2 studies). All are very-low-certainty evidence.

Measured in: The smaller exercise-type subgroups within the same Cochrane review

This is an absence of evidence, not evidence of absence, and the difference matters. The review ran no meta-analysis comparing exercise types against each other and states that these trials were underpowered to detect differences between them. Five small trials in 327 people with I-squared of 67% cannot establish that strength work does not help. The defensible reading is that balance-challenging exercise is the only type with high-certainty evidence behind it for falls, and that resistance training alone has barely been tested for this particular outcome.

Sherrington et al., exercise for preventing falls in older people living in the community · Cochrane Database Syst Rev 2019;1(1):CD012424

The evidence conflicts. One US cohort (Tinetti 2014) found about 40% higher injurious-fall risk on antihypertensives at moderate intensity, while other analyzes have found neutral or even lower risk for several drug classes, so the evidence does not point one way.

Measured in: 4,961 community-living adults over 70 with hypertension in the Medicare Current Beneficiary Survey

Two bodies of evidence point opposite ways, and neither is strong enough to settle the question. What both sides agree on is narrower and more actionable: a drop in blood pressure on standing raises fall risk, and that is measurable in a chair and a stand. Nobody should stop a blood pressure medicine on the strength of a web page.

What could explain it instead: People taking antihypertensives are sicker than those taking none: more cardiovascular disease, more diabetes, more diabetic neuropathy, more coexisting drugs. Propensity adjustment narrows that gap and cannot close it. Randomized trials of intensive blood pressure lowering have not found the same increase in injurious falls.

Tinetti et al., antihypertensive medications and serious fall injuries in a nationally representative sample of older adults · JAMA Intern Med 2014;174(4):588-595

At a threshold of 13.5 seconds or more, the test had specificity 0.74 and sensitivity 0.31 across the 10 studies that could be pooled, out of 25 reviewed. The review’s own regression found the score was not a significant predictor of falls (OR 1.01, 95% CI 1.00 to 1.02, p=0.05).

Measured in: 25 studies of community-dwelling older adults

Sensitivity of 0.31 means the test misses most people who go on to fall, so a fast time is close to meaningless as reassurance. A slow time is informative; a fast one is not. The pooled analysis rests on 10 studies, and the review could not show the score predicts falls at all.

Barry et al., is the timed up and go test a useful predictor of risk of falls in community-dwelling older adults? · BMC Geriatr 2014;14:14

Tai chi reduced the number of people experiencing a fall by 20% (risk ratio 0.80, 95% CI 0.70 to 0.91; 8 trials, 2,677 participants, high-certainty evidence) and the rate of falls by 19% (rate ratio 0.81, 95% CI 0.67 to 0.99; 7 trials, 2,655 participants, low-certainty evidence).

Measured in: Community-dwelling adults aged 60 and over, within a Cochrane review of 108 exercise trials and 23,407 participants across 25 countries

High-certainty evidence, and the number to trust on this page. A separate review reports a much larger reduction; that one pools a different and more heavily Chinese-language trial set, and the language split documented under blood pressure below is the reason to prefer the more conservative figure. This review’s Tai Chi subgroup also contains the 2005 trial cited under fear of falling, so those two are not independent of each other.

Sherrington et al., exercise for preventing falls in older people living in the community (Cochrane review) · Cochrane Database Syst Rev 2019;1:CD012424

Two 60-minute classes a week of the 8-form therapeutic routine (Tai Ji Quan: Moving for Better Balance) for 24 weeks cut falls 58% against stretching (incidence rate ratio 0.42, 95% CI 0.31 to 0.56) and 31% against a multimodal exercise program combining balance, aerobic, strength and flexibility work (IRR 0.69, 95% CI 0.52 to 0.94). Raw counts were 152 falls in 85 people, 218 in 112, and 363 in 127.

Measured in: 670 adults aged 70 and over with a fall in the previous year or impaired mobility, mean age 77.7, 65% women, 92% white, in Oregon

Participants and instructors knew their assignment and falls were self-recorded on monthly calendars. The intervention is a therapeutic adaptation built specifically to load balance and gait, so the result belongs to that routine rather than to a general tai chi class.

Li et al., therapeutic Tai Ji Quan vs multimodal exercise to prevent falls in older adults at high risk · JAMA Intern Med 2018;178(10)

Pooled across 13 randomized trials, fear of falling improved by a standardized mean difference of -0.54 (95% CI -1.09 to 0.01), dynamic balance by -2.23 (-3.89 to -0.75), and fall incidence sat at a risk ratio of 0.48 (0.35 to 0.65). Static balance did not improve significantly. In a separate 256-person trial of three sessions a week for six months, fear of falling fell significantly (p < 0.001) and the gain held at six-month follow-up.

Measured in: Older adults across 13 trials, plus 256 inactive adults aged 70 to 92 (mean 77.5) in Portland, Oregon

The trial behind this is also inside the Cochrane Tai Chi subgroup cited above, so it is one trial appearing in two rows rather than two findings agreeing.

Meta-analysis of tai chi on fear of falling and balance in older adults · Geriatr Nurs 2023 Li et al., Tai Chi and fall reductions in older adults: a randomized controlled trial · J Gerontol A Biol Sci Med Sci 2005;60(2):187-94

Against supervised aerobic exercise, the standard exercise recommendation for fibromyalgia, the combined tai chi groups improved the revised Fibromyalgia Impact Questionnaire 5.5 points more at 24 weeks (95% CI 0.6 to 10.4, p = 0.03). At matched intensity and duration the difference was 16.2 points (p < 0.001), and 24 weeks of tai chi beat 12 weeks by 9.6 points (p = 0.007). Secondary gains covered patient global assessment, anxiety, self-efficacy and coping. In the earlier 66-person trial, tai chi ran 18.4 points ahead of wellness education plus stretching at 12 weeks (p < 0.001), holding at 24 weeks.

Measured in: 226 adults with fibromyalgia, mean duration of pain around 12 years, randomized to Tai Chi at several doses or to supervised aerobic exercise twice weekly for 24 weeks

The comparator was supervised aerobic exercise progressing to 60 to 70% of maximum heart rate, which is the standard recommendation rather than a weak control, and that is what makes this result worth reporting. Only 14 men were in the Tai Chi arms, 17 in the whole trial, so this is effectively a finding in women.

Wang et al., tai chi versus aerobic exercise for fibromyalgia: comparative effectiveness randomized controlled trial · BMJ 2018;360 Wang et al., a randomized trial of tai chi for fibromyalgia · N Engl J Med 2010

Twelve weeks of classical Yang style tai chi twice a week improved WOMAC by 167 points (95% CI 145 to 190) against 143 points (119 to 167) for standard physical therapy, a between-group difference of 24 points (95% CI -10 to 58) that was not statistically significant. Depression and the physical component of quality of life improved more in the tai chi group.

Measured in: 204 adults with symptomatic knee osteoarthritis, mean age 60, 70% women, 53% white

This is an equivalence result against an active comparator, so it says tai chi performed about as well as the treatment already recommended, not that it does something physical therapy cannot. Participants reported their own pain without blinding, and the physical therapy arm switched to home exercise at week six, which weakens the comparator in the second half.

Wang et al., comparative effectiveness of tai chi versus physical therapy for knee osteoarthritis · Ann Intern Med 2016

Sixty-minute sessions twice a week for 24 weeks, using six movements built into an 8-form routine. Tai chi beat resistance training on maximum excursion by 5.55 percentage points and stretching by 11.98, and on directional control by 10.45 and 11.38 points. Falls ran at 0.22 per participant-month against 0.62 with stretching (incidence rate ratio 0.33, 95% CI 0.16 to 0.71); against resistance training the ratio was 0.47 (0.21 to 1.00). Gains persisted three months after the program ended.

Measured in: 195 patients with stage 1 to 4 Parkinson's disease, mean age 68; the 65-person tai chi arm was 31% women

Nothing measured here changes the disease itself, only balance and fall rate while living with it. The margin over resistance training was borderline, with an upper confidence limit of exactly 1.00, and participants knew which of the three programs they were attending.

Li et al., tai chi and postural stability in patients with Parkinson's disease · N Engl J Med 2012

In hypertensive samples, and in a model that controls for publication bias, the same meta-analysis returns about 10.4 mmHg systolic from English-language trials and about 18.6 mmHg from Chinese-language trials for the same intervention, against non-exercising controls.

Measured in: 342 adults with prehypertension at two tertiary hospitals in China, mean age 49.3, 176 women and 166 men; plus 3,223 middle-aged adults (mean 56.6) in the pooled analysis

That gap is publication bias measured rather than argued about, and it is the reason to treat every large Tai Chi effect in this literature carefully. Both figures are against people doing no exercise, so neither is a comparison against other training. They apply to hypertensive samples and come from a bias-adjusted model, so they are an estimate of what the trials would have shown, not a raw pooled result.

Li et al., effect of tai chi vs aerobic exercise on blood pressure in patients with prehypertension · JAMA Netw Open 2024;7(2):e2354937 Wu et al., Tai Ji Quan as antihypertensive lifestyle therapy: systematic review and meta-analysis · J Sport Health Sci 2021

Twelve weeks of tai chi twice a week changed neither six-minute walk distance (median 115 feet (35 metres) against 2 in controls, p = 0.95) nor peak oxygen uptake (1.1 against -0.5 mL/kg/min, p = 0.81). Quality of life did improve (-19 against 1 on the Minnesota Living With Heart Failure Questionnaire, p = 0.02), along with mood (p = 0.01) and exercise self-efficacy (p < 0.001).

Measured in: 100 outpatients with systolic heart failure, NYHA class I to III, mean ejection fraction 29%, mean age 67, 64 men and 36 women

The trial was powered to detect a 262-foot (80-metre) difference in six-minute walk and observed about 108 feet (33 metres), so this rules out a large effect rather than establishing there is none. The authors do not oversell it.

Yeh et al., tai chi exercise in patients with chronic heart failure: a randomized clinical trial · Arch Intern Med 2011

Of 153 randomized trials, 50 (33%) reported adverse events at all and 18 (12%) described an explicit monitoring protocol. Reported events were minor and expected, primarily musculoskeletal, with knee and back pain the commonest. No serious adverse event was attributed to the intervention in any trial.

Measured in: Participants across 153 English-language randomized trials of tai chi published through March 2013, spanning healthy older adults and clinical populations

Two thirds of the trials reported nothing about adverse events, so what this establishes is that serious harm has not surfaced where anyone looked. The review's own conclusion is that reporting is too poor and inconsistent to support a firm safety statement.

Wayne et al., what do we really know about the safety of tai chi? A systematic review of adverse event reports in randomized trials · Arch Phys Med Rehabil 2014

Across 12 randomized trials and 984 participants (811 analyzed), tai chi added 97 feet (29.64 metres) of six-minute walk distance over usual care and improved measured pulmonary function. Added on top of other interventions it showed no superiority, and effects on breathlessness and quality of life were inconclusive. No adverse events were reported in any included trial.

Measured in: 318 participants across the pooled trials

Certainty across these Tai Chi versus usual care outcomes was graded from very low to moderate, so this is weak evidence. The pooled walk-distance gain rests on 318 people rather than the full review population.

Ngai et al., tai chi for chronic obstructive pulmonary disease (Cochrane review) · Cochrane Database Syst Rev 2016;CD009953

A pooled effect of g=0.90 (four trials, 421 people, heterogeneity I-squared 92%) falls to g=0.39 when one high-risk-of-bias trial is removed, and that lower estimate rests on three trials with no remaining heterogeneity (I-squared 0%).

Measured in: The higher-quality estimate rests on 4 randomized trials and 421 participants, within a review covering 2,553 people in total

Heterogeneity of 92% means the trials are not measuring one thing, and the effect more than halves once quality is accounted for. The larger participant count belongs to the whole review, not to the estimate quoted, and pairing the two would overstate this considerably.

Wayne et al., effect of tai chi on cognitive performance in older adults: systematic review and meta-analysis · J Am Geriatr Soc 2014

The sham arm, which was matched gentle movement without the breath and attention components, also beat the education-support control on sleep and cognition.

Measured in: 167 breast cancer survivors, all women, mean age 59.9

We could not confirm any direct comparison between the full practice and the sham from the published record, so no claim is made about one beating the other. What the trial does show is that matched gentle movement alone outperformed education support, which is the comparison that matters for deciding how much of the benefit is specific to the method.

Randomized controlled trial of Tai Chi Easy/Qigong and sham qigong on breast cancer survivors' fatigue and associated symptoms · Complement Ther Clin Pract 2025

60 older people using wheelchairs in residential care were randomized to 26 weeks of seated tai chi or usual activities. The seated tai chi group had significantly lower depression scores and higher quality of life at the end of the program.

Measured in: 60 older residential-care residents who use wheelchairs

One small unblinded trial with self-reported outcomes, run in a setting where group attention is itself an intervention and the comparator was usual activities. Seated practice has not been shown to improve standing balance, which is the outcome the falls evidence rests on.

Seated tai chi versus usual activities in older people using wheelchairs: a randomized controlled trial · Complement Ther Med 2016

Odds of survival were 50% higher for people with stronger social relationships (OR 1.50, 95% CI 1.42 to 1.59).

Measured in: 148 studies, 308,849 people, mean age 64, followed an average of 7.5 years

That pooled figure hides a wide spread: measures capturing several dimensions of social integration gave 1.91, while the crude question of whether someone lives alone gave 1.19. It is also odds of survival, not a reduction in risk of death, and those are different quantities.

What could explain it instead: Observational throughout. Illness shrinks social networks, so reverse causation is unresolved, and isolation and poor health reinforce each other.

Holt-Lunstad et al., social relationships and mortality risk, a meta-analytic review · PLoS Med 2010;7(7):e1000316

With confounders statistically controlled, the weighted average effects were social isolation OR 1.29, loneliness OR 1.26 and living alone OR 1.32, corresponding to 29%, 26% and 32% higher likelihood of death. The review tested objective against subjective measures and found no difference between them.

Measured in: Studies published between January 1980 and February 2014 providing quantitative mortality data on loneliness, social isolation or living alone

Results differed by participant age in a direction most coverage inverts: social deficits predicted death more strongly in samples with a mean age under 65 than in older samples. Initial health status also influenced the findings, which is the reverse-causation problem showing up inside the meta-analysis itself.

What could explain it instead: Observational throughout. Poor health reduces contact and membership before it kills, and the review reports that baseline health status changed the size of the association.

Holt-Lunstad et al., loneliness and social isolation as risk factors for mortality · Perspect Psychol Sci 2015;10(2):227-237

Social isolation carried a hazard ratio of 1.73 (95% CI 1.65 to 1.82) adjusted only for age, sex, ethnicity and chronic disease, falling to 1.26 (1.20 to 1.33) after full adjustment. Loneliness carried 1.38 (1.30 to 1.47) minimally adjusted and 0.99 (0.93 to 1.06) fully adjusted, a measured null.

Measured in: 466,901 UK Biobank participants, mean age 56.5 at baseline, mean follow-up 6.5 years

The full adjustment included depressive symptoms, which is plausibly a step on the causal path from loneliness to death rather than a confounder. Adjusting for a mediator removes real effect, so the loneliness null is a ceiling on the direct effect rather than a measurement of the total one.

What could explain it instead: Socioeconomic position, smoking, alcohol, physical activity, depressive symptoms and existing chronic disease all cluster with isolation and each independently predicts death; the paper's own design is an attempt to quantify how much of the association they account for.

Elovainio et al., contribution of risk factors to excess mortality in isolated and lonely individuals, UK Biobank · Lancet Public Health 2017;2(6):e260-e266

After adjustment for demographics and baseline health, the most isolated fifth had a hazard ratio for death of 1.26 (95% CI 1.08 to 1.48). Loneliness in the same adjusted model gave 0.92 (0.78 to 1.09), and adding it left the isolation estimate unchanged, so loneliness was not the pathway carrying isolation's effect.

Measured in: 6,500 men and women aged 52 and over in the English Longitudinal Study of Ageing, followed a mean of 7.25 years to March 2012

A single national cohort of older English adults, with isolation measured once at baseline from contact with family and friends plus civic participation. One measurement cannot capture a social life that changes with health over seven years.

What could explain it instead: Reverse causation from illness present at baseline: the adjustment for baseline health is the paper's attempt to handle it, and self-reported health is a blunt instrument for early disease.

Steptoe et al., social isolation, loneliness, and all-cause mortality in older men and women · Proc Natl Acad Sci USA 2013;110(15):5797-5801

Poor social relationships were associated with a 29% higher risk of incident coronary heart disease (pooled RR 1.29, 95% CI 1.04 to 1.59) and a 32% higher risk of stroke (1.32, 1.04 to 1.68).

Measured in: 16 longitudinal datasets in high-income countries, 4,628 coronary events and 3,002 strokes, follow-up 3 to 21 years; 11 studies pooled for coronary disease and 8 for stroke

Both confidence intervals come close to 1.04 at the lower bound, so the pooled estimates are consistent with much smaller effects than those percentages suggest. The exposure is a mixture of loneliness and isolation measures that the review pooled together.

What could explain it instead: Subclinical cardiovascular disease reduces activity and social participation years before an event is recorded, and the shared upstream causes are the same ones that drive cardiac risk: smoking, deprivation, depression and inactivity.

Valtorta et al., loneliness and social isolation as risk factors for coronary heart disease and stroke · Heart 2016;102(13):1009-1016

Social isolation carried a 1.26-fold risk of dementia (95% CI 1.15 to 1.37) after adjustment for demographic, socioeconomic, biological, cognitive, behavioral and psychological factors including loneliness and depression. Loneliness gave 1.04 (0.94 to 1.16), and 75% of the loneliness association was attributable to depressive symptoms.

Measured in: 462,619 UK Biobank participants, mean age 57.0 at baseline, mean follow-up 11.7 years, 4,998 incident all-cause dementia cases

The imaging arm reporting lower gray matter volume in isolated people is cross-sectional and cannot separate a cause of isolation from a consequence of it. The dementia prodrome runs longer than most cohorts follow, so an 11.7-year mean follow-up does not remove reverse causation.

What could explain it instead: Prodromal dementia causes withdrawal years before diagnosis through apathy, word-finding difficulty and losing the thread of conversation, which produces exactly this association with no causal link from isolation to disease.

Shen et al., associations of social isolation and loneliness with later dementia · Neurology 2022;99(2):e164-e175

Poor social network RR 1.59 (95% CI 1.31 to 1.96) and poor social support RR 1.28 (1.01 to 1.62). In studies following people 10 years or more, good social engagement was modestly protective, RR 0.88 (0.80 to 0.96). Loneliness gave RR 1.38 with an interval of 0.98 to 1.94 that crosses 1.

Measured in: 31 cohort and 2 case-control studies, 2,370,452 participants, searched to May 2017

The structural measures reach significance and the subjective one does not, on an interval wide enough that a real loneliness effect is not excluded. Social support pooling showed substantial heterogeneity (I-squared 55.5%), so the studies behind that estimate disagree with each other.

What could explain it instead: Reverse causation from the dementia prodrome, which the review addresses only by reporting long-follow-up studies separately rather than by excluding early cases.

Penninkilampi et al., the association between social engagement, loneliness, and risk of dementia · J Alzheimers Dis 2018;66(4):1619-1633

Compared with married people, widowed people had a mean hazard ratio of 1.23 (95% CI 1.19 to 1.28) among estimates adjusted for age and further covariates with a high quality score. The effect was larger for men, 1.27 (1.19 to 1.35), than for women, 1.15 (1.08 to 1.22), and the gap narrowed as mean age rose.

Measured in: 1,381 mortality risk estimates extracted from 124 publications, covering more than 500 million person-records

Estimate magnitude also varied with sample size, geographic region, level of statistical adjustment and study quality, which means the pooled figure is sensitive to which studies are included. Shared exposures within a marriage, from diet to smoking to housing, are not separable from bereavement itself.

What could explain it instead: Couples share health behaviors, income and environment, and a spouse's illness often precedes their death by years during which the survivor is caregiving, so the exposure period is not clean.

Shor et al., widowhood and mortality, a meta-analysis and meta-regression · Demography 2012;49(2):575-606

Psychological interventions reduced loneliness against control groups with a small to medium effect, Hedges' g = 0.43, pooled across 28 randomized trials.

Measured in: 31 randomized trials in the systematic review (N = 3,959), 28 pooled (N = 3,039), spanning a range of cultures, ages and populations; interventions were mostly face to face, group based and weekly, and the most common type was cognitive behavioral therapy

Considerable heterogeneity between trials, and the included studies were of mixed quality on the Cochrane risk-of-bias tool. Type of intervention only approached significance as a moderator, so this pooled figure does not identify which therapy is doing the work.

Hickin et al., the effectiveness of psychological interventions for loneliness · Clin Psychol Rev 2021;88:102066

Pooling five cohort studies, volunteers had a relative risk of death of 0.78 (95% CI 0.66 to 0.90). Cohort studies also showed favorable effects on depression, life satisfaction and wellbeing, but not on physical health.

Measured in: 40 papers selected: 5 randomized trials (7 papers), 4 non-randomized trials and 17 cohort studies (29 papers); the mortality pooling rests on 5 of the cohorts

The experimental studies did not confirm the wellbeing and depression benefits the cohorts showed, and the review found insufficient evidence that type or intensity of volunteering changed outcomes. Five cohorts is a thin base for a mortality estimate.

What could explain it instead: Healthy-volunteer selection in its most literal form: being well enough to leave the house and commit to regular unpaid work is itself a marker of health, mobility, income and cognition.

Jenkinson et al., is volunteering a public health intervention? Systematic review and meta-analysis · BMC Public Health 2013;13:773

Across 20 randomized comparison studies the mean reduction in loneliness was 0.198 standard deviations (95% CI 0.08 to 0.32), with 14 of the 20 individually showing no change. Within that group, interventions addressing maladaptive social cognition reduced loneliness by 0.598 (4 studies), enhancing social support by 0.162 (12 studies), improving social skills by 0.017 (2 studies, p = 0.90) and increasing opportunities for social contact by 0.062 (2 studies, p = 0.67).

Measured in: Trials targeting loneliness in adults, adolescents and children; 12 single-group pre-post studies (mean reduction 0.367), 18 non-randomized comparisons (0.459) and 20 randomized comparisons (0.198)

The subgroup difference is a moderator analysis, not a head-to-head trial, it reached significance at p = 0.05 exactly, and the winning cell contains four studies. Effect sizes fell steadily as study design got stronger, which is the shape of a literature inflating itself through weak designs.

Masi et al., a meta-analysis of interventions to reduce loneliness · Pers Soc Psychol Rev 2011;15(3):219-266

In cross-lagged panel models over five years, loneliness predicted subsequent change in depressive symptoms while depressive symptoms did not predict subsequent change in loneliness. The temporal ordering held after accounting for gender, ethnicity, education, physical functioning, medications, social network size, neuroticism, stressful life events, perceived stress and social support.

Measured in: 229 men and women aged 50 to 68 at study onset, population-based and ethnically diverse, Chicago Health, Aging and Social Relations Study

A single cohort of 229 people in one American city, and cross-lagged panel models establish temporal precedence rather than causation. Loneliness and depressive symptom scales share item content, which inflates the association between them by construction.

What could explain it instead: An unmeasured third factor changing before both, such as early cognitive decline, an undiagnosed illness or a change in hearing, would produce this ordering without loneliness causing anything.

Cacioppo, Hawkley and Thisted, perceived social isolation makes me sad, 5-year cross-lagged analyzes · Psychol Aging 2010;25(2):453-463

Video calls made little to no difference to UCLA Loneliness Scale scores against usual care: mean difference 0.44 points on a 20 to 80 scale at three months (95% CI 3.28 better to 2.41 worse), 0.34 at six months and 2.40 at twelve months. Depression scores were similarly unchanged at three and six months.

Measured in: 3 cluster quasi-randomized trials, 201 participants in total, all comparing video calls to usual care in nursing homes, none conducted during the COVID-19 pandemic

Certainty was downgraded three levels for study limitations, imprecision and indirectness, so this is an absence of usable evidence as much as a demonstrated null. No included study reported social isolation as an outcome, and nursing-home residents are not the population most readers belong to.

Noone et al., video calls for reducing social isolation and loneliness in older people, Cochrane rapid review · Cochrane Database Syst Rev 2020;5:CD013632

Breathing near six breaths a minute matches the natural frequency of the baroreflex, the loop that adjusts heart rate to changes in blood pressure. Breathing at that frequency drives heart rate into large smooth oscillations, many times bigger than at rest. Biofeedback training at that rate shows small to moderate benefit for anxiety, among its largest effects.

Measured in: 58 randomized studies pooled for the outcomes; the mechanism comes from separate physiological work

The effect does not require breath and heartbeat to line up precisely. One study by the same group found they often do not, more so with age, and the large oscillations appeared anyway. Someone who cannot breathe to a metronome still gets the effect. The authors note the number of studies per condition is small and call for confirmation.

Lehrer et al., heart rate variability biofeedback improves emotional and physical health and performance, systematic review and meta-analysis · Appl Psychophysiol Biofeedback 2020;45(3):109-129 Lehrer & Gevirtz, heart rate variability biofeedback, how and why does it work (the baroreflex mechanism) · Front Psychol 2014;5:756 Lehrer, Vaschillo & Vidali, heart rate and breathing are not always in phase during resonance frequency breathing · Appl Psychophysiol Biofeedback 2020;45(3):145-152

Mindfulness programs produced small improvements: anxiety 0.38, depression 0.30 and pain 0.33 at eight weeks, shrinking to about 0.22 by three to six months. No trial showed them beating active comparators such as exercise or medication.

Measured in: 47 randomized trials, 3,515 participants

The moderate grade describes how confident we can be, not how large the effect is, and these effects are small. It applies to mindfulness specifically; mantra programs were graded insufficient. And that no trial showed superiority over exercise or medication is not the same as equivalence, since few trials made that comparison.

Goyal et al., meditation programs for psychological stress and well-being · JAMA Intern Med 2014;174(3):357-368

Breathing at six breaths a minute for a short controlled period lowered blood pressure in people with essential hypertension, from 149.7 to 141.1 mmHg systolic and from 82.7 to 77.8 mmHg diastolic, and nearly doubled arterial baroreflex sensitivity, from 5.8 to 10.3 ms/mmHg. Carbon dioxide did not fall, so this was slow breathing rather than over-breathing.

Measured in: 20 adults with essential hypertension, mean age 56, compared with 26 controls, mean age 52

This is the acute response measured over minutes in a laboratory, not a treatment result. Nothing here says the pressure stays down between sessions, and the trial literature that tried to answer that question is where the picture turns mixed.

Joseph et al., slow breathing improves arterial baroreflex sensitivity and decreases blood pressure in essential hypertension · Hypertension 2005;46(4):714-718

Resonance frequency was measured directly in 56 people by pacing breathing across a range of rates and finding where heart-rate oscillations peaked. It fell with increasing height, sat lower in men than in women, and did not vary with age, weight or the presence of asthma. It stayed constant across ten training sessions. In the earlier mapping study the largest heart-rate oscillations appeared between 0.055 and 0.11 Hz, roughly 3.3 to 6.6 breaths a minute.

Measured in: 32 adults with asthma and 24 healthy adults for the resonance measurements; the frequency mapping came from 5 healthy men

Six breaths a minute is the middle of a distribution rather than a universal constant, and the studies that located it are small. The height relationship suggests blood volume sets the loop delay, which is an interpretation the authors offer rather than a demonstrated cause. The frequency mapping rests on five men.

Vaschillo, Vaschillo & Lehrer, characteristics of resonance in heart rate variability stimulated by biofeedback · Appl Psychophysiol Biofeedback 2006;31(2):129-142 Vaschillo et al., heart rate variability biofeedback as a method for assessing baroreflex function · Appl Psychophysiol Biofeedback 2002;27(1):1-27

Pooled across 24 studies and 484 participants, heart rate variability biofeedback improved self-reported stress and anxiety with Hedges' g of 0.81 before-to-after and 0.83 against control groups. That is a large effect, and it is roughly two and a half times the pooled effect for breathing practice generally.

Measured in: 484 participants across 24 studies, clinical and non-clinical

This effect is larger than the pooled effects for breathing practice and for biofeedback in other reviews. Biofeedback cannot be blinded: the participant sees their own heart rhythm respond, the trainer knows who is in which arm, and the outcomes are all self-reported. Control conditions were mostly inactive. Expectancy is inside this estimate and there is no way to subtract it from the published data.

Goessl, Curtiss & Hofmann, the effect of heart rate variability biofeedback training on stress and anxiety, a meta-analysis · Psychol Med 2017;47(15):2578-2586

Across 14 randomized trials and 794 adults, heart rate variability biofeedback improved depressive symptoms with Hedges' g of 0.38 (95% CI 0.16 to 0.60), a small to moderate effect, with moderate heterogeneity (I-squared 45%).

Measured in: 794 adults across 14 randomized trials, in a mix of psychological and physical health conditions

A separate research group from the anxiety meta-analysis, and about half the size of that estimate, in a design with the same unavoidable blinding problem. Publication year and the choice of depression questionnaire both moderated the result, so the effect is not stable across how it is measured.

Pizzoli et al., a meta-analysis on heart rate variability biofeedback and depressive symptoms · Sci Rep 2021;11(1):6650

Pooling 44 studies and 21,438 healthy adults to establish reference values found between-person variation of up to 260,000% for the frequency-domain measures, and short-term values consistently lower than the standing Task Force norms.

Measured in: 21,438 healthy adults across 44 studies

The reviewers set out to produce normal ranges and concluded the spread was too wide for them to be useful. Some of that is biological, from age, height, breathing rate, and posture, and some is measurement method, since the included studies did not record under one protocol. Either way, a reading only carries information against your own earlier readings.

Nunan, Sandercock & Brodie, a quantitative systematic review of normal values for short-term heart rate variability in healthy adults · Pacing Clin Electrophysiol 2010;33(11):1407-1417 Task Force of the European Society of Cardiology and the North American Society of Pacing and Electrophysiology, heart rate variability standards of measurement, physiological interpretation and clinical use · Circulation 1996;93(5):1043-1065

In 808 people monitored about 11 days after a myocardial infarction and followed a mean of 31 months, those with 24-hour heart rate variability below 50 ms had 5.3 times the mortality of those above 100 ms.

Measured in: 808 survivors of acute myocardial infarction, Holter-monitored at 11 days and followed 31 months

This is why heart rate variability is taken seriously as a physiological signal, and it does not transfer to a healthy person's morning reading. The prediction was made at the low end of the range, in people who had just had a heart attack, across a group rather than for any individual.

What could explain it instead: Low heart rate variability after an infarct travels with the size of the infarct and with worse left ventricular function, so it may be marking how much heart muscle was lost rather than acting on survival itself. Medication at the time, particularly beta-blockade, also shifts the measure.

Kleiger et al., decreased heart rate variability and its association with increased mortality after acute myocardial infarction · Am J Cardiol 1987;59(4):256-262

Coronial records across Australia from July 2002 to June 2015 identified 22 drowning deaths attributed to hypoxic blackout. All 22 were male and 17 were aged 18 to 34. Twelve happened in swimming pools, domestic or public, and nine in the ocean or a harbour, five of those spearfishing and four freediving.

Measured in: 22 fatalities in the Royal Life Saving National Fatal Drowning Database, drawn from the Australian National Coronial Information System, over 13 years

A count with no denominator, so it says nothing about risk per session, and coronial coding will miss cases where the breath-holding was never recorded. It is one country over thirteen years. What it establishes is that these deaths happen in ordinary swimming pools to fit young people, not only to competitive freedivers. Not all were healthy young men: of the twelve pool deaths, four had a pre-existing condition, mostly cardiac, and three had a blood alcohol level above 0.05.

Franklin, Peden & Pearn, drowning deaths in Australia caused by hypoxic blackout, 2002-2015 · Med J Aust 2018;208(6):271

Hyperventilation is used deliberately in standard EEG recording to provoke abnormal activity. It brings out epileptiform discharges more often than full seizures, and it triggers a clinical seizure in up to 50% of people with generalized epilepsy. It works better in children with absence seizures than in adults, and less well in focal epilepsy.

Measured in: People undergoing routine EEG, across the reviewed literature

These figures come from people already diagnosed with epilepsy, hyperventilating to instruction for about three minutes under supervision. What has not been quantified is the risk in someone with an undiagnosed seizure threshold doing a much longer breathing protocol at home.

Rana et al., hyperventilation and seizures, not a new sense, a literature review · Neuropediatrics 2023;54(6):359-364

Six healthy volunteers hyperventilated until symptoms appeared. As alveolar carbon dioxide fell by roughly 20 mmHg, tingling developed in the hands, face and trunk, and the sensory and muscle responses to a fixed nerve stimulus grew progressively larger, showing a rise in axonal excitability. The authors attribute it to the fall in plasma calcium that follows respiratory alkalosis acting on the axon membrane.

Measured in: 6 healthy volunteers hyperventilating to the point of paraesthesiae, with nerve recordings

Six people, and a deliberate provocation rather than an observed practice. What it settles is the mechanism: the tingling and the clawed hands are nerves firing because carbon dioxide has fallen, and they resolve when breathing returns to normal.

Macefield & Burke, paraesthesiae and tetany induced by voluntary hyperventilation, increased excitability of human cutaneous and motor axons · Brain 1991;114(Pt 1B):527-540

Forty-eight people with panic disorder hyperventilated to instruction. Twenty-four of them, exactly half, judged the sensations produced to be like their own panic attacks. Respiratory physiology before and during the test did not separate those who recognized the sensations from those who did not.

Measured in: 48 adults meeting DSM-III-R criteria for panic disorder

The authors' own reading is that what makes hyperventilation feel like panic is how the sensations are interpreted rather than the respiratory changes themselves, since the physiology was the same in both halves of the group. That does not make the experience less real for the half who had it. It means a breathing practice that produces those sensations is a live risk for this group and there is no breathing measurement that predicts who.

Spinhoven et al., the hyperventilation provocation test in panic disorder · Behav Res Ther 1992;30(5):453-461

Across 83 studies and 6,703 participants, the pooled prevalence of adverse events was 8.3% (95% CI 5% to 12%). It was 3.7% in experimental studies and 33.2% in observational ones. The most commonly reported were anxiety (33% of reports), depression (27%) and cognitive anomalies (25%), with gastrointestinal problems and suicidal behavior each at 11%. Events occurred in people with no prior mental health history.

Measured in: 6,703 participants across 83 studies of meditation practices and meditation-based therapies

The experimental and observational figures differ nine-fold: trials screen people out and observational studies ask afterwards, so neither number is the rate for a person following a video. This covers meditation broadly rather than breathing practice specifically, and intensive breath-based methods sit inside that category rather than separated from it.

Farias et al., adverse events in meditation practices and meditation-based therapies, a systematic review · Acta Psychiatr Scand 2020;142(5):374-393

Five minutes a day of cyclic sighing improved mood and lowered resting breathing rate more than mindfulness meditation of the same duration, over one month.

Measured in: 108 adults randomized across four arms; this comparison rests on 30 who sighed and 24 who meditated

The mood result covers the whole trial, while the specific comparison described here rests on about fifty people. Results were reported as p-values without effect sizes, so the size of the difference is not established. A first signal, not yet a settled finding.

Yilmaz Balban et al., brief structured respiration practices enhance mood and reduce physiological arousal · Cell Rep Med 2023;4(1):100895

Across 17 randomized trials of slow breathing at ten breaths a minute or fewer, done at least three days a week for at least four weeks, systolic pressure fell 5.62 mmHg and diastolic 2.97 mmHg against control. Two other analyzes of the same field disagree. Pooling eight trials of a breathing-pacer device gave 3.67 mmHg systolic, and removing the five trials sponsored by or involving the device manufacturer left no effect at all. An individual-patient analysis restricted to blinded, actively controlled trials found 2.2 mmHg in favor of the controls.

Measured in: Adults with hypertension or prehypertension; the individual-patient analysis drew on 3 blinded trials with usable data

The three analyzes disagree because they included different things. The largest number comes from the loosest inclusion criteria, and the effect shrinks toward zero as the comparison gets more demanding: sham devices and music instead of usual care, and independent funding instead of manufacturer involvement. The direction here is recorded as mixed for that reason. The 2.2 mmHg figure did not reach significance (95% CI -2.7 to 7.0).

Chaddha et al., device and non-device-guided slow breathing to reduce blood pressure, systematic review and meta-analysis · Complement Ther Med 2019;45:179-184 Mahtani, Nunan & Heneghan, device-guided breathing exercises in the control of human blood pressure, systematic review and meta-analysis · J Hypertens 2012;30(5):852-860 Landman et al., efficacy of device-guided breathing for hypertension in blinded, randomized, active-controlled trials, meta-analysis of individual patient data · JAMA Intern Med 2014;174(11):1815-1821

In seven anesthetized dogs whose heart rate was paced artificially, adding a respiratory rhythm to the pacing cut the ratio of physiological dead space to tidal volume by about 10% and reduced intrapulmonary shunt by 51% compared with steady pacing. The interpretation is that matching heartbeat to breath sends more blood through the lung while it holds fresh air.

Measured in: 7 anesthetized dogs

This is the study behind the claim that the heart-breath rhythm improves gas exchange, and it was done in dogs under anesthesia with artificial pacing. The equivalent has not been shown in awake humans, and a later modelling analysis argues the efficiency account does not fully explain why the rhythm exists.

Hayano et al., respiratory sinus arrhythmia, a phenomenon improving pulmonary gas exchange and circulatory efficiency · Circulation 1996;94(4):842-847 Ben-Tal, Shamailov & Paton, evaluating the physiological significance of respiratory sinus arrhythmia, looking beyond ventilation-perfusion efficiency · J Physiol 2012;590(8):1989-2008

Forty-two people were randomized to fifteen days of daily Wim Hof practice or to no intervention. Nothing moved. Heart rate, RMSSD, SDNN, systolic and diastolic pressure, pulse wave velocity, perceived stress, positive and negative affect, and both vitality scales all showed no group-by-time interaction, and neither did the responses to a cold pressor test.

Measured in: 42 adults randomized to intervention or control for 15 days

Fifteen days is short, forty-two people is small, and the control group did nothing rather than something matched, which would ordinarily bias toward finding an effect. It did not find one. A single null does not close a question; it does place the burden on the next trial.

Ketelhut et al., the effectiveness of the Wim Hof method on cardiac autonomic function, blood pressure, arterial compliance, and different psychological parameters · Sci Rep 2023;13(1):17517

Twenty-four healthy young men were randomized, twelve to ten days of training in meditation, cyclic hyperventilation followed by breath retention, and cold immersion, and twelve to nothing. All then received intravenous E. coli endotoxin. The trained group's practice produced intermittent respiratory alkalosis and hypoxia with sharply raised adrenaline, and after the endotoxin they showed higher anti-inflammatory IL-10, lower TNF-alpha, IL-6 and IL-8, and fewer flu-like symptoms.

Measured in: 24 healthy young men, 12 trained and 12 untrained

This is the study almost every claim made for the method rests on. Twenty-four young men, one artificial challenge in a laboratory, no clinical outcome measured, and no follow-up. The training bundled meditation, breathing and cold immersion, so the breathing component was never isolated. The control group did nothing, and the surge in adrenaline it turns on is the same mechanism that makes the practice dangerous near water.

Kox et al., voluntary activation of the sympathetic nervous system and attenuation of the innate immune response in humans · Proc Natl Acad Sci U S A 2014;111(20):7379-7384

Sixteen weeks of nightly wrist-worn recording in Olympic water polo players gave a mean weekly coefficient of variation of 5.4% for log-transformed RMSSD and 7.6% for heart rate. The authors place that within or below the 3 to 13% day-to-day variability seen with other measurement protocols.

Measured in: 11 elite male water polo players, mean age 28.8, recorded nightly for 16 weeks

Eleven elite male athletes sleeping on a controlled schedule is close to the best case for this measurement, and the figure is for the log-transformed value, which compresses the spread considerably. On the raw scale most devices display, and in a person with ordinary sleep and drinking habits, the night-to-night movement is larger. It is why the athlete literature recommends reading a rolling weekly average rather than a single morning.

Bellenger et al., evaluating the typical day-to-day variability of WHOOP-derived heart rate variability in Olympic water polo athletes · Sensors (Basel) 2022;22(18):6723

The paranasal sinuses produce nitric oxide continuously at high concentration, and it is drawn into the lungs on the in-breath. In eight healthy adults alternating nasal and oral breathing, transcutaneous oxygen tension was 10% higher during nasal breathing in six of the eight. In six long-term intubated patients, whose noses were bypassed, adding nasal air to the ventilator circuit raised arterial oxygen tension by 18%. Humming, which flushes the sinuses, raises exhaled nasal nitric oxide about fifteen-fold in healthy adults.

Measured in: 8 healthy adults and 6 intubated patients in the oxygenation work; 10 healthy adults for the humming measurement

Very small numbers, acute measurements of an oxygen tension rather than any health outcome, and the largest effect came from patients whose own noses had been surgically bypassed, which is the opposite of the situation a reader is in. The mechanism is established. No study has shown that breathing through your nose instead of your mouth, in a healthy person going about a day, changes anything you would notice.

Lundberg et al., inhalation of nasally derived nitric oxide modulates pulmonary function in humans · Acta Physiol Scand 1996;158(4):343-347 Lundberg et al., high nitric oxide production in human paranasal sinuses · Nat Med 1995;1(4):370-373 Weitzberg & Lundberg, humming greatly increases nasal nitric oxide · Am J Respir Crit Care Med 2002;166(2):144-145

Nine adults completed anaerobic power testing breathing nasally and orally in a within-subject comparison. No power output or performance measure differed between the two. Nasal breathing did keep the respiratory exchange ratio below 1.0, meaning it prevented hyperventilation, and it came with a higher heart rate at the same work.

Measured in: 9 adults, 7 men and 2 women

Nine people and a single anaerobic test. It is enough to say that forcing the airway smaller did not improve power output, and not enough to settle anything about endurance, adaptation over months, or breathing pattern during ordinary training.

Recinto et al., effects of nasal or oral breathing on anaerobic power output and metabolic responses · Int J Exerc Sci 2017;10(4):506-514

Ten studies covering 213 patients were located. Two reported a statistically significant improvement in an established sleep apnea marker, either the apnea-hypopnea index or oxygen desaturation. Others documented risk, specifically asphyxiation where nasal obstruction was present. The reviewers concluded the practice carries a potentially serious risk of harm and that the evidence base is inadequate.

Measured in: 213 patients across 10 studies, mostly with mild obstructive sleep apnea or mouth breathing

Two hundred and thirteen people across ten heterogeneous studies is a thin base for a practice this widely recommended, and the reviewers' safety concern is specific rather than observed: they discuss the potential for asphyxiation if the nose is obstructed, which is a reasoned caution rather than a recorded event. If your nose is blocked, taping the mouth removes the airway you have left.

Rhee et al., mouth taping for mouth breathing and sleep disordered breathing, a systematic review · PLoS One 2025;20(5):e0323643

A 36-year-old woman with no predisposing condition developed a haze, floaters and a dark blob in the left eye after straining in a yoga backbend; the authors noted no breathing exercise or headstand was involved. Examination found retinal hemorrhage below the optic disc, the pattern of Valsalva retinopathy, where a sharp rise in chest and abdominal pressure, whether from a hard physical strain or a forceful breath-hold against a closed glottis, can burst small retinal vessels. Vision was 20/20 throughout and the bleed had almost completely cleared a month later.

Measured in: A single case report in an otherwise healthy adult

One case, and it resolved. It shows the mechanism: a hard strain or breath-hold with the throat shut drives pressure into the eye, and the people for whom that matters are the ones who already have diabetic retinopathy, high myopia, a retinal tear, or recent eye surgery. Nobody has counted how often it happens.

Parvus et al., Valsalva retinopathy after yoga in a patient with no clear predisposing condition · J Vitreoretin Dis 2023;7(4):337-339

Protein supplementation added about 5.5 lb (2.5 kg) to one-rep-max strength and about 0.7 lb (0.3 kg) of fat-free mass on top of resistance training. The widely quoted plateau at 1.6 g per kilogram per day comes from this review, where it was not statistically significant (p = 0.079) with a range running from 1.03 to 2.20.

Measured in: 1,863 people across 49 randomized trials; the plateau analysis rests on 723 of them

Treat 1.6 as the rough center of a wide range rather than a threshold. The authors themselves suggest about 2.2 g/kg/day for anyone wanting to be certain. The absolute effects are small, and the authors are blunt that the training is the far more potent stimulus and the protein is the minor variable.

Morton et al., protein supplementation and resistance training-induced gains in muscle mass and strength, systematic review and meta-analysis · Br J Sports Med 2018;52(6):376-384

Pooling 23 studies and 525 participants for muscle size and 20 studies and 478 participants for strength, protein taken close to training showed a small to moderate effect on hypertrophy and no significant effect on strength. Once total daily protein intake was entered into the model, the difference between the timed and control conditions disappeared for both outcomes. Total protein intake was the strongest predictor of effect size.

Measured in: 525 participants across 23 hypertrophy trials and 478 across 20 strength trials, mostly young untrained or recreationally trained adults in short training blocks

The apparent timing benefit came from the timed groups eating more protein overall, not from when they ate it. The included trials are mostly 6 to 12 weeks in young adults, and only a minority directly compared matched total protein at different times of day, so this is a covariate-adjusted null rather than a set of head-to-head trials designed to answer the question.

Schoenfeld, Aragon and Krieger, the effect of protein timing on muscle strength and hypertrophy, a meta-analysis · J Int Soc Sports Nutr 2013;10(1):53

Pooling laboratory data on 34 young and 72 older men at rest and 35 young and 40 older men after a meal, post-prandial muscle protein synthesis rates were 16% lower in the older group, and synthesis was more than three-fold more responsive to protein ingestion in the young. A separate retrospective analysis put the per-meal protein dose that maximizes myofibrillar synthesis at 0.24 g per kilogram of body mass in men around 22 and 0.40 g per kilogram in men around 71, roughly 67% more.

Measured in: Pooled trial data on 181 men in the synthesis comparison; retrospective dose analysis in healthy younger and older men

Every participant in both analyzes was male, and the pooled data come from studies run by the same groups using their own protocols rather than from a prospective comparison. Synthesis rate is a signal measured over hours, not muscle gained. The per-meal dose estimate carries a wide spread (0.40 plus or minus 0.19 g/kg in the older men), so it is a central tendency rather than a threshold. The pooled participant count double-counts overlapping cohorts, so the effective sample is smaller than it reads.

Wall et al., aging is accompanied by a blunted muscle protein synthetic response to protein ingestion · PLoS One 2015;10(11):e0140903 Moore et al., protein ingestion to stimulate myofibrillar protein synthesis requires greater relative protein intakes in healthy older versus younger men · J Gerontol A Biol Sci Med Sci 2015;70(1):57-62

Across four cohorts and four national dietary surveys covering 8,107 community-dwelling older adults, 21.5% (95% CI 14.0 to 30.1) ate less than 0.8 g of protein per kilogram of adjusted body weight per day, 46.7% (38.3 to 55.3) less than 1.0 g/kg, and 70.8% (65.1 to 76.3) less than 1.2 g/kg. Prevalence was higher in women at every cut-off.

Measured in: 8,107 community-dwelling older adults across eight datasets in the PROMISS consortium, spanning several countries

Intake is estimated from dietary recall and food records, which under-report, so the true shortfall could sit either side of these figures. The cut-offs are recommendations rather than measured individual requirements, and the surveys were run in different years with different instruments, which is most of the spread in the confidence intervals. The claim that the shortfall is worse in women is verifiable at the 0.8 g/kg cut-off; the other cut-offs do not publish it separately.

Hengeveld et al., prevalence of protein intake below recommended in community-dwelling older adults, a meta-analysis across cohorts from the PROMISS consortium · J Cachexia Sarcopenia Muscle 2020;11(5):1212-1222

The PROT-AGE expert group, working from a systematic review of the evidence, recommends an average of 1.0 to 1.2 g of protein per kilogram of body weight per day for healthy adults over 65, at least 1.2 g/kg for those exercising, and 1.2 to 1.5 g/kg during acute or chronic illness, against a general adult recommended allowance of 0.8 g/kg.

Measured in: Healthy and ill adults over 65; a position statement drawn from the trial and observational literature rather than a single trial

This is a consensus recommendation, not a measured requirement, and the group states that the evidence for the higher figures is weaker than for the lower ones. Severe kidney disease is an explicit exception requiring individual medical supervision. Recommendations of this kind have historically moved when better nitrogen balance and tracer data arrived. The source is an expert position paper rather than a systematic review, which is a weaker kind of document than the effect sizes around it.

Bauer et al., evidence-based recommendations for optimal dietary protein intake in older people, the PROT-AGE position paper · J Am Med Dir Assoc 2013;14(8):542-59

Across 36 studies and 1,682 non-frail community-dwelling older adults, protein supplementation produced no significant change in lean body mass, muscle cross-sectional area, grip strength, lower-limb strength, walking speed or chair-rise performance, and added nothing measurable when combined with resistance training. In the frailty trial that separated the two, ten weeks of progressive resistance training raised strength 113% while multinutrient supplementation alone changed no primary outcome.

Measured in: 1,682 non-frail community-dwelling older adults across 36 trials; 100 frail nursing home residents, 63 women and 37 men, mean age 87, in the four-arm trial

Habitual protein intake in most included trials was already adequate, so this is a null for adding protein on top of enough rather than evidence that protein does not matter to someone eating too little. The authors say so directly, and call for trials in people with low baseline intake. One included trial had its protein powder supplied by a manufacturer whose sponsored body the first author heads, and another discloses fees from a meat industry association.

Ten Haaf et al., effects of protein supplementation on lean body mass, muscle strength and physical performance in nonfrail community-dwelling older adults, a systematic review and meta-analysis · Am J Clin Nutr 2018;108(5):1043-1059 Fiatarone et al., exercise training and nutritional supplementation for physical frailty in very elderly people · N Engl J Med 1994;330(25):1769-75

Across 28 trials and 1,358 healthy adults, the change in glomerular filtration rate did not differ between higher-protein and lower or normal-protein diets (standardized mean difference 0.11, p = 0.16). Post-intervention GFR was trivially higher on the high-protein diets (0.19), which the authors read as normal adaptation rather than damage. Higher protein was defined as at least 1.5 g/kg per day, at least 20% of energy, or at least 100 g per day.

Measured in: 1,358 adults without kidney disease across 28 randomized and controlled trials

Most trials ran weeks to months rather than years, GFR was estimated by several different methods across the included studies, and none of them enrolled anyone with existing kidney disease. This says nothing about that group, and it cannot rule out an effect over decades that a few months would not reveal.

Devries et al., changes in kidney function do not differ between healthy adults consuming higher- compared with lower- or normal-protein diets, a systematic review and meta-analysis · J Nutr 2018;148(11):1760-1775

Across 17 studies and 2,996 adults with non-diabetic chronic kidney disease, a very low protein diet (roughly 0.3 to 0.4 g/kg per day, usually with keto-acid analogues) probably reduced the number of people reaching end-stage kidney disease compared with a low or normal protein diet, at moderate certainty across 10 studies and 1,010 participants. A standard low protein diet against a normal protein diet may make little or no difference to that outcome, at low certainty. Neither comparison changed the death rate.

Measured in: 2,996 adults with non-diabetic chronic kidney disease, mostly stages 3 to 5, across 17 trials

The benefit belongs to supervised very low protein diets in advanced disease, not to eating less protein in general, and the effect on measured GFR was very low certainty. Adherence, adverse events and quality of life were sparsely reported, and a very low protein diet carries its own risk of losing muscle in a population already prone to it. This is a diet run with a renal dietitian, not a decision to make from a web page.

Hahn, Hodson and Fouque, low protein diets for non-diabetic adults with chronic kidney disease · Cochrane Database Syst Rev 2020;10(10):CD001892

Across 16 randomized trials pooled from 18 selected for review, protein source did not change absolute lean mass or muscle strength. Animal protein was favored for percent lean mass overall, and in adults under 50 it was favored for absolute lean mass as well. Resistance training did not moderate the comparison. In a 12-week trial matching both groups at 1.6 g/kg per day, 19 habitual vegans supplementing with soy and 19 omnivores supplementing with whey showed no between-group difference in leg lean mass, fiber cross-sectional area or strength.

Measured in: 16 pooled trials in adults of both sexes; 38 young men, 19 vegan and 19 omnivorous, mean age 26, in the matched-intake trial

The pooled trials rarely matched leucine content or total daily protein between arms, so part of the animal-protein edge on percent lean mass is a dose difference rather than a source difference. The cleanest comparison, where total protein was matched at 1.6 g/kg, found nothing, and it ran in 38 young men for 12 weeks.

Lim et al., animal protein versus plant protein in supporting lean mass and muscle strength, a systematic review and meta-analysis of randomized controlled trials · Nutrients 2021;13(2):661 Hevia-Larrain et al., high-protein plant-based diet versus a protein-matched omnivorous diet to support resistance training adaptations, a comparison between habitual vegans and omnivores · Sports Med 2021;51(6):1317-1330

Across 24 randomized trials and 1,063 adults on energy-restricted diets, a higher-protein version of the same calorie-cut diet mitigated the loss of fat-free mass by 0.9 lb (0.43 kg) (95% CI 0.09 to 0.78) and produced greater falls in body weight (-1.7 lb (-0.79 kg), 95% CI -1.50 to -0.08) and fat mass (-1.9 lb (-0.87 kg), 95% CI -1.26 to -0.48) than a standard-protein diet matched for fat and calories. Mean diet duration was about 12 weeks.

Measured in: 1,063 adults aged 18 and over of both sexes across 24 randomized controlled trials of isocaloric energy-restricted diets

The effects are modest and the confidence intervals for body weight and fat-free mass run close to zero, so this is a steady direction rather than a large effect. The trials averaged about 12 weeks, the diets were prescribed to match calories rather than always achieving it, and this is retention of existing lean mass during a deficit, not muscle gained. The design swaps carbohydrate for the extra protein, so it does not isolate protein on its own.

Wycherley et al., effects of energy-restricted high-protein, low-fat compared with standard-protein, low-fat diets, a meta-analysis of randomized controlled trials · Am J Clin Nutr 2012;96(6):1281-98

Eight healthy adults eating about 30 g of protein at each of breakfast, lunch and dinner ran a 24-hour muscle protein synthesis rate about 25% higher than the same total protein skewed toward dinner (10.7 g, 16.0 g and 63.4 g), on both day 1 and day 7. In 24 trained men given 80 g of whey over 12 hours after resistance exercise, four 20 g doses every 3 hours produced 31 to 48% more myofibrillar protein synthesis than eight 10 g doses every 90 minutes or two 40 g doses every 6 hours.

Measured in: 8 healthy adults of both sexes, mean age 37, in the distribution crossover; 24 trained men, 8 per arm, in the post-exercise timing trial

Both outcomes are synthesis rates measured over hours or a day, not muscle gained over months, and acute synthesis has repeatedly failed to predict long-term growth. Eight participants in the distribution study. No long trial has tested whether redistributing the same total protein changes lean mass across a training block, so the step from these numbers to a practical rule has not been measured.

Mamerow et al., dietary protein distribution positively influences 24-h muscle protein synthesis in healthy adults · J Nutr 2014;144(6):876-80 Areta et al., timing and distribution of protein ingestion during prolonged recovery from resistance exercise alters myofibrillar protein synthesis · J Physiol 2013;591(9):2319-31

Thirty-six young men were randomized to 0 g, 25 g or 100 g of intrinsically labeled milk protein after resistance exercise and tracked with a quadruple isotope tracer approach. The 100 g dose produced a larger myofibrillar protein synthesis response that was still running beyond 12 hours, and dietary amino acid incorporation into muscle protein rose across the dose range. Protein ingestion had a negligible effect on whole-body protein breakdown or amino acid oxidation, so the excess was not simply burned off.

Measured in: 36 healthy recreationally active men aged 18 to 40, 12 per group, single post-exercise feeding

One meal in 36 young men after a training session, measuring synthesis rates rather than muscle gained. It shows that a large dose is not wasted; it does not show that two large meals build more muscle than four moderate ones over months, which is a different experiment nobody has run. The 25 g comparator was chosen because it is the conventional dose, so the study cannot locate a ceiling between the two.

Trommelen et al., the anabolic response to protein ingestion during recovery from exercise has no upper limit in magnitude and duration in vivo in humans · Cell Rep Med 2023;4(12):101324

Leucine alone raised muscle protein synthesis in the exercised leg by more than whey did, and failed to reach significance at rest where whey succeeded.

Measured in: 22 healthy older women, 11 per group, mean age 69, with stable isotope infusion and muscle biopsies

The two papers cited here report on one registered trial, not two independent studies. The direction is not the simple one usually told: leucine did more after exercise and less without it.

Devries et al., leucine, not total protein, content of a supplement is the primary determinant of muscle protein anabolic responses in healthy older women · J Nutr 2018;148(7):1088-1095 Devries et al., protein leucine content is a determinant of shorter- and longer-term muscle protein synthetic responses at rest and following resistance exercise in healthy older women · Am J Clin Nutr 2018;107(2):217-226

Forty young men in a 40% energy deficit for four weeks, training with resistance and interval work six days a week, gained 2.6 lb (1.2 kg) of lean mass on 2.4 g of protein per kilogram per day against 0.2 lb (0.1 kg) on 1.2 g/kg, and lost 10.6 lb (4.8 kg) of fat against 7.7 lb (3.5 kg).

Measured in: 40 young men, 20 per group, in a supervised residential-style feeding and training protocol

Four weeks at a 40% energy deficit with six training days a week is a laboratory protocol, not a way of living, and 2.4 g/kg sits well above the plateau estimate everywhere else on this page. Young men only. Gaining lean mass while losing that much fat is a beginner-and-deficit phenomenon that does not continue.

Longland et al., higher compared with lower dietary protein during an energy deficit combined with intense exercise promotes greater lean mass gain and fat mass loss, a randomized trial · Am J Clin Nutr 2016;103(3):738-46

Nine randomized trials in 408 healthy women aged 18 to 73 found that multi-ingredient protein supplementation alongside exercise raised fat-free mass by 1.0 lb (0.45 kg) (95% CI 0.19 to 0.71), muscle size (Hedges' g 0.35, CI 0.05 to 0.65) and strength (g 0.50, CI 0.06 to 0.95). The review was undertaken because women make up only 0 to 22% of the samples in the existing meta-analyzes of protein supplementation with exercise.

Measured in: 408 healthy women aged 18 to 73 across nine randomized trials

Multi-ingredient supplements contain creatine, vitamin D and other actives alongside protein, so the protein contribution cannot be isolated from these trials. Nine studies is a small evidence base, the confidence intervals are wide, and the fat-free mass gain of 1.0 lb (0.45 kg) is in the same modest range as the male-dominated literature it was written to supplement.

Zhou et al., effects of multi-ingredient protein supplementation combined with exercise intervention on body composition and muscle fitness in healthy women, a systematic review with multilevel meta-analysis · Front Nutr 2025;12:1678433 Noh and Park, effects of resistance training and protein supplementation interventions on muscle volume and muscle function, sex differences in humans · Phys Act Nutr 2023;27(4):15-25

Fourteen resistance-trained men, mean age 26 with about nine years of training, spent roughly six months on their habitual diet at 2.51 g of protein per kilogram per day and six months at 3.32 g/kg. Blood lipids, liver function and kidney function measures did not change adversely, and fat mass did not rise despite the higher energy intake.

Measured in: 14 resistance-trained men, mean age 26, crossover over one year

Fourteen men who were already eating 2.5 g/kg at baseline, so this compares high with higher rather than normal with high. Diet was self-reported, kidney function rested on routine blood chemistry rather than measured GFR, and one year in young athletes says nothing about decades or about anyone with reduced kidney function to begin with.

Antonio et al., a high protein diet has no harmful effects, a one-year crossover study in resistance-trained males · J Nutr Metab 2016;2016:9104792

Across 99 randomized trials and 6,582 adults, time-restricted eating produced no greater weight loss than continuous calorie restriction. Alternate-day fasting was the one variant with an edge, at 2.8 lb (1.29 kg) (95% CI 0.59 to 1.99), and even that was gone in trials running 24 weeks or longer. HbA1c and HDL cholesterol showed no differences across any comparison.

Measured in: 6,582 adults across 99 randomized trials, 4,336 of them women (66%)

A network meta-analysis pools protocols that differ in window length, window placement and trial duration, so an advantage confined to one variant could be diluted. The authors graded certainty as moderate for the body-weight comparisons and low to moderate for the lipid ones.

Semnani-Azad et al., intermittent fasting strategies and body weight and cardiometabolic risk factors, network meta-analysis · BMJ 2025;389:e082007

Under two 8-day laboratory protocols separating circadian phase from behavior, identical meals produced 17% higher postprandial glucose in the biological evening than in the biological morning, with early-phase insulin 27% lower. Circadian misalignment on top of that added a further 6%.

Measured in: Healthy adults in a randomized crossover with a 12-hour behavioural inversion, so the same meals were eaten at opposite circadian phases

A behavioural inversion rather than a forced-desynchrony protocol; the authors contrast this design with their own earlier forced-desynchrony work. It isolates circadian phase from meal content, and it does not isolate it from the disruption of inverting a day.

Morris et al., endogenous circadian system and circadian misalignment impact glucose tolerance via separate mechanisms in humans · Proc Natl Acad Sci U S A 2015;112(17):E2225-34

Across 20 randomized trials and 1,242 adults, an eating window with no calorie target took off 3.5 lb (1.59 kg) against no restriction (95% CI 1.15 to 2.02) and 2.1 lb (0.93 kg) of fat mass (95% CI 0.63 to 1.22). Added on top of calorie restriction it was worth a further 2.1 lb (0.94 kg) (95% CI 0.31 to 1.58).

Measured in: 1,242 adults across 20 randomized trials, 15 of which enrolled both sexes and 4 of which enrolled only women

The review's stated conclusion is that combining a window with calorie restriction adds nothing, which sits awkwardly beside its own subgroup result of 2.1 lb (0.94 kg) at p = 0.004. Most included trials ran under 12 weeks, and what happens to the weight afterwards is not measured here.

Sun et al., efficiency of time-restricted eating and energy restriction on anthropometrics and body composition, systematic review and meta-analysis · Int J Behav Nutr Phys Act 2025;22(1):121 Jin et al., counting hours or calories? metabolic regulatory role of time-restricted eating, systematic review and meta-analysis · Crit Rev Food Sci Nutr 2025;65(21):4065-4079

Adults given a 4-hour or a 6-hour window and told nothing at all about calories ate 528 and 566 kcal a day less than at baseline, roughly 30% less, and lost about 3.2% of body weight in 8 weeks. In a separate trial that cut eating windows by four and a half hours in metabolic syndrome, intake fell about 350 kcal a day, from carbohydrate and fat rather than protein.

Measured in: 49 adults with obesity completing an 8-week three-arm trial, 44 of them women, plus 108 adults with metabolic syndrome in a 3-month trial

Intake was estimated from self-reported records, which under-report. The size of the drop is the point: it accounts for the weight change on its own, with no timing mechanism required.

Cienfuegos et al., effects of 4- and 6-h time-restricted feeding on weight and cardiometabolic health, randomized controlled trial · Cell Metab 2020;32(3):366-378.e3 Manoogian et al., time-restricted eating in adults with metabolic syndrome, randomized controlled trial · Ann Intern Med 2024;177(11):1462-1470

Both arms were prescribed the same 1,200 to 1,800 kcal a day for 12 months, and one also ate only between 8am and 4pm. Weight fell 17.6 lb (8.0 kg) with the window and 13.9 lb (6.3 kg) without it, a difference of 4.0 lb (1.8 kg) that did not reach significance (95% CI -4.0 to 0.4, P = 0.11). Waist, BMI, body fat, lean mass, blood pressure and metabolic risk factors all followed the primary outcome.

Measured in: 139 adults with obesity in Guangzhou, China, randomized for 12 months, 118 of whom completed

The confidence interval still permits an 8.8 lb (4 kg) advantage, so this is a trial that failed to find a difference rather than one that ruled a useful difference out. It remains the longest calorie-matched comparison in existence.

Liu et al., calorie restriction with or without time-restricted eating in weight loss · N Engl J Med 2022;386(16):1495-1504

With weight-loss counselling matched in both arms, eating between 7am and 3pm produced 5.1 lb (2.3 kg) more weight loss over 14 weeks than a window of 12 hours or more (95% CI 0.9 to 3.7), and dropped diastolic blood pressure 4 mmHg further. Body fat did not differ (-3.1 lb (-1.4 kg), 95% CI -2.9 to 0.2). Fatigue, vigour and depression scores improved.

Measured in: 90 adults with obesity at a weight-loss clinic in Alabama, 72 of them women (80%), mean age 43

Weight moved further than fat did, which points at part of the extra loss being lean tissue or fluid. Both arms were counselled to restrict energy but intake was not enforced, so the early group may simply have eaten less; the authors put the effect at about 214 kcal a day.

Jamshed et al., effectiveness of early time-restricted eating for weight loss, fat loss and cardiometabolic health in adults with obesity · JAMA Intern Med 2022;182(9):953-962

In the late window, the sleep midpoint occurred 15 minutes later (P<0.001). The 24-minute shift in the monocyte circadian marker did not reach significance (95% CI -5 to 54, P=0.10).

Measured in: 31 women with overweight or obesity in a randomized crossover, adherence 96 to 98%

The monocyte circadian shift is a non-significant trend, not a measured effect, so it should not be read as one.

Peters et al., intended isocaloric time-restricted eating shifts circadian clocks but does not improve cardiometabolic health in women with overweight · Sci Transl Med 2025;17(822):eadv6787

Fat-free mass fell about 1.3 lb (0.57 kg) on average across 20 randomized trials, whether the window was used alone (-1.3 lb (-0.58 kg), 95% CI -0.81 to -0.36) or added to calorie restriction (-1.2 lb (-0.56 kg), 95% CI -0.81 to -0.31). In the 12-week trial with body composition measured, 2.4 lb (1.10 kg) of the 3.7 lb (1.70 kg) lost was lean mass, about 65%, against the 20 to 30% the authors give as normal during weight loss.

Measured in: 1,242 adults across 20 randomized trials, plus a 50-person body-composition subgroup of a 116-person trial (70 men, 46 women)

Fat-free mass includes water and glycogen, both of which fall early in any energy deficit, so short trials overstate muscle loss. The 65% figure comes from 25 people per arm with no resistance training prescribed and no protein target set, which is the condition that produces it. That trial also carries two published errata and three critique letters.

Sun et al., efficiency of time-restricted eating and energy restriction on anthropometrics and body composition, systematic review and meta-analysis · Int J Behav Nutr Phys Act 2025;22(1):121 Lowe et al., effects of time-restricted eating on weight loss and other metabolic parameters, the TREAT randomized clinical trial · JAMA Intern Med 2020;180(11):1491-1499

Pooling 8 trials of intermittent fasting alongside resistance training, fat-free mass did not change (-0.6 lb (-0.27 kg), 95% CI -0.82 to 0.28) while fat mass fell 3.0 lb (1.36 kg) and body fat percentage fell 1.49 points. In 34 resistance-trained men with calories matched and protein at 22% of energy, lean mass, limb muscle area and maximal strength held. In 40 resistance-trained women on matched calories and 1.6 g/kg of protein, fat-free mass rose 2 to 3% on a 7.5-hour window, the same as on a 13-hour one.

Measured in: 221 adults across 8 pooled trials, plus a 34-man trial and a 40-woman trial, all resistance-trained

These are trained people lifting under supervision with protein deliberately held high, which is the condition the finding depends on rather than an incidental detail. In the men's trial, total testosterone and IGF-1 fell significantly over the 8 weeks.

Ashtary-Larky et al., effects of intermittent fasting combined with resistance training on body composition, systematic review and meta-analysis · Physiol Behav 2021;237:113453 Moro et al., eight weeks of time-restricted feeding (16/8) in resistance-trained males · J Transl Med 2016;14(1):290 Tinsley et al., time-restricted feeding plus resistance training in active females, randomized trial · Am J Clin Nutr 2019;110(3):628-640

In adults with type 2 diabetes on metformin or other glucose-lowering drugs, adding two very-low-calorie fasting days a week doubled the rate of hypoglycemia (RR 2.05, 95% CI 1.17 to 3.52) even though medication was reduced in advance and participants were educated about it. The absolute rate was 1.4 events over 12 weeks.

Measured in: 41 adults with type 2 diabetes, BMI 30 to 45, HbA1c 6.7 to 10%, randomized for 12 weeks

This tested two fasting days a week rather than a daily eating window, and it does not separate insulin from sulfonylureas. The rate stayed low because medication was proactively reduced under supervision, which is the condition most people doing this at home will not have.

Corley et al., intermittent fasting in type 2 diabetes mellitus and the risk of hypoglycaemia, randomized controlled trial · Diabet Med 2018;35(5):588-594

Across 12,243 people with diabetes in 13 countries, severe hypoglycemic episodes ran at 0.14 a month in type 1 diabetes during Ramadan against 0.03 outside it, close to a five-fold increase, and 0.03 against 0.004 in type 2 diabetes, roughly seven-fold. Episodes were more frequent in those who changed their oral drug or insulin dose or their activity level, and fewer than half changed their treatment at all.

Measured in: 12,243 people with diabetes across 13 countries, 1,070 of them with type 1

Ramadan fasting includes fluid restriction and a large night-time meal, so it is not the same exposure as a daily eating window with water allowed. It remains the largest human dataset on what happens when people on insulin and sulfonylureas stop eating for long stretches.

What could explain it instead: Ramadan changes sleep, physical activity, and the timing and composition of meals all at once, so the fast is not the only exposure that differs. Episode counts were collected retrospectively, which favors recall of the more recent fasting month.

Salti et al., a population-based study of diabetes and its characteristics during the fasting month of Ramadan in 13 countries (EPIDIAR) · Diabetes Care 2004;27(10):2306-2311 Hassanein et al., diabetes and Ramadan: practical guidelines (IDF and DAR International Alliance) · Diabetes Res Clin Pract 2017;126:303-316

Narrowing the daily eating window to eight to ten hours for three months improved HbA1c by 0.10 percentage points over standard nutritional counseling alone (95% CI -0.19 to -0.003).

Measured in: 108 adults with metabolic syndrome already on medication, 52% women, mean age 59

Calorie intake was not prescribed or matched between the groups, so this trial cannot separate the effect of when people ate from the effect of eating less. Compressing an eating window plausibly does both. The authors call the glycemic effect modest and treat broader cardiometabolic benefit as possible rather than shown, and the upper bound of the confidence interval nearly touches zero.

Manoogian et al., time-restricted eating in adults with metabolic syndrome, randomized controlled trial · Ann Intern Med 2024;177(11):1462-1470

Under supervised feeding that held weight constant, a 6-hour window with dinner before 3pm improved insulin sensitivity and beta cell responsiveness, cut mean insulin by 26 mU/l, lowered systolic blood pressure 11 mmHg and diastolic 10 mmHg, and reduced the oxidative stress marker 8-isoprostane by 11 pg/ml.

Measured in: 8 men with prediabetes, mean age 56, in a randomized crossover with 5 weeks on each schedule

Eight men. This is the most-quoted evidence that the window does something independent of weight, and it is the smallest study on this page. Morning fasting triglycerides rose 57 mg/dl, which the authors attribute to the 18-hour fast before the blood draw rather than to the schedule.

Sutton et al., early time-restricted feeding improves insulin sensitivity, blood pressure and oxidative stress even without weight loss in men with prediabetes · Cell Metab 2018;27(6):1212-1221.e3

Over 5 weeks, eating between 6am and 3pm cut HOMA-IR by 1.08 while eating between 11am and 8pm raised it by 0.39, with the early group also losing 3.5 lb (1.6 kg) against 0.4 lb (0.2 kg) and 1.7 lb (0.76 kg) of fat mass against 0.7 lb (0.30 kg). Both windows cut intake by a similar amount, 240 and 159 kcal a day, so calories do not explain the difference between them.

Measured in: 82 healthy adults without obesity in China, 64 of them women, mean age around 31

Five weeks, one center, participants unblinded, and in people who were not overweight to begin with. HOMA-IR is a calculated index rather than a clamp measurement and it moves easily.

Xie et al., randomized controlled trial for time-restricted eating in healthy volunteers without obesity · Nat Commun 2022;13(1):1003

A 9-hour window starting at 8am and one starting at noon improved glucose tolerance equally, with no interaction between meal timing and the window in any variable measured. An 8-week trial in 37 adults on a matched 25% energy deficit found no difference between an 8am to 4pm window, a noon to 8pm window and ordinary calorie restriction on weight, fat mass or fasting glucose.

Measured in: 15 men at risk of type 2 diabetes in a 7-day randomized crossover, plus 37 adults aged 20 to 40 in an 8-week randomized trial

Both are small and short, and the second describes itself as exploratory. They test whether placement matters once calories are controlled, which is a narrower question than whether placement matters in ordinary life, where it also changes how much people eat.

Hutchison et al., time-restricted feeding improves glucose tolerance in men at risk for type 2 diabetes, randomized crossover trial · Obesity (Silver Spring) 2019;27(5):724-732 Queiroz et al., cardiometabolic effects of early versus delayed time-restricted eating plus energetic restriction, exploratory randomised clinical trial · Br J Nutr 2023;129(4):637-649

In muscle biopsies from 50 women on 8 weeks of intermittent fasting, autophagy markers were reduced rather than raised: p62, which falls when autophagy is running, rose after a 24-hour fast, and BECLIN1, SQSTM1 and LAMP2 messenger RNA fell. In 8 men after a 72-hour water fast, LC3B-II rose about 30% while p62 rose about 10% alongside it, which the authors state makes autophagic flux uninterpretable. mTOR phosphorylation fell about 50% in the same men.

Measured in: 50 women aged around 51 in a randomized trial of 24-hour fasts three days a week, and 8 healthy men in a 72-hour fasting crossover

Muscle is one tissue. The same paper found fasting did raise autophagy markers in mouse liver, so this is a statement about human skeletal muscle rather than about every human tissue.

Chaudhary et al., intermittent fasting activates markers of autophagy in mouse liver, but not muscle from mouse or humans · Nutrition 2022;101:111662 Vendelbo et al., fasting increases human skeletal muscle net phenylalanine release and this is associated with decreased mTOR signaling · PLoS One 2014;9(7):e102031

Over 6 months, an 8-hour window from noon to 8pm produced more weight loss than calorie counting and lowered HbA1c, with no difference between groups in time spent in hypoglycemic or hyperglycemic ranges, no difference in medication effect score, and no serious adverse events.

Measured in: 75 adults with type 2 diabetes in Chicago, 53 of them women (71%), mean age 55, mean HbA1c 8.1%

Medication was actively managed by the study team throughout, which is the condition that makes this result safe rather than a demonstration that the practice is safe unsupervised. The senior author reports author fees for a fasting book and NIH data safety monitoring board fees.

Pavlou et al., effect of time-restricted eating on weight loss in adults with type 2 diabetes, randomized clinical trial · JAMA Netw Open 2023;6(10):e2339337

In 2,762 Canadian adolescents and young adults, intermittent fasting in the past 12 months was reported by 47.7% of women, 38.4% of men and 52.0% of transgender and gender non-conforming participants, and was significantly associated with eating disorder psychopathology in all three groups, most consistently in women. In a separate sample of 64 people already fasting, 31% scored at or above the clinical cut-off on a standard eating disorder questionnaire.

Measured in: 2,762 Canadian adolescents and young adults, plus 64 self-selected intermittent fasters recruited online (44 women, 20 men)

What these establish is that the two travel together, which is enough to matter for anyone with a history. Effect sizes for the larger study are reported in the full text rather than the abstract, so the association is described here by direction and significance.

What could explain it instead: Reverse causation. Fasting is an accessible and socially approved way to restrict intake, so people already experiencing disordered eating adopt it. Neither study can put the two in time order, and the smaller one is a self-selected online sample.

Ganson et al., intermittent fasting: describing engagement and associations with eating disorder behaviors and psychopathology among Canadian adolescents and young adults · Eat Behav 2022;47:101681 Cuccolo et al., intermittent fasting implementation and association with eating disorder symptomatology · Eat Disord 2022;30(5):471-491

Across 16 studies of eating windows of 12 hours or less, disordered eating questionnaire scores did not change with time-restricted eating, with one study finding lower hunger. Bedtime hunger was higher under a window, and the qualitative work surfaced fear of hunger, eating in the absence of hunger, and eating-related stress.

Measured in: Adults enrolled in 16 time-restricted eating studies with windows of 12 hours or less

Trials screen out people with an eating disorder history at recruitment, so the group most at risk is the group these studies do not contain. A null result in a screened population does not transfer to an unscreened one, and this review does not pool its studies quantitatively.

Vizthum et al., the impact of time restricted eating on appetite and disordered eating in adults, mixed methods systematic review · Appetite 2023;183:106452

Four days of an 8am to 2pm window raised expression of the autophagy gene LC3A in whole blood cells by 22% before breakfast, and SIRT1 by 10%. In the same participants, expression of MTOR, which suppresses autophagy, rose 9% in the evening. Mean 24-hour glucose fell 4 mg/dl and glycemic excursions 12 mg/dl.

Measured in: 11 overweight adults, 7 men and 4 women, mean age 32, in a 4-day randomized crossover

This is messenger RNA for one gene in circulating blood cells at a single time point. Autophagy is a flux, a rate of production and clearance, and the consensus methods guideline for the field states that a static marker reading cannot establish it. The authors' own wording is that the schedule may increase autophagy.

Jamshed et al., early time-restricted feeding improves 24-hour glucose levels and affects markers of the circadian clock, aging and autophagy in humans · Nutrients 2019;11(6):1234 Klionsky et al., guidelines for the use and interpretation of assays for monitoring autophagy, 4th edition · Autophagy 2021;17(1):1-382

In the one human trial to measure autophagic flux properly, using a chloroquine block on peripheral blood mononuclear cells, a 5-day fasting-mimicking diet produced a significant between-group difference against control at the end of the 6-day intervention, alongside changes in weight, fasting glucose, ketones and HOMA-IR. The difference was not significant across all time points.

Measured in: 30 healthy adults, mean age 49, randomized to two fasting-mimicking formulations or control for 8 days

This is a 5-day fasting-mimicking diet rather than a daily eating window, so it does not tell you what a 16-hour overnight fast does. It is a pilot of 30 people, and it was sponsored by the company that sells the product tested.

Espinoza et al., effect of fasting-mimicking diet on markers of autophagy and metabolic health in human subjects · GeroScience 2025

Mice eating the same number of high-fat calories inside an 8-hour window instead of around the clock were protected from obesity, hyperinsulinemia, fatty liver and inflammation, with improved CREB, mTOR and AMPK pathway function and restored circadian clock oscillation. A follow-up found the benefit scaled with the length of the fast, survived two ad-libitum days a week, and partly reversed established obesity and diabetes.

Measured in: Male mice on high-fat and other challenge diets

Mice are nocturnal, eat in bouts, and run a mass-specific metabolic rate many times ours, so an 8-hour window in a mouse is a far larger intervention than an 8-hour window in a person. Neither paper measured autophagic flux. The calorie-matched human trials are where the size of this gap shows.

Hatori et al., time-restricted feeding without reducing caloric intake prevents metabolic diseases in mice fed a high-fat diet · Cell Metab 2012;15(6):848-860 Chaix et al., time-restricted feeding is a preventative and therapeutic intervention against diverse nutritional challenges · Cell Metab 2014;20(6):991-1005

In 19,831 US adults, an average eating duration under 8 hours carried 2.35 times the cardiovascular mortality of a 12 to 14 hour duration (95% CI 1.39 to 3.98) over a median 8.1 years. All-cause and cancer mortality showed no robust association. The finding held across 8 subgroups and 14 sensitivity analyzes.

Measured in: 19,831 adults in NHANES 2003 to 2018, linked to the National Death Index through December 2019

This reached the public in March 2024 as a conference abstract reporting a 91% increase, presented at the American Heart Association EPI/Lifestyle Sessions and not peer reviewed at the time. It was published 16 months later with a narrower reference category and a larger hazard ratio. Cardiovascular deaths rose while all-cause deaths did not, which is hard to reconcile, and the authors' own conclusion asks whether the association is the eating duration or residual confounding.

What could explain it instead: Reverse causation and confounding by circumstance. In a general population survey, a habitually short eating window marks illness, appetite loss, shift work, food insecurity and smoking at least as often as it marks a deliberate practice. Eating duration was taken from two 24-hour dietary recalls at enrolment and then treated as the person's pattern for the next eight years, with no repeated measurement.

Chen et al., association of eating duration less than 8 h with all-cause, cardiovascular and cancer mortality · Diabetes Metab Syndr 2025;19(7):103278 American Heart Association newsroom, 8-hour time-restricted eating linked to a 91% higher risk of cardiovascular death (conference abstract P192, EPI/Lifestyle Scientific Sessions) · AHA EPI/Lifestyle Scientific Sessions 2024

After 8 weeks of a 4 to 6 hour eating window, testosterone, androstenedione and sex hormone binding globulin were unchanged in both premenopausal and postmenopausal women, and estradiol, estrone and progesterone were unchanged in the postmenopausal group. DHEA fell about 14% in premenopausal and 13% in postmenopausal women, remaining inside the normal range. Weight fell 3 to 4%.

Measured in: 23 women with obesity, 12 premenopausal and 11 postmenopausal, as a secondary analysis of an 8-week trial

Twenty-three women over 8 weeks, analyzed after the fact. Menstrual cycle regularity was not an outcome, so the specific worry about cycle disruption was not tested. Estradiol and progesterone were not measured in the premenopausal group because they move across the cycle. A review of reproductive hormones across human fasting trials closes by saying too few studies exist to draw conclusions. Perimenopausal women were excluded from both women-only analyzes, so the group with the most volatile hormonal picture has no data at all.

Kalam et al., effect of time-restricted eating on sex hormone levels in premenopausal and postmenopausal females · Obesity (Silver Spring) 2023;31(Suppl 1):57-62 Cienfuegos et al., effect of intermittent fasting on reproductive hormone levels in females and males, review of human trials · Nutrients 2022;14(11):2343

In female rats, severe (40%) calorie restriction caused emaciation, stopped estrous cycling altogether, and produced endocrine masculinization, while males on the same restriction held more of their body weight and changed behavior less. The study also included an intermittent-fasting arm.

Measured in: Male and female rats over 6 months on usual, 20% restricted, 40% restricted, intermittent fasting or high-fat high-glucose diets

Rats, and the severe reproductive shutdown came from 40% calorie restriction rather than from the fasting arm, where the finding was irregular cycling. A rat's energy turnover relative to body size makes a 24-hour fast a far larger stress than it is for a person. No human trial has tested whether an eating window disrupts menstrual cycles.

Martin et al., sex-dependent metabolic, neuroendocrine and cognitive responses to dietary energy restriction and excess · Endocrinology 2007;148(9):4318-4333

On the same 4 to 6 hour window for 8 weeks, weight fell 3.3% in premenopausal women and 3.3% in postmenopausal women, with no interaction between group and time. Fat mass, lean mass, fasting insulin, insulin resistance and 8-isoprostane fell similarly in both. Visceral fat, blood pressure, lipids, glucose, HbA1c, TNF-alpha and IL-6 did not change in either.

Measured in: 32 women with obesity, 13 premenopausal and 19 postmenopausal, adherence 6.2 days a week

Thirty-two women over 8 weeks, as a secondary analysis of a trial not designed to answer this question. Perimenopausal women were excluded, which is the group with the most volatile hormonal picture and the most reason to want the answer.

Cienfuegos et al., changes in body weight and metabolic risk during time restricted feeding in premenopausal versus postmenopausal women · Exp Gerontol 2021;154:111545

Pooling nine cohort studies covering more than eight million people in seven countries, each 0.1 increase in vegetation greenness within 547 yards (500 meters) of home was associated with 4% lower all-cause mortality (HR 0.96, 95% CI 0.94 to 0.97).

Measured in: 8,324,652 people across nine cohort studies in seven countries

Heterogeneity between the studies was severe, so treat this as a rough central tendency rather than a stable dose-response, and two of the nine found no association at all. The unit is a satellite vegetation index, which is not the same thing as living near a park.

What could explain it instead: Residential confounding. Greener neighborhoods differ from less green ones in nearly every way that also affects health.

Rojas-Rueda et al., green spaces and mortality, systematic review and meta-analysis of cohort studies · Lancet Planet Health 2019;3(11):e469-e477

Across the 16 randomized trials inside a review of 50 studies, structured outdoor nature activities improved depressed mood (SMD -0.64, 95% CI -1.05 to -0.23) and positive affect (SMD 0.95, 95% CI 0.59 to 1.31), with a large but statistically fragile effect on anxiety (SMD -0.94, upper CI near zero) and a smaller effect on negative affect. Programs running 8 to 12 weeks with sessions of 20 to 90 minutes did best. Physical health outcomes moved much less.

Measured in: Community adults with and without existing mental or physical health problems, across 50 studies

Every intervention bundles time outdoors with exercise, a group, a leader and a routine, and none of the trials separates the setting from the rest of the package. Risk of bias was low to moderate in the randomized subset and moderate to high in the 34 non-randomized studies.

Coventry et al., nature-based outdoor activities for mental and physical health, systematic review and meta-analysis · SSM Popul Health 2021;16:100934

Pooling 49 outcome measures from 42 studies across eight cognitive domains, exposure to natural environments improved working memory and cognitive flexibility at low to moderate effect sizes, and attentional control less reliably. The other five domains tested did not improve. Real places produced larger effects than photographs or video.

Measured in: 42 studies published from July 2013 onward, largely undergraduates in laboratory and short field settings

Most exposures last minutes and the outcomes are laboratory tasks, so this does not show a change in how someone works the next day. The real-place exposures were also longer than the screen-based ones, which the authors say they cannot separate from the setting.

Stevenson, Schilhab and Bentsen, Attention Restoration Theory II, systematic review and meta-analysis · J Toxicol Environ Health B Crit Rev 2018;21(4):227-268

943,027 Danes born 1985 to 2003, with satellite greenness measured in a 230 by 230 yard (210 by 210 meter) square around every childhood address up to age 10. Those in the lowest green space band carried up to 55% higher risk of later psychiatric disorders than those in the highest, strongest for depression and for stress-related disorders. Intellectual disability and schizoaffective disorder showed no clear association.

Measured in: 943,027 people in Denmark, followed from their tenth birthday

Cumulative green space across childhood predicted more strongly than green space in any single year, which fits an accumulating exposure and also fits a stable family staying in one neighborhood. Two of the disorders examined showed no association, which is worth reading alongside the ones that did.

What could explain it instead: Where a family lives is not assigned. The analysis adjusted for parental income, education, employment, psychiatric history and urbanization, and the authors still name green space quality, crime rates and neighborhood disadvantage as unmeasured.

Engemann et al., residential green space in childhood is associated with lower risk of psychiatric disorders from adolescence into adulthood · Proc Natl Acad Sci U S A 2019;116(11):5188-5193

Five randomized trials, four of them randomized by school, covering 10,733 children aged six to nine. Adding outdoor time cut two-year myopia incidence from 26.7% to 22.5% (RR 0.84, 95% CI 0.72 to 0.98) and slowed the refractive shift by 0.13 diopters over two years. The single trial that ran three years reported 30.5% against 39.8%.

Measured in: 10,733 primary schoolchildren, mostly first and second grade, in trials run in East Asia

Certainty was moderate at two years and rests on a single trial at three years, and neither children nor teachers can be blinded to extra time outside. The exposure is daylight and distance viewing rather than woodland, so this belongs to being outdoors more than to nature.

Kido, Miyake and Watanabe, interventions to increase time spent outdoors for preventing incidence and progression of myopia in children (Cochrane review) · Cochrane Database Syst Rev 2024;6(6):CD013549

Compared with no time in nature, 120 minutes a week or more was associated with better self-reported health (OR 1.59) and wellbeing (OR 1.23). Benefit plateaued between 200 and 300 minutes. It made no difference whether the time came in one long visit or several short ones.

Measured in: 19,806 adults in England, from a nationally representative survey

A single survey snapshot with self-reported outcomes, so healthier people getting outside more explains part of it. The association held among older adults and people with long-term conditions, and it survived adjustment for local green space, neighborhood, physical activity and demographics.

What could explain it instead: Cross-sectional, so healthier people spending more time outdoors explains part of it.

White et al., spending at least 120 minutes a week in nature is associated with good health and wellbeing · Sci Rep 2019;9(1):7730

Of 25 included studies, most compared the same activity, usually a walk or a run, in a natural setting against a built or indoor one. Pooled self-reported emotion favored the natural setting. The attention advantage did not hold after adjusting for differences between the groups before the walk, and blood pressure and cortisol showed no consistent difference between settings.

Measured in: Twenty-five crossover and controlled studies, mostly single short exposures in students and other young adults

Nobody can be blinded to which environment they are walking in, and the outcome that moved most is the self-reported one, which is the most open to expectation. The review dates from 2010 and the field has grown a great deal since, though the design problem it identified is still there.

Bowler et al., systematic review of evidence for the added benefits to health of exposure to natural environments · BMC Public Health 2010;10:456

Ninety-two studies identified, 28 pooled. Referral programs lowered systolic blood pressure by 4.8 mmHg (95% CI -8.9 to -0.7) and diastolic by 3.8 mmHg (95% CI -6.5 to -1.2), improved depression scores (SMD -0.50, 95% CI -0.84 to -0.16) and anxiety scores (SMD -0.57, 95% CI -1.12 to -0.03), and added about 900 steps a day (95% CI 790 to 1,010).

Measured in: Adults referred by a health or social professional to a structured outdoor program, across 92 studies in mostly high-income countries

Most of the included studies carry a moderate to high risk of bias, the programs range from a walking group to a therapeutic garden, and someone who accepts a referral to spend time outdoors differs from someone who declines one.

Nguyen et al., effect of nature prescriptions on cardiometabolic and mental health, and physical activity, systematic review and meta-analysis · Lancet Planet Health 2023;7(4):e313-e328

In the same review, weekly time spent in moderate physical activity did not improve (mean difference 25.9 minutes, 95% CI -10.3 to 62.1), while daily step counts in the same programs rose by about 900.

Measured in: Adults in referral-based outdoor programs, from the subset of the 92 studies reporting physical activity

The two measures disagree because steps usually come from a device and weekly moderate activity is usually recalled on a questionnaire. Either the added walking sat below moderate intensity, or the questionnaires did not capture it.

Nguyen et al., effect of nature prescriptions on cardiometabolic and mental health, and physical activity, systematic review and meta-analysis · Lancet Planet Health 2023;7(4):e313-e328

Thirty-five quantitative studies of lakes, rivers, canals and coast. The balance of evidence favored better mental health and wellbeing (12 studies) and higher physical activity (13 studies). General health, obesity and cardiovascular outcomes were less consistent across the remaining studies.

Measured in: 35 observational and experimental studies, mostly in European populations

No pooling was possible because exposure was measured differently in almost every study, from straight-line distance to the coast to self-reported visits. The reviewers call for longitudinal work and natural experiments before causation is read into it.

What could explain it instead: Coastal and waterside housing tracks income in most of the countries studied, and people who can walk to water can usually also afford other things that protect health.

Gascon et al., outdoor blue spaces, human health and well-being, a systematic review of quantitative studies · Int J Hyg Environ Health 2017;220(8):1207-1221

Twenty-two studies giving 76 comparisons found a positive pooled effect of gardening across a wide spread of outcomes, including lower depression, anxiety and body mass index, and higher life satisfaction and sense of community.

Measured in: 22 studies published after 2001, most from the United States, then Europe, Asia and the Middle East

Egger's test detected publication bias, and the effect held at a smaller size after a trim-and-fill correction. Many comparisons are before-and-after in the same people with no control group, and gardening supplies physical work, daylight and often company alongside the plants.

Soga, Gaston and Yamaura, gardening is beneficial for health, a meta-analysis · Prev Med Rep 2016;5:92-99

Small studies report changes in inflammation markers, heart-rate variability and blood measures after direct skin contact with the ground or a grounded surface.

Measured in: Individual studies of eight to fifty-eight people, mostly pilots

The samples are very small, and a large share of this literature comes from the same few authors. Several of the authors have financial ties to a company that sells earthing products and funded the research. Walking barefoot outside is pleasant and harmless. The specific electrical mechanism has not been tested at any useful scale.

Chevalier et al., earthing, health implications of reconnecting the human body to the Earth's surface electrons · J Environ Public Health 2012;2012:291541

38 healthy city dwellers, 18 of them women, mean age 27, randomly assigned to a 90-minute 3.3 mile (5.3 km) walk through oak grassland or along a four-lane road. Self-reported rumination fell in the grassland group and did not change in the road group, and blood perfusion in the subgenual prefrontal cortex fell alongside it. Heart rate and respiration rate did not differ between the groups, so the effect was not a difference in exertion.

Measured in: 38 healthy adults living and working in urban parts of the San Francisco Bay Area, with no psychiatric or neurological history

Nineteen people per group, one walk each, no follow-up, so this says nothing about a lasting change. Rumination was self-reported by people who knew which walk they had taken, and the brain measure was available for fewer participants than the questionnaire.

Bratman et al., nature experience reduces rumination and subgenual prefrontal cortex activation · Proc Natl Acad Sci U S A 2015;112(28):8567-8572

Twenty studies covering 2,257 participants, most of them in Asia and Europe. Shinrin-yoku reduced symptoms in the short term, with the clearest signal on anxiety and a weaker one on depression. No included study followed participants beyond a few weeks.

Measured in: 2,257 participants across 20 studies, mostly single-visit designs in healthy adults

The authors found indications of publication bias in nearly every analysis they ran, and the included studies are small, short, and mostly without an active comparison condition, so the pooled figures likely sit above the real effect.

Kotera, Richardson and Sheffield, effects of shinrin-yoku (forest bathing) and nature therapy on mental health, systematic review and meta-analysis · Int J Ment Health Addict 2022;20:337-361, published online 2020

Eight studies pooled out of 22 reviewed. Salivary cortisol was lower in the forest groups than the urban groups after the visit (MD -0.05 ug/dl, 95% CI -0.06 to -0.04). It was also lower before the visit began (MD -0.08 ug/dl, 95% CI -0.11 to -0.05).

Measured in: Eight small single-visit studies, participants mostly young adults in Japan and Korea

The groups already differed before anyone entered a forest, which limits how much of the after-visit difference can be attributed to the forest, and the review's own authors attribute part of the effect to anticipation. Heterogeneity after the intervention was 88%.

Antonelli, Barbieri and Donelli, effects of forest bathing (shinrin-yoku) on levels of cortisol as a stress biomarker · Int J Biometeorol 2019;63(8):1117-1134

Twelve men took a three-day forest trip and, separately, a three-day city trip with activity levels matched. Natural killer activity rose roughly 50% after the forest trip and was still raised more than seven days later. It did not rise after the city trip. Urinary adrenaline fell after the forest trip only. Alpha-pinene and beta-pinene were measurable in forest air and close to absent in city air.

Measured in: Twelve healthy Japanese men aged 35 to 56, recruited from large Tokyo companies

Twelve people, no randomization, no blinding, and the trips ran in a fixed order. Almost all of the natural killer work on forest visits comes from one group at Nippon Medical School. A three-day trip away from a Tokyo office differs from an ordinary working day in many ways besides the trees, and no study has followed whether the cell-count change alters anything a person experiences.

Li et al., visiting a forest, but not a city, increases human natural killer activity and expression of anti-cancer proteins · Int J Immunopathol Pharmacol 2008;21(1):117-127 Li et al., a forest bathing trip increases human natural killer activity and expression of anti-cancer proteins in female subjects · J Biol Regul Homeost Agents 2008;22(1):45-55

Records of 46 surgical patients at one suburban Pennsylvania hospital between 1972 and 1981, matched into 23 pairs. Patients whose window faced trees left hospital sooner, took fewer strong painkillers, and drew fewer negative comments in nurses' notes than their matched pair whose window faced a brick wall.

Measured in: 46 cholecystectomy patients in a single hospital, matched into 23 pairs

Twenty-three pairs, one hospital, one surgical procedure, and a review of old records rather than an experiment. It has not been repeated at scale in the four decades since, and it is quoted far more often than anything else in this field.

What could explain it instead: Room assignment was not random. Which room a patient received depended on what was free when they arrived, and rooms that face grounds differ from rooms that face a wall in ward position, noise and light as well as view.

Ulrich, view through a window may influence recovery from surgery · Science 1984;224(4647):420-421

This is the best-evidenced use of photobiomodulation. A systematic review that fed the MASCC/ISOO clinical practice guidelines concluded that photobiomodulation reduces the severity and pain of oral mucositis, the painful mouth and throat ulceration caused by chemotherapy and radiotherapy, and it is recommended to prevent mucositis in defined settings, including adults having a hematopoietic stem-cell transplant with high-dose conditioning and patients receiving radiotherapy for head and neck cancer. A later randomized, double-blind, sham-controlled trial of an intraoral device in head and neck radiotherapy adds direct trial support.

Measured in: Adults having hematopoietic stem-cell transplantation or head and neck radiotherapy, across randomized trials pooled in the guideline review

The recommendation is strongest for the specific cancer-treatment settings studied; wavelength, dose and schedule differ between trials, so the protocol matters and results outside those settings are less certain.

Zadik et al., systematic review of photobiomodulation for the management of oral mucositis in cancer patients and clinical practice guidelines · Supportive Care in Cancer 2019;27(10):3969-3983 Randomized, double-blind, sham-controlled trial of an intraoral photobiomodulation device for oral mucositis due to radiotherapy for head and neck cancer · Supportive Care in Cancer 2026

A meta-analysis of randomized controlled trials found that low-level laser therapy increased hair density compared with sham devices in adults with androgenetic alopecia. The signal is consistent across several trials, which is why this is the strongest of the cosmetic uses, but the individual studies are small and short and many were funded or run by device makers, so the size of the benefit and how long it lasts are not settled.

Measured in: Adults with androgenetic alopecia, men and women, across randomized sham-controlled trials

The trials are small and short and many involved device manufacturers, so the true effect size and durability are uncertain, and it has not been shown to match the established treatments over the long term.

Liu et al., comparative effectiveness of low-level laser therapy for adult androgenic alopecia: a systematic review and meta-analysis of randomized controlled trials · Lasers in Medical Science 2019;34(6):1063-1069

In a controlled trial, people who completed a course of red and near-infrared light sessions showed improved skin complexion and feeling, reduced fine lines, wrinkles and roughness, and a measurable rise in intradermal collagen density on ultrasound. The active groups improved significantly compared with an untreated control group. The trial lacked a blinded sham-light placebo and the controls were unmasked, so some of the improvement could reflect expectation, and the study was modest in size.

Measured in: Adults treated for facial photoaging in a single controlled trial

The trial included an untreated control group but no blinded sham-light placebo, and the controls were unmasked, so part of the benefit may reflect expectation; with a modest sample this is early rather than confirmed.

Wunsch & Matuschka, a controlled trial to determine the efficacy of red and near-infrared light treatment in patient satisfaction, reduction of fine lines, wrinkles, skin roughness, and intradermal collagen density increase · Photomedicine and Laser Surgery 2014;32(2):93-100

A systematic review of six randomized trials of LED photobiomodulation for chronic lower-limb wounds concluded it may improve healing, particularly in diabetic foot ulcers, though the overall certainty of the evidence was rated very low and the trials could not be pooled. The benefit appeared at lower light doses, while the single trial using a very high dose (126 J/cm2) showed no benefit, which fits a therapeutic window rather than more light being better. The trials are small and their dosing varies widely, so this is a promising signal in a hard-to-treat problem rather than a settled standard of care.

Measured in: Patients with chronic lower-limb wounds across randomized trials

The evidence is very low certainty, the trials are small and could not be pooled, and dosing varies widely, so the benefit is an emerging signal and the best protocol is not established; it is used alongside standard wound care, not instead of it.

Clinical dosimetry and efficacy of LED photobiomodulation for chronic lower-limb wound healing: a systematic review of randomized trials · Lasers in Medical Science 2026

A meta-analysis of randomized, placebo-controlled trials found that low-level laser therapy reduced knee osteoarthritis pain compared with placebo, by roughly 15 mm on a 100 mm pain scale at the end of therapy, with the benefit concentrated in trials that used an adequate light dose. Trials using too low a dose showed little effect, which is a clear example of how much the dose matters.

Measured in: Adults with knee osteoarthritis across randomized placebo-controlled trials

Benefit depended heavily on using an adequate dose, and protocols varied widely between trials, so results with an under-dosed device or an unclear setting are unreliable.

Stausholm et al., efficacy of low-level laser therapy on pain and disability in knee osteoarthritis: systematic review and meta-analysis of randomised placebo-controlled trials · BMJ Open 2019;9(10):e031142

A meta-analysis of randomized placebo and active-controlled trials found that low-level laser therapy reduced neck pain, with relief that continued for up to 22 weeks after the treatment course ended in chronic neck pain. The trials were mostly short and the protocols varied, so the finding supports it as a reasonable option for neck pain rather than a first-line standard.

Measured in: Adults with acute and chronic neck pain across randomized controlled trials

Most trials were short and used differing wavelengths and doses, so the best protocol and the benefit beyond a few months are not settled.

Chow et al., efficacy of low-level laser therapy in the management of neck pain: a systematic review and meta-analysis of randomised placebo or active-treatment controlled trials · Lancet 2009;374(9705):1897-1908

A systematic review with meta-analysis of low-level laser therapy for tendinopathy found it can reduce pain and improve function when an appropriate wavelength and dose are used, but results across trials were mixed and heavily dependent on getting the parameters right. Studies using doses outside the effective range showed little benefit, which accounts for much of the inconsistency.

Measured in: Adults with various tendinopathies across randomized trials

Benefit hinged on using the right wavelength and dose, results across trials were mixed, and samples were small, so it is best seen as an adjunct to loading exercise rather than a standalone fix.

Tumilty et al., low level laser treatment of tendinopathy: a systematic review with meta-analysis · Photomedicine and Laser Surgery 2010;28(1):3-16

A systematic review with meta-analysis found that phototherapy (low-level laser and LED light) applied before exercise improved measures of muscle performance, such as repetitions and time to exhaustion, and reduced markers of muscle fatigue and damage afterward. Effects were small to moderate, timing and dose mattered, and much of the work comes from one research group, so the finding needs wider replication.

Measured in: Healthy exercising adults across randomized trials, largely young and male in the original studies

Effects were small to moderate, depended on applying the light before exercise, and much of the trial base comes from a single research group, so wider independent replication is still needed.

Leal-Junior et al., effect of phototherapy (low-level laser therapy and light-emitting diode therapy) on exercise performance and markers of exercise recovery: a systematic review with meta-analysis · Lasers in Medical Science 2015;30(2):925-939

A systematic review of low-level laser therapy for body circumference found small reductions in waist and other circumferences in short trials, on the order of a few centimeters. Circumference is not the same as fat mass or lasting weight loss, the trials were short and often industry-linked, and there is no evidence of a durable change in body fat, so the popular fat-burning claim runs well ahead of what the data show.

Measured in: Adults in short body-contouring trials, largely women

Circumference is not fat mass or weight, the trials were short and often industry-linked, and no durable fat or weight change has been shown, so this is the most overstated use.

Low-level laser therapy for reducing body circumferences: a systematic review · Lasers in Medical Science 2025

The leading proposed mechanism is that red and near-infrared light is absorbed by cytochrome c oxidase, an enzyme in the mitochondrial energy chain, briefly increasing energy production and signaling molecules that influence inflammation and cell repair. This is a plausible, actively studied mechanism drawn largely from cell and laboratory work; it explains why the light band matters, but a mechanism does not by itself establish a clinical benefit.

Measured in: Cell and laboratory models

This is the working mechanistic model, drawn mainly from cell and animal studies; a plausible mechanism does not by itself establish a clinical benefit, which each use has to earn in trials.

de Freitas & Hamblin, proposed mechanisms of photobiomodulation or low-level light therapy · IEEE Journal of Selected Topics in Quantum Electronics 2016;22(3):7000417

Photobiomodulation follows a biphasic dose response: too little light does nothing, a middling dose produces the effect, and too much can cancel it or turn it the other way. This pattern, described across laboratory and animal studies, is why device settings, wavelength, irradiance, distance and time, matter more than raw power, and why cranking the dose up does not increase the benefit.

Measured in: Cell and animal dose-response studies

The dose window is derived largely from laboratory and animal work and varies by tissue and target, so a single ideal setting cannot be stated; the reliable point is that more is not better.

Huang et al., biphasic dose response in low level light therapy · Dose-Response 2009;7(4):358-383

Across 572 studies (486,538 cancer cases), heavy drinkers versus non and occasional drinkers had RR 5.13 for oral and pharyngeal cancer, 4.95 for esophageal squamous cell carcinoma, 2.65 for laryngeal, 2.07 for liver, 1.61 for breast and 1.44 for colorectal cancer, each with a clear dose-risk gradient. Risk was also raised for stomach (1.21) and pancreatic (1.19) cancer.

Measured in: Adults pooled across 572 observational studies of alcohol and cancer

These are pooled observational relative risks, strongest where the dose-response gradient and a biological mechanism agree, as they do for the upper-airway, esophageal, liver, colorectal and breast sites. The gradient means the risk is not confined to heavy drinking: for several of these cancers it begins to rise at light intake. Alcohol is classified by IARC as a Group 1 carcinogen, which is why this holds at strong.

Bagnardi V, Rota M, Botteri E, et al. Alcohol consumption and site-specific cancer risk: a comprehensive dose-response meta-analysis · Br J Cancer 2015;112(3):580-93

Across 107 cohort studies (4,838,825 participants, 425,564 deaths), after adjusting for former-drinker bias and study-quality, occasional drinking (>0 to <1.3 g/day) carried RR 0.96 (95% CI 0.86-1.06) and low-volume drinking (1.3-24 g/day) RR 0.93 (P=.07), neither a significant reduction versus lifetime nondrinkers. Risk rose at 25-44 g/day (RR 1.05), and significantly at 45-64 (1.19) and 65+ g/day (1.35), starting at lower intake for women.

Measured in: Adults across 107 prospective cohorts, pooled and re-adjusted for the biases that separate abstainers from drinkers

This is pooled observational data, so it describes association, and the correction for former-drinker bias is a modeling adjustment rather than a randomized comparison. What it removes is the artifact that made light drinkers look protected; what it leaves is no measurable mortality benefit at low intake and clear excess at higher intake.

What could explain it instead: Abstainer bias: former and occasional drinkers reclassified as lifetime abstainers make the reference group sicker than it should be, flattering light drinkers by comparison. The re-analysis adjusts for exactly this, along with sampling and study-quality differences.

Zhao J, Stockwell T, Naimi T, et al. Association Between Daily Alcohol Intake and Risk of All-Cause Mortality: A Systematic Review and Meta-analyzes · JAMA Netw Open 2023;6(3):e236185

In 87 studies (3,998,626 people, 367,103 deaths), the unadjusted pool reproduced the classic J-curve: low-volume drinkers RR 0.86 (95% CI 0.83-0.90), while former drinkers carried elevated risk RR 1.22 (1.14-1.31). After adjusting for abstainer misclassification and study quality, the low-volume advantage fell to RR 0.97 (0.88-1.07), no longer significant, and higher-quality bias-free studies also found no reduced mortality.

Measured in: Adults across 87 prospective cohorts of alcohol use and all-cause mortality

The result depends on the study design: whether the group of abstainers it compares drinkers against is contaminated with people who quit because they were already sick. The J-curve is real in the raw data and mostly vanishes once that contamination is removed, so this finding is design-dependent rather than pointing one way.

What could explain it instead: Abstainer bias, also called the sick-quitter effect: former and occasional drinkers reclassified as lifetime abstainers make the abstainer reference group unhealthier than a true never-drinker group, so light drinkers appear to benefit. This meta-regression exists to isolate that bias.

Stockwell T, Zhao J, Panwar S, et al. Do Moderate Drinkers Have Reduced Mortality Risk? A Systematic Review and Meta-Analysis of Alcohol Consumption and All-Cause Mortality · J Stud Alcohol Drugs 2016;77(2):185-98

The Global Burden of Disease 2016 analysis, spanning 195 countries and territories, found the level of consumption that minimized total health loss across all causes combined was zero standard drinks per week. For 2016, alcohol was the leading risk factor for death and disability among people aged 15 to 49 worldwide.

Measured in: Pooled global data across 195 countries and territories, 1990-2016

This is population-level modeling that applies relative risks to consumption estimates, so it describes total health loss across a whole population rather than the risk to one person on one night. Low, occasional intake carries a small absolute risk for an individual; the population-minimizing figure being zero and an individual night being low-risk are both true and answer different questions.

GBD 2016 Alcohol Collaborators. Alcohol use and burden for 195 countries and territories, 1990-2016 · Lancet 2018;392(10152):1015-1035

In 105,986 women in the Nurses Health Study followed 1980-2008 (7,690 invasive breast cancers), risk rose significantly at intake as low as 5.0-9.9 g/day, about 3 to 6 drinks per week: RR 1.15 (95% CI 1.06-1.24). Binge drinking added risk independent of total intake; drinking frequency alone did not.

Measured in: Women in a long-running US nurses cohort

As a cohort it can adjust for measured factors but not fully separate drinking from the other habits that travel with it. The signal is consistent with the pooled cancer meta-analysis and with a dose-response, which is why it holds at moderate; the practical point is that the increase begins at a few drinks a week, not only at heavy intake.

What could explain it instead: Women who drink a few drinks a week differ from non-drinkers in diet, hormone use, body weight and screening frequency, several of which independently affect breast cancer risk; the cohort adjusts for many but residual confounding remains.

Chen WY, Rosner B, Hankinson SE, Colditz GA, Willett WC. Moderate alcohol consumption during adult life, drinking patterns, and breast cancer risk · JAMA 2011;306(17):1884-90

An estimated 741,300 new cancer cases in 2020, about 4.1% of all cancers, were attributable to alcohol; esophageal (189,700), liver (154,700) and breast (98,300) cancers contributed most. Men accounted for 568,700 cases (76.7%). Heavier drinking caused most of the burden, though moderate intake under 20 g/day still contributed.

Measured in: Global modelled estimate applying pooled relative risks to national consumption and GLOBOCAN 2020 incidence

This is a population attributable-fraction estimate: it applies pooled relative risks to national consumption data, so the case counts carry wide uncertainty intervals and assume those relative risks are causal. The relative risks come from systematic reviews and are supported by alcohol being an established (IARC Group 1) carcinogen.

Rumgay H, Shield K, Charvat H, et al. Global burden of cancer in 2020 attributable to alcohol consumption: a population-based study · Lancet Oncol 2021;22(8):1071-1080

In 371,463 UK Biobank participants (46% men, mean age 57), the healthier lifestyles of light-to-moderate drinkers explained most of their apparent cardiac benefit. Using Mendelian randomization, a 1-SD higher genetically predicted intake raised hypertension risk 1.3-fold (95% CI 1.2-1.4) and coronary artery disease 1.4-fold (1.1-1.8). The relationship was nonlinear: light intake added minimal risk, heavier intake raised risk exponentially, with no protective level.

Measured in: UK Biobank adults, mostly middle-aged and of European ancestry

Mendelian randomization uses inherited genetic variants as a natural experiment to reduce the lifestyle confounding that dogs observational alcohol research, which is its strength here. Its main limits are horizontal pleiotropy (a gene affecting risk through a path other than alcohol) and a population that is mostly middle-aged and of European ancestry.

What could explain it instead: Standard observational analyzes of alcohol and heart disease are confounded by the healthier lifestyles of light drinkers; Mendelian randomization is built to break that link, so the residual concern is genetic pleiotropy and population selection rather than lifestyle.

Biddinger KJ, Emdin CA, Haas ME, et al. Association of Habitual Alcohol Intake With Risk of Cardiovascular Disease · JAMA Netw Open 2022;5(3):e223849

Across 36 trials (2,865 participants, 2,464 men and 401 women), reducing alcohol lowered blood pressure in people drinking more than two drinks a day, with larger falls at heavier baseline intake; in people drinking two or fewer a day, reduction produced no significant change. Because these are randomized reductions, they show the blood-pressure effect of alcohol above about two drinks a day is causal.

Measured in: Adults in randomized alcohol-reduction trials, about 86% male

The trials tested reducing alcohol rather than drinking it, and the sample was about 86% male, so the size of the effect is best established in men. The finding is dose-dependent: below roughly two drinks a day the blood-pressure change was not significant; above it the reduction was clear and grew with baseline intake.

Roerecke M, Kaczorowski J, Tobe SW, et al. The effect of a reduction in alcohol consumption on blood pressure: a systematic review and meta-analysis · Lancet Public Health 2017;2(2):e108-e120

In a crossover polysomnography study of healthy volunteers, a bedtime dose raising blood alcohol to 0.10% shortened sleep latency, reduced waking, and increased slow-wave sleep in the first half of the night, resembling a short-acting sedative. A lower dose (0.03%) produced no clear change.

Measured in: Healthy adult volunteers in a sleep laboratory

The onset effect is real and is why alcohol is used as a sleep aid, but the same study and later work show the benefit is confined to the first hours and is repaid later in the night (the paired finding below). The sample was small, so the size of the onset effect is approximate.

Feige B, Gann H, Brueck R, et al. Effects of alcohol on polysomnographically recorded sleep in healthy subjects · Alcohol Clin Exp Res 2006;30(9):1527-37

At 0.10% blood alcohol, the same crossover study found rebound effects in the second half of the night (increased light sleep) and suppressed REM density early, with REM rebounding once the alcohol cleared. A 2024 crossover polysomnography study of 30 adults across three consecutive nights found presleep alcohol reduced total REM sleep and slowed its accumulation each night.

Measured in: Healthy adults across two crossover sleep-laboratory studies

Both samples are small, but the direction is consistent and mechanistically expected: alcohol front-loads deep sleep and then fragments the second half and cuts REM sleep, so a night can feel like it started well and still leave you underslept.

Feige B, Gann H, Brueck R, et al. Effects of alcohol on polysomnographically recorded sleep in healthy subjects · Alcohol Clin Exp Res 2006;30(9):1527-37 McCullar KS, Barker DH, McGeary JE, et al. Altered sleep architecture following consecutive nights of presleep alcohol · Sleep 2024;47(4):zsae003

In 550 adults in the Whitehall II study followed 30 years with MRI, higher alcohol intake predicted dose-dependent hippocampal atrophy: over 30 units a week carried 5.8-fold odds versus abstainers (95% CI 1.8-18.6), and even moderate intake (14-21 units a week) tripled the odds of right-sided atrophy (OR 3.4, 1.4-8.1). Light drinking (1 to under 7 units a week) gave no protection over abstinence, and higher intake tracked faster decline in lexical fluency.

Measured in: Community-dwelling UK adults in the Whitehall II imaging substudy

As an observational cohort it cannot prove alcohol caused the shrinkage, though the dose-response and the 30-year follow-up strengthen the case. Alcohol was self-reported and the cohort was relatively healthy and advantaged, which may understate effects in the wider population.

What could explain it instead: Heavier drinkers differ in smoking, diet, education and cardiovascular risk, all of which affect brain structure; the study adjusts for measured factors, but self-reported intake and residual confounding remain.

Topiwala A, Allan CL, Valkanova V, et al. Moderate alcohol consumption as risk factor for adverse brain outcomes and cognitive decline: longitudinal cohort study · BMJ 2017;357:j2353

In a 25-year New Zealand birth cohort (1,055 followed to age 25), alcohol abuse or dependence and major depression were strongly linked; after fixed-effects adjustment for shared confounders, structural-equation modeling found the best-fitting causal direction ran from alcohol problems to depression, not from depression to self-medication with alcohol.

Measured in: A birth cohort followed from childhood to age 25

This addresses heavier, problem drinking rather than the occasional drink, and even with fixed-effects modelling a cohort cannot fully exclude a shared underlying vulnerability. What it argues against is the belief that drinking treats low mood; the modelled arrow points the other way.

What could explain it instead: Alcohol problems and depression share risk factors such as adversity, other substance use, unemployment and life stress; the fixed-effects design controls stable confounders, but time-varying shared causes cannot be entirely ruled out.

Fergusson DM, Boden JM, Horwood LJ. Tests of causal links between alcohol abuse or dependence and major depression · Arch Gen Psychiatry 2009;66(3):260-6

In a multicenter open-label randomized trial, 140 regular drinkers (85% men, mean age 62) with paroxysmal or persistent atrial fibrillation were assigned to abstain from alcohol or keep drinking. The abstinence group cut intake from 16.8 to 2.1 standard drinks a week; over six months, atrial fibrillation recurred in 53% of them versus 73% of controls (hazard ratio 0.55, 95% CI 0.36-0.84), and the share of time spent in atrial fibrillation fell to a median 0.5% from 1.2%.

Measured in: Regular drinkers (10 or more standard drinks a week) with existing atrial fibrillation, recruited at six Australian hospitals

This is a single open-label trial in people who already had atrial fibrillation and drank regularly, so it speaks to reducing recurrence in that group rather than to first onset at low intake. Because it randomized the change in drinking, it gives causal evidence that alcohol drives the arrhythmia in regular drinkers, alongside the genetic and blood-pressure findings.

Voskoboinik A, Kalman JM, De Silva A, et al. Alcohol Abstinence in Drinkers with Atrial Fibrillation · N Engl J Med 2020;382(1):20-28

Across 26 randomized trials in about 170,000 people, each 39 mg/dL (1.0 mmol/L) reduction in LDL cholesterol lowered major vascular events by about 22% per 1.0 mmol/L reduction, with the benefit tracking the size of the LDL drop however it was achieved.

Measured in: About 170,000 adults across 26 randomized statin trials

The trials tested LDL lowering by statins, so the clean dose-response applies most directly to that route; the number is an average across many populations and does not fix any one person's risk.

Cholesterol Treatment Trialists Collaboration, efficacy and safety of more intensive lowering of LDL cholesterol: a meta-analysis of data from 170,000 participants in 26 randomised trials · Lancet 2010;376(9753):1670-1681

A European Atherosclerosis Society consensus synthesizing genetic, epidemiological and trial evidence concluded that LDL causes atherosclerotic cardiovascular disease, with lifelong genetically lower LDL matched by proportionally lower lifetime risk, so the effect of LDL is cumulative over years.

Measured in: Synthesis of genetic, cohort and trial evidence across large mixed populations

A consensus synthesis reflects the weight of evidence rather than a single measured effect size, and causal certainty for LDL does not extend automatically to every other lipid particle.

Ference et al., low-density lipoproteins cause atherosclerotic cardiovascular disease. 1. Evidence from genetic, epidemiologic, and clinical studies. A consensus statement from the European Atherosclerosis Society Consensus Panel · Eur Heart J 2017;38(32):2459-2472

In 11,092 adults without diabetes followed about 15 years, higher HbA1c predicted the later diagnosis of diabetes and predicted cardiovascular disease at least as well as fasting glucose, giving a single blood draw that reads recent months of blood sugar.

Measured in: 11,092 adults without diabetes in a US community cohort

HbA1c can read falsely high or low with anemia, recent blood loss, pregnancy or some hemoglobin variants, so a surprising value is confirmed before it is acted on.

What could explain it instead: People with higher HbA1c also tend to carry more body fat and be less active, factors that independently raise both blood sugar and cardiovascular risk, so part of the association reflects those shared traits rather than the marker itself.

Selvin et al., glycated hemoglobin, diabetes, and cardiovascular risk in nondiabetic adults · N Engl J Med 2010;362(9):800-811

In a collaborative meta-analysis of general-population cohorts totalling about 105,000 people, an estimated GFR below 60 and a higher urine albumin-to-creatinine ratio each independently predicted all-cause and cardiovascular death in a graded way, and the two together classify kidney risk better than either alone.

Measured in: About 105,872 adults across 14 general-population cohorts

A single low eGFR can reflect dehydration or a temporary insult, and albumin can rise transiently after exercise, fever or infection, so kidney staging rests on repeated measurements over months, not one result.

Chronic Kidney Disease Prognosis Consortium (Matsushita et al.), association of estimated glomerular filtration rate and albuminuria with all-cause and cardiovascular mortality in general population cohorts: a collaborative meta-analysis · Lancet 2010;375(9731):2073-2081

Across 17 randomized trials and 251,891 participants, offering healthy adults a package of general health checks made little or no difference to all-cause mortality (risk ratio 1.00, 95% CI 0.97 to 1.03), nor to cancer or cardiovascular death, on high-certainty evidence.

Measured in: 251,891 adults across 17 randomized trials of general health checks

This concerns the strategy of broad routine screening in the well; a targeted test chosen because of age, symptoms, family history or a known risk factor is a different question with its own evidence.

Krogsboll et al., general health checks in adults for reducing morbidity and mortality from disease · Cochrane Database Syst Rev 2019;(1):CD009009

A meta-analysis found ApoB, which counts the number of atherogenic particles, discriminated cardiovascular risk better than LDL cholesterol or non-HDL cholesterol, and a large biobank analysis found particle number rather than cholesterol content was the principal driver of heart attack risk.

Measured in: Pooled marker-comparison studies plus a UK Biobank cohort

The added value of ApoB over LDL cholesterol is real but incremental for most people; where the two agree, either measure carries the same information.

Sniderman et al., a meta-analysis of low-density lipoprotein cholesterol, non-high-density lipoprotein cholesterol, and apolipoprotein B as markers of cardiovascular risk · Circ Cardiovasc Qual Outcomes 2011;4(3):337-345 Marston et al., association of apolipoprotein B-containing lipoproteins and risk of myocardial infarction in individuals with and without atherosclerosis · JAMA Cardiol 2022;7(3):250-256

In a meta-analysis of long-term cohorts totalling about 126,000 people, each 3.5-fold higher lipoprotein(a) carried about 1.13 times the risk of coronary heart disease, independent of other lipids, and genetic studies link the same particle to coronary disease, so the level is largely set by inheritance.

Measured in: About 126,000 adults in pooled prospective cohorts, plus genetic association data

Elevated lipoprotein(a) identifies added risk but there is no widely approved therapy that lowers it and changes outcomes yet, so the value is in refining overall risk and prompting attention to the factors that can be changed.

Emerging Risk Factors Collaboration, lipoprotein(a) concentration and the risk of coronary heart disease, stroke, and nonvascular mortality · JAMA 2009;302(4):412-423 Clarke et al., genetic variants associated with Lp(a) lipoprotein level and coronary disease · N Engl J Med 2009;361(26):2518-2528

Pooling individual data on 160,309 people, each standard-deviation higher log C-reactive protein was associated with about 1.63 times the risk of coronary heart disease after age and sex adjustment, falling to 1.37 after conventional risk factors and 1.23 after further adjustment for fibrinogen, so hs-CRP adds modest independent information.

Measured in: 160,309 adults without prior vascular disease, in pooled prospective cohorts

CRP rises with any infection, injury or inflammatory flare, so a single high value taken when someone is unwell reflects that illness rather than baseline cardiovascular risk and should be repeated when well.

Emerging Risk Factors Collaboration, C-reactive protein concentration and risk of coronary heart disease, stroke, and mortality: an individual participant meta-analysis · Lancet 2010;375(9709):132-140

The US Preventive Services Task Force found insufficient evidence to weigh the benefits and harms of screening for thyroid dysfunction in non-pregnant adults who have no symptoms, while TSH remains the appropriate first test when symptoms do suggest a thyroid problem.

Measured in: Asymptomatic non-pregnant adults, per the evidence review

The insufficient-evidence finding applies to routine screening; pregnancy and clear symptoms are separate situations where thyroid testing is indicated.

LeFevre, US Preventive Services Task Force, screening for thyroid dysfunction: US Preventive Services Task Force recommendation statement · Ann Intern Med 2015;162(9):641-650

In a national survey of 991 US physicians, almost all reported that incidental or borderline findings had set off cascades of further tests, referrals and procedures, and most reported these cascades causing patient harm, wasted resources, and patient anxiety.

Measured in: National survey of 991 US physicians reporting on their practice

This describes the downstream pattern reported by clinicians, not a randomized measurement of net harm, and a cascade started by a truly important finding can also be appropriate.

What could explain it instead: The findings come from physicians recalling their own experience in a survey, so they reflect self-report and the salience of memorable cases rather than a directly measured rate of harm.

Ganguli et al., cascades of care after incidental findings in a US national survey of physicians · JAMA Netw Open 2019;2(10):e1913325

HOMA-IR, calculated from fasting glucose and fasting insulin, estimates insulin resistance and tracks against the gold-standard clamp measurement, offering a window on metabolism that can shift before HbA1c or fasting glucose move.

Measured in: Model validated against clamp studies; interpretation guidance from methods reviews

Because insulin assays are not standardized and cut-offs are not agreed, HOMA-IR is best used to track change in one person over time rather than compared against a fixed number.

Wallace, Levy and Matthews, use and abuse of HOMA modeling · Diabetes Care 2004;27(6):1487-1495

The US Preventive Services Task Force reviewed the evidence and concluded it is insufficient to weigh the benefits and harms of screening asymptomatic adults for vitamin D deficiency, in part because the threshold that defines deficiency is itself unsettled.

Measured in: Asymptomatic community-dwelling non-pregnant adults, per the evidence review

An insufficient-evidence verdict is about routine screening of people without symptoms; it does not speak to testing someone with a bone disorder, malabsorption or another clear indication.

US Preventive Services Task Force, screening for vitamin D deficiency in adults: US Preventive Services Task Force recommendation statement · JAMA 2021;325(14):1436-1442

In STEP 1, adults with overweight or obesity and no diabetes lost a mean of about 14.9% of body weight over 68 weeks on weekly semaglutide 2.4 mg, against about 2.4% on placebo, and roughly 86% lost at least 5%.

Measured in: Adults with overweight or obesity and no diabetes, majority women

This is a licensed medicine studied in a specific population; the weight effect and the trial safety data do not carry over to unregulated peptides sold outside the medical system.

Wilding et al., Once-Weekly Semaglutide in Adults with Overweight or Obesity · N Engl J Med 2021;384(11):989-1002

In SUSTAIN-6, semaglutide lowered HbA1c by roughly 0.7 to 1.0 percentage points versus placebo over two years in adults with type 2 diabetes at high cardiovascular risk, alongside weight reduction.

Measured in: Adults with type 2 diabetes at high cardiovascular risk

The effect is established for a prescription drug in diagnosed type 2 diabetes and does not describe peptides taken without medical oversight.

Marso et al., Semaglutide and Cardiovascular Outcomes in Patients with Type 2 Diabetes (SUSTAIN-6) · N Engl J Med 2016;375(19):1834-1844

In SELECT, semaglutide cut major adverse cardiovascular events by about 20% (hazard ratio 0.80) over roughly three years in people with obesity and established heart disease but no diabetes, and a 2025 meta-analysis found consistent cardiovascular and kidney benefit across the GLP-1 receptor agonist class.

Measured in: Adults with cardiovascular disease or diabetes across pooled trials

The cardiovascular benefit is established only for licensed GLP-1 receptor agonists in studied populations and says nothing about unregulated peptides.

Lincoff et al., Semaglutide and Cardiovascular Outcomes in Obesity without Diabetes (SELECT) · N Engl J Med 2023;389(24):2221-2232 Badve et al., Effects of GLP-1 receptor agonists on kidney and cardiovascular disease outcomes: a meta-analysis of randomised controlled trials · Lancet Diabetes Endocrinol 2025

In SURMOUNT-1, adults with obesity and no diabetes lost a mean of about 20.9% of body weight over 72 weeks on the top 15 mg weekly dose of tirzepatide, against about 3.1% on placebo, with about 15.0% lost at the lowest 5 mg dose.

Measured in: Adults with obesity and no diabetes

This is a licensed medicine studied in adults with obesity; the weight effect does not carry over to unregulated peptides sold outside the medical system.

Jastreboff et al., Tirzepatide Once Weekly for the Treatment of Obesity (SURMOUNT-1) · N Engl J Med 2022;387(3):205-216

Even the approved GLP-1 peptides commonly cause nausea, vomiting, diarrhea and constipation, which were the leading reasons participants stopped semaglutide in its weight-loss trial.

These effects are well characterized for the approved drug under supervision; self-dosing outside that setting removes the titration that keeps them manageable.

Wilding et al., Once-Weekly Semaglutide in Adults with Overweight or Obesity (STEP 1) · N Engl J Med 2021;384(11):989-1002

In healthy adults, the growth-hormone-releasing peptide CJC-1295 raised circulating growth hormone and IGF-1 for days after a single injection, with IGF-1 rising well above baseline.

Measured in: Healthy adult volunteers in a dose-finding study

The trial measured only hormone levels and safety over a short period, so it establishes the biomarker effect and nothing about longer-term benefit or harm.

Teichman et al., Prolonged stimulation of growth hormone and IGF-I secretion by CJC-1295 in healthy adults · J Clin Endocrinol Metab 2006;91(3):799-805

Peptides sold as research chemicals or through the grey market carry documented risks of falsification, wrong or absent active ingredient, and contamination, because they are made and sold outside pharmaceutical controls.

The specific contents of any given grey-market vial are unknown, which is itself the hazard; documented cases show wrong dose, wrong molecule and contamination.

Janvier et al., Falsification of biotechnology drugs: current dangers and/or future disasters? · J Pharm Biomed Anal 2018;161:175-191

Across observational studies, higher circulating IGF-1 is associated with an increased risk of some cancers, chiefly prostate cancer and premenopausal breast cancer, which is a concern for peptides taken specifically to raise IGF-1.

The link is observational and about naturally varying IGF-1, so it indicates a plausible hazard from deliberately raising IGF-1 rather than a proven causal effect of the peptides.

What could explain it instead: People with higher IGF-1 may differ in growth, body size, diet and other cancer risk factors, so the association may partly reflect those rather than IGF-1 itself.

Renehan et al., Insulin-like growth factor (IGF)-I, IGF binding protein-3, and cancer risk: systematic review and meta-regression analysis · Lancet 2004;363(9418):1346-1353

The human studies of growth-hormone-releasing peptides measured hormone levels rather than muscle, strength, fat loss, recovery or any healthspan outcome, so a benefit to the body is not established.

Measured in: Healthy adults in short biomarker-focused studies

This describes the absence of outcome data rather than a trial showing no effect, so the accurate reading is that benefit is untested, not disproven.

Teichman et al., Prolonged stimulation of growth hormone and IGF-I secretion by CJC-1295 in healthy adults · J Clin Endocrinol Metab 2006;91(3):799-805

BPC-157, a synthetic gastric peptide fragment, improved tendon, muscle and wound healing in rat studies, but no completed randomized human trials of it exist.

All of the healing evidence is from animals; there is no completed human randomized trial, so the effect in people is untested.

Japjec et al., BPC-157 and tendon/ligament healing (animal) · Biomedicines 2021;9(11):1547 Seiwerth et al., BPC-157 in tissue healing and cytoprotection (review) · Front Pharmacol 2021;12:627533

Thymosin beta-4, the molecule TB-500 is based on, shows tissue-repair and regenerative activity in preclinical models, with no approved human use for recovery or anti-aging.

The regenerative evidence is preclinical; efficacy and safety of TB-500 for its marketed human uses are not demonstrated.

Bock-Marquette et al., Thymosin beta-4 and anti-aging regenerative therapies (review) · Int Immunopharmacol 2023;116:109741

The copper peptide GHK-Cu promotes skin-matrix remodeling and wound-related activity in laboratory and preclinical work, with limited human cosmetic data.

Most of the mechanism is shown in the lab; robust human evidence for a meaningful anti-aging skin effect is limited.

Dou et al., The potential of GHK as an anti-aging peptide · Aging Pathobiol Ther 2020;2(1):58-61

Melanotan II, an injectable tanning peptide sold online, has been linked in case reports to priapism and to renal infarction, alongside nausea and changes in moles.

This evidence is case reports rather than controlled data, but the harms are serious and the product is unregulated.

Dreyer et al., Melanotan-induced priapism: a hard-earned tan (case report) · BMJ Case Rep 2019;12(2):bcr-2018-227644 Peters et al., Melanotan II: a possible cause of renal infarction (case report) · CEN Case Rep 2020;9(2):159-161

Each 11 lb (5 kg) lower grip strength came with a 16% higher rate of death (HR 1.16, 95% CI 1.13 to 1.20) across 139,691 people in 17 countries.

Measured in: 139,691 adults aged 35 to 70 across 17 countries, 58% women (the PURE study)

Grip is a marker of overall condition rather than a target in itself, so training your hands will not move the outcome. The often-repeated line that grip beats blood pressure as a predictor is narrower than it sounds: it was a post-hoc comparison, measured per standard deviation, and it holds for death but not for new cardiovascular disease, where blood pressure predicted better. Median follow-up was only four years, and these were adults from 35 upward rather than an older-adult sample.

What could explain it instead: Observational. Low grip strength is often an early sign of illness rather than its cause, which the short four-year follow-up makes more likely rather than less.

Leong et al., prognostic value of grip strength, findings from the PURE study · Lancet 2015;386(9990):266-273

Every 0.1 m/s faster usual walking speed came with about 12% lower mortality (HR 0.88, 95% CI 0.87 to 0.90).

Measured in: 34,485 community-dwelling adults aged 65 and over, pooled from nine cohorts, about 60% women

A pooled analysis of individual participant data from nine selected cohorts rather than a systematic review, so it carries none of the publication-bias checks that phrase usually implies. Samples were mostly white and US-based. Predicted ten-year survival at 75 ran from 19% to 87% in men and 35% to 91% in women across the speed range, a remarkable spread for a test that takes seconds.

What could explain it instead: Observational, and slow walking reflects existing disease as much as it forecasts new disease.

Studenski et al., gait speed and survival in older adults · JAMA 2011;305(1):50-58

The pooled survival analysis was built on a stopwatch and a 13-foot (4-metre) course: from a standing start, walk at your usual pace, as if walking down the street, with no further encouragement. Mean speed in the 34,485 participants was 0.92 m/s (SD 0.27).

Measured in: 34,485 community-dwelling adults aged 65 and over, mean age 73.5, 59.6% women, 79.8% white, pooled from nine cohorts collected between 1986 and 2000

The contributing cohorts did not all use a 13-foot (4-metre) course. Walk distances ran from 8 feet to 20 feet (6 metres) and were converted to a 13-foot (4-metre) equivalent by formula, including one conversion derived from just 61 people walking both distances. Reproducing the protocol at home gets you closer to the published figures than any variation of it, and the pooled number still carries that conversion inside it.

What could explain it instead: Observational. Slow walking is a common final pathway for heart failure, arthritis, neuropathy and depression, so the speed reads existing disease as much as it forecasts new disease.

Studenski et al., gait speed and survival in older adults · JAMA 2011;305(1):50-58

Against a correctly sized cuff, a regular cuff read 19.5 mmHg higher systolic (95% CI 16.1 to 22.9) in people whose arm required an extra-large cuff, 4.8 mmHg higher (3.0 to 6.6) in those requiring a large cuff, and 3.6 mmHg lower (-5.6 to -1.7) in those requiring a small cuff.

Measured in: 195 community-dwelling adults in Baltimore across a range of mid-arm circumferences, mean age 54, 34% male, 68% Black, 51% with hypertension

Readings were taken with an automated oscillometric device in a research setting, so the figures isolate cuff error with every other variable controlled. At home the cuff error stacks with arm position and rest time rather than replacing them. Single-site and heavily Black, and mid-arm circumference distribution decides who is affected.

Ishigami et al., effects of cuff size on the accuracy of blood pressure readings: the Cuff(SZ) randomized crossover trial · JAMA Intern Med 2023;183(10):1061-1068

Against the arm supported on a desk at heart height, resting the hand on the lap overestimated systolic pressure by 3.9 mmHg (95% CI 2.5 to 5.2) and diastolic by 4.0 mmHg (3.1 to 5.0). An unsupported arm at the side overestimated systolic by 6.5 mmHg (5.1 to 7.9) and diastolic by 4.4 mmHg (3.4 to 5.4).

Measured in: 133 adults aged 18 to 80 recruited in Baltimore, mean age 57, 53% female; 36% had systolic pressure of 130 mmHg or above and 41% had a BMI of 30 or above

Single-site trial with triplicate automated readings under research conditions, so it isolates arm position with everything else held constant. It does not say how often each position is used in practice, which is the quantity that decides the population-level error.

Liu et al., arm position and blood pressure readings: the ARMS crossover randomized clinical trial · JAMA Intern Med 2024;184(12):1436-1442

At twelve months, systolic pressure was 3.5 mmHg lower than usual care with self-monitoring alone (95% CI -5.8 to -1.2) and 4.7 mmHg lower with telemonitoring added (-7.0 to -2.4). Telemonitoring did not beat self-monitoring alone (-1.2 mmHg, 95% CI -3.5 to 1.2). Adverse events were similar across all three groups.

Measured in: 1,182 patients over 35 with blood pressure above 140/90 mmHg, across 142 UK general practices, who were willing to self-monitor; 1,003 (85%) were included in the primary analysis

The effect came from GPs titrating medication on the readings, not from the readings existing, so self-monitoring with nobody adjusting anything is a different intervention with no reason to expect this result. Participants were selected for willingness to self-monitor and the trial was unmasked. The monitor manufacturer was among the funders.

McManus et al., efficacy of self-monitored blood pressure, with or without telemonitoring, for titration of antihypertensive medication (TASMINH4) · Lancet 2018;391(10124):949-959

After adjustment for BMI, men in the highest fifth of waist circumference had 2.05 times the risk of death of those in the lowest (95% CI 1.80 to 2.33) and women 1.78 times (1.56 to 2.04). For waist-to-hip ratio the figures were 1.68 and 1.51. BMI remained significantly associated with death in models that also contained waist.

Measured in: 359,387 participants from nine European countries in the EPIC cohort, mean follow-up 9.7 years, 14,723 deaths; lowest mortality risk sat at a BMI of 25.3 in men and 24.3 in women

Waist and BMI are each adding information the other misses, so this is a case for measuring both rather than replacing one. The models adjust for education, smoking, alcohol, physical activity and height, and cannot adjust away the underlying problem that illness causes weight loss before it causes death.

What could explain it instead: Reverse causation through illness-related weight loss. Undiagnosed cancer, COPD and heart failure shrink the waist years before they kill, which loads the leanest group with sick people and steepens the apparent gradient.

Pischon et al., general and abdominal adiposity and risk of death in Europe · N Engl J Med 2008;359(20):2105-2120

General health checks had little or no effect on total mortality (risk ratio 1.00, 95% CI 0.97 to 1.03; 11 trials, 233,298 participants, 21,535 deaths, high certainty, I-squared 0%) or cancer mortality (1.01, 0.92 to 1.12, high certainty), and probably little or none on cardiovascular mortality (1.05, 0.94 to 1.16, moderate certainty). Ischaemic heart disease events were unchanged (0.98, 0.94 to 1.03, high certainty).

Measured in: 17 randomized trials, 15 reporting outcome data, 251,891 participants unselected for disease or risk factors; geriatric trials were excluded by design

This is about packaged screening of more than one organ system in people with no particular reason to be tested. It says nothing about a specific test ordered for a specific reason, nothing about established single-disease screening programmes, and nothing about older adults, who were excluded. Several included trials are old enough that the treatments available on a positive result have since improved.

Krogsboll, Jorgensen and Gotzsche, general health checks in adults for reducing morbidity and mortality from disease · Cochrane Database Syst Rev 2019;1(1):CD009009

Progressive resistance training improved gait speed by 0.08 m/s (24 trials, 1,179 participants, 95% CI 0.04 to 0.12) and produced a moderate to large improvement in getting out of a chair (11 trials, 384 participants, SMD -0.94, 95% CI -1.49 to -0.38). Muscle strength improved substantially (73 trials, 3,059 participants, SMD 0.84). Physical ability improved slightly (33 trials, SMD 0.14).

Measured in: 121 randomized trials, 6,700 participants, mostly older adults, with training typically two to three times a week at high intensity

These are improvements in the test scores, not in the outcomes the test scores predict, and no trial here measured survival. Adverse events were poorly recorded across the literature, though joint pain and muscle soreness were commonly reported where they were tracked, and serious events were rare and not attributed to the exercise.

Liu and Latham, progressive resistance strength training for improving physical function in older adults · Cochrane Database Syst Rev 2009;(3):CD002759

Risk-adjusted all-cause mortality fell across every fitness band with no upper limit found. The lowest fitness group carried 5.04 times the mortality of elite performers (95% CI 4.10 to 6.20), and below-average against above-average carried 1.41 (1.34 to 1.49). Those figures are comparable to or larger than coronary artery disease (1.29), smoking (1.41) and diabetes (1.40) in the same model.

Measured in: 122,007 adults referred for symptom-limited exercise treadmill testing at a US tertiary center between 1991 and 2014, mean age 53.4, 59.2% male, median follow-up 8.4 years, 13,637 deaths

Everyone in this cohort was referred for a treadmill test, which means everyone had a clinical reason to be there, so it is not a general-population sample. Fitness was estimated from peak metabolic equivalents on a treadmill rather than measured by gas exchange, and none of it can be reproduced with anything you own.

What could explain it instead: Reverse causation. Undiagnosed illness lowers treadmill performance before it becomes a diagnosis, and the least fit group in a referral population is enriched for people who were already unwell on the day they were tested.

Mandsager et al., association of cardiorespiratory fitness with long-term mortality among adults undergoing exercise treadmill testing · JAMA Netw Open 2018;1(6):e183605

One in five could not hold the 10-second stance. Among them 17.5% died over the follow-up, against 4.6% of those who could, an adjusted hazard ratio of 1.84 (95% CI 1.23 to 2.78).

Measured in: 1,702 adults aged 51 to 75 attending a private exercise-medicine clinic in Brazil, 68% men

Much the smallest of our three self-measures, and not a general-population sample: these were self-selected attendees at one clinic. The confidence interval is wide. Treat it as a free signal worth acting on rather than a precise multiplier, and not as equal in weight to the grip and walking-speed findings.

What could explain it instead: Observational. Neurological and musculoskeletal disease impairs balance and independently raises mortality, and a clinic population differs from the general public in ways adjustment cannot remove.

Araujo et al., successful 10-second one-legged stance performance predicts survival · Br J Sports Med 2022;56(17):975-980

Of six health and function measures tracked quarterly for a year, only improved gait speed predicted eight-year survival. Mortality was 31.6% in those who improved by at least 0.1 m/s, 41.2% in those who improved transiently, and 49.3% in those who never improved (adjusted hazard ratio 0.42, 95% CI 0.29 to 0.61).

Measured in: 439 adults aged 65 and over recruited through a Medicare health maintenance organization and Veterans Affairs primary care programs in the United States

This shows that people whose walking speed improves live longer, not that making your walking speed improve makes you live longer. The authors end by calling for research into whether interventions that raise gait speed affect survival, which had not been done. The sample is small, single-country, and drawn from two specific health systems.

What could explain it instead: Observational. Recovery from an acute illness or a hospitalization raises gait speed and independently improves survival, so improvement partly marks who was temporarily unwell at baseline rather than who got fitter.

Hardy et al., improvement in usual gait speed predicts better survival in older adults · J Am Geriatr Soc 2007;55(11):1727-1734

For the 30-second chair stand, test-retest intraclass correlation was 0.84 in men and 0.92 in women, and chair-stand count correlated with maximum weight-adjusted leg press at r = 0.78 in men and r = 0.71 in women. Performance fell significantly across each decade from the 60s to the 80s and was lower in low-active than high-active participants.

Measured in: 76 community-dwelling volunteers over 60, mean age 70.5, for the reliability and validity work; the criterion-referenced independence standards were later derived from 2,140 moderate-functioning older adults aged 60 to 94

The validation sample was 76 generally active volunteers, and the authors limit their conclusion to generally active community-dwelling older adults. The published cut-offs describe the capacity associated with staying physically independent, which is a different question from whether you will fall, and neither paper measured falls.

What could explain it instead: Volunteer selection. People who agree to perform two maximal leg-press tests are fitter and more mobile than the population the test is used to screen, which inflates both the reliability and the correlation with leg press.

Jones, Rikli and Beam, a 30-s chair-stand test as a measure of lower body strength in community-residing older adults · Res Q Exerc Sport 1999;70(2):113-119 Rikli and Jones, development and validation of criterion-referenced clinically relevant fitness standards for maintaining physical independence in later years · Gerontologist 2013;53(2):255-267

In the same cohort where each 11 lb (5 kg) less grip carried a 16% higher rate of death, grip strength showed no significant association with incident diabetes, with hospital admission for pneumonia or COPD, with injury due to a fall, or with fracture.

Measured in: 139,691 adults aged 35 to 70 across 17 countries, 58% women, grip measured with a Jamar dynamometer, median follow-up 4.0 years

A null over a median four years of follow-up is a weaker statement than a null over fifteen, particularly for outcomes as uncommon as fracture in a cohort starting at 35. The finding is that grip is a general marker rather than a specific one, and it does not license the reverse claim that hand strength is unrelated to falls.

What could explain it instead: Observational. Fracture and fall injury depend heavily on bone density, medication and home environment, none of which grip strength measures, so a null here reflects what the test does not read rather than an absence of relationship in the body.

Leong et al., prognostic value of grip strength: findings from the PURE study · Lancet 2015;386(9990):266-273

Each entered alone, office pressure (hazard ratio 1.13 per 10 mmHg systolic) and home pressure (1.23) both predicted cardiovascular events. Entered into the same model, only home pressure still predicted events (1.22, 95% CI 1.09 to 1.37) and office pressure did not (1.01, 0.92 to 1.12). Systolic home pressure was the sole predictor of total mortality.

Measured in: 2,081 randomly selected Finnish adults aged 45 to 74, followed a mean of 6.8 years, with 162 cardiovascular events and 118 deaths

162 events is a modest number for a model comparison, and the confidence intervals overlap more than a single point estimate suggests. Home readings in this study were taken under instruction with a supplied validated device, which is not the same as readings taken on whatever monitor a person owns. Ambulatory monitoring, not home monitoring, remains the reference standard for out-of-office pressure.

What could explain it instead: Observational. People willing and able to self-measure at home differ in health literacy, medication adherence and comorbidity from those who are not, and that difference tracks cardiovascular risk independently of the reading.

Niiranen et al., home-measured blood pressure is a stronger predictor of cardiovascular risk than office blood pressure: the Finn-Home study · Hypertension 2010;55(6):1346-1351 Muntner et al., measurement of blood pressure in humans: a scientific statement from the American Heart Association · Hypertension 2019;73(5):e35-e66

Each 10 beats per minute higher resting heart rate carried a relative risk of 1.09 (95% CI 1.07 to 1.12) for all-cause mortality and 1.08 (1.06 to 1.10) for cardiovascular mortality. Against the lowest category, 60 to 80 beats per minute carried 1.12 and above 80 beats per minute carried 1.45 (1.34 to 1.57). All-cause mortality risk rose linearly from 45 beats per minute upward.

Measured in: 46 prospective cohort studies in general populations, 1,246,203 participants and 78,349 deaths for all-cause mortality; 848,320 participants and 25,800 deaths for cardiovascular mortality

The authors report substantial heterogeneity between studies and detected publication bias, which is why this sits at moderate rather than strong despite the size. Resting heart rate is also measured inconsistently across cohorts, and the cardiovascular risk increase only reached significance at 90 beats per minute.

What could explain it instead: Observational. Fitness, thyroid disease, anemia, fever, subclinical heart failure and beta blocker use all set resting heart rate and independently affect mortality, and adjustment for traditional cardiovascular risk factors does not touch most of them.

Zhang, Shen and Qi, resting heart rate and all-cause and cardiovascular mortality in the general population: a meta-analysis · CMAJ 2016;188(3):E53-E63

Against people under 70 beats per minute at both measurements (8.2 ischaemic heart disease deaths per 10,000 person-years), those under 70 at first measurement and above 85 ten years later had an adjusted hazard ratio of 1.9 (95% CI 1.0 to 3.6), and those going from 70 to 85 up to above 85 had 1.8 (1.2 to 2.8). The association was not linear (p = 0.003 for quadratic trend), and a decrease in resting heart rate showed no general mortality benefit.

Measured in: 13,499 men and 15,826 women in Norway without known cardiovascular disease, measured twice about ten years apart in the Nord-Trondelag health study, mean 12 years of follow-up, 3,038 deaths of which 388 from ischaemic heart disease

388 ischaemic heart disease deaths across many change categories makes the individual hazard ratios imprecise, and the confidence interval on the largest of them touches 1.0. The finding that a falling rate carried no general benefit is the part most relevant to anyone trying to train the number down, and it is a null within an observational study rather than a trial of lowering it.

What could explain it instead: Observational. A resting heart rate that climbs over a decade is a signal of developing illness, weight gain, deconditioning or new medication, all of which raise cardiac mortality on their own, so the rise marks the process rather than causing it.

Nauman et al., temporal changes in resting heart rate and deaths from ischemic heart disease · JAMA 2011;306(23):2579-2587

Pooling receiver operating characteristic data, waist circumference improved discrimination of adverse cardiometabolic outcomes by 3% over BMI and waist-to-height ratio by 4 to 5% over BMI. Within-study comparison of the area under the curve found waist-to-height ratio significantly better than waist circumference for diabetes, hypertension, cardiovascular disease and all outcomes combined, in both sexes.

Measured in: 31 papers covering more than 300,000 adults across several ethnic groups

These are discrimination statistics for detecting risk factors that are already present, not prospective prediction of events, and a 4 to 5% improvement in area under the curve is a modest gain over a measure most people already have. The included studies used different waist landmarks and different outcome definitions.

Ashwell, Gunn and Gibson, waist-to-height ratio is a better screening tool than waist circumference and BMI for adult cardiometabolic risk factors: systematic review and meta-analysis · Obes Rev 2012;13(3):275-286

Across 120 studies and 236 samples, the measurement protocol had no substantial influence on the association between waist circumference and all-cause or cardiovascular mortality, cardiovascular disease or diabetes. Significant associations were found in 65% of samples, and the common protocols (minimal waist, midpoint between rib and hip, umbilicus) were distributed similarly among the significant and non-significant ones.

Measured in: 120 studies, 236 samples, analyzed by an expert panel across sample size, sex, age, race and ethnicity

This says the site does not change the risk association at population level. It does not say the sites give the same number: they do not, and switching landmarks between your own measurements will produce a change that is entirely artefact. The review also notes that no scientific rationale has been published for any of the protocols recommended by major health authorities.

Ross et al., does the relationship between waist circumference, morbidity and mortality depend on measurement protocol for waist circumference? · Obes Rev 2008;9(4):312-325

Across the top four quintiles of the Sleep Regularity Index against the least regular quintile, all-cause mortality was 20% to 48% lower, cancer mortality 16% to 39% lower and cardiometabolic mortality 22% to 57% lower. Sleep regularity predicted all-cause mortality more strongly than sleep duration did, on both model comparisons run.

Measured in: 60,977 UK Biobank participants, mean age 62.8, 55.0% female, with more than 10 million hours of accelerometer data and 1,859 deaths over a mean 6.3 years

Six years of follow-up in a cohort with a median regularity index of 81 leaves little room to separate irregular sleep from the shift work, illness and social circumstances that produce it. The comparison with duration rests on nested model tests whose p values (0.14 to 0.20) show duration adding nothing, rather than showing regularity to be a cause. UK Biobank participants are healthier and less deprived than the UK population.

What could explain it instead: Observational. Shift work, chronic pain, depression and terminal illness all fragment sleep timing and independently raise mortality, so irregularity partly marks who is unwell or working nights.

Windred et al., sleep regularity is a stronger predictor of mortality risk than sleep duration: a prospective cohort study · Sleep 2024;47(1):zsad253

Average measured sleep was 6.0 hours against 6.8 hours reported, a gap of about 0.8 hours. Reports rose by only 31 minutes for each additional hour of measured sleep, and the correlation between reported and measured duration was 0.45. Modelled, people sleeping 5 hours over-reported by 1.3 hours and those sleeping 7 hours over-reported by 0.3 hours.

Measured in: 669 middle-aged adults at the Chicago site of the CARDIA study, 82% of those invited, measured with three days each of wrist actigraphy, a sleep log and questions about usual sleep duration in two waves

Actigraphy is itself an estimate: it infers sleep from movement and tends to score quiet wakefulness as sleep, so the true gap between belief and sleep may differ from 0.8 hours in either direction. One city, one age band, and three nights per wave.

What could explain it instead: Health, sociodemographic and sleep characteristics all altered the size of the discrepancy in this sample, so the over-reporting is not a constant offset that can be subtracted, and groups differing in health differ in how wrong their estimate is.

Lauderdale et al., self-reported and measured sleep duration: how similar are they? · Epidemiology 2008;19(6):838-845

Epoch-by-epoch sensitivity for detecting sleep was high across all devices (0.93 or better) while specificity for detecting wake was low to medium (0.18 to 0.54). Sleep stage comparisons were mixed, and devices performed worse on the disrupted-sleep night. Most devices matched or beat research actigraphy on sleep and wake; two Garmin devices performed worse.

Measured in: 34 healthy young adults, 22 women, mean age 28.1, tested on three consecutive nights in a sleep laboratory including one disrupted-sleep condition, against polysomnography

Healthy young adults sleeping in a laboratory is close to the easiest condition a sleep tracker will ever face, and the devices already failed at detecting wake. Anyone with insomnia, apnea or fragmented sleep is outside what was tested. Consumer firmware is also revised continuously, so a device tested in 2021 is not the device sold now.

Chinoy et al., performance of seven consumer sleep-tracking devices compared with polysomnography · Sleep 2021;44(5):zsaa291

Across 158 publications covering nine commercial device brands, Fitbit, Apple and Samsung devices counted steps accurately in controlled laboratory conditions. Heart rate accuracy varied by brand, with Apple Watch and Garmin most accurate and Fitbit tending to underestimate. No brand was accurate for energy expenditure.

Measured in: 158 publications examining nine commercial wearable device brands, in laboratory and free-living conditions

Accuracy figures come mostly from controlled laboratory testing, which flatters step counting: slow walking, pushing a trolley and wheelchair use are the conditions where counters fail, and they are underrepresented. Devices are redesigned faster than they are validated, so a brand-level verdict ages quickly.

Fuller et al., reliability and validity of commercially available wearable devices for measuring steps, energy expenditure, and heart rate: systematic review · JMIR Mhealth Uhealth 2020;8(9):e18694

Thirty practitioners rating ten tongue slides reached the 80% agreement threshold on 17.3% of occasions in the first session and 19.1% in the second, and almost all of those were simple yes-or-no questions: with more complex response choices, 80% agreement was reached 5% of the time. In a later study using a formal classification scheme, inter-rater reliability across 17 clinicians was moderate for tongue coating (Gwet AC2 0.49 to 0.55) and fair for tongue body color and other body features (0.34).

Measured in: 30 traditional Chinese medicine practitioners rating 10 tongue slides across two sessions; separately, 17 clinicians in Hong Kong and mainland China rating 24 representative smartphone tongue images

Both studies conclude that the problem is the definitions rather than the observation: agreement rose when practitioners used an explicit operating classification scheme, and the same-rater agreement between looking at a person and looking at the photograph was good to very good. This measures agreement about a description, and it does not test whether tongue signs carry clinical information.

What could explain it instead: Terminology rather than perception. Practitioners trained in different lineages and languages use different operational definitions and different tongue regions, so disagreement about a word is scored as disagreement about a tongue.

Kim, Cobbin and Zaslawski, traditional Chinese medicine tongue inspection: an examination of the inter- and intrapractitioner reliability for specific tongue characteristics · J Altern Complement Med 2008;14(5):527-536 Wang et al., intra-rater and inter-rater reliability of tongue coating diagnosis in traditional Chinese medicine using smartphones: quasi-Delphi study · JMIR Mhealth Uhealth 2020;8(7):e16018

Scored 0 to 10 from how many hand or knee supports you need to lower yourself to the floor and stand back up, lower scores carried higher mortality. Against the top band of 8 to 10, multivariate-adjusted hazard ratios were 5.44 (95% CI 3.1 to 9.5) for scores of 0 to 3, 3.44 (2.0 to 5.9) for 3.5 to 5.5 and 1.84 (1.1 to 3.0) for 6 to 7.5. Each one-point increase carried a 21% improvement in survival.

Measured in: 2,002 adults aged 51 to 80, 68% men, attending the CLINIMEX exercise-medicine clinic in Rio de Janeiro, median follow-up 6.3 years, 159 deaths (7.9%)

The same self-selected clinic population as the ten-second balance finding, so it is not a general-population sample and the confidence intervals are wide. The score combines strength, flexibility and body weight, so a low result does not point to any one cause, and the test asks you to lower yourself to the floor, which is not safe to attempt alone if you are frail or have fallen in the past year.

What could explain it instead: Observational. The conditions that make it hard to get off the floor, arthritis, obesity, and neurological and cardiac disease among them, independently raise mortality, so the score reads existing illness as much as it forecasts new illness.

Brito et al., ability to sit and rise from the floor as a predictor of all-cause mortality · Eur J Prev Cardiol 2014;21(7):892-898

At the usual threshold of 13.5 seconds or more, pooled specificity was 0.74 (95% CI 0.52 to 0.88) and pooled sensitivity 0.31 (95% CI 0.13 to 0.57), so the test misses roughly two thirds of the people who go on to fall. The review's own logistic regression found the score was not a significant predictor of falls (OR 1.01, 95% CI 1.00 to 1.02, p = 0.05).

Measured in: 25 studies reviewed in community-dwelling older adults, of which 10 could be pooled for the meta-analysis

This is about predicting falls in community-dwelling older adults specifically. The test remains a valid measure of mobility, correlating with the Berg Balance Scale and with activities-of-daily-living scores in its original validation, and the review's conclusion is that it should not be used in isolation to identify people at high risk of falling rather than that it measures nothing.

Barry et al., is the Timed Up and Go test a useful predictor of risk of falls in community dwelling older adults: a systematic review and meta-analysis · BMC Geriatr 2014;14:14 Podsiadlo and Richardson, the timed Up and Go: a test of basic functional mobility for frail elderly persons · J Am Geriatr Soc 1991;39(2):142-148

Wearable users reported higher rates of symptom monitoring and preoccupation (p = 0.03) and more treatment concerns (p = 0.02) than non-users. 20% of wearable users experienced anxiety and always contacted their doctor in response to an irregular rhythm notification. After matching, atrial-fibrillation-specific health care use was significantly greater in users (p = 0.04), including more ECGs, echocardiograms and ablations.

Measured in: 172 patients with atrial fibrillation at a US academic center, mean age 72.6, 42% women, of whom 83 used a wearable, followed over 9 months of merged survey and electronic health record data

Retrospective and propensity-matched, so the direction of causation is open: more anxious patients may be more likely to buy a wearable in the first place. 172 people in a US academic health system, and the authors call for prospective randomized study before concluding anything about net effect. This is a cardiac population, and it says nothing directly about readiness scores in healthy users.

What could explain it instead: Selection by health anxiety. People already preoccupied with their rhythm are more likely to buy a monitor and more likely to call a doctor, which would produce this entire association without the device causing anything.

Rosman et al., wearable devices, health care use, and psychological well-being in patients with atrial fibrillation · J Am Heart Assoc 2024;13(15):e033750

Twelve eligible studies were found: three evaluated both intra- and inter-rater reliability and nine only inter-rater. Acceptable agreement was achieved where the method carried concrete operational definitions. Poor agreement traced to unclear definitions and terminology in the classical descriptions and to imprecise descriptions persisting in standardized systems. Most studies did not assess intra-rater reliability at all.

Measured in: 12 studies of manual pulse diagnosis at the radial artery by human testers, published in English

A narrative review with no pooling, over a literature the authors describe as consistently limited by small samples and by testers who often knew the participants beforehand. It reports how far practitioners agree with each other, which is a prerequisite for a diagnostic method rather than evidence about what the method detects.

Bilton and Zaslawski, reliability of manual pulse diagnosis methods in traditional East Asian medicine: a systematic narrative literature review · J Altern Complement Med 2016;22(8):599-609

Core body temperature falls by roughly half a degree to a degree Celsius from its early-evening peak as sleep approaches, and the timing of that fall lines up with sleep onset; a cooler room and warm hands and feet both help the body release the heat that lets it happen.

This is the well-mapped physiology of sleep onset, not a measure of how much any single bedroom adjustment is worth.

Harding, Franks and Wisden, The Temperature Dependence of Sleep · Front Neurosci 2019;13:336

In a controlled physiology study, the degree of warming in the hands and feet, the body's main route for shedding heat, was the strongest single predictor of how quickly people fell asleep, ahead of melatonin level or how sleepy people reported feeling.

A physiological predictor measured in a controlled setting, which explains the direction of the effect rather than sizing any home routine.

Krauchi, Cajochen, Werth and Wirz-Justice, Warm feet promote the rapid onset of sleep · Nature 1999;401(6748):36-37

Pooled across studies, a warm bath or shower at 104–108.5°F (40–42.5°C) taken one to two hours before bed shortened the time to fall asleep by roughly 9 minutes and improved self-rated sleep quality and sleep efficiency.

Measured in: adults across the pooled controlled studies

Effect sizes vary across a modest set of studies, and the benefit is clearest for people who are slow to fall asleep.

Haghayegh et al., Before-bedtime passive body heating by warm shower or bath to improve sleep: a systematic review and meta-analysis · Sleep Med Rev 2019;46:124-135

In a review of thermal-environment sleep studies, heat exposure under normal bedding raised night-time wakefulness and reduced both slow-wave (deep) sleep and REM (dreaming) sleep, with humid heat making it worse; cold was better tolerated once bedding and clothing were used.

A narrative review of the mechanism and the experimental literature, describing the direction and the sleep stages affected rather than a single pooled number.

Okamoto-Mizuno and Mizuno, Effects of thermal environment on sleep and circadian rhythm · J Physiol Anthropol 2012;31(1):14

In a randomized trial of 74 women on hormone-blocking breast-cancer therapy with hot flushes and insomnia, adding a cool pad pillow topper to standard care improved sleep efficiency almost twice as much as standard care alone and improved depression scores, while the reduction in hot flushes did not reach significance (p=0.09).

Measured in: 74 women on endocrine therapy for breast cancer, pre- and postmenopausal

A single small randomized trial in one specific group, so it supports the night-sweat use-case rather than a general sleep benefit.

Marshall-McKenna et al., A randomised trial of the cool pad pillow topper versus standard care for sleep disturbance and hot flushes in women on endocrine therapy for breast cancer · Support Care Cancer 2016;24(4):1821-1829

Measured in older adults' own homes, sleep was most efficient and restful when the bedroom stayed between about 68–77°F (20–25°C), with a 5 to 10 percent drop in sleep efficiency as the room warmed from 77–86°F (25–30°C).

Measured in: community-dwelling older adults, monitored in their own homes

Observational, in older adults, with large person-to-person variation, so the band is a place to start rather than a fixed rule.

What could explain it instead: Bedroom temperature was observed rather than assigned, and it rises with warmer seasons and with behaviours such as leaving windows open or changing bedding, so season and habits move alongside it and cannot be fully separated.

Baniassadi, Manor, Yu, Travison and Lipsitz, Nighttime ambient temperature and sleep in community-dwelling older adults · Sci Total Environ 2023;899:165623

Across a large United States survey, warmer-than-normal nighttime temperatures tracked with more nights of insufficient sleep, with the effect largest among lower-income and older respondents and during summer.

Measured in: roughly 765,000 United States survey respondents

Observational and based on self-reported sleep, so it is a population signal consistent with the experimental work rather than confirmation of cause for an individual.

What could explain it instead: Warmer nights coincide with summer, longer daylight, open-window noise and higher humidity, and sleep was self-reported, so heat cannot be cleanly separated from these other seasonal disruptors.

Obradovich, Migliorini, Mednick and Fowler, Nighttime temperature and human sleep loss in a changing climate · Sci Adv 2017;3(5):e1601555

In a randomized crossover trial, a temperature-controlled mattress cover produced a large improvement in how well people felt they slept and how comfortable they were, while measured sleep from wrist actigraphy and embedded sensors did not significantly change.

Measured in: 34 healthy adults, 20 women and 14 men

The improvement was in how sleep felt, not in measured sleep, and the authors flag expectancy as a likely part of it.

Stevenson, Suppiah, Mundel and Driller, Under the Covers: the effect of a temperature-controlled mattress cover on sleep and perceptual measures in healthy adults · Clocks Sleep 2025;7(4):55

A dual-zone mattress with a cooler center and warmer edges, paired with a pillow that gently warmed the neck, raised the distal-to-proximal skin gradient by about 2 degrees Celsius and lowered core body temperature by about 0.15 degrees Celsius in the first half hour after lights-out, compared with a control night.

Measured in: 11 young healthy male normal sleepers

A tiny early-stage study measuring the thermal signals, not sleep outcomes, and authored in part by the system developers.

Haghayegh et al., Novel temperature-controlled sleep system to improve sleep: a proof-of-concept study · J Sleep Res 2022;31(6):e13662

In a single-arm pilot of 15 perimenopausal and postmenopausal women with frequent hot flashes, eight weeks of a cooling mattress pad system was followed by a 52 percent drop in hot-flash frequency and a fall in the Pittsburgh Sleep Quality Index from 11.1 to 7.9.

Measured in: 15 perimenopausal and postmenopausal women aged 45 to 59

A small single-arm pilot with no control group, so it points to a use-case rather than proving the size of the effect. The cooling devices and an unrestricted gift were provided by the manufacturer.

Avis, Levine and Coeytaux, Results of a pilot study of a cooling mattress pad to reduce vasomotor symptoms and improve sleep · Menopause 2022;29(8):973-978

A traditional Finnish sauna heats the air to roughly 176–212°F (80–100°C) and warms the body from the outside, with the option of raising humidity by throwing water on hot stones. An infrared cabin runs cooler, roughly 113–140°F (45–60°C), using radiant panels that warm the body directly. Both raise core temperature, which is the stimulus the body adapts to, so an infrared cabin can reach that stimulus at a lower, more tolerable air temperature.

This describes how the two work rather than measuring an outcome; the size of any benefit is graded on the outcome claims.

Hussain & Cohen, clinical effects of regular dry sauna bathing: a systematic review · Evid Based Complement Alternat Med 2018;2018:1857413

Heat raises heart rate and lowers blood pressure, so the hotter traditional sauna at 176–212°F (80–100°C) imposes a larger acute cardiovascular load than a cooler infrared cabin. A broad review reports sauna bathing is well tolerated by most healthy people, with the main cautions being dehydration, alcohol, pregnancy, and unstable cardiovascular disease, where either form of heat is a real load.

The strain difference follows from the temperature difference and general heat physiology; individual response varies with health and with how hot and long the session is.

Laukkanen et al., cardiovascular and other health benefits of sauna bathing: a review of the evidence · Mayo Clin Proc 2018;93(8):1111-1121

In a randomized crossover trial in 10 healthy women, an infrared sauna raised tympanic temperature by about 1.05 degrees Celsius yet produced no significant change in blood pressure, arterial stiffness or heart rate variability, unlike a matched bout of exercise. The authors read the effect as thermoregulatory rather than an exercise-level cardiovascular activation, which fits the picture of infrared as a gentler heat load.

A single small crossover study of 10 women, so it maps the acute load rather than settling how infrared compares over weeks.

Hussain et al., infrared sauna as exercise-mimetic? Physiological responses to infrared sauna vs exercise in healthy women: a randomized controlled crossover trial · Complement Ther Med 2022;64:102798

In 25 men with coronary risk factors, two weeks of a daily 140°F (60°C) far-infrared sauna for 15 minutes improved flow-mediated dilation, a measure of how well the arteries open, from about 4.0 to 5.8 percent, moving toward the range seen in healthy controls.

Small, short, and measured in men only, using an artery-function marker rather than a clinical event.

Imamura et al., repeated thermal therapy improves impaired vascular endothelial function in patients with coronary risk factors · J Am Coll Cardiol 2001;38(4):1083-1088

A meta-analysis of nine studies of Waon therapy, a far-infrared protocol of 140°F (60°C) for 15 minutes followed by rest, five times a week for 2 to 4 weeks, found short-term improvements in heart-failure markers: lower B-type natriuretic peptide, a smaller cardiothoracic ratio, and higher left-ventricular ejection fraction. The authors call for long-term studies.

The pooled trials are small, short and largely from one research tradition, and they measure heart-failure markers rather than long-term survival.

Kallstrom et al., effects of sauna bath on heart failure: a systematic review and meta-analysis · Clin Cardiol 2018;41(11):1491-1501

A review of nine papers found limited-to-moderate evidence that far-infrared sauna helps normalize blood pressure and supports the treatment of congestive heart failure, and fair single-study evidence for chronic pain. The same review found the evidence did not support claims that far-infrared sauna lowers cholesterol.

The evidence base is small, several conclusions rest on single studies, so this is an early signal rather than a settled effect.

Beever, far-infrared saunas for treatment of cardiovascular risk factors: summary of published evidence · Can Fam Physician 2009;55(7):691-696

In 34 patients with rheumatoid arthritis or ankylosing spondylitis, eight infrared sessions over four weeks produced significant short-term drops in pain and stiffness, and the treatment was well tolerated with no adverse effects reported. The improvement carried across the whole treatment period was felt but did not reach statistical significance.

Small and uncontrolled, with no comparison group, so the short-term relief is an early signal rather than a proven treatment effect.

Oosterveld et al., infrared sauna in patients with rheumatoid arthritis and ankylosing spondylitis · Clin Rheumatol 2009;28(1):29-34

In 10 active men, far-infrared bathing after an endurance session improved counter-movement jump recovery at 30 minutes compared with no sauna, and heart rate during the session was much lower in the far-infrared cabin (about 71 beats a minute) than in a traditional sauna (about 92), a direct sign of the gentler load. After strength training the infrared cabin added nothing over no sauna.

A very small crossover study of 10 men, so the recovery signal is preliminary and the numbers may not carry to women or to other training.

Mero et al., effects of far-infrared sauna bathing on recovery from strength and endurance training sessions in men · SpringerPlus 2015;4:321

In 2,315 middle-aged Finnish men followed a median of about 21 years, those who used a traditional sauna 4 to 7 times a week had far lower risk than once-a-week users: sudden cardiac death about 63 percent lower (hazard ratio 0.37), with fatal cardiovascular disease and all-cause death also lower. This depth of long-term outcome data sits behind the traditional sauna and has no equivalent for infrared.

This is observational, so it shows an association rather than proving the sauna caused the lower risk, and it was measured only in Finnish men.

What could explain it instead: Healthy-user bias and reverse causation: men who are already healthier and more active can tolerate frequent hot saunas, and men with early illness sauna less, which can inflate the apparent benefit.

Laukkanen et al., association between sauna bathing and fatal cardiovascular and all-cause mortality events · JAMA Intern Med 2015;175(4):542-548

In the same 2,315 Finnish men, using a traditional sauna 4 to 7 times a week tracked with roughly a third the risk of dementia (hazard ratio 0.34) and of Alzheimer's disease (0.35) compared with once-a-week use, over about 21 years. As with the mortality figures, this long-term cognitive outcome data exists for the traditional sauna and not for infrared.

Observational and measured only in Finnish men, so it shows an association rather than a proven cause, and the size may not transfer to other groups.

What could explain it instead: Reverse causation and healthy-user bias: early, undiagnosed cognitive decline can reduce how often someone uses a sauna, which can make frequent use look more protective than it is.

Laukkanen et al., sauna bathing is inversely associated with dementia and Alzheimer's disease in middle-aged Finnish men · Age Ageing 2017;46(2):245-249

75% fewer hot flashes than placebo (95% CI 64.3 to 82.3), about 18 fewer per week. Severity odds ratio 0.13 (0.07 to 0.23).

Measured in: 24 randomized trials, 3,329 participants, oral estrogen and combined estrogen-progestogen

The placebo arms of the same trials improved by 57.7%, so most of what a woman notices in the first weeks of any treatment is not the drug. The review is from 2004 and predates the low-dose and transdermal preparations now most often prescribed.

MacLennan et al., oral oestrogen and combined oestrogen/progestogen therapy versus placebo for hot flushes · Cochrane Database Syst Rev 2004

Placebo arms in the pooled hormone trials cut hot flashes by 57.7% (95% CI 45.1 to 67.7). In a pooled analysis of placebo-treated women, 33% were significantly improved by week 8, and 77% of those were still improved at week 11, three weeks after treatment stopped.

Measured in: 3,329 participants across 24 hormone trials; 247 placebo and 297 actively treated women in the pooled placebo analysis

A placebo arm captures regression to the mean, the natural waxing and waning of symptoms, and the effect of daily symptom diaries, not only expectation. It is why an uncontrolled trial or a personal before-and-after cannot tell you whether something worked.

MacLennan et al., oral oestrogen and combined oestrogen/progestogen therapy versus placebo for hot flushes · Cochrane Database Syst Rev 2004 Freeman et al., placebo improvement in pharmacologic treatment of menopausal hot flashes: time course, duration, and predictors · Psychosom Med 2015

Per 10,000 women per year: 7 more coronary events, 8 more strokes, 8 more pulmonary emboli, 8 more invasive breast cancers, against 6 fewer colorectal cancers and 5 fewer hip fractures. Hazard ratios 1.29 for coronary heart disease, 1.41 stroke, 2.13 pulmonary embolism, 1.26 breast cancer.

Measured in: 16,608 postmenopausal women aged 50 to 79 with an intact uterus, mean 5.2 years of follow-up

Mean age at entry was 63 and most participants were more than a decade past their final period, which is not the woman who asks about hormone therapy for symptoms. One oral formulation was tested: conjugated equine estrogen with medroxyprogesterone acetate. The breast cancer hazard ratio's confidence interval touched 1.00.

Rossouw et al., risks and benefits of estrogen plus progestin in healthy postmenopausal women: principal results from the Women's Health Initiative randomized controlled trial · JAMA 2002

Over 6.8 years: 12 more strokes and 6 fewer hip fractures per 10,000 women per year, and no increase in coronary events (HR 0.91). Breast cancer HR 0.77 (0.59 to 1.01) during the trial and 0.79 (0.65 to 0.97) across 13 years of cumulative follow-up, a reduction rather than an increase.

Measured in: 10,739 postmenopausal women aged 50 to 79 with prior hysterectomy

This arm only describes women without a uterus. A woman with a uterus needs a progestogen to protect the endometrium, and the progestogen is where most of the breast cancer signal sits. The stroke increase appears in both arms and does not go away with age stratification.

Anderson et al., effects of conjugated equine estrogen in postmenopausal women with hysterectomy: the Women's Health Initiative randomized controlled trial · JAMA 2004 Manson et al., menopausal hormone therapy and health outcomes during the intervention and extended poststopping phases of the Women's Health Initiative randomized trials · JAMA 2013

27.1% of women assigned hormone therapy died against 27.6% assigned placebo, HR 0.99 (95% CI 0.94 to 1.03). Cardiovascular mortality HR 1.00, cancer mortality HR 1.03. During the treatment years the ratio of hazard ratios for ages 50 to 59 against 70 to 79 was 0.61 (0.43 to 0.87).

Measured in: 27,347 women aged 50 to 79 at randomization, 18 years of cumulative follow-up, 7,489 deaths

A null on mortality is not a null on every outcome. The same trials found differences in stroke, clot and breast cancer, and a treatment can move those without moving the death rate. The age comparison is a subgroup finding within a trial not designed to test it.

Manson et al., menopausal hormone therapy and long-term all-cause and cause-specific mortality: the Women's Health Initiative randomized trials · JAMA 2017

Five years of therapy started at age 50 adds roughly one breast cancer per 50 users of estrogen with daily progestagen, one per 70 with intermittent progestagen, and one per 200 with estrogen alone, counted across ages 50 to 69. Ten years is about twice that. Relative risk in current users of 5 to 14 years: 2.08 for estrogen-progestagen, 1.33 for estrogen alone.

Measured in: 108,647 postmenopausal women who developed breast cancer in prospective studies, 55,575 of them hormone therapy users, mean age at diagnosis 65

Individual participant data pooled from prospective observational studies, not from randomized trials, and the direction disagrees with the WHI estrogen-alone arm. Vaginal estrogens were the one preparation with no excess. Some excess risk persisted more than a decade after stopping.

What could explain it instead: Detection bias: women on hormone therapy are seen and screened more often, and mammographic density rises on combined therapy, which changes both how much is found and how easily it is found.

Collaborative Group on Hormonal Factors in Breast Cancer, type and timing of menopausal hormone therapy and breast cancer risk: individual participant meta-analysis of the worldwide epidemiological evidence · Lancet 2019

Across 30 randomized trials and 6,235 women, intravaginal estrogen improved symptoms against placebo, with odds ratios between 4.10 and 12.67 depending on preparation. Creams, tablets and rings did not differ from each other in effect.

Measured in: 6,235 postmenopausal women with vaginal atrophy

Most trials ran only about 12 weeks, so long-term endometrial safety rests on observational data rather than on these trials. The outcome measures for dryness and discomfort varied enough that pooling them is approximate.

Lethaby et al., local oestrogen for vaginal atrophy in postmenopausal women · Cochrane Database Syst Rev 2016

Total fractures HR 0.76 (95% CI 0.69 to 0.83) and 5 fewer hip fractures per 10,000 women per year. Total hip bone density rose 3.7% over three years against 0.14% on placebo.

Measured in: 16,608 postmenopausal women aged 50 to 79

The fracture reduction held across women at every level of baseline fracture risk, and it sat inside a trial whose overall balance in that age group was unfavorable. The authors' own conclusion was that there was no net benefit. Bone protection alone is not a reason to start hormone therapy at 70.

Cauley et al., effects of estrogen plus progestin on risk of fracture and bone mineral density: the Women's Health Initiative randomized trial · JAMA 2003

Across 129 studies, 9% of women with postmenopausal bleeding turned out to have endometrial cancer (95% CI 8 to 11), and 91% of women with endometrial cancer had reported bleeding (87 to 93).

Measured in: 34,432 women with postmenopausal bleeding; 6,358 women with endometrial cancer

Most women who bleed do not have cancer. The 9% is the reason every episode is investigated rather than watched, and the 91% is why bleeding is the symptom that finds the disease early. Prevalence varied with hormone therapy use, which causes benign bleeding of its own.

Clarke et al., association of endometrial cancer risk with postmenopausal bleeding in women: a systematic review and meta-analysis · JAMA Intern Med 2018

Stratified by age, estrogen alone at 50 to 59 ran 26 fewer events per 10,000 women per year, against 33 more at 70 to 79, over 13 years of cumulative follow-up.

Measured in: 27,347 women aged 50 to 79 across both Women's Health Initiative hormone trials

These are the 13-year cumulative figures. The intervention-phase numbers, gathered while women were actually taking the hormones, are different and the two are routinely quoted interchangeably. The age gradient is the finding that changed prescribing, and it is an observation within a randomized trial rather than a randomized comparison of ages.

Manson et al., menopausal hormone therapy and health outcomes during the intervention and extended poststopping phases of the Women's Health Initiative randomized trials · JAMA 2013

In women less than 6 years past menopause, carotid intima-media thickness rose 0.0044 mm per year on oral estradiol against 0.0078 on placebo (P=0.008). In women 10 or more years past, 0.0100 against 0.0088 (P=0.29). Interaction P=0.007.

Measured in: 643 healthy postmenopausal women, stratified by time since menopause

Carotid wall thickness is a surrogate. The trial was not powered for heart attacks or strokes and did not measure them as endpoints, so this shows a difference in a marker rather than in events. Participants were healthy volunteers without cardiovascular disease.

Hodis et al., vascular effects of early versus late postmenopausal treatment with estradiol (ELITE) · N Engl J Med 2016

Oral hormone therapy was associated with venous thromboembolism, adjusted odds ratio 1.58 (95% CI 1.52 to 1.64); oral estrogen alone 1.40, oral combined preparations 1.73. Transdermal preparations showed no increase, 0.93 (0.87 to 1.01).

Measured in: 80,396 UK women with venous thromboembolism matched to 391,494 controls in primary care records

Nested case-control drawn from prescribing records, so exposure is what was dispensed rather than what was swallowed or worn. Route was not randomized, and no trial has randomized oral against transdermal with clot as the endpoint.

What could explain it instead: Confounding by indication: prescribers already steer women with obesity, a previous clot or a family history toward patches, so the transdermal group carries a different baseline risk than the oral group. Here that bias would work against the patch, which strengthens the finding rather than explaining it away.

Vinogradova et al., use of hormone replacement therapy and risk of venous thromboembolism: nested case-control studies using the QResearch and CPRD databases · BMJ 2019

Among 45,663 postmenopausal women, users of vaginal estrogen had no higher risk of stroke, heart attack, pulmonary embolism, hip fracture, breast cancer, colorectal cancer or endometrial cancer than non-users. In women with an intact uterus the composite index favored users, adjusted HR 0.68 (95% CI 0.55 to 0.86).

Measured in: 45,663 postmenopausal women aged 50 to 79 in the Women's Health Initiative Observational Study

Observational, not randomized, so it cannot rule out a small effect. The finding describes the low-dose preparations in use during that study rather than every product now sold, and women with a history of breast cancer were not the population studied.

What could explain it instead: Healthy-user selection: a woman prescribed vaginal estrogen has usually been examined, has no contraindication and is in regular contact with a doctor, all of which predict better outcomes on their own and could produce the favorable composite result.

Crandall et al., breast cancer, endometrial cancer, and cardiovascular events in participants who used vaginal estrogen in the Women's Health Initiative Observational Study · Menopause 2018

In its phase 3 trial (SKYLIGHT 1), fezolinetant 45 mg reduced daily hot flashes by about 2.55 in frequency and 2.39 in severity more than placebo by week 12. In its phase 3 trials (OASIS 1 and 2), elinzanetant 120 mg reduced them by about 3.2 by week 12. Together the two drugs enrolled about 1,300 women.

Measured in: About 1,300 women aged 40 to 65 with moderate to severe vasomotor symptoms across the SKYLIGHT 1 and OASIS phase 3 trials

Both drugs were tested against placebo, not against hormone therapy, so nothing here establishes which works better. Fezolinetant carries a liver-monitoring schedule, at baseline and months one, two, three, six, and nine.

Lederman et al., fezolinetant for treatment of moderate-to-severe vasomotor symptoms associated with menopause (SKYLIGHT 1): a phase 3 randomised controlled study · Lancet 2023 Pinkerton et al., elinzanetant for the treatment of vasomotor symptoms associated with menopause: OASIS 1 and 2 randomized clinical trials · JAMA 2024

Group CBT and a self-help CBT booklet both beat usual care on the hot flash and night sweat problem rating at 6 weeks, adjusted mean differences 2.12 and 2.08 on a 10-point scale, holding at 26 weeks. After breast cancer treatment, group CBT gave a mean difference of 1.67 (95% CI 0.91 to 2.43) at 9 weeks.

Measured in: 140 women with problematic hot flashes and night sweats (MENOS 2); 96 women with symptoms after breast cancer treatment (MENOS 1)

What moves is how much the flashes bother you, not how many you have; measured frequency changed much less than the problem rating. Neither trial could blind participants, and both were run by the group that developed the intervention.

Ayers et al., effectiveness of group and self-help cognitive behavior therapy in reducing problematic menopausal hot flushes and night sweats (MENOS 2): a randomized controlled trial · Menopause 2012 Mann et al., cognitive behavioural treatment for women who have menopausal symptoms after breast cancer treatment (MENOS 1): a randomised controlled trial · Lancet Oncol 2012

Telephone-delivered CBT for insomnia dropped the Insomnia Severity Index by 9.9 points against 4.7 in the control arm, a difference of 5.2 points at 8 weeks. By 24 weeks, 84% of the CBT group scored in the no-insomnia range against 43% of controls.

Measured in: 106 perimenopausal and postmenopausal women, mean age 54 to 55, with insomnia and hot flashes

Small, one research network, and the control arm was menopause education rather than an equally intensive sleep program, so some of the gap is attention. Hot flash frequency itself barely changed; what improved was the sleep around them.

McCurry et al., telephone-based cognitive behavioral therapy for insomnia in perimenopausal and postmenopausal women with vasomotor symptoms: a MsFLASH randomized clinical trial · JAMA Intern Med 2016

In a Cochrane review, the sham-controlled comparison pooled 8 trials and 414 women and found no difference from sham needling in hot flash frequency, MD -1.13 per day (95% CI -2.55 to 0.29). Against no treatment, in 3 trials and 463 women, acupuncture did better. The largest single trial, 327 women, found no difference from sham at the end of treatment, MD 0.33 (-1.87 to 2.52).

Measured in: 8 trials and 414 women in the sham-controlled comparison, 3 trials and 463 in the no-treatment comparison; 327 women aged 40 and over in the single largest trial

Against no treatment or a waiting list, acupuncture did better in the same review. That gap between the sham-controlled and untreated comparisons is the signature of a large non-specific response, and it is why a hot-flash trial without a sham arm cannot answer the question. Adverse event reporting was sparse. The sham-controlled estimate rests on 8 trials and 414 women and the no-treatment estimate on only 3 trials and 463, so both carry less weight than the review's headline count suggests.

Dodin et al., acupuncture for menopausal hot flushes · Cochrane Database Syst Rev 2013 Ee et al., acupuncture for menopausal hot flashes: a randomized trial · Ann Intern Med 2016

22 randomized trials and 2,902 women: no difference from placebo in hot flashes per day, MD 0.00 (95% CI -0.88 to 0.89).

Measured in: 2,902 women, most trials conducted in China

Twenty of the twenty-two trials gave one fixed formula to everyone enrolled, which is not how the medicine is prescribed, and no trial at all tested the pattern-matched, individualized prescribing the tradition actually claims works. Adverse events were incompletely reported; those recorded were mild diarrhea, breast tenderness, gastric discomfort and an unpleasant taste.

Zhu et al., Chinese herbal medicine for menopausal symptoms · Cochrane Database Syst Rev 2016

16 randomized trials and 2,027 women: no difference from placebo in hot flash frequency, MD 0.07 per day (95% CI -0.43 to 0.56).

Measured in: 2,027 perimenopausal and postmenopausal women

The reviewers judged safety reporting too poor to draw a conclusion from, which is a separate matter from the efficacy result. Trials used standardized extracts, while retail products vary in species, plant part, extraction and dose.

Leach and Moore, black cohosh (Cimicifuga spp.) for menopausal symptoms · Cochrane Database Syst Rev 2012

Five trials, 733 women: no difference between exercise and no active treatment in the frequency or intensity of vasomotor symptoms, SMD -0.10.

Measured in: 733 perimenopausal and postmenopausal women

This measured hot flashes and nothing else. The same women still get the bone, cardiovascular, sleep and mood effects of exercise, none of which this review looked at. Trials were small and the exercise programs differed enough that pooling them is rough.

Daley et al., exercise for vasomotor menopausal symptoms · Cochrane Database Syst Rev 2014

Median total duration of frequent vasomotor symptoms was 7.4 years, with a median 4.5 years continuing after the final period. Women whose symptoms started while still premenopausal or in early perimenopause had a median duration above 11.8 years; those whose symptoms started after the final period, 3.4 years.

Measured in: 1,449 women with frequent vasomotor symptoms in the Study of Women's Health Across the Nation, followed up to 17 years

This describes women with frequent symptoms, so it is not the average experience of every woman. Duration varied by ethnicity, with a median of 10.1 years in African American participants, and by education and baseline stress.

What could explain it instead: Symptom onset was self-reported at annual visits, so recall sets the start date, and women with the worst symptoms are the most likely to have started hormone therapy and left the untreated analysis, which would shorten the observed duration rather than lengthen it.

Avis et al., duration of menopausal vasomotor symptoms over the menopause transition · JAMA Intern Med 2015

Cumulative 10-year lumbar spine bone density loss was 10.6%, of which 7.38% happened during the transmenopause, the window from one year before the final period to two years after. Femoral neck loss was 9.1%, with 5.8% in that same window.

Measured in: 862 women followed through the decade around their final menstrual period: 242 African American, 384 white, 117 Chinese, 119 Japanese

Rates differed by body size and ancestry: higher BMI and African American heritage went with slower loss, Chinese and Japanese ancestry with faster. Bone density is a risk marker, not a fracture.

What could explain it instead: Ordinary age-related bone loss runs at the same time as the menopause-related loss, and an observational cohort cannot fully separate the two, so some of the attributed loss belongs to aging.

Greendale et al., bone mineral density loss in relation to the final menstrual period in a multiethnic cohort: results from the Study of Women's Health Across the Nation (SWAN) · J Bone Miner Res 2012

Lumbar spine bone density rose 2.9% in the training group against a 1.2% loss in controls (P<0.001); femoral neck 0.3% against a 1.9% loss (P=0.004). The program was 30 minutes twice a week for 8 months.

Measured in: 101 postmenopausal women with osteopenia or osteoporosis, mean age around 65

One site, supervised throughout, and the comparison was a low-intensity home program rather than nothing. The loads were heavy and the technique needs coaching, so this is not a description of what unsupervised gym training does.

Watson et al., high-intensity resistance and impact training improves bone mineral density and physical function in postmenopausal women with osteopenia and osteoporosis: the LIFTMOR randomized controlled trial · J Bone Miner Res 2018

More than 50 published cases of clinically apparent liver injury. Latency typically 2 to 12 weeks, range 1 to 48 weeks, hepatocellular pattern with jaundice, ranging from asymptomatic enzyme rises to acute liver failure needing transplant. LiverTox assigns likelihood category A.

Measured in: Published case reports, almost entirely women taking the product for menopausal symptoms

Causality is contested. Adulteration and misidentified species are documented alternative explanations, and the number of total users is unknown, so no rate can be calculated from these cases. Systematic causality assessment has downgraded many of the reports.

LiverTox: clinical and research information on drug-induced liver injury, black cohosh chapter · National Institute of Diabetes and Digestive and Kidney Diseases, last update 15 April 2025 Teschke et al., herbal hepatotoxicity: suspected cases assessed for alternative causes · Eur J Gastroenterol Hepatol 2013

Cochrane pooled the red clover extract trials and found no difference from placebo, MD -0.93 hot flashes per day (95% CI -1.95 to 0.10). A 2026 review of 9 trials reported a small effect, SMD -0.446 (-0.807 to -0.084).

Measured in: 4,364 participants across 43 phytestrogen trials; 9 trials of women aged 40 to 65 in the later review

The two reviews disagree and overlap in the trials they include. The newer one reports a standardized effect size rather than flashes per day, so it cannot be read as a count, and a standardized difference of 0.45 against a placebo arm that itself falls by half is a small signal.

Lethaby et al., phytoestrogens for menopausal vasomotor symptoms · Cochrane Database Syst Rev 2013 Jiang and Wu, the effectiveness of red clover on hot-flash in menopausal women: a GRADE-assessed systematic review and meta-analysis · Eur J Obstet Gynecol Reprod Biol 2026

In new episodes, average pain fell from 52 out of 100 at onset to 23 at six weeks, 12 at six months and 6 at one year. In already-persistent pain the curve is flatter: 51 at baseline, 33 at six weeks, 26 at six months, 23 at one year.

Measured in: 33 inception cohorts, 11,166 people with acute or persistent low back pain, pooled by mixed models

These are cohorts, not treated-versus-untreated comparisons, so the improvement includes whatever care people happened to receive. Almost all cohorts were recruited in primary care in high-income countries, where most people do get some treatment.

What could explain it instead: Regression to the mean. People enroll in these cohorts at the moment their pain is bad enough to seek care, which is by definition near their personal peak, so some of the fall would occur with no intervention and no natural healing at all.

da C Menezes Costa et al., the prognosis of acute and persistent low-back pain: a meta-analysis · CMAJ 2012;184(11):E613-24

Among people with no back pain, disc degeneration ran 37% at age 20 rising to 96% at age 80; disc bulge 30% to 84%; disc protrusion 29% to 43%; annular fissure 19% to 29%. Prevalence rises steadily with each decade.

Measured in: 3,110 asymptomatic individuals across 33 studies using CT or MRI

The pooled age-specific figures are modeled estimates fitted across studies that used different scanners, sequences and reporting definitions, so a single decade figure carries more uncertainty than the smooth curve suggests. It does not follow that a disc finding is never the cause of a given person's pain; it follows that the finding on its own cannot establish that it is.

Brinjikji et al., systematic literature review of imaging features of spinal degeneration in asymptomatic populations · AJNR Am J Neuroradiol 2015;36(4):811-6

No difference in pain at up to 3 months (SMD 0.19, 95% CI -0.01 to 0.39) or at 6 to 12 months (SMD -0.04, -0.15 to 0.07), and none in function, quality of life, mental health, overall improvement or satisfaction.

Measured in: 6 randomized trials, 1,804 people with low back pain and no features suggesting serious underlying disease

This applies only where red flags are absent, which is the entire premise of the trials. Imaging modality varied between radiography, CT and MRI, and the trials predate current MRI protocols.

Chou et al., imaging strategies for low-back pain: systematic review and meta-analysis · Lancet 2009;373(9662):463-72

Against no treatment, usual care or placebo, exercise reduced pain by 15.2 points on a 0 to 100 scale (95% CI -18.3 to -12.2), which clears the threshold for a change a patient notices. Function improved by only 6.8 points (-8.3 to -5.3), which does not. Against other conservative treatments, pain fell a further 9.1 points and function 4.1, neither clinically important, and there was no difference against manual therapy (MD 1.0).

Measured in: 249 randomized trials in adults with low back pain lasting more than 12 weeks, worldwide

Moderate certainty, downgraded for risk of bias: exercise cannot be blinded, so expectation is present in every trial. The comparison that matters most to a reader, exercise against another active treatment, is the one where the benefit stops being clinically important. Adverse events ran 33% in exercise arms against 29% in comparison arms, mostly increased soreness.

Hayden et al., exercise therapy for chronic low back pain · Cochrane Database Syst Rev 2021;9(9):CD009790

Across 39 trials and 20,827 patients with chronic pain, acupuncture beat no-acupuncture control by about 0.5 standard deviations and sham by about 0.2, with roughly a 15% decay at one year. For back pain specifically, the analysis reports 0.17 (95% CI 0.07 to 0.26) against sham and 0.46 (0.41 to 0.50) against no acupuncture.

Measured in: 39 randomized trials, 20,827 patients, individual patient data rather than published summaries

The back-pain stratum rests on only seven trials against sham. A 0.2 standard deviation advantage over sham is small and it is consistently present rather than absent, which is a different claim from clinically decisive. These three acupuncture rows are not three independent bodies of evidence. The Vickers individual-patient analysis and the Cochrane review share at least six back-pain trials, including GERAC itself, and Cochrane reports that seven German studies account for more than 67% of its participants. Read them as three views of one largely German trial set.

Vickers et al., acupuncture for chronic pain: update of an individual patient data meta-analysis · J Pain 2018;19(5):455-474

Across 229,230 patients receiving an average of 10.2 treatments, 8.6% reported at least one adverse effect and 2.2% one that needed treatment. Bleeding or bruising accounted for 6.1% of patients and 58% of all events, pain 1.7%, vegetative symptoms such as faintness 0.7%. Two patients had a pneumothorax, one requiring hospital treatment. The longest-lasting event was a lower limb nerve lesion at 180 days.

Measured in: 229,230 German patients treated by physicians for osteoarthritis, low back pain, neck pain, headache, allergic rhinitis, asthma or dysmenorrhea

Every practitioner here was a physician working within German insurance-funded care, which is a higher training floor than acupuncture is delivered at in many countries, so this is close to a best case. Adverse effects were self-reported by patients after treatment, which will undercount events noticed later and overcount coincidental symptoms.

What could explain it instead: No control group. Some of the reported symptoms, particularly faintness, tiredness and headache, occur at background rates in people with chronic pain and would have been reported by an untreated comparison group too. Without one, the 8.6% figure attributes all of them to the needling.

Witt et al., safety of acupuncture: results of a prospective observational study with 229,230 patients · Forsch Komplementmed 2009;16(2):91-7

A difference of 0.5 points on a 0 to 100 pain scale against placebo, with 0.4 points on disability and 0.4 on quality of life. Graded high quality. Patients on paracetamol were 3.8 times more likely to have abnormal liver function tests.

Measured in: Randomized placebo-controlled trials of paracetamol in spinal pain and hip or knee osteoarthritis; the back pain analysis rests on three trials

Measured on a 0 to 100 pain scale, where the difference from placebo was about half a point on high-quality evidence. GlaxoSmithKline funded the main trial behind this and four of the review’s authors, and the result still went against paracetamol, which makes it harder rather than easier to dismiss.

Machado et al., efficacy and safety of paracetamol for spinal pain and osteoarthritis · BMJ 2015;350:h1225

Six people need treating for one additional person to reach a clinically important pain reduction (95% CI 4 to 10). In only 3 of 14 analyzes by pain type, outcome and timepoint did the pooled effect even marginally clear a 10-point threshold on a 0 to 100 scale. Gastrointestinal adverse events rose 2.5 fold (95% CI 1.2 to 5.2).

Measured in: 35 randomized placebo-controlled trials in spinal pain, median trial duration 7 days

A median of seven days means the gastrointestinal risk figure is for short courses. Longer use raises it, and adds cardiovascular and renal risk that these trials were far too short to measure. The authors' own conclusion is that NSAIDs are effective but that the difference from placebo is not clinically important.

Machado et al., non-steroidal anti-inflammatory drugs for spinal pain: a systematic review and meta-analysis · Ann Rheum Dis 2017;76(7):1269-1278

Workers scanned within 30 days of onset came off disability at much lower rates and incurred roughly $12,948 to $13,816 more in medical costs, whether or not radiculopathy was present. Even among those with minimal disability the excess cost ran $7,643 to $8,584.

Measured in: 555 workers with acute, work-related, disabling low back pain in a US workers' compensation cohort

Four of the five authors work at an insurer's research institute analyzing that insurer's own claims data, which bears on a claim about time off work.

What could explain it instead: Confounding by indication. Clinicians order early MRI for the patients who look worse or who are more distressed, so some of the longer disability belongs to the reason for the scan rather than to the scan. The authors matched on measured severity, which does not remove unmeasured severity.

Webster et al., iatrogenic consequences of early magnetic resonance imaging in acute, work-related, disabling low back pain · Spine (Phila Pa 1976) 2013;38(22):1939-46

Past malignancy was the single item that meaningfully raised the probability of a spinal tumor, to around 33% in the settings studied. For fracture, long-term corticosteroids raised it to around 33% and visible bruising or grazing over the spine to around 62%. One combination rule reached about 90%, on a confidence interval running from 34% to 99%.

Measured in: 14 diagnostic accuracy studies across primary, secondary and tertiary care

Scoped to fracture and malignancy only. It says nothing about cauda equina or spinal infection, whose red flags rest on guideline consensus rather than on diagnostic accuracy data, which is why the emergency box above does not cite it. Only 5 of the 14 studies examined combinations of flags at all, so the combination figure is one rule from one study rather than a general property.

Downie et al., red flags to screen for malignancy and fracture in patients with low back pain: systematic review · BMJ 2013;347:f7095

For acute low back pain, advice to stay active beat advice to rest in bed on pain (SMD 0.22, 95% CI 0.02 to 0.41) and function (SMD 0.29, 0.09 to 0.49). For sciatica there was little or no difference on either (pain SMD -0.03; function SMD 0.19).

Measured in: 10 randomized trials; the acute back pain comparison rests on 2 trials and 401 people, the sciatica comparison on more

The acute low back pain result is graded moderate quality. Bed rest is the comparator, so this establishes that resting in bed is worse rather than that any particular activity is best.

Dahm et al., advice to rest in bed versus advice to stay active for acute low-back pain and sciatica · Cochrane Database Syst Rev 2010;(6):CD007612

Both CFT arms beat usual care by 4.6 points on the 0 to 24 Roland-Morris activity limitation scale at 13 weeks (95% CI -5.9 to -3.4), and effect sizes were similar at 52 weeks. Societal costs were AU$5,276 to AU$8,211 lower per person. Adding movement sensor biofeedback added nothing.

Measured in: 492 adults with chronic disabling low back pain across Australian primary care physiotherapy

The effect held at 52 weeks, where the table shows differences slightly larger than at 13 weeks, which is unusual in this literature. The developers of the therapy declare speaker fees for teaching it and two are clinical directors of a clinic that uses it, which belongs alongside a result this favorable.

Kent et al., cognitive functional therapy with or without movement sensor biofeedback versus usual care (RESTORE) · Lancet 2023;401(10391):1866-1877

At 26 weeks, clinically meaningful improvement in function reached 60.5% on mindfulness-based stress reduction and 57.7% on CBT, against 44.1% on usual care. Pain bothersomeness improved in 43.6% and 44.9% respectively, against 26.6%. The two active arms did not differ.

Measured in: 342 adults aged 20 to 70 with chronic low back pain, Seattle, eight weekly two-hour groups

Usual care participants received no group contact at all, so attention and group membership are not controlled for. Roughly half of those assigned to the group programs attended six or more of the eight sessions, so this is closer to an offer-of-treatment effect than a full-dose effect.

Cherkin et al., effect of mindfulness-based stress reduction vs cognitive behavioral therapy or usual care on back pain and functional limitations · JAMA 2016;315(12):1240-9

In subacute pain of four weeks to three months, high Fear Avoidance Beliefs Questionnaire scores predicted failure to return to work, with odds ratios from 1.05 to 4.64 across four cohorts of 258 to 1,068 patients. The signal was weak or absent in very acute and in long-established chronic pain.

Measured in: 21 studies from 2,031 screened references, adults with nonspecific low back pain

Prognostic, not causal. The odds ratio range is wide enough that the practical size of the effect is uncertain, and it holds only in the subacute window. Whether treating the belief changes the outcome is a separate question, answered better by the CFT trial than by this review.

What could explain it instead: Pain severity. People whose pain is worse are both more afraid of movement and less likely to return to work, so part of the association is severity showing up twice. Most included cohorts adjusted for baseline pain, and adjustment cannot fully separate the two.

Wertli et al., the role of fear avoidance beliefs as a prognostic factor for outcome in patients with nonspecific low back pain: a systematic review · Spine J 2014;14(5):816-36

Moderate-certainty evidence of greater pain relief and better back-specific function immediately after a course of acupuncture compared with no treatment. Against usual care the results were mixed.

Measured in: 33 trials, 8,270 participants with chronic nonspecific low back pain, in Europe, Asia, North and South America

Compared against no treatment or a waiting list, so it measures the whole encounter rather than the needles, and the trials cannot be blinded on that comparison. These three acupuncture rows are not three independent bodies of evidence. The Vickers individual-patient analysis and the Cochrane review share at least six back-pain trials, including GERAC itself, and Cochrane reports that seven German studies account for more than 67% of its participants. Read them as three views of one largely German trial set.

Mu et al., acupuncture for chronic nonspecific low back pain · Cochrane Database Syst Rev 2020;12(12):CD013814

Cochrane concluded acupuncture may not be more clinically meaningful than sham for pain immediately after treatment. In the largest single trial, response at 6 months reached 47.6% on real acupuncture against 44.2% on sham, a 3.4 point difference that did not reach significance (P = .39), while guideline drug and physiotherapy care reached 27.4%.

Measured in: 33 trials and 8,270 participants for the pooled result; 1,162 German outpatients with a mean of 8 years of back pain for the single trial

Sham acupuncture is not an inert control: it usually means a practitioner, a private room, half an hour of attention and something touching the skin. GERAC found 47.6% responding to real needling against 44.2% to sham, a gap of 3.4 points at P=0.39, while conventional drug and physiotherapy care managed 27.4% (P<0.001 against needling). These three acupuncture rows are not three independent bodies of evidence. The Vickers individual-patient analysis and the Cochrane review share at least six back-pain trials, including GERAC itself, and Cochrane reports that seven German studies account for more than 67% of its participants. Read them as three views of one largely German trial set.

Mu et al., acupuncture for chronic nonspecific low back pain · Cochrane Database Syst Rev 2020;12(12):CD013814 Haake et al., German Acupuncture Trials (GERAC) for chronic low back pain · Arch Intern Med 2007;167(17):1892-8

Pain-related function was the same at 12 months (BPI interference 3.4 opioid against 3.3 non-opioid). Pain intensity was significantly better on non-opioids (BPI severity 4.0 against 3.5, P = .03). Medication side effects were twice as common on opioids (1.8 against 0.9 symptoms).

Measured in: 240 US Veterans Affairs primary care patients with moderate to severe chronic back pain or hip or knee osteoarthritis despite analgesic use; 97.5% completed

A VA population with a much higher burden of psychiatric comorbidity and substance use history than general primary care. It tests a treat-to-target opioid strategy, not any single drug, and it excluded people already on long-term opioids, which is the group clinicians most often ask about.

Krebs et al., effect of opioid vs nonopioid medications on pain-related function: the SPACE randomized clinical trial · JAMA 2018;319(9):872-882

Heat wrap therapy gave a small short-term reduction in pain, significant at five days against placebo in two trials of 258 people, and adding exercise to the heat wrap reduced pain further at seven days. Evidence on cold was insufficient, with only poor-quality studies found.

Measured in: Adults with acute, sub-acute and chronic low back pain across the review’s included trials

This review is from 2006 and the heat wrap trials were largely manufacturer-funded. The effect is small and short-lived, and nothing here establishes a benefit for chronic pain. The absence of good cold trials is a gap in the literature, not a finding that cold does nothing.

French et al., superficial heat or cold for low back pain · Cochrane Database Syst Rev 2006;(1):CD004750

Against other recommended therapies, manipulation gave no clinically meaningful short-term pain difference (MD -3.17, 95% CI -7.85 to 1.51) and a small functional advantage (SMD -0.25, -0.41 to -0.09). Against non-recommended therapies the pain difference was -7.48 points, still below the clinical threshold. One adequately powered trial found no excess adverse events against sham manipulation (RR 1.24, 0.85 to 1.81).

Measured in: 47 randomized trials, 9,211 adults with chronic low back pain, mean ages 35 to 60

The review records one serious adverse event judged possibly related to spinal manipulation, alongside the transient soreness that is common. Two conflicts belong alongside this result: the lead author’s post was funded by European, Belgian and Netherlands chiropractic bodies, and two of the authors practice as chiropractors.

Rubinstein et al., benefits and harms of spinal manipulative therapy for chronic low back pain · BMJ 2019;364:l689

Median time to a recurrence that limited activity was 208 days in the walking group against 112 days in the control group. The program was cost-effective at a willingness-to-pay threshold of AU$28,000 per quality-adjusted life year.

Measured in: 701 Australian adults who had recently recovered from an episode of low back pain; 81% women, mean age 54

The control group received nothing at all, not an alternative program, so attention and monitoring are part of the effect. Lower limb adverse events were more common in the walking group. Six physiotherapist sessions over six months is more support than most people get, and the trial cannot say how much of the benefit came from the walking rather than the coaching.

Pocovi et al., effectiveness and cost-effectiveness of an individualised, progressive walking and education intervention for the prevention of low back pain recurrence (WalkBack) · Lancet 2024;404(10448):134-144

The review judges it likely that earlier surgery is more beneficial for compressed nerves, while stating that both early and delayed surgery may result in improved neurological outcomes. Its stronger conclusion is that severity at the time of surgery, incomplete syndrome against established urinary retention, is probably the most significant determinant of prognosis.

Measured in: A principally qualitative review of the animal and human clinical literature, with no pooling

No randomized evidence exists on timing and none ethically can, so this rests on a qualitative reading. The widely quoted 48-hour window has no strong basis in it. That argues for going sooner rather than for treating any deadline as safe, and it is why the advice above is to go in rather than to wait for an appointment.

Chau et al., timing of surgical intervention in cauda equina syndrome: a systematic critical review · World Neurosurg 2014;81(3-4):640-50 Hoeritzauer et al., what is the incidence of cauda equina syndrome? A systematic review · J Neurosurg Spine 2020;32(6):832-841

Against inactive controls, massage improved function in sub-acute and chronic low back pain (SMD -0.72, 95% CI -1.05 to -0.39, 725 participants). Short-term pain relief was found in acute pain, but the largest of those figures rests on a single trial of 51 people.

Measured in: 25 randomized trials, 3,096 participants, mostly subacute or chronic low back pain

The evidence is graded low to very low throughout, and the benefits are short-term. The strongest-looking acute pain figure comes from one trial of 51 people, so it should not be read as the expected result.

Furlan et al., massage for low-back pain · Cochrane Database Syst Rev 2015;(9):CD001929

Four studies measuring intralumbar angles found no difference in peak lumbar flexion during lifting between people with and without back pain. Seven cross-sectional studies using thoracopelvic angles found people with back pain lifted with 6.0 degrees less lumbar flexion. Nine of eleven studies reported no significant between-group difference.

Measured in: 11 studies of lifting biomechanics in adults with and without low back pain

Low-quality evidence, so this loosens a widely held belief rather than reversing it. This review shares four authors with the cognitive functional therapy trial cited above, so two claims on this page come from the same group rather than from independent lines of work.

Saraceni et al., to flex or not to flex? Is there a relationship between lumbar spine flexion during lifting and low back pain? · J Orthop Sports Phys Ther 2020;50(3):121-130

Topical Capsicum frutescens (cayenne) reduced pain more than placebo. Devil's claw, white willow bark, comfrey, Brazilian arnica and lavender oil appeared to reduce pain more than placebo on weaker evidence. No significant adverse events were recorded in the included trials.

Measured in: 14 randomized trials, 2,050 adults with acute, subacute or chronic nonspecific low back pain

The herbs assessed are cayenne, devil’s claw, willow bark, comfrey, arnica, lavender and Solidago chilensis, which come from American, southern African, European and Brazilian traditions. No Chinese herb or formula appears anywhere in the review, so nothing here transfers to Du Huo Ji Sheng Tang or the blood-stasis formulas. One included trial delivered lavender oil by acupressure rather than topically, which makes it a poor test of the herb alone.

Gagnier et al., herbal medicine for low back pain: a Cochrane review · Spine (Phila Pa 1976) 2016;41(2):116-33

Against a placebo condition, CBT improved the symptoms of the target disorder by a Hedges g of 0.56, a moderate effect, with an odds ratio of 2.97 for treatment response. Effects were large for OCD, generalized anxiety disorder and acute stress disorder, and smaller for PTSD, social anxiety disorder and panic disorder. Exposure-based versions produced larger effects than cognitive or combined versions, though the difference was not statistically significant.

Measured in: 2,835 patients across 41 randomized placebo-controlled trials covering acute stress disorder, GAD, OCD, panic disorder, PTSD and social anxiety disorder

The comparator is an active placebo condition rather than a waitlist, which is the harder test and gives a smaller number than the figures usually quoted for CBT. Tolerability is not uniform across diagnoses: in the PTSD trials, dropout was higher on CBT (29.0%) than on placebo (17.2%).

Carpenter et al., cognitive behavioral therapy for anxiety and related disorders, meta-analysis of randomized placebo-controlled trials · Depress Anxiety 2018

Exposure-based treatment produced large effect sizes against no treatment, and also beat placebo conditions and other active psychotherapies. Treatments involving real contact with the feared thing beat imaginal and virtual-reality exposure at the end of treatment, though not at follow-up. More sessions predicted better outcomes, and the type of phobia did not change the result.

Measured in: 33 randomized treatment studies of specific phobia

Placebo conditions themselves beat no treatment, so part of what any phobia treatment delivers is expectation. The record gives the effects as large without printing the numbers, and the review dates from 2008, so it predates most of the virtual-reality literature.

Wolitzky-Taylor et al., psychological approaches in the treatment of specific phobias, meta-analysis · Clin Psychol Rev 2008

Duloxetine (Hamilton Anxiety mean difference -3.13, 95% CrI -4.13 to -2.13), pregabalin (-2.79), venlafaxine (-2.69) and escitalopram (-2.45) beat placebo with relatively good acceptability. Quetiapine had the largest effect (-3.60) and was poorly tolerated. Paroxetine and the benzodiazepines were effective and also poorly tolerated. Mirtazapine, sertraline, fluoxetine, buspirone and agomelatine looked effective on smaller samples.

Measured in: 25,441 adult outpatients randomized across 89 trials to 22 active drugs or placebo, published 1994 to 2017

Two to three points on the 56-point Hamilton Anxiety scale is a modest average, and an average conceals the split between people who respond well and people who get nothing. Trial durations are short next to how long these drugs are actually taken, and industry sponsorship is common across this literature.

Slee et al., pharmacological treatments for generalised anxiety disorder, systematic review and network meta-analysis · Lancet 2019

St John's wort activates the pregnane-X receptor, which induces CYP3A4 and P-glycoprotein and lowers blood levels of the drugs they clear. Documented interactions include cyclosporine (two heart transplant patients rejected their grafts in 2000), tacrolimus, warfarin, digoxin, simvastatin, indinavir, alprazolam and oral contraceptives. The degree of CYP3A4 induction tracks the hyperforin content of the preparation.

Hyperforin content varies widely between products and is rarely on the label, so the size of any given interaction cannot be predicted from the packet. The serotonin-syndrome risk when St John's wort is combined with an SSRI or SNRI runs by a different route, addition rather than induction, and is not covered by this mechanism. Three of the four authors of the reassuring interaction review are employees of a St John’s wort manufacturer.

Nicolussi et al., clinical relevance of St. John's wort drug interactions revisited · Br J Pharmacol 2020;177(6):1212-1226

Exercise reduced anxiety symptoms more than control conditions, with a standardized mean difference of -0.582 (p = 0.02), a moderate effect.

Measured in: 262 adults across six randomized controlled trials; exercise arm 132 people, mean age 34.7, control arm 130, mean age 37.3

Six trials and 262 people is a small base for a headline number. The confidence interval as printed in the abstract, -1.0 to -0.76, does not contain the reported point estimate of -0.582, so the published interval cannot be read at face value. Exercise trials cannot blind participants, and control arms were usual treatment rather than an equally engaging activity.

Stubbs et al., anxiolytic effects of exercise for people with anxiety and stress-related disorders, meta-analysis · Psychiatry Res 2017

Resistance training reduced anxiety symptoms with an effect size of 0.31 (95% CI 0.17 to 0.44). Healthy participants improved more (0.50) than participants with a physical or mental illness (0.19). Sex, age, program length, session intensity and frequency did not change the result, and neither did whether people actually got stronger.

Measured in: 922 participants across 16 randomized controlled trials, mean age 43, 68% female and 32% male; 486 assigned to resistance training and 436 to non-active control

In people who already had a physical or mental illness the effect was 0.19, which is small, so most of the pooled benefit comes from healthy samples. Strength gains did not track anxiety gains, which means whatever is producing the effect is not the strength itself.

Gordon et al., the effects of resistance exercise training on anxiety, meta-analysis and meta-regression of randomized controlled trials · Sports Med 2017

Eight weeks of mindfulness-based stress reduction was non-inferior to escitalopram 10 to 20 mg. Clinical Global Impression severity fell 1.35 points with MBSR and 1.43 with escitalopram, a difference of -0.07 (95% CI -0.38 to 0.23), inside the pre-set margin of 0.495. Study-related adverse events occurred in 78.6% of the escitalopram group, with 8% stopping because of them, against 15.4% of the MBSR group and nobody stopping.

Measured in: 208 adults completing the trial, mean age 33; 102 assigned to MBSR and 106 to escitalopram

Non-inferiority is not superiority, and the margin was chosen in advance. The MBSR arm was eight weekly classes, a day-long retreat and daily home practice, a time commitment the trial supported and an ordinary week may not. Participants knew which arm they were in.

Hoge et al., mindfulness-based stress reduction vs escitalopram for the treatment of adults with anxiety disorders, randomized clinical trial · JAMA Psychiatry 2023

Response rates were 70.8% (about 71%) for CBT, 54.2% (about 54%) for Kundalini yoga and 33.0% for stress education. Yoga beat the control condition. The pre-specified non-inferiority test did not find yoga as effective as CBT, a difference of 16.6% (p = 0.42 for non-inferiority).

Measured in: 226 adults with primary generalized anxiety disorder, mean age 33.4; 158 women (69.9%) and 68 men (30.1%)

One style, Kundalini yoga, taught by trained instructors over twelve weeks, so it does not describe other styles or a video followed at home. The failed non-inferiority test is the finding, and it is why the authors kept CBT as first-line.

Simon et al., efficacy of yoga vs cognitive behavioral therapy vs stress education for the treatment of generalized anxiety disorder, randomized clinical trial · JAMA Psychiatry 2021

A single 480 mg dose of caffeine, roughly four to five cups of brewed coffee taken at once, triggered a panic attack in 17 of 28 people with panic disorder (60.7%, about 61%) and 10 of 19 with performance-type social anxiety (52.6%), against 4 of 25 with generalized social anxiety (16.0%) and 0 of 26 healthy controls (0%).

Measured in: 98 adults in four groups: panic disorder, generalized social anxiety disorder, performance social anxiety disorder and healthy controls, randomized double-blind against placebo

480 mg at once is a provocation dose chosen to produce an effect, and it says nothing directly about a normal cup of coffee. It shows that susceptibility exists and is concentrated in panic disorder. It does not show that ordinary intake carries that risk.

Nardi et al., panic disorder and social anxiety disorder subtypes in a caffeine challenge test · Psychiatry Res 2009

People with insomnia had roughly three times the odds of developing an anxiety disorder later, OR 3.23 (95% CI 1.52 to 6.85), pooled from six longitudinal studies with at least 12 months of follow-up.

Measured in: Six longitudinal studies of adults for the anxiety outcome, within a review of insomnia as a predictor of several mental disorders

The interval is wide and rests on six studies. Insomnia and early anxiety overlap in how they are measured, so some of the anxiety counted as new may have been present in a subclinical form at baseline. Prediction is not causation, and nothing in this pool tested whether treating the insomnia prevents the anxiety.

Hertenstein et al., insomnia as a predictor of mental disorders, systematic review and meta-analysis · Sleep Med Rev 2019

Sixteen weeks of kava standardized to 120 mg of kavalactones twice daily did not reduce anxiety more than placebo in generalized anxiety disorder, with placebo slightly ahead. Liver function test abnormalities were significantly more frequent on kava, though no participant met the criteria for herb-induced liver injury. Memory impairment and tremor were reported more often on kava.

Measured in: 171 adults with generalized anxiety disorder, randomized double-blind over 16 weeks

The largest trial of kava for anxiety, though not the longest: an earlier trial ran 25 weeks. It was co-sponsored by industry and still returned a null, which makes the null more credible rather than less. Liver function abnormalities were significantly more frequent on kava, in an aqueous extract.

Sarris et al., kava for generalised anxiety disorder, a 16-week double-blind randomised placebo-controlled study · Aust N Z J Psychiatry 2020

More than 100 published cases of liver injury, including acute liver failure and deaths. The standard reference database scores the likelihood of kava causing clinically apparent liver injury at its highest category.

Measured in: Published case reports and regulatory case series worldwide, mainly from ethanolic and acetonic extracts sold in Europe

The reassuring version of this story, that only solvent-extracted preparations are implicated, is contradicted by the source usually cited for it: injury occurred independently of the solvent used. The aqueous extract that story treats as safe is the one that produced significantly more liver function abnormalities in the trial cited above. Causality in individual cases is contested and the absolute rate is low against how widely kava is taken. Neither of those makes the traditional-preparation-is-safe claim supportable.

LiverTox: clinical and research information on drug-induced liver injury, Kava Kava monograph, NIDDK · LiverTox, NIH National Library of Medicine, updated 2018 Teschke, kava hepatotoxicity, a clinical review · Ann Hepatol 2010;9(3):251-265

Adding CBT to a gradual taper roughly doubled the chance of discontinuing: risk ratio 1.96 (95% CI 1.29 to 2.98) at three months, number needed to treat 3.2, and 2.16 (95% CI 1.41 to 3.32) at six to twelve months, number needed to treat 2.8.

Measured in: Three randomized controlled trials in patients with anxiety disorders taking benzodiazepine anxiolytics

Three trials is a thin base for a pooled risk ratio and the authors say larger studies are needed. It shows that tapering goes better with structured support, and it says nothing about how many of those prescriptions should have been written.

Takeshima et al., does cognitive behavioral therapy for anxiety disorders assist the discontinuation of benzodiazepines, systematic review and meta-analysis · Psychiatry Clin Neurosci 2021;75(4):119-127

Hyperventilating before a breath-hold lowers carbon dioxide without adding meaningful oxygen, so the urge to breathe arrives only after oxygen has fallen far enough to cause loss of consciousness. Underwater this is fatal without immediate rescue, and it happens at any depth in any body of water.

Measured in: Case reports, autopsy series and forensic reconstructions of breath-hold swimming and diving fatalities

This rests on case reports, forensic reconstructions and respiratory physiology rather than on any trial, and there is no denominator, so the risk per episode is not quantified. The mechanism itself is not in dispute.

Bart, Murray and Lau, shallow water blackout, StatPearls · StatPearls [Internet], Treasure Island (FL), updated 2026

Therapist-supported internet-delivered CBT produced clinically important improvement in anxiety far more often than a waiting list or information-only control, risk ratio 3.75 (95% CI 2.51 to 5.60) across 11 trials and 866 participants, and lowered disorder-specific symptoms with a large effect, SMD -1.06 (95% CI -1.29 to -0.82) across 28 trials and 2,147 participants. Against face-to-face CBT it showed no clear difference in clinically important improvement, risk ratio 1.09 (95% CI 0.89 to 1.34) across 4 trials and 365 participants.

Measured in: 3,214 adults across 38 randomized controlled trials covering social phobia, panic disorder, generalized anxiety disorder, PTSD, OCD and specific phobia

The Cochrane authors graded the evidence low to moderate quality, largely because participants and outcome assessors cannot be blinded in psychotherapy trials, and the head-to-head against face-to-face CBT rests on 4 trials and 365 people. The programs carried therapist support by email or phone, which is a different thing from an unguided self-help app. Adverse events were rarely reported.

Olthuis et al., therapist-supported internet cognitive behavioural therapy for anxiety disorders in adults, Cochrane systematic review with meta-analysis · Cochrane Database Syst Rev 2016

Measured against non-mindful exercise rather than against nothing, tai chi came out ahead on anxiety by d = 0.28 (95% CI 0.08 to 0.48). Depression improved by d = 0.20 and general mental health by d = 0.40.

Measured in: 950 participants across 10 trials for the anxiety outcome, within a review of 23 randomized controlled trials and 4,370 participants

Comparing against another form of exercise is the demanding test, and the interval nearly touches zero. Style, session length, teacher and duration vary a lot between the included trials, and the authors call for better-standardized interventions. Nobody in these trials was blinded.

Yin et al., comparative effects of tai chi versus non-mindful exercise on anxiety, depression and general mental health, systematic review and meta-analysis · J Affect Disord 2023

Pooled across 20 randomized trials, breathwork improved anxiety with a small effect, g = -0.32. Subjective stress improved by g = -0.35 across 12 trials and 785 adults, and depressive symptoms by g = -0.40 across 18 trials.

Measured in: 785 adults for the primary stress outcome; the anxiety and depression analyzes draw on 20 and 18 trials respectively

Most included studies carried a moderate risk of bias and the authors themselves urged caution and called for low-risk designs. The category pools slow paced breathing with fast and hold-based techniques, which are different physiology, so the pooled number does not belong to any single method. Control conditions were mostly inactive, so attention and expectation are inside the effect.

Fincham et al., effect of breathwork on stress and mental health, meta-analysis of randomised controlled trials · Sci Rep 2023

Across 63 studies, reduced alcohol consumption was associated with improvement in anxiety and depression symptoms, fewer withdrawal symptoms, fewer psychiatric episodes, shorter inpatient stays, lower psychosocial stress and better mental quality of life.

Measured in: 63 studies in hazardous, harmful and alcohol-dependent drinkers, including people with psychiatric comorbidity

The review pools trials with observational studies and reports no single pooled effect size, so the size of the anxiety change is not established. People who succeed in cutting down differ from those who do not in motivation, severity and social support, and that difference travels with the result.

What could explain it instead: Anxiety that lifts when someone cuts down may have been withdrawal anxiety between drinks all along, which makes the improvement the removal of a cause rather than the treatment of a disorder.

Charlet and Heinz, harm reduction, a systematic review on effects of alcohol reduction on physical and mental symptoms · Addict Biol 2017;22(5):1119-1159

Manual acupuncture beat sham acupuncture on anxiety scores, SMD -1.06 (95% CI -1.74 to -0.39), and beat usual care or waitlist, SMD -1.35 (95% CI -2.26 to -0.44). The advantage over sham persisted at follow-up (SMD -0.78); the advantage over usual care did not. Seventy-nine adverse events were reported, all transient discomfort, minor bleeding or local pain.

Measured in: 1,462 participants across 20 randomized controlled trials; 14 trials rated low risk of bias, the rest with allocation-concealment or blinding concerns

An SMD above 1.0 against sham is larger than most drug treatments manage against placebo, which is a reason to treat the number with caution. Heterogeneity across trials is high, sham needling is not an inert control, and most follow-up is short.

Jang et al., acupuncture for anxiety, systematic review and meta-analysis of randomized controlled trials · J Clin Psychol 2026

Compared against anxiolytic drugs rather than against placebo, Chinese herbal medicine improved Hamilton Anxiety scores by a further 1.50 points (95% CI 0.78 to 2.21) in generalized anxiety disorder, with fewer adverse events (incidence rate ratio 0.33, 95% CI 0.24 to 0.45).

Measured in: 92 randomized controlled trials of Chinese herbal medicine against anxiolytics, screened from 9,805 reports across English and Chinese-language databases to January 2025

The comparator is an active drug, so this measures a difference between two treatments and not a difference against no treatment. A 1.50-point gain on the 56-point Hamilton scale sits below what is usually called clinically meaningful. No trial at low risk of bias tested Chinese herbal medicine for panic disorder or OCD, so the result covers generalized anxiety only.

Birling et al., Chinese herbal medicine for anxiety disorders and obsessive-compulsive disorder, systematic review with meta-analysis · J Psychiatr Res 2025

Across 20 studies, nearly all found significant comorbidity between anxiety disorders and thyroid disorders. About half found subtle thyroid dysfunction, mainly a blunted response to thyroid hormone stimulation testing, and self-reported anxiety ran inversely with TSH. The authors support routine screening for thyroid disorders in patients with anxiety disorders.

Measured in: 20 studies of hypothalamic-pituitary-thyroid axis function in adults with anxiety disorders

The pattern is subtle and the studies are small and varied. This supports checking thyroid function once rather than treating a borderline number as the explanation. An overactive thyroid produces anxiety symptoms directly; a marginal TSH in an anxious person usually does not.

What could explain it instead: Anxious arousal itself alters HPA and HPT axis output, so a lower TSH in an anxious person can be a consequence of the anxiety rather than its cause.

Fischer and Ehlert, hypothalamic-pituitary-thyroid (HPT) axis functioning in anxiety disorders, systematic review · Depress Anxiety 2018

Xiao Yao San added to an anxiolytic beat the anxiolytic alone on response rate (RR 1.19, 95% CI 1.13 to 1.26) with fewer adverse events (RR 0.44, 95% CI 0.28 to 0.82). Xiao Yao San alone against anxiolytics alone gave RR 5.41 (95% CI 2.23 to 13.11).

Measured in: 1,256 participants across 14 randomized controlled trials

Eight of the fourteen trials were at high risk of bias and the remaining six at moderate. A relative risk of 5.41 for a herbal formula against an active drug is far outside anything the comparable literature produces, which points at trial quality rather than at the formula. Xiao Yao San is also prescribed for one pattern, and these trials recruited by Western diagnosis. That figure rests on three of the fourteen trials and 324 patients, not on the full set.

Wang et al., efficacy and safety of Xiao Yao San for treating anxiety, systematic review with meta-analysis and trial sequential analysis · Front Pharmacol 2023

Across 108 studies and 47,989 participants, for moderate apnea (AHI 15 or more) pooled sensitivity and specificity were 90% and 36% for STOP-BANG, 77% and 44% for the Berlin questionnaire, and 47% and 62% for the Epworth Sleepiness Scale. For mild apnea (AHI 5 or more), STOP-BANG was 88% sensitive and 42% specific, the Berlin 76% and 59%, and Epworth 54% and 65%.

Measured in: 47,989 participants across 108 diagnostic accuracy studies, mostly clinic populations referred for suspected apnea

The review found sex to be a significant moderator of both sensitivity and specificity without publishing stratified estimates, which is a stronger version of this page’s point than saying the question went unexamined. On the instrument itself, only the neck circumference and male sex items are unambiguously weighted toward men: by our own cited clinic series, women presenting with apnea are older and heavier, so the age and BMI items score in their favor.

Chiu et al., diagnostic accuracy of the Berlin questionnaire, STOP-BANG, STOP and Epworth sleepiness scale in detecting obstructive sleep apnea: a bivariate meta-analysis · Sleep Med Rev 2017;36:57-70

The AASM clinical practice guideline recommends polysomnography, or home sleep apnea testing with a technically adequate device, for uncomplicated adults with signs and symptoms of moderate to severe apnea, and recommends (strong) against using clinical tools, questionnaires or prediction algorithms to diagnose apnea in the absence of a sleep study. If a single home test is negative, inconclusive or technically inadequate, polysomnography should be performed.

Measured in: Adults being assessed for obstructive sleep apnea; GRADE assessment of the underlying diagnostic literature

A guideline is a synthesis with expert judgement layered on top, and the underlying evidence for several recommendations was rated low certainty. The home-testing recommendation is limited to uncomplicated patients: significant cardiorespiratory disease, neuromuscular weakness, opioid use, hypoventilation or suspected non-respiratory sleep disorders take you back to the laboratory study.

Kapur et al., clinical practice guideline for diagnostic testing for adult obstructive sleep apnea, American Academy of Sleep Medicine · J Clin Sleep Med 2017;13(3):479-504

Positive airway pressure compared with no treatment produced clinically significant reductions in disease severity, sleepiness, blood pressure and motor vehicle accidents, and clinically significant improvement in sleep-related quality of life. Auto-adjusting and bilevel modes gave no clinically significant advantage over standard CPAP; nasal masks gave better adherence and slightly greater severity reduction than oronasal masks.

Measured in: 336 studies met inclusion, 184 provided data suitable for meta-analysis; adults with obstructive sleep apnea

The comparator is no treatment, so this measures what the therapy does when used rather than what it delivers across everyone prescribed it. Sleepiness gains are largest in people who were sleepy to begin with, and a large share of people who meet the AHI threshold are not. Trial adherence is generally better than adherence at home.

Patil et al., treatment of adult obstructive sleep apnea with positive airway pressure, AASM systematic review, meta-analysis and GRADE assessment · J Clin Sleep Med 2019;15(2):301-334

Across 51 trials and 4,888 patients, CPAP lowered systolic blood pressure by 2.5 mmHg (95% CI 1.5 to 3.5) and diastolic by 2.0 mmHg against inactive control. Mandibular advancement devices lowered systolic by 2.1 mmHg and diastolic by 1.9 mmHg, with no significant difference between the two treatments. Each additional hour of nightly CPAP use was associated with a further 1.5 mmHg systolic reduction.

Measured in: 4,888 patients across 51 randomized trials of CPAP or mandibular devices in obstructive sleep apnea

2.5 mmHg is a small average, roughly a third of what a first-line antihypertensive delivers, and the dose-response with hours of use means the average hides people who got considerably more and people who got nothing. Most included trials were short, so this does not establish that the reduction persists for years. The two treatments matching is likely to be reduced airway benefit from the device offset by more hours of use.

Bratton et al., CPAP vs mandibular advancement devices and blood pressure in patients with obstructive sleep apnea: a systematic review and meta-analysis · JAMA 2015;314(21):2280-93

2,717 adults aged 45 to 75 with moderate to severe apnea and established coronary or cerebrovascular disease were randomized to CPAP plus usual care or usual care alone. Mean AHI fell from 29.0 to 3.7 events per hour. Over 3.7 years the primary composite (cardiovascular death, myocardial infarction, stroke, hospitalisation for unstable angina, heart failure or TIA) occurred in 17.0% on CPAP and 15.4% on usual care, hazard ratio 1.10 (95% CI 0.91 to 1.32). Snoring, daytime sleepiness, quality of life and mood all improved significantly.

Measured in: 2,717 adults with moderate to severe OSA and established cardiovascular disease, across seven countries

Mean CPAP use was 3.3 hours a night, below the 4 hours conventionally used to define adherence, so this is a test of CPAP as people actually used it. Participants were selected for minimal sleepiness, because randomizing a sleepy patient to no treatment for four years is not permissible, and sleepiness may be the marker of who benefits. This is secondary prevention in people who already had cardiovascular disease, not a statement about untreated apnea in the general population.

McEvoy et al., CPAP for prevention of cardiovascular events in obstructive sleep apnea (SAVE) · N Engl J Med 2016;375(10):919-31

Ten randomized trials, 7,266 participants, 80.5% men, mean age 60.9. Positive airway pressure showed no significant association with major adverse cardiovascular events (RR 0.77, 95% CI 0.53 to 1.13), cardiovascular death (RR 1.15, 95% CI 0.88 to 1.50) or all-cause death (RR 1.13, 95% CI 0.99 to 1.29). Meta-regression found no association between outcome and apnea severity, follow-up duration or adherence to the device.

Measured in: 7,266 participants across 10 trials (9 CPAP, 1 adaptive servo-ventilation), predominantly men with existing cardiovascular disease

The absence of an adherence signal in meta-regression is the part that complicates the usual explanation for these nulls. The confidence interval on major events (0.53 to 1.13) is wide enough to include a meaningful benefit as well as no effect, so this is an unresolved question rather than a settled negative. Every included trial enrolled people with existing cardiovascular disease and low sleepiness.

Yu et al., association of positive airway pressure with cardiovascular events and death in adults with sleep apnea: a systematic review and meta-analysis · JAMA 2017;318(2):156-166

In adults with heart failure with reduced ejection fraction and predominantly central sleep apnea, adaptive servo-ventilation brought mean AHI to 6.6 events per hour at 12 months but did not change the primary composite endpoint (54.1% vs 50.8%, hazard ratio 1.13, 95% CI 0.97 to 1.31). All-cause mortality was higher on adaptive servo-ventilation (hazard ratio 1.28, 95% CI 1.06 to 1.55) as was cardiovascular mortality (hazard ratio 1.34, 95% CI 1.09 to 1.65).

Measured in: Adults with symptomatic heart failure with reduced ejection fraction and predominantly central sleep apnea

This applies specifically to adaptive servo-ventilation in reduced-ejection-fraction heart failure with predominantly central events. It does not transfer to CPAP, and it does not transfer to obstructive apnea, where the treatment picture is different. The mechanism of the excess mortality was not established by the trial, and the finding remains debated. The trial was funded by the device manufacturer, which makes an unfavorable result harder to dismiss rather than easier.

Cowie et al., adaptive servo-ventilation for central sleep apnea in systolic heart failure (SERVE-HF) · N Engl J Med 2015;373(12):1095-105

In a sample of 4,925 employed adults, with in-laboratory polysomnography on a subset of 1,090, an estimated 93% of women and 82% of men with moderate to severe sleep apnea syndrome had never been clinically diagnosed.

Measured in: 4,925 employed adults in Wisconsin with good access to healthcare, 1,090 of whom had laboratory sleep studies

These are 1997 figures from a single employed US population, and diagnosis has become easier since home testing became routine, so the absolute rates are dated. The 11-point sex gap is the part that later presentation studies keep reproducing. Sex-specific substudies do exist on weight and on glycemia, including one in which women on usual care fared worse. What no trial reports by sex is its CARDIOVASCULAR outcome, which is the narrower and defensible version of this gap.

What could explain it instead: Diagnosed status came from self-reported doctor diagnosis, so anyone told they had 'a snoring problem' without a formal label counts as undiagnosed. Health-care-seeking behavior also differs by sex independently of symptom severity, which contributes to the gap without being a failure of the clinician.

Young et al., estimation of the clinically diagnosed proportion of sleep apnea syndrome in middle-aged men and women · Sleep 1997;20(9):705-6

Among 2,827 patients with diagnosed obstructive sleep apnea (2,052 men, 775 women), men reported witnessed apneas more often, while nocturnal choking, morning headache, fatigue, insomnia symptoms, impaired memory, mood disturbance, reflux and nocturia were all more frequent in women (all p<0.0001). Women were 5.7 years older at diagnosis (56.1 vs 50.4) and heavier (BMI 36.3 vs 31.8) despite lower AHI, with higher REM AHI and lower supine AHI.

Measured in: 2,827 consecutive sleep-clinic patients with polysomnography-confirmed OSA

A sleep-clinic sample describes women who were referred; women who never got a referral are absent by construction, which would if anything widen the real gap. The women were also older and heavier than the men, and some of the symptom differences could follow from that rather than from sex.

What could explain it instead: Age and body mass index differed substantially between the sexes in this sample, and both independently produce fatigue, nocturia and mood symptoms, so the symptom profile is not cleanly attributable to sex alone.

Gender differences in clinical and polysomnographic features of obstructive sleep apnea: a clinical study of 2827 patients · Sleep Breath 2018;22(1):241-249

In 589 women with laboratory polysomnography, adjusted odds of an AHI of 15 or more were 3.5 times higher after menopause than before it (95% CI 1.4 to 8.8), and 2.6 times higher for an AHI of 5 or more (95% CI 1.4 to 4.8). In a separate two-phase population sample, clinically defined sleep apnea affected 0.6% of premenopausal women and 2.7% of postmenopausal women not taking hormone therapy, against 3.9% of men.

Measured in: 589 women in the Wisconsin Sleep Cohort; and 1,000 women plus 741 men in a Pennsylvania population sample

Both are cross-sectional, so this describes an association with menopausal status rather than a transition observed in the same women over time. The hormone therapy finding in the second study is observational, and women who take hormone therapy differ from women who do not in ways that also affect apnea risk.

What could explain it instead: Menopausal status travels with age and with weight gain, both of which independently raise apnea risk. The Wisconsin estimates are adjusted for age, body habitus and smoking, and statistical adjustment does not fully separate variables that move together this tightly. Menopausal status was assigned partly from self-reported vasomotor symptoms.

Young et al., menopausal status and sleep-disordered breathing in the Wisconsin Sleep Cohort Study · Am J Respir Crit Care Med 2003;167(9):1181-5 Bixler et al., prevalence of sleep-disordered breathing in women: effects of gender · Am J Respir Crit Care Med 2001;163(3 Pt 1):608-13

Pooled across controlled case-control and cohort studies, the mean crash-rate ratio associated with obstructive sleep apnea is likely to fall in the range 1.21 to 4.89, roughly 1.2 to 4.9 times. Body mass index, apnea-hypopnea index, oxygen saturation and possibly daytime sleepiness predicted crash within the apnea group.

Measured in: Controlled studies of drivers with obstructive sleep apnea, with a focus on commercial motor vehicle operators

The review reports a range rather than a point estimate because heterogeneity across studies was high: they agree on direction and disagree on size. All included studies are observational, and drivers with apnea differ from those without in body mass index, age, shift patterns and accumulated sleep debt, any of which independently raises crash risk.

What could explain it instead: Obesity, age, occupational driving hours and chronic sleep restriction all travel with apnea and all independently raise crash risk, so part of the pooled ratio is attributable to the company apnea keeps.

Tregear et al., obstructive sleep apnea and risk of motor vehicle crash: systematic review and meta-analysis · J Clin Sleep Med 2009;5(6):573-81

Across nine observational studies of drivers with obstructive sleep apnea compared before and after CPAP, crash risk fell to about a quarter of the pre-treatment rate (risk ratio 0.278, 95% CI 0.22 to 0.35). Daytime sleepiness improves significantly after a single night of treatment and simulated driving performance improves within two to seven days.

Measured in: Drivers with obstructive sleep apnea across nine before-and-after observational studies

Every included study is a before-and-after comparison with no control group, so regression to the mean, changed driving behavior after a diagnosis, and the tendency to crash less in the period following any medical scare are all in the estimate. A risk ratio of 0.278 from uncontrolled data should be read as a clear direction rather than a precise size. No crash data were available to establish how quickly the reduction appears.

What could explain it instead: People starting CPAP have just been diagnosed with a condition they were told affects their driving, and diagnosis alone changes driving behavior independently of the therapy.

Tregear et al., continuous positive airway pressure reduces risk of motor vehicle crash among drivers with obstructive sleep apnea: systematic review and meta-analysis · Sleep 2010;33(10):1373-80

In 709 Wisconsin Sleep Cohort participants with baseline polysomnography, adjusted odds of hypertension at four years rose with baseline AHI: 1.42 (95% CI 1.13 to 1.78) at 0.1 to 4.9 events per hour, 2.03 (1.29 to 3.17) at 5.0 to 14.9, and 2.89 (1.46 to 5.64) at 15 or more, relative to no events. Adjusted for baseline hypertension, body mass index, neck and waist circumference, age, sex, alcohol and cigarettes.

Measured in: 709 adults in the Wisconsin Sleep Cohort followed four years, 184 of them for eight

Observational with a four-year window, so this establishes a dose-response association rather than causation. The dose-response and the adjustment for three separate measures of body habitus are what make it more persuasive than the cross-sectional literature it replaced, and neither removes the objection.

What could explain it instead: Obesity causes both sleep apnea and hypertension, and it is the standing objection to this whole literature. Adjusting for body mass index plus neck and waist circumference is the strongest available response and leaves residual confounding by body composition and visceral fat distribution untouched.

Peppard et al., prospective study of the association between sleep-disordered breathing and hypertension · N Engl J Med 2000;342(19):1378-84

Obese adults with type 2 diabetes randomized to an intensive lifestyle intervention lost 23.8 lb (10.8 kg) at one year against 1.3 lb (0.6 kg) with diabetes support and education, and their AHI fell by an adjusted 9.7 events per hour more than the control group's. More than three times as many reached total remission of apnea, and severe apnea was half as prevalent in the intervention group. Weight loss of 22 lb (10 kg) or more gave the largest AHI reductions.

Measured in: Obese adults with type 2 diabetes and objectively measured sleep apnea, the Sleep AHEAD sub-study of Look AHEAD

Everyone enrolled had both obesity and type 2 diabetes, so the result does not transfer to people with apnea from craniofacial anatomy at a normal weight. Remission was the exception rather than the rule: most participants still had apnea after losing 23.8 lb (10.8 kg). Lifestyle weight loss at this scale is difficult to maintain outside a trial that funds the coaching and the contact hours.

Foster et al., a randomized study on the effect of weight loss on obstructive sleep apnea among obese patients with type 2 diabetes: the Sleep AHEAD study · Arch Intern Med 2009;169(17):1619-26

Apnea-hypopnea index fell by 20.0 and 23.8, about 20 to 24, events per hour against placebo across the two trials.

Measured in: 469 adults with moderate to severe OSA and obesity, 142 women and 327 men, over 52 weeks

Placebo-adjusted, which is the comparison that matters; the within-arm changes are larger and are the numbers usually quoted. Sponsored by the manufacturer, entry BMI 30 or above, and 142 of 469 participants were women. A surrogate endpoint: nothing here measures symptoms or events.

Malhotra et al., tirzepatide for the treatment of obstructive sleep apnea and obesity (SURMOUNT-OSA) · N Engl J Med 2024;391(13):1193-1205 SURMOUNT-OSA baseline sex breakdown, posted results for NCT05412004 · ClinicalTrials.gov, NCT05412004

126 participants, 83% men, mean age 54.5, mean BMI 28.4. Median AHI fell 68% from 29.3 to 9.0 events per hour at 12 months, and the oxygen desaturation index fell 70% from 25.4 to 7.4. In the randomized therapy-withdrawal phase, the group switched off went from 7.6 at 12 months back to a mean AHI of 25.8, while the maintenance group stayed at 8.9. Procedure-related serious adverse events under 2%.

Measured in: 126 adults who could not accept or adhere to CPAP, with BMI under 32 and a favorable airway collapse pattern on drug-induced sleep endoscopy

The 12-month result comes from a single-group design with no control arm, so only the withdrawal sub-study is controlled. The eligibility criteria are the fine print: CPAP failure, BMI under 32, and a specific pattern of collapse on endoscopy exclude a large share of the people in whom apnea is diagnosed. 83% of participants were men. This is a surgically implanted device with a battery that eventually needs replacing.

What could explain it instead: Participants were selected for a favorable airway phenotype and then followed without a comparison group, so the response rate reflects the selection as much as the device.

Strollo et al., upper-airway stimulation for obstructive sleep apnea (STAR trial) · N Engl J Med 2014;370(2):139-49

102 adults with moderate or severe apnea in whom conventional treatment had failed, randomized to modified uvulopalatopharyngoplasty plus tongue volume reduction or to ongoing medical management. AHI went 47.9 to 20.8 with surgery and 45.3 to 34.5 with medical management, a baseline-adjusted between-group difference of -17.6 events per hour (95% CI -8.4 to -26.8). Epworth scores went 12.4 to 5.3 and 11.1 to 10.5, a difference of -6.7 points.

Measured in: 102 adults, 18 of them women, mean age 44.6, all of whom had already failed device treatment

102 people across a small number of centers, open-label with no sham surgery, so the 6.7-point Epworth difference carries whatever a major operation contributes by expectation. The mean AHI after surgery was still 20.8, which remains moderate apnea. Two of 51 in the surgical arm had a serious adverse event, one a myocardial infarction on day 5. Only 18% of participants were women.

Mackay et al., effect of multilevel upper airway surgery vs medical management on the apnea-hypopnea index and patient-reported daytime sleepiness (SAMS randomized clinical trial) · JAMA 2020;324(12):1168-1179

Across 21 comparative epidemiological studies, higher alcohol consumption was associated with a 25% higher risk of sleep apnea (RR 1.25, 95% CI 1.13 to 1.38). The estimate held across different definitions of both exposure and outcome, and across study designs. In a separate eight studies, mean alcohol intake was two units a week higher in people with apnea, which was not statistically significant.

Measured in: Adults across 21 observational studies, 1985 to 2015

Heterogeneity was high (I2 82%) and Egger's test showed evidence of publication bias (p = 0.001), so treat the exact figure loosely and the direction as reasonably settled. All studies are observational: alcohol travels with weight, smoking and late bedtimes. The pooled estimate cannot separate a drink tonight worsening tonight's breathing from long-term drinking raising underlying risk, and the mechanism (reduced pharyngeal dilator muscle tone) supports both.

What could explain it instead: Body weight is the dominant shared cause: heavier drinkers are heavier, and weight is the strongest modifiable risk factor for obstructive apnea. Smoking and shift work also cluster with drinking.

Simou et al., alcohol and the risk of sleep apnoea: a systematic review and meta-analysis · Sleep Med 2018;42:38-46

Across 82 papers reporting adherence between 1994 and 2015, the overall CPAP non-adherence rate was 34.1%, with no significant improvement over the twenty-year span. Behavioral intervention improved adherence by roughly one hour per night on average.

Measured in: 82 studies of CPAP adherence in adults with obstructive sleep apnea, 1994 to 2015

Adherence was defined against a 7-hour sleep time, and definitions vary across the included studies, so the 34.1% figure is a synthesis of measurements that were not made the same way. Study populations are mostly sleep-clinic patients on structured follow-up, which is likely to be better supported than routine care. The null here is about the trend over time, not about whether any individual intervention helps.

Rotenberg et al., trends in CPAP adherence over twenty years of data collection: a flattened curve · J Otolaryngol Head Neck Surg 2016;45(1):43

145 adults with an AHI of 15 or more and comorbid insomnia were randomized to four sessions of cognitive behavioral therapy for insomnia before starting CPAP, or to treatment as usual. The CBT-I group averaged 61 more minutes of nightly CPAP use at six months (95% CI 9 to 113, d = 0.38) and higher initial acceptance of CPAP (99% vs 89%). Insomnia severity and dysfunctional sleep beliefs improved more in the CBT-I group.

Measured in: 145 adults with moderate to severe obstructive sleep apnea and co-occurring insomnia

One trial of 145 people, unblindable by design, with the CBT-I group receiving four extra sessions of clinician contact that the control group did not. There were no between-group differences in sleep outcomes or daytime impairment at six months, so what was shown is better use of the device rather than a better outcome from it.

Sweetman et al., cognitive and behavioral therapy for insomnia increases the use of continuous positive airway pressure therapy in obstructive sleep apnea participants with comorbid insomnia: a randomized clinical trial · Sleep 2019;42(12):zsz178

In a network meta-analysis of 80 randomized trials, CPAP lowered the apnea-hypopnea index most, by 25.27 events per hour against control (95% CI 22.03 to 28.52). A mandibular advancement device lowered it too, ranking below CPAP: CPAP reduced the index about 10 events an hour more than the device (between-treatment difference 10.06 events per hour, 95% CI 5.91 to 14.21). Daytime sleepiness on the Epworth Sleepiness Scale did not differ significantly between the device and CPAP.

Measured in: Adults with obstructive sleep apnea across 80 randomized trials of CPAP, mandibular devices, exercise training or dietary weight loss

A mandibular device lowers apnea severity less than CPAP does, which is why the ranking places it below the mask. Its appeal is that people often wear it more readily, and that fits the finding that daytime sleepiness came out similar between the two. The device suits mild to moderate apnea and CPAP-intolerant patients more than severe disease, and a network meta-analysis compares treatments partly through indirect links rather than head to head throughout, so the between-treatment figures are less certain than a single head-to-head trial would be.

Iftikhar et al., comparative efficacy of CPAP, MADs, exercise-training, and dietary weight loss for sleep apnea: a network meta-analysis · Sleep Med 2017;30:7-14

The CPAP treatment effect was greater in high-risk apnea, defined by heart-rate response and hypoxic burden, than in low-risk apnea: interaction hazard ratio 0.69 (95% CI 0.50 to 0.95). The same analysis reports HARM in low-risk apnea, with the authors concluding the harm may counteract the benefit. The 0.69 is the differential effect between the two groups, not the within-high-risk-group treatment hazard ratio.

Measured in: 3,549 participants pooled from three randomized CPAP cardiovascular outcome trials

Both halves come from one post-hoc analysis with tertiles defined after the fact, so this is a hypothesis about who benefits rather than a demonstration. Quoting the favorable half alone would misrepresent the paper.

Cardiovascular benefit of continuous positive airway pressure according to high-risk obstructive sleep apnoea: a multi-trial analysis · Eur Heart J 2026;47(17):2077-2089

Eight randomized studies, 323 participants. Against inactive control, positional therapy reduced AHI by 7.38 events per hour (95% CI 4.7 to 10.06, low certainty) and Epworth scores by 1.58 points (95% CI 0.29 to 2.89, moderate certainty). Against CPAP, CPAP reduced AHI by a further 6.4 events per hour (95% CI 3.00 to 9.79, low certainty), while subjective adherence favored positional therapy by 2.5 hours a night (moderate certainty). Epworth scores did not differ between CPAP and positional therapy.

Measured in: 323 participants, mostly with supine-predominant obstructive sleep apnea; devices ranged from vibration alarms to backpacks and pillows

Cochrane rated the AHI comparisons low certainty and the trials are small and short. The devices pooled are not equivalent to each other: a vibrotactile alarm and a tennis ball sewn into a shirt are different interventions. Adherence data came from one study and was self-reported. This only applies to people whose events are supine-predominant, and none of the trials followed people long enough to say whether the devices are still worn after a year.

Srijithesh et al., positional therapy for obstructive sleep apnoea · Cochrane Database Syst Rev 2019;5:CD010990

Across nine adult studies and 120 patients, AHI fell from a mean of 24.5 to 12.3 events per hour (mean difference -14.26, 95% CI -20.98 to -7.54), about 50%. Lowest oxygen saturation improved from 83.9% to 86.6%, snoring fell from 14.1% to 3.9% of total sleep time, and Epworth scores fell from 14.8 to 8.2.

Measured in: 120 adults across nine studies, plus 25 children across two studies

Nine studies and 120 adults, most of them before-and-after with no control arm, so regression to the mean and the effect of being observed are not excluded. A 50% fall from a mean AHI of 24.5 still leaves 12.3, which is mild to moderate apnea rather than resolution. The therapy protocols differed between studies and adherence to a daily exercise regimen over months is its own problem.

What could explain it instead: Before-and-after designs in a condition whose severity varies night to night will overstate benefit, since people enrol during a bad period and are re-measured later.

Camacho et al., myofunctional therapy to treat obstructive sleep apnea: a systematic review and meta-analysis · Sleep 2015;38(5):669-75

Nine randomized trials, 584 participants, covering manual acupuncture and electroacupuncture. AHI fell by 6.18 events per hour (95% CI -9.58 to -2.78) and Epworth scores by 2.84 points (95% CI -4.80 to -0.16). Lowest oxygen saturation improved by 5.29 percentage points. Larger AHI reductions were seen in the moderate (-9.44) and severe (-10.09) subgroups.

Measured in: 584 adults with obstructive sleep apnea across nine trials, most conducted in China

The authors rate the quality of evidence low to very low. The sleepiness result reversed direction under sensitivity analysis, which means it does not survive removing the weakest studies. Most included trials come from Chinese-language databases where acupuncture trials report positive results at an unusually high rate. A 6-point fall in AHI does not move someone with severe apnea out of the range where airway treatment is indicated.

Wang et al., acupuncture for obstructive sleep apnea (OSA) in adults: a systematic review and meta-analysis · Biomed Res Int 2020;2020:6972327

58 randomized trials, 4,590 participants. Against placebo, Chinese herbal medicine reduced AHI by 7.10 events per hour (95% CI -11.95 to -2.25) across 7 studies and 583 participants. Added to CPAP, it reduced AHI by a further 4.71 events per hour against CPAP alone (95% CI -5.62 to -3.80) across 28 studies and 2,267 participants. Sleepiness, quality of life, body weight, oxidative and inflammatory markers, cognition and blood pressure also improved.

Measured in: 4,590 adults with obstructive sleep apnea, the great majority in trials from Chinese-language databases

Only 7 of the 58 trials had a placebo comparison, and the 51 that did not are the ones carrying most of the participants. The formulas differ from study to study, so the pooled figure describes a category of treatment rather than a prescription, and nothing in it identifies which pattern a given formula suited. Trials from this literature carry known risks of selective reporting and inadequate allocation concealment. The effect size is smaller than any device produces.

Chinese herbal medicine for obstructive sleep apnoea: a systematic review with meta-analysis · Sleep Breath 2024;29(1):56

Measured under forced desynchrony, with the imposed day length outside the range the clock can lock onto and light controlled by the protocol rather than by the participant, the intrinsic period averages 24.18 hours, about 24.2, in young and older adults alike, with a distribution as tight as any other species tested. Across 157 adults studied in month-long inpatient protocols the standard deviation was about 0.2 hours, roughly 12 minutes. The 25-hour figure still in general circulation came from earlier isolation studies in which participants switched their own room lighting, so self-selected light was resetting the clock being measured.

Measured in: Healthy young and older adults in controlled-lighting inpatient protocols. The pooled period estimate covers 157 people aged 18 to 74, 52 women and 105 men.

Forced desynchrony measures the pacemaker's own period by imposing a schedule nobody lives on, and it costs weeks of inpatient time per person, which is why the total number of people ever measured this way is in the low hundreds. It gives the period of the clock, not the timing anyone actually keeps, which is set by light on top of that period.

Czeisler et al., stability, precision, and near-24-hour period of the human circadian pacemaker · Science 1999;284(5423):2177-81 Duffy et al., sex difference in the near-24-hour intrinsic period of the human circadian timing system · Proc Natl Acad Sci USA 2011;108 Suppl 3:15602-8

Two independent action spectra for nocturnal melatonin suppression fit a photopigment distinct from rods and cones. One, from 22 volunteers across 215 monochromatic light exposures, fit best to a rhodopsin template peaking at 459 nm, with rod, cone and cryptochrome spectra all fitting poorly. The other, from 72 people across 627 suppression tests, found 446 to 477 nm the most potent region and fit an opsin template with an R-squared of 0.91. The pigment is melanopsin, carried by a small population of intrinsically photosensitive retinal ganglion cells that project to the suprachiasmatic nucleus.

Measured in: 22 volunteers over 215 light exposure trials, and 72 volunteers (37 women, 35 men, mean age 24.5) over 627 melatonin suppression tests

Action spectra are built from melatonin suppression at night, which is one output of the pathway rather than the pathway itself, and both used narrow-band light at intensities and durations nobody encounters. Melanopsin's own peak sensitivity in the intact eye sits nearer 480 nm than these melatonin figures, because rods and cones also feed into the same ganglion cells.

Thapan, Arendt and Skene, an action spectrum for melatonin suppression · J Physiol 2001;535(Pt 1):261-7 Brainard et al., action spectrum for melatonin regulation in humans · J Neurosci 2001;21(16):6405-12

The clock is a transcription-translation feedback loop. CLOCK and BMAL1 bind and drive transcription of the Period and Cryptochrome genes; the PER and CRY proteins accumulate, complex, return to the nucleus and repress the very activators that made them, and are then phosphorylated by kinases and degraded so the cycle restarts. The loop is cell-autonomous and keeps running in isolated cells. In humans, a single serine-to-glycine substitution in PER2, inside the casein kinase I epsilon binding region, produces autosomal dominant familial advanced sleep phase syndrome, with a four-hour advance of the sleep, temperature and melatonin rhythms.

Measured in: Molecular work in flies, mice and cultured cells; the human genetic evidence comes from a kindred with familial advanced sleep phase syndrome

The loop was worked out mostly outside humans, and the human genetic evidence rests on a small number of families carrying rare high-effect mutations. That shows the loop sets human sleep timing. It does not show how much of ordinary chronotype variation runs through these same genes, and the genome-wide work suggests common variation moves timing by minutes rather than hours.

Takahashi, transcriptional architecture of the mammalian circadian clock · Nat Rev Genet 2017;18(3):164-79 Toh et al., an hPer2 phosphorylation site mutation in familial advanced sleep phase syndrome · Science 2001;291(5506):1040-3

Pooling three prospective UK cohorts with seven previously published prospective studies, covering about 1.4 million women and 4,660 breast cancers among shift workers, the relative risk of breast cancer for any night shift work was 0.99 (95% CI 0.95 to 1.03). Long-term night shift work showed no excess either. The authors concluded that night shift work, including long-term shift work, has little or no effect on breast cancer incidence.

Measured in: About 1.4 million women across 10 prospective studies, with night shift work reported by questionnaire

Night shift work was captured by questionnaire, mostly at a single point in time, so exposure misclassification would push a real association towards the null. It tests breast cancer only: the prostate and colorectal associations IARC cites are not addressed here, and neither is the animal or mechanistic evidence that carries much of the classification.

Travis et al., night shift work and breast cancer incidence, three prospective studies and meta-analysis · J Natl Cancer Inst 2016;108(12):djw169

Women's intrinsic period averaged 24.09 hours against men's 24.19 hours (P < 0.01), a difference of about 6 minutes, and 35% of women had a period shorter than 24.0 hours against 14% of men. In a separate constant-routine study of 28 women and 28 men matched on habitual wake time, the melatonin and core body temperature rhythms occurred earlier relative to sleep in women, with higher melatonin amplitude and lower temperature amplitude, so women were sleeping at a later biological phase on the same clock time.

Measured in: 157 adults aged 18 to 74 (52 women, 105 men) for the period comparison; 56 adults aged 18 to 30 (28 women, 28 men) for the phase angle comparison

Six minutes of period difference and a phase difference well under an hour: replicated across two datasets, and small next to what an alarm clock imposes. Both datasets come from the same research group and overlapping laboratory protocols, so these are related analyzes rather than independent confirmation. Menstrual cycle phase and hormonal contraception were not controlled across the whole period sample.

Duffy et al., sex difference in the near-24-hour intrinsic period of the human circadian timing system · Proc Natl Acad Sci USA 2011;108 Suppl 3:15602-8 Cain et al., sex differences in phase angle of entrainment and melatonin amplitude in humans · J Biol Rhythms 2010;25(4):288-96

Light given after the core body temperature minimum advances the clock, and a bright pulse produces a phase-response curve with a peak-to-trough amplitude of about 5 hours under laboratory conditions, and light given before it delays the clock. The crossover sits at the temperature minimum, which falls about two hours before habitual waking in a regular sleeper, and the curve is nearly flat through the middle of the biological day. Response is steeply non-linear with duration: a one-hour pulse of about 8,000 lux produced a curve with a peak-to-trough amplitude of 2.20 hours, roughly 40% of what a 6.7-hour exposure produced on 15% of the exposure time.

Measured in: 21 adults aged 19 to 44 (16 men, 7 women) for the three-cycle curve; 36 adults randomized to bright white or dim light for the one-hour curve

The human phase response literature is small, young and laboratory-bound, and both curves were built from single pulses against a dim controlled background rather than from the mixed light of an ordinary day. It measures the clock moving. The step from a phase shift to sleeping better is an inference from mechanism, not something these protocols measured.

Khalsa et al., a phase response curve to single bright light pulses in human subjects · J Physiol 2003;549(Pt 3):945-52 St Hilaire et al., human phase response curve to a 1 h pulse of bright white light · J Physiol 2012;590(13):3035-45

Among 11 patients with no conscious perception of light, plasma melatonin fell during bright light exposure in three, and those three were normally entrained and slept normally. Seven showed no melatonin response to light and reported histories of insomnia and circadian rhythm disturbance; one had undetectable melatonin. Six sighted controls suppressed normally.

Measured in: 11 blind patients with no conscious light perception and 6 sighted controls

Eleven patients with different causes and degrees of blindness, so this establishes that photic input to the clock and conscious vision can come apart, and it gives no reliable proportion for how often. It also predates the identification of melanopsin by seven years, so the mechanism was inferred here rather than measured.

Czeisler et al., suppression of melatonin secretion in some blind patients by exposure to bright light · N Engl J Med 1995;332(1):6-11

Applying an algorithm that reconstructs the temporal order of samples with no time-of-day information available, across 13 tissues from 632 human donors, nearly half of all protein-coding genes were found cycling in at least one tissue. About 1,000 of those encode drug targets, or the proteins that transport and metabolize drugs.

Measured in: 13 tissues from 632 human donors in a post-mortem genotype-expression resource

The method orders samples within a reconstructed cycle rather than against clock time, so it produces phase relationships between genes and between tissues and not an hour of the day for any organ. Donor tissue also varies in agonal state, cause of death and time of collection, and the reconstruction is an inference from expression rather than a rhythm watched in a living person.

Ruben et al., a database of tissue-specific rhythmically expressed human genes · Sci Transl Med 2018;10(458):eaat8806

Restricting food to the light phase shifted the phase of clock gene expression in mouse peripheral tissues by up to 12 hours while leaving cyclic gene expression in the suprachiasmatic nucleus unaffected. The liver reset fastest, with kidney, heart and pancreas following more slowly, and one week of daytime feeding synchronized every tissue examined.

Measured in: Mice on restricted feeding schedules, with clock gene expression measured in liver, kidney, heart, pancreas and the suprachiasmatic nucleus

Mice are nocturnal, so feeding them by day is a larger insult than late eating is to a person, and the human version of the same manipulation produced a peripheral shift a fraction of this size. Animal work establishes that the route exists rather than what it does in people.

Damiola et al., restricted feeding uncouples circadian oscillators in peripheral tissues from the central pacemaker in the suprachiasmatic nucleus · Genes Dev 2000;14(23):2950-61

On a 28-hour day protocol that walked eating and sleeping across every circadian phase, behavior running roughly 12 hours out of phase with the internal clock raised glucose 6% (P < 0.001), insulin 22% (P = 0.006) and mean arterial pressure 3% (P = 0.001), lowered leptin 17% (P < 0.001) and completely reversed the daily cortisol rhythm. Three of eight participants showed postprandial glucose in the prediabetic range while misaligned. Sleep efficiency fell 20%.

Measured in: 10 adults, 5 women and 5 men, through a 10-day inpatient forced desynchrony protocol

Ten people over ten days, with food, activity and sleep opportunity held constant, which isolates misalignment itself and says nothing about what years of it do. Sleep efficiency also fell 20%, so disturbed sleep sits inside the misaligned condition rather than beside it, and the two cannot be separated in this design.

Scheer et al., adverse metabolic and cardiovascular consequences of circadian misalignment · Proc Natl Acad Sci USA 2009;106(11):4453-8

A week camping in summer with no electrical light raised light exposure to about four times the usual and moved the internal biological night roughly two hours earlier, so that it began near sunset and ended near sunrise. Late chronotypes advanced most, ending the week closer to the early types. Repeated under a natural winter light-dark cycle, the biological night and sleep began earlier still and the biological night expanded. A weekend of natural light achieved about 69% of the shift a full week produced.

Measured in: 8 adults, 6 men and 2 women, mean age 30, in the summer study, with a separate winter and weekend study by the same group

Eight people in the summer study, and camping removes electrical light along with work, commuting, screens, indoor temperature and the whole daily routine, so the light change is confounded with everything else that changes on a camping trip. The two studies come from one research group, so they are related work rather than independent replication.

Wright et al., entrainment of the human circadian clock to the natural light-dark cycle · Curr Biol 2013;23(16):1554-8 Stothard et al., circadian entrainment to the natural light-dark cycle across seasons and the weekend · Curr Biol 2017;27(4):508-13

A genome-wide association study of 697,828 people raised the number of loci associated with being a morning person from 24 to 351. The effect on timing is smaller than that number suggests: the 5% of participants carrying the most morningness alleles had accelerometer-measured sleep timing about 25 minutes earlier than the 5% carrying the fewest. Associated genes were enriched for circadian regulation, cAMP, glutamate and insulin signaling, and for expression in retina, hindbrain, hypothalamus and pituitary. Mendelian randomization supported a causal path from morning preference to better mental health, and found no effect on BMI or type 2 diabetes.

Measured in: 697,828 adults from UK Biobank and 23andMe, with activity-monitor sleep timing in 85,760 of them

351 loci that together move sleep timing by about 25 minutes between the extreme twentieths, so common genetic variation shifts chronotype by minutes for most people rather than by hours. Mendelian randomization assumes the genetic instrument affects the outcome only through chronotype, which cannot be verified directly, and the mental health outcomes were themselves self-reported.

What could explain it instead: Chronotype was self-reported by most participants, so the phenotype is partly a report of the schedule a person keeps, which work and family impose, rather than the clock underneath it. UK Biobank and 23andMe participants are also healthier, wealthier and more European in ancestry than the general population, the standard population-stratification problem in large genetic studies.

Jones et al., genome-wide association analyzes of chronotype in 697,828 individuals · Nat Commun 2019;10(1):343

Across 53,689 American time-use diaries, chronotype gets later through adolescence, reaches its latest point at 18.4 years in females and 19.2 years in males, and moves earlier from then on for the rest of life. The largest sex difference is 0.27 hours, about 16 minutes, at 21.6 years, with men later than women before about 40 and women later after it. The difference disappears at 41.4 years and again at 79.1. Chronotype variability also narrows with age.

Measured in: 53,689 respondents to the American Time Use Survey, pooled across 2003 to 2014

A snapshot of different people at different ages rather than the same people followed, so an age effect cannot be separated from a cohort effect. The original European description of the adolescent peak, in about 25,000 questionnaires, placed maximum lateness nearer 20 than 19, so the exact turning point moves with the sample and the instrument.

What could explain it instead: Mid-sleep on free days from a time-use diary is a behavioral proxy for chronotype, and free-day sleep is still shaped by work, caring duties, school schedules and shift patterns, all of which change with age, so part of the age curve is life stage rather than clock.

Fischer et al., chronotypes in the US, influence of age and sex · PLoS One 2017;12(6):e0178782 Roenneberg et al., a marker for the end of adolescence · Curr Biol 2004;14(24):R1038-9

In UK Biobank, 433,268 adults followed a mean 6.5 years with 10,534 deaths, definite evening types had a hazard ratio for all-cause mortality of 1.10 (95% CI 1.02 to 1.18) against definite morning types, after adjustment for age, sex, ethnicity, smoking, BMI, sleep duration, socioeconomic status and comorbidity. In the Finnish Twin Cohort, 23,854 people followed 37 years with 8,728 deaths, the adjusted hazard ratio was 1.09 (1.01 to 1.18) and it attenuated mainly through smoking and alcohol, with no excess mortality among non-smokers who were at most light drinkers. Those authors concluded there is little or no independent contribution of chronotype to mortality.

Measured in: 433,268 UK adults aged 38 to 73 and 23,854 Finnish adults, both sexes, followed 6.5 and 37 years respectively

Chronotype was a single self-reported item on a four-point scale, collected once, in both cohorts. A hazard ratio near 1.10 is small enough that unmeasured differences in behavior are a sufficient explanation, and the two cohorts differ in follow-up length by a factor of five, so they are not measuring the same exposure window.

What could explain it instead: Evening preference travels with smoking, heavier drinking, shorter sleep, shift work and lower socioeconomic position. The Finnish stratified analysis is the informative one here: the association disappeared among people who did not smoke and drank little, rather than merely shrinking, which is what residual confounding by those behaviors looks like.

Knutson and von Schantz, associations between chronotype, morbidity and mortality in the UK Biobank cohort · Chronobiol Int 2018;35(8):1045-53 Hublin and Kaprio, chronotype and mortality, a 37-year follow-up study in Finnish adults · Chronobiol Int 2023;40(7):841-849

The IARC working group that met in June 2019 classified night shift work in Group 2A, probably carcinogenic to humans. The classification rests on three separate evidence streams: limited evidence in humans, with positive associations for cancers of the breast, prostate, colon and rectum; sufficient evidence in experimental animals for the carcinogenicity of alteration of the light-dark schedule; and strong mechanistic evidence in experimental systems, based on effects consistent with immunosuppression, chronic inflammation and cell proliferation.

Measured in: A working group evaluating the published human, animal and mechanistic literature on night shift work, rather than a participant sample

In IARC's own vocabulary, limited evidence in humans means a positive association was observed but chance, bias and confounding could not be ruled out as explanations. A Group 2A classification is a hazard identification: it says an agent can cause cancer under some conditions and carries no estimate of how much risk any particular job or schedule involves. Much of the weight sits on animal and mechanistic work rather than on the human studies.

IARC Monographs Volume 124, night shift work · IARC Monographs on the Identification of Carcinogenic Hazards to Humans, Vol 124, Lyon 2020 IARC Monographs Vol 124 group, carcinogenicity of night shift work · Lancet Oncol 2019;20(8):1058-9

Across 34 studies and 2,011,935 people, shift work was associated with myocardial infarction (RR 1.23, 95% CI 1.15 to 1.31), coronary events (RR 1.24, 1.10 to 1.39) and ischemic stroke (RR 1.05, 1.01 to 1.09), with no association with all-cause mortality. Across 12 studies and 226,652 people, shift work was associated with diabetes at a pooled odds ratio of 1.09 (1.05 to 1.12), higher in men (1.37, 1.20 to 1.56) than in women (1.09, 1.04 to 1.14) and higher for rotating shifts.

Measured in: 2,011,935 people across 34 studies for vascular events, and 226,652 people with 14,595 diabetes cases across 12 studies for diabetes

Both pool observational studies, in which shift work travels with lower socioeconomic position, more smoking, shorter and more fragmented sleep, irregular eating and less daylight, none of which the pooled analyzes can separate from the clock. The relative risks are modest, and the absence of any association with mortality sits oddly beside a raised risk of myocardial infarction.

Vyas et al., shift work and vascular events, systematic review and meta-analysis · BMJ 2012;345:e4800 Gan et al., shift work and diabetes mellitus, a meta-analysis of observational studies · Occup Environ Med 2015;72(1):72-8

A systematic review and meta-analysis of 40 studies (49 intervention comparisons) found that light therapy, most often bright morning light via a light box, produced small but significant benefits versus control: improved sleep continuity (effect size -0.23, where 0.2 counts as a small effect, versus control, p < 0.001), reduced self-reported sleep disturbance (ES -0.32, p = 0.014), and advancement of delayed sleep timing so people fell asleep earlier (ES -0.34, p = 0.010). Avoiding light in the evening was associated with greater gains in total sleep time.

Measured in: Adults with intrinsic circadian rhythm sleep disorders (delayed or advanced sleep phase) and people with affective, psychotic, or dementing illness.

Effect sizes are small, the pooled trials were clinically diverse, and morning light works best when the dose (roughly 30 minutes to 2 hours of bright light) is timed to the individual clock; benefit in psychiatric populations rests on fewer studies.

Faulkner SM, Bee PE, Meyer N, Dijk DJ, Drake RJ. Light therapies to improve sleep in intrinsic circadian rhythm sleep disorders and neuro-psychiatric illness: A systematic review and meta-analysis. · Sleep Med Rev 2019;46:108-123

A meta-analysis of nine clinical trials found that bright light therapy reduced depressive symptoms in non-seasonal depression compared with control (standardized mean difference -0.62, p < 0.001), a mild-to-moderate effect. Benefit was larger when light was used on its own (SMD -0.71) and when treatment ran 2 to 5 weeks (SMD -0.78). Perinatal depression showed no clear effect (SMD -0.17).

Measured in: Adults with non-seasonal depressive disorders, treated with bright light either alone or added to usual care.

Trials were relatively small and varied in light dose, timing, and whether light was used alone or alongside medication; the perinatal subgroup did not benefit, so results should not be generalized to every form of depression.

Al-Karawi D, Jubair L. Bright light therapy for nonseasonal depression: Meta-analysis of clinical trials. · J Affect Disord 2016;198:64-71

Delaying every meal by five hours delayed the plasma glucose rhythm by 5.69 hours (P < 0.001) and the PER2 messenger RNA rhythm in adipose tissue by 0.97 hours (P < 0.01), while plasma melatonin and cortisol, the markers of the master clock, did not shift at all. Average glucose concentration fell 4.9 mg/dL (0.27 mmol/L). Insulin and triglyceride rhythms, whole-blood clock gene expression, sleep and the rhythms of subjective hunger and sleepiness did not move.

Measured in: 10 healthy young men, mean age 22.9, in a crossover with early meals and then meals delayed by five hours

Ten men, one laboratory, and the two schedules ran in a fixed order rather than randomized, so an order effect cannot be excluded. The adipose clock gene moved by about an hour against a five-hour meal delay, a far smaller peripheral shift than the same manipulation produces in rodents, and adipose tissue is a proxy for the liver and gut clocks that cannot be biopsied as easily.

Wehrens et al., meal timing regulates the human circadian system · Curr Biol 2017;27(12):1768-75

Serial muscle biopsies taken through the day in humans showed extensive rhythmic transcription, and a large part of that rhythmicity was lost when primary myotubes from the same source were synchronized and studied in culture. Disrupting the clock in those cells with siRNA changed the expression of about 8% of all genes.

Measured in: Human volunteers undergoing serial in vivo muscle biopsies, compared with primary human myotubes synchronized in vitro

The comparison is between muscle inside a living body and cells in a dish, which differ in far more than their clocks, so the rhythmicity missing in vitro shows that systemic daily signals contribute without quantifying how much. Participant numbers and sex composition are not recoverable from the abstract we checked.

Perrin et al., transcriptomic analyzes reveal rhythmic and CLOCK-driven pathways in human skeletal muscle · eLife 2018;7:e34114

In a large chronotype database, social jetlag, defined as the difference between mid-sleep on work days and mid-sleep on free days, was associated with higher body mass index over and above sleep duration. The association was carried by people who were already overweight or obese.

Measured in: Respondents to the Munich ChronoType Questionnaire, a large self-selected online sample of both sexes

Cross-sectional, with self-reported sleep times and self-reported height and weight, so it shows an association rather than cause. The full text is paywalled, so only the abstract-level finding can be verified here: social jetlag is associated with higher BMI beyond the effect of sleep duration, especially in the overweight and obese.

What could explain it instead: Social jetlag is largest in people whose work starts early relative to their own chronotype, which tracks shift work, manual occupations and lower socioeconomic position, each independently associated with higher BMI. Reverse causation is also open, since sleeping in on free days is a response to accumulated sleep debt, which higher body weight and sleep-disordered breathing both worsen.

Roenneberg et al., social jetlag and obesity · Curr Biol 2012;22(10):939-43

In a randomized crossover controlled-feeding trial in 8 men with prediabetes, eating within an early 6-hour window (finishing dinner before 3 p.m.) for 5 weeks improved insulin sensitivity and beta-cell responsiveness, lowered blood pressure, and reduced oxidative stress compared with a 12-hour window, with meals matched so the gains occurred without weight loss. A separate 14-week parallel randomized trial in 90 adults with obesity found early time-restricted eating (roughly 7 a.m. to 3 p.m., an 8-hour window) added to calorie reduction produced greater weight loss (-5.1 lb (-2.3 kg) vs control; 95% CI -3.7 to -0.9) and lower diastolic blood pressure (-4 mm Hg; 95% CI -8 to 0).

Measured in: Adults with prediabetes (men) and adults with obesity (men and women) eating the bulk of their calories earlier in the day.

The clearest circadian-alignment signal comes from a very small 8-person crossover trial in men only, and the larger 90-person trial combined early eating with calorie restriction, so part of the benefit reflects eating less; longer trials in broader populations are needed to confirm the metabolic gain from timing alone.

Sutton EF, Beyl R, Early KS, Cefalu WT, Ravussin E, Peterson CM. Early Time-Restricted Feeding Improves Insulin Sensitivity, Blood Pressure, and Oxidative Stress Even without Weight Loss in Men with Prediabetes. · Cell Metab 2018;27(6):1212-1221.e3 Jamshed H, Steger FL, Bryan DR, et al. Effectiveness of Early Time-Restricted Eating for Weight Loss, Fat Loss, and Cardiometabolic Health in Adults With Obesity: A Randomized Clinical Trial. · JAMA Intern Med 2022;182(9):953-962

Across 51 older adults aged 59 to 75 and 48 younger adults aged 18 to 30 who performed one hour of moderate treadmill exercise, at 65 to 75% of heart rate reserve, on three consecutive days at one of eight times of day, exercise produced a phase-response curve in the melatonin metabolite 6-sulfatoxymelatonin. Exercise at 7am and again from 1pm to 4pm advanced the clock, while exercise from 7pm to 10pm delayed it, a curve that adds a second afternoon advance region not seen in the light curve. Exercise is a weaker timing signal than light and has been shown to act alongside it.

Measured in: 99 aerobically fit adults, 51 aged 59 to 75 and 48 aged 18 to 30, studied on a 90-minute ultrashort sleep-wake laboratory protocol for up to 5.5 days

A single laboratory study, in aerobically fit people, on an ultrashort sleep-wake schedule rather than ordinary days. It measures a phase shift in a melatonin metabolite, which is the clock moving, and does not measure sleep or any downstream outcome. Exercise is a milder zeitgeber than light, and the practical use described for it is alongside light rather than instead of it.

Youngstedt, Elliott and Kripke, human circadian phase-response curves for exercise · J Physiol 2019;597(8):2253-68

An artery relaxes to acetylcholine only while its inner lining is intact. The lining releases a short-lived factor, later identified as nitric oxide, that tells the surrounding muscle to relax and widens the vessel.

Measured in: Isolated rabbit artery; the endothelial release of nitric oxide was later confirmed across species, including humans.

The original 1980 demonstration was in isolated rabbit aorta, and the factor was not identified as nitric oxide until later work. That the endothelium releases nitric oxide to relax vessels is established physiology; this paper is the origin of the mechanism, not a human outcome.

Furchgott & Zawadzki, the obligatory role of endothelial cells in the relaxation of arterial smooth muscle by acetylcholine · Nature 1980;288(5789):373-376

Nitrate from vegetables is concentrated in saliva, reduced to nitrite by mouth bacteria, and converted to nitric oxide in the body, a second supply route that does not depend on the eNOS enzyme.

Measured in: Human physiology, characterised in a peer-reviewed review.

This is a review of the pathway rather than an outcome trial. The route is well characterised; whether a given dose of food nitrate helps a given person is the separate question the blood-pressure claims address.

Lundberg et al., the nitrate-nitrite-nitric oxide pathway in physiology and therapeutics · Nat Rev Drug Discov 2008;7(2):156-167

Pooling 16 randomized trials, inorganic nitrate and beetroot juice lowered systolic blood pressure by 4.4 mmHg (95% CI 2.8 to 5.9). The diastolic change of 1.1 mmHg was not statistically significant. Beetroot juice showed the larger effect.

Measured in: 254 participants across 16 controlled trials, mostly short-term.

The trials were short, most lasting days to a few weeks, and blood pressure is a marker rather than a count of heart attacks. A 4 to 5 mmHg systolic fall is real and on the scale of a single medication, but the diastolic effect was not significant and long-term outcome trials have not been done.

Siervo et al., inorganic nitrate and beetroot juice supplementation reduces blood pressure in adults, a systematic review and meta-analysis · J Nutr 2013;143(6):818-826

In hypertensive patients, four weeks of daily beetroot juice cut 24-hour ambulatory blood pressure by about 7.7/5.2 mmHg, improved endothelial function by roughly 20%, and reduced arterial stiffness by 0.59 m/s, against a nitrate-depleted placebo juice.

Measured in: 68 people with high blood pressure, 34 on no medication and 34 already treated.

A single four-week trial, though the placebo was nitrate-depleted beetroot juice, a control that isolates the nitrate. The effect held in people already on blood-pressure medication as well as the untreated, but four weeks is not long-term and cardiovascular events were not the endpoint.

Kapil et al., dietary nitrate provides sustained blood pressure lowering in hypertensive patients, a randomized, phase 2, double-blind, placebo-controlled study · Hypertension 2015;65(2):320-327

Training raised flow-mediated dilation progressively at two, four and six weeks. When the exercise-driven rise in blood flow and wall shear stress was blunted in one arm with a cuff, that arm did not improve, identifying shear stress as the signal behind the adaptation.

Measured in: Healthy young men.

A small mechanistic study in young men. It establishes the pathway, that movement raises blood flow, blood flow raises wall shear stress, and shear stress upregulates eNOS, rather than a long-term outcome. It does not measure how much habitual exercise changes disease risk.

Tinken et al., shear stress mediates endothelial adaptations to exercise training in humans · Hypertension 2010;55(2):312-318

Pooling randomized controlled trials, exercise improved flow-mediated dilation across every modality: aerobic training by 2.79 percentage points (95% CI 2.12 to 3.45), resistance training by 2.52 (95% CI 1.11 to 3.93), and combined training by 2.07 (95% CI 0.70 to 3.44). Higher aerobic intensity produced larger gains in a dose-response pattern.

Measured in: Adults aged 18 and over across randomized controlled trials of at least four weeks, comparing exercise with non-exercise, usual-care or sedentary groups.

Flow-mediated dilation is a marker of endothelial function rather than a count of heart attacks, so a training-driven rise in the marker is a strong physiological signal rather than a demonstrated reduction in events. The pooled trials varied in exercise type, intensity and duration.

Ashor et al., exercise modalities and endothelial function, a systematic review and dose-response meta-analysis of randomized controlled trials · Sports Med 2015;45(2):279-296

Flow-mediated dilation fell with age (r = -0.34). In men the decline began around age 40, at about 0.21% per year; in women it began later, in the early 50s, then fell faster, at about 0.49% per year.

Measured in: 238 healthy volunteers, 103 men and 135 women, aged 15 to 72.

A snapshot across ages rather than people followed over time. The sex pattern, men declining earlier and women later but faster after menopause, fits the loss of estrogen's support of eNOS, but the study describes the association and cannot prove the cause.

What could explain it instead: Cross-sectional design: different ages are different people, not the same people followed over time, so a generational difference in diet, smoking or activity can masquerade as an ageing effect. Cardiovascular risk factors also accumulate with age, so this is ageing plus everything that travels with it.

Celermajer et al., aging is associated with endothelial dysfunction in healthy men years before the age-related decline in women · J Am Coll Cardiol 1994;24(2):471-476

Across 14 studies and 5,547 people, each 1% higher flow-mediated dilation was associated with a 13% lower risk of a future cardiovascular event (pooled relative risk 0.87, 95% CI 0.83 to 0.91).

Measured in: 5,547 people pooled from 14 cohorts, both population-based and clinical.

This validates flow-mediated dilation as a marker that tracks risk, which is why it is used throughout this page. It does not follow that raising it with a treatment lowers events by the same amount: a marker that predicts an outcome is a different thing from one that, when moved, moves the outcome.

What could explain it instead: The pooled data are observational cohorts, so people with better endothelial function also tend to have fewer risk factors overall. Flow-mediated dilation predicts events but is not shown to cause them, and it is a surrogate rather than the outcome itself.

Inaba et al., prediction of future cardiovascular outcomes by flow-mediated vasodilatation of brachial artery, a meta-analysis · Int J Cardiovasc Imaging 2010;26(6):631-640

Adding L-arginine, the raw material eNOS uses to build nitric oxide, to standard care after a heart attack produced no improvement in vascular stiffness or heart function, and the trial was stopped when 6 of the L-arginine group had died against none on placebo.

Measured in: 153 people after a heart attack, 104 men, mean age 60.

One trial in a specific group, older heart-attack survivors, and the mortality difference came from small numbers. It is a caution against assuming that swallowing the substrate raises useful nitric oxide: food nitrate and exercise raise it where and when a vessel signals for it, which a steady oral dose of substrate does not reproduce.

Schulman et al., L-arginine therapy in acute myocardial infarction, the VINTAGE MI randomized clinical trial · JAMA 2006;295(1):58-64

Three hours of uninterrupted sitting reduced flow-mediated dilation in the superficial femoral artery in the thigh. Taking vitamin C beforehand prevented the decline at one, two and three hours, pointing to oxidative stress consuming nitric oxide as the cause.

Measured in: 11 healthy young men, mean age about 24.

A small crossover study of eleven young men, measuring the artery behind the knee over three hours. It shows the mechanism, that stillness lets shear stress fall and oxidative stress rise, rather than a lasting harm, and the vitamin C arm is a probe of that mechanism, not a recommendation to supplement.

Thosar et al., antioxidant vitamin C prevents decline in endothelial function during sitting · Med Sci Monit 2015;21:1015-1021

Placebo analgesia after dental surgery was reversed by naloxone, an opioid blocker, indicating that the pain relief was mediated by the body's own endogenous opioids.

Measured in: Adults undergoing extraction of impacted mandibular third molars; sex distribution not reported.

This identifies a mechanism for placebo analgesia specifically. It does not imply the same chemistry mediates placebo responses in other symptoms.

Levine, Gordon & Fields, The mechanism of placebo analgesia · Lancet 1978;2(8091):654-657

Across 202 trials that compared a placebo against no treatment, placebos produced a modest effect on patient-reported continuous outcomes, clearest for pain, and no important effect on binary or objectively measured outcomes.

Measured in: Pooled participants across 202 randomized trials in many conditions, both sexes.

The effect is modest and concentrated in what the patient feels and reports. It is not evidence that a placebo alters the underlying disease.

Hrobjartsson & Gotzsche, Placebo interventions for all clinical conditions · Cochrane Database Syst Rev 2010;(1):CD003974

In an analysis of 114 trials comparing placebo with no treatment, placebo had no significant effect on binary outcomes and no meaningful effect on objective continuous outcomes; any signal was confined to subjective, patient-reported measures such as pain.

Measured in: Pooled participants across 114 randomized trials in many conditions, both sexes.

A null on objective disease is not a null on symptoms: the same analyzes find an effect on self-reported pain. The boundary is the finding, not the absence.

Hrobjartsson & Gotzsche, Is the placebo powerless? · N Engl J Med 2001;344(21):1594-1602

In Parkinson's disease, the expectation of receiving an active drug triggered measurable dopamine release in the striatum on PET imaging, comparable to that from a therapeutic dose of levodopa.

Measured in: Patients with Parkinson's disease studied with PET imaging.

This shows a neurochemical correlate of expectation in one disease. It concerns symptomatic response, not the progression of Parkinson's.

de la Fuente-Fernandez et al., Expectation and dopamine release · Science 2001;293(5532):1164-1166

Under placebo analgesia, fMRI showed reduced activity in pain-processing regions during the experience of pain and increased prefrontal activity during its anticipation, tracking the expectation of relief.

Measured in: Healthy volunteers in an experimental placebo analgesia paradigm with fMRI.

This is experimental pain in healthy volunteers, which isolates the mechanism cleanly but is a step removed from clinical pain conditions.

Wager et al., Placebo-induced changes in FMRI in the anticipation and experience of pain · Science 2004;303(5661):1162-1167

In irritable bowel syndrome, adequate relief rose stepwise across a waiting list, a limited sham treatment, and the same sham delivered inside a warm, attentive relationship, at roughly 28%, 44% and 62%.

Measured in: 262 adults with irritable bowel syndrome, both sexes, IBS cohorts skew female.

The outcome was self-reported symptom relief in a symptom-defined condition, exactly where placebo responses are largest. It does not show an effect on objective disease.

Kaptchuk et al., Components of placebo effect: randomised controlled trial in IBS · BMJ 2008;336(7651):999-1003

Among people who had stopped statins because of side effects, 90% of the symptom burden recorded during statin months was also present during identical placebo months.

Measured in: 60 patients who had discontinued statins for adverse effects, both sexes.

This concerns symptom attribution in people already intolerant, and does not mean statin side effects are never pharmacological. It shows how large the nocebo share can be.

Wood, Howard et al., N-of-1 trial of a statin, placebo, or no treatment to assess side effects (SAMSON) · N Engl J Med 2020;383(22):2182-2184 Colloca & Barsky, Placebo and nocebo effects · N Engl J Med 2020;382(6):554-561

Men told about the sexual side effects of finasteride reported them roughly three times as often as men not informed, at about 44% versus 15% on the same drug.

Measured in: 120 men taking finasteride for benign prostatic hyperplasia.

The mechanism is general; this particular figure is specific to informed men on finasteride. It argues for careful wording of warnings, not for withholding them.

Mondaini et al., Finasteride 5 mg and sexual side effects: how many are a nocebo phenomenon · J Sex Med 2007;4(6):1708-1712

Patients with IBS openly given pills described as inert reported higher global improvement than patients given no treatment, roughly 59% versus 35%.

Measured in: 80 adults with irritable bowel syndrome, both sexes, IBS cohorts skew female.

Small and short, in a condition where placebo responses are large. Promising and early, not a settled or general result.

Kaptchuk et al., Placebos without deception: a randomized controlled trial in IBS · PLoS One 2010;5(12):e15591

A meta-analysis of open-label placebo trials found a large pooled benefit over no treatment (standardized mean difference about 0.9) across symptom-driven conditions, though the trials were few and small.

Measured in: Pooled participants across a small set of open-label placebo trials in symptom-driven conditions, both sexes.

Few trials, small samples and self-reported outcomes. The direction is consistent; the size and durability are not yet settled.

Charlesworth et al., Effects of placebos without deception compared with no treatment · J Evid Based Med 2017;10(2):97-107

Across 237,979 people in 15 countries, blood pressure, cholesterol, body-mass and waist all sat higher in winter than summer. In the northern hemisphere the winter-to-summer gap ran about 2.9 mmHg systolic and 1.4 mmHg diastolic blood pressure, 3.9 mg/dL (0.10 mmol/L) total cholesterol, 4.3 mg/dL (0.11 mmol/L) LDL cholesterol and 0.24 inches (0.6 cm) on the waist.

Measured in: 237,979 adults pooled from 24 population studies across 15 countries, in both hemispheres

The gaps are small at the individual level, and the design compares different people measured in different seasons rather than tracking the same person through the year, so it shows a population pattern and not a personal trajectory.

What could explain it instead: Cold months change behavior as much as physiology: less activity, heavier eating, more time indoors and more winter infection all move these same numbers, and a cross-sectional comparison cannot separate them from the season itself.

Marti-Soler et al., seasonality of cardiovascular risk factors across 15 countries · Heart 2014;100(19):1517-23

Pooling 19 randomized trials (610 patients for symptom scores, 559 for response), bright light therapy beat placebo for seasonal affective disorder with a standardized mean difference of -0.37 (95% CI -0.63 to -0.12) on depression ratings and a 42% higher chance of responding to treatment (risk ratio 1.42, 95% CI 1.08 to 1.85).

Measured in: 610 patients with seasonal affective disorder across 19 randomized controlled trials

A small-to-medium effect drawn from trials the authors describe as methodologically heterogeneous with moderate-to-high risk of bias, so the direction is consistent while the exact size is uncertain.

Pjrek et al., efficacy of light therapy in seasonal affective disorder, meta-analysis of RCTs · Psychother Psychosom 2020;89(1):17-24

In 1,897 people wearing personal temperature loggers alongside 24-hour blood-pressure monitoring, daytime systolic pressure fell as personal temperature rose, while both night-time blood pressure and the morning blood-pressure surge tracked the number of daylight hours. Temperature and daylight acted as two independent influences on blood pressure.

Measured in: 1,897 adults undergoing 24-hour ambulatory blood-pressure monitoring with individual temperature loggers

The associations are modest and come from a single observational dataset, so it shows that blood pressure moves with temperature and daylight, not that deliberately changing your light exposure will change your blood pressure.

What could explain it instead: People are not randomly assigned to seasons, and the factors that shift with temperature and daylight, activity levels, salt intake, indoor heating and sleep timing, also move blood pressure, so the daylight relationship cannot be cleanly separated from them.

Modesti et al., seasonal blood-pressure changes track temperature and daylight hours independently · Hypertension 2013;61(4):908-14

In 188 adults recorded across a full year, total sleep time ran up to 60 minutes longer in winter than summer (this difference did not reach statistical significance), REM sleep was about 30 minutes longer in winter than in spring, and deep sleep dipped in autumn.

Measured in: 188 adults attending a Berlin sleep clinic for insomnia, depression or breathing disorders; mean age 46.6, 52% female

These were people with sleep problems rather than healthy sleepers, and the longer-total-sleep finding did not reach statistical significance while the REM and deep-sleep shifts did, so it points to a real seasonal signal in sleep measured within a clinical group.

What could explain it instead: A retrospective clinic sample averaged by month cannot separate the season itself from who happened to attend in which month, or from the light, temperature and schedule changes that travel with the calendar.

Seidler et al., seasonality of human sleep from clinic polysomnography · Front Neurosci 2023;17:1105233

Water turnover measured by deuterium dilution in 5,604 people aged 8 days to 96 years across 26 countries varied with age, body size, body composition, physical activity, athletic status, pregnancy, socioeconomic status, latitude, altitude, air temperature and humidity. People in low human development index countries turned over more water than people in high-index ones. The output is a set of prediction equations. There is no single number in it.

Measured in: 5,604 people, 8 days to 96 years old, in 26 countries, pooled into the International Atomic Energy Agency doubly labeled water database

This produced prediction equations driven by body size, composition, activity and climate rather than a single daily number, which is why no universal target appears on this page. The first author declares a related patent. At 3,729 women and 1,875 men it is the best sex-represented finding on the page.

What could explain it instead: Turnover was measured in people living their ordinary lives, so the factors it identifies travel together: physical activity, athletic status, income and climate are correlated, and a low human development index brings heat, manual work, different housing and different diet all at once. The equations describe association in a pooled convenience sample, not a controlled comparison, so no single factor in them is isolated.

Yamada et al., variation in human water turnover associated with environmental and lifestyle factors · Science 2022;378(6622):909-915

In children admitted to hospital with acute diarrhea, a reduced-osmolarity oral rehydration solution needed fewer unscheduled intravenous infusions than the standard WHO formula: odds ratio 0.59, 95% CI 0.45 to 0.79, pooled across 8 trials with no heterogeneity. Stool output was lower and vomiting less frequent, with no excess hyponatremia. Changing the sodium and glucose concentrations changed the clinical result, which is what identifies the formula rather than the packaging as the working part.

Measured in: Randomized trials in children hospitalized with acute diarrhea; 11 trials reported the primary outcome and 8 were pooled for it

This compares two oral rehydration formulas against each other, not oral rehydration against plain water, and it is in hospitalized children rather than adults managing a stomach bug at home. The glucose-linked sodium cotransport mechanism is textbook physiology, not a finding of this review. That a sugar-free electrolyte powder cannot do this job follows from the formulas rather than from a trial that tested one against the other. Cochrane has not updated the review since 2002.

Hahn, Kim and Garner, reduced osmolarity oral rehydration solution for treating dehydration caused by acute diarrhoea in children · Cochrane Database Syst Rev 2002;(1):CD002847

Drinking beyond thirst during prolonged exercise can dilute blood sodium far enough to cause cerebral and pulmonary edema. At least twelve confirmed deaths are documented in the consensus literature, in settings including football practice, military training, calisthenics and an 11.8 mile (19 km) bike ride, and the authors state that athletes continue to die of it. Substantial weight gain during the race, the marker of overdrinking, carried an odds ratio of 4.2 (95% CI 2.2 to 8.2) for hyponatremia.

Measured in: 488 non-elite runners in the 2002 Boston Marathon who provided a usable blood sample at the finish, out of 766 enrolled

The confidence interval on that odds ratio is very wide. This is rare in absolute terms and it is the reason the modern instruction is to drink to thirst rather than on a schedule. Smaller athletes, slower finishers and longer events carry more risk, and the early symptoms are the same as dehydration, which is what makes it dangerous: the instinct is to drink more.

What could explain it instead: Fluid intake was recalled on a questionnaire after the race, which is exactly the kind of measure that shifts with how the runner felt at the end. Slower runners drink more and also have more time to accumulate a deficit or a surplus, so race time carries both the exposure and other things that go with a four-hour marathon: less training, more aid-station stops and different weather exposure over a longer day.

Almond et al., hyponatremia among runners in the Boston Marathon · N Engl J Med 2005;352(15):1550-6 Hew-Butler et al., statement of the third international exercise-associated hyponatremia consensus development conference · Clin J Sport Med 2015;25(4):303-20

After 24 hours without fluid, scores on a portrait memory test fell from 34 to 27, vigor from 11.8 to 9.2 and self-esteem from 7.8 to 6.4, while fatigue and total mood disturbance rose. Drinking plain purified water then improved processing speed, working memory and mood, and 1,000 mL was the best of the three volumes tested against 500, 200 and none, measured 90 minutes later.

Measured in: 76 healthy Chinese college students aged 18 to 23, equal numbers of men and women in each arm, randomized to four rehydration volumes

This is recovery from 24 hours of deliberate water restriction, a state most readers will never be in, and it says nothing about topping up someone already normally hydrated. Measurement stops 90 minutes after drinking. The fluid was plain water with nothing added, which is the part of the result that gets left out when the study is used to sell something. Funded by the National Natural Science Foundation of China, with no industry involvement.

Zhang et al., effects of water restriction and supplementation on cognitive performances and mood among young adults · Nutrients 2021;13(10):3645

Sweat sodium averaged 42.9 plus or minus 18.7 mmol/L across 157 runners in one marathon and ranged from 7.0 to 95.5, a fourteen-fold spread in the same race and the same weather. Potassium sat in a narrow band, 6.0 plus or minus 0.9 mmol/L, range 3.1 to 8.0. Women averaged 33.9 mmol/L against 44.0 in men. One runner in five was above 60 mmol/L, and no woman was.

Measured in: 157 experienced marathon runners, 141 men and 16 women, sweat collected on forearm patches during a race at 75.9°F (24.4°C) and 28% humidity

A field study during one marathon in one climate, so it captures between-person spread rather than how stable any individual is across conditions. 20% of participants exceeded 60 mmol/L, and no female runner in the sample did.

What could explain it instead: The source reports no significant correlation between sweat sodium concentration and age, body size, training experience, training volume, OR sweat rate. So the variation is not explained by how hard or how much someone sweats, which removes the obvious mechanism and leaves the spread unexplained.

Lara et al., interindividual variability in sweat electrolyte concentration in marathoners · J Int Soc Sports Nutr 2016;13:31

Across 1,944 sweat tests from 1,304 people, regression models explained only 17 to 23% of the variation in whole-body sweat sodium. Age group, race or ethnicity, relative humidity, exercise duration, pre-exercise urine specific gravity, exercise fluid balance and dietary sodium intake had no significant effect in any model. What did register was season of the year, energy expenditure, exercise mode, whole-body sweating rate, body mass and sex.

Measured in: 1,304 people contributing 1,944 sweat tests, compiled retrospectively from a sports science institute's testing records; the largest model predicted whole-body sweat sodium from forearm patches

Every author is employed by PepsiCo R&D at the Gatorade Sports Science Institute, which sells the category this finding is used to discuss. The result cuts two ways: no lookup table will tell you where you sit, and the study identified the failure of the obvious predictors, not a mechanism. Sex did register as a predictor, so individual characteristics explain nothing overstates it. It is a retrospective compilation of tests collected for other purposes, so nothing was controlled.

What could explain it instead: The tests were run on people who sought out or were recruited into sweat testing at an institute facility, mostly athletes, so the sample is selected toward the trained and the already worried about salt. Season of the year is used as a proxy for heat acclimatization rather than measured, and it carries training load, clothing and diet along with it, so the strongest predictor in the model is the one whose meaning is least pinned down.

Baker et al., explaining variation in sweat sodium concentration · J Appl Physiol (1985) 2022;133(6):1250-1259

In young adults the beverage hydration index rose progressively with beverage sodium, independent of glucose or amino acid content. The highest, a 60 mmol/L sodium amino-acid drink, reached 1.24 plus or minus 0.10 against water at 1.00 by two hours, about 24% more fluid retained. In older adults the top performer was the 30 mmol/L drink at 1.20 plus or minus 0.13, and retention in the 60 mmol/L drink was still changing past two hours.

Measured in: 24 euhydrated adults: 12 young (23 plus or minus 3 years, 7 men and 5 women) and 12 older (67 plus or minus 6 years, 5 men and 7 women), each drinking 1 L of five beverages on separate visits

This is not the independent comparison it is usually described as. It was partly funded by an unrestricted gift from Entrinsic Health Solutions, which manufactures the two amino-acid beverages that scored highest, alongside an NIH training grant, though the authors report no conflict. The outcome is urine volume over two to four hours and nothing else: not energy, cognition, symptoms or performance, and the paper does not claim otherwise. Twelve people per age group is small.

Clarke et al., a randomized trial to assess beverage hydration index in healthy older adults · Am J Clin Nutr 2019;109(6):1640-1647

The authors' own sentence: the addition of electrolytes alone, in the range found in sports drinks, did not consistently improve the beverage hydration index versus water. Carbohydrate-electrolyte and dipeptide-electrolyte drinks did beat water, both reaching 1.15, at 120 and 240 minutes respectively. Plain water produced more reported stomach bloating than either.

Measured in: 19 healthy young adults, 15 men and 4 women, each completing four randomized sessions of 1 L with urine and body mass collected hourly to 240 minutes

Funded by The Coca-Cola Company under grant AWD000506, which a page citing it should say out loud. The outcome is again four hours of urine. The same paper reports that electrolyte content made the largest single absolute contribution to the index, over 12%, and concludes that electrolytes matter most of the three additives, so electrolytes do nothing is not what it found. What it found is that electrolytes on their own did not consistently beat water.

Millard-Stafford et al., the beverage hydration index, influence of electrolytes, carbohydrate and protein · Nutrients 2021;13(9):2933

A systematic search screened 1,464 records and found 18 randomized trials in the entire published literature testing a change in daily water intake against any health outcome. Ten reported at least one positive result and eight reported negative ones. Two outcomes held up: recurrent kidney stones, about 15 fewer stone events per 100 people over five years, and weight loss, 44 to 100% more than control. Single trials each suggested benefit for migraine, urinary tract infection, diabetes control and low blood pressure.

Measured in: 18 trials in varied adult populations, 15 parallel-group and 3 crossover, with interventions lasting from 4 days to 5 years

The review does not pool the trials statistically, so these are their findings summarized rather than a combined effect size, and the authors call both the quantity and the quality of the evidence limited. Most included trials recruited people who already had the condition being studied, so the kidney stone result belongs to recurrent stone formers rather than to the general reader. Nothing in the set tested energy, mood, skin or the other things water is popularly credited with.

Hakam et al., outcomes in randomized clinical trials testing changes in daily water intake, a systematic review · JAMA Netw Open 2024;7(11):e2447621

In 50 habitual coffee drinkers, three days of four cups of coffee a day (4 mg/kg caffeine) left total body water no different from three days of water: 113.5 versus 113.3 lb (51.5 plus or minus 1.4 versus 51.4 plus or minus 1.3 kg). Twenty-four-hour urine volume matched too, 2,409 versus 2,428 mL, as did blood and urine hydration markers and body mass. Urinary sodium ran higher on coffee (p = 0.02).

Measured in: 50 men who habitually drank 3 to 6 cups of coffee a day, each completing two three-day trials, coffee or water, in a counterbalanced crossover with food, fluid and physical activity controlled

Everyone was a male habitual coffee drinker taking a moderate 4 mg/kg dose, so it does not describe someone who rarely drinks coffee or takes far more, and tea was not tested. One author is employed by PepsiCo. Total body water was measured by deuterium dilution, the same isotope method behind the daily-target finding, which makes this a measured match rather than an absence of measurement.

Killer, Blannin and Jeukendrup, moderate daily coffee intake and hydration, a counterbalanced crossover in a free-living population · PLoS One 2014;9(1):e84154

The figure in general circulation has no primary measurement behind it that we could locate. The document it appears to descend from is indexed under the MeSH major topic Models, Educational, and its own indexed opening describes it as offering a theoretical exercise to students.

Measured in: Not a study of participants as far as the indexed record shows; the full text is behind a publisher wall we could not pass

We could not obtain the full text: the publisher and every mirror refused it, so the description above rests on the indexed record rather than on our having read it. Tracing the popular figure to this document is our own reading of the citation trail, not something the document claims. It has been cited six times in twenty-one years, which is hard to reconcile with it being the source of a globally repeated number, so the plain position is that the figure’s provenance is unclear rather than settled.

Weissenberg, insensible water loss during sleep, a theoretical exercise · Adv Physiol Educ 2005;29(4):213-5

The one study that monitored skin evaporation continuously through a whole night reported it in watts per square meter and never converted it to milliliters. In thermoneutral conditions evaporation fell across the night by about 12 W/m2, alongside falling skin and rectal temperature. In warm conditions it stayed steady. Evaporation also stopped in bursts during each REM period, which recurred every 80 to 90 minutes.

Measured in: Sleeping adults in a climate chamber run from 89.6–103.1°F (32–39.5°C) at 45% relative humidity. The indexed record gives neither a participant count nor a sex breakdown.

The falls-across-the-night pattern held in the thermoneutral conditions only; in the warm conditions evaporation held level, so the popular figure's steady-rate assumption is wrong in one condition and right in the other. Thermoneutral here means a chamber at 89.6–93.2°F (32–34°C), which is neutral for an uncovered sleeper and far warmer than a bedroom, so none of these numbers describe a person under a duvet at 64.4°F (18°C). It is a 1977 paper and we found no replication.

Henane et al., variations in evaporation and body temperatures during sleep in man · J Appl Physiol Respir Environ Exerc Physiol 1977;42(1):50-5

It has been tested, in narrow forms. In one trial, 64 young adults were water-restricted for 12 hours overnight, then randomized to 500, 200, 100 or no milliliters on waking, with cognition and mood as outcomes; rehydrating helped. A separate 12-week trial had Japanese adults raise their daily water intake against no change, tracking health markers.

Measured in: 64 young adults in Baoding, China, fasted 12 hours overnight then randomized to 500, 200, 100 or no milliliters, with cognition and mood measured 90 minutes later; and, separately, Japanese adults in a 12-week daily water-intake trial

What has not been tested is the habit as most people practice it: drinking on waking after a normal night in which you also drank normally the evening before. The trial that tested drinking on waking withheld evening fluid first, so it measures recovery from a mild overnight deficit rather than the value of the glass itself. The habit costs nothing and there is no argument here for stopping it. It is more widely recommended than the evidence supports.

Zhang et al., different amounts of water supplementation after 12 h water restriction, cognition and mood in young adults · Int J Environ Res Public Health 2020;17(21):7792 effect of increased daily water intake and hydration on health in Japanese adults · Nutrients 2020;12(4):1191

The double-strength drink performed no better than the single. The single-strength drink beat plain water on urine output and lost less body weight (0.8 lb (0.38 kg) lost versus 1.3 lb (0.57 kg) on water, p=0.02); at double strength the difference was not significant (0.9 lb (0.43 kg) lost, p=0.08).

Measured in: 18 exercise-trained adults, 8 men and 10 women, fasted and euhydrated, drinking 1 L on three occasions about a week apart

Funded in part by Nuun and Company, Inc. The quoted sentence is narrower than it reads on its own. In the paper's discussion, additional means the double-strength dose rather than electrolytes as such, and the same study found single-strength Nuun improved fluid balance against water. Everything measured is urine, body weight and blood pressure over four hours in people sitting still. Nothing about how anyone felt, thought or performed was recorded.

Pence and Bloomer, impact of Nuun electrolyte tablets on fluid balance in active men and women · Nutrients 2020;12(10):3030

Over four weeks of daily use, the group-by-time interactions were in tolerability, bloating and cramping, and also in mean corpuscular volume, estimated glomerular filtration rate and potassium.

Measured in: 40 adults, men and women, randomized four ways with 10 per group, over four weeks with visits at 0, 2 and 4 weeks

The authors describe the physiological changes as little to no change rather than none, which is the more careful phrasing and the one to use. The trial was funded by the manufacturer of the product it tested, and it set out to measure safety rather than benefit.

Bloomer et al., randomized trial to assess the safety and tolerability of daily intake of an allulose amino acid-based hydration beverage · Nutrients 2024;16(11):1766

Half of the maximum melatonin suppression was reached at roughly 50 to 130 lux, with the response saturating near 200 lux. Phase resetting was half-maximal at 80 to 160 lux and saturated near 550 lux. Both half-maximal figures sit inside the range of ordinary room illumination, and because the curve is steep at the bottom, the first lamp turned off does more than the last one.

Measured in: 23 healthy young men and women studied as inpatients on a fixed schedule, each given a single measured evening light exposure at a set circadian phase

The exposure is one controlled episode against a dim laboratory background, which is a cleaner light history than any evening at home produces. It measures melatonin and clock phase, and it does not report how the participants then slept.

Zeitzer et al., sensitivity of the human circadian pacemaker to nocturnal light · J Physiol 2000;526(Pt 3):695-702

The group half-maximal dose was 24.6 lux (95% CI 21.3 to 28.4), but individual half-maximal doses ran from 6.0 lux in the most sensitive person to 349.8 lux in the least. That is more than a fifty-fold spread inside one young healthy sample.

Measured in: 55 healthy young adults analyzed (56 completed, 61 enrolled), each measured across a range of evening light intensities within-subject

Everyone measured was young and healthy, and sensitivity also moves with age, lens transmission and recent light history, so the spread in a general population is likely wider rather than narrower. A reader cannot find out where they sit on this curve at home.

Phillips et al., high sensitivity and interindividual variability in the response of the human circadian system to evening light · PNAS 2019;116(24):12019-12024

The consensus recommendation is a maximum of 10 melanopic EDI lux at the eye in the three hours before bed, and 1 lux or less during sleep. A visual display on its own, with no other light in the room, can deliver over 60 melanopic EDI lux. Adjusting the spectral content of a current display, which is what a night mode does, cuts melanopic output by about 50% and leaves luminance where it was.

Measured in: Not a participant study. An international expert consensus setting exposure limits from pooled laboratory dose-response work, expressed in melanopic equivalent daylight illuminance

The roughly 50% melanopic reduction is for adjusting a display’s spectral content, not for wearing a filter, and the paper makes no claim about what happens to perceived brightness. Its authors disclose funding and consultancy across the lighting industry, including Signify, Lutron, Samsung, Velux, f.lux and BIOS.

Brown et al., recommendations for daytime, evening and nighttime indoor light exposure in healthy adults · PLoS Biol 2022;20(3):e3001571

Across 17 trials, the review found the evidence on sleep quality indeterminate; that outcome rests on 6 trials and 148 participants. For visual fatigue it reports there may be little or no difference, at low certainty.

Measured in: 17 randomized controlled trials in adults, 5 to 156 participants each, with follow-up from less than one day to five weeks

Very low certainty for the sleep outcome, which is the floor of the grading scale, and the sleep-quality finding rests on 6 of the 17 trials rather than the full set. These two lens rows are not independent: all three trials in the pooled crossover set are included studies inside the Cochrane review, and two of them carry its sleep-quality outcome. Same trials, two analyzes.

Singh et al., blue-light filtering spectacle lenses for visual performance, sleep and macular health in adults · Cochrane Database Syst Rev 2023;8:CD013244

Sleep outcomes did not differ across the three conditions. In a post-hoc split, participants averaging more than 6.8 hours of sleep a night did better on sleep efficiency and wake after sleep onset in the no-phone condition than with the phone in either setting, and the two phone settings did not differ from each other.

Measured in: 167 emerging adults aged 18 to 24, 71.3% female, randomized to Night Shift on, Night Shift off or no phone use in the hour before bed

The one positive comparison is a post-hoc subgroup and it separates having the phone from not having it, rather than separating warm light from cool. Young adults who already sleep well are also the group with the least room to improve.

Duraccio et al., does iPhone Night Shift mitigate negative effects of smartphone use on sleep outcomes in emerging adults · Sleep Health 2021;7(4):478-484

Pooled objective effects were small: sleep efficiency g=0.31 (95% CI -0.05 to 0.66, k=7) and total sleep time g=0.32 (95% CI 0.01 to 0.63, k=6). Self-reported sleep quality on the Pittsburgh index moved much further, g=-1.25 (95% CI -2.39 to -0.11, k=3), and the larger combined effects sat with sleep-disordered and psychiatric patients rather than with healthy sleepers.

Measured in: 12 studies of interventions that reduce short-wavelength light at night, in adults spanning healthy sleepers and clinical groups

The self-reported figure rests on three studies with an interval that nearly touches zero, so its size should not be read closely. Amber lenses and screen filters are hard to blind, so expectation sits inside the self-reported result, and a person who reports sleeping better is still reporting something they experienced.

Shechter et al., interventions to reduce short-wavelength light exposure at night and their effects on sleep · Sleep Adv 2020;1(1):zpaa002

A 16-person panel working from 522 empirical articles reached consensus that screen use impairs sleep health in children and adolescents, and that the content consumed before sleep does too. It did not reach consensus that light from presleep screen use impairs sleep, and it reached no consensus on any screen statement for adults.

Measured in: Modified Delphi RAND/UCLA panel of 16 experts across five meetings, screening 2,209 articles down to 522 empirical papers and 52 reviews, with 35 experimental studies carrying the causal question

This is a modified Delphi expert panel rather than pooled trial data, so it reports what a body of specialists could and could not agree on. The useful part is what the panel would not endorse: consensus was NOT reached that evening light affects sleep, for any of the three age groups, and adults failed all three statements.

Hartstein et al., the impact of screen use on sleep health across the lifespan, a National Sleep Foundation consensus statement · Sleep Health 2024;10(4):373-384

Light measured near eye level in people's own homes before bed was bright enough in 48% of homes to suppress melatonin by at least 50% for an average sensitivity. Applying the measured range of individual sensitivities to a single average home gave predicted suppression anywhere from 0 to 87%. Homes lit with energy-efficient bulbs delivered about twice the melanopic illuminance of homes lit with incandescents.

Measured in: 59 community-dwelling adults aged 18 to 65 in Melbourne, Australia, 163 evenings of wearable spectrophotometry at eye level with actigraphy alongside

Suppression here is predicted from the measured light using dose-response curves borrowed from a separate laboratory sample, not assayed in the people living in those homes. Housing stock, lamp choice and evening habits differ between countries, and the measurement ran across the southern autumn and winter.

What could explain it instead: Homes that are brightly lit in the evening differ from dim ones in household size, working hours, screen habits and how late the occupants stay up, and each of those shifts sleep timing on its own. The brightness and the behavior arrive together, and a single evening of measurement cannot separate them.

Cain et al., evening home lighting adversely impacts the circadian system and sleep · Sci Rep 2020;10:19110

A pre-registered evening light manipulation produced about 14% melatonin suppression and no matching difference in sleep architecture, slow-wave activity, subjective sleep quality, sleepiness, sensory processing or next-day vigilance. Suppression and sleep came apart in the same participants on the same nights.

Measured in: 29 healthy volunteers aged 18 to 30, 15 of them women, in a laboratory protocol with polysomnography

Twenty-nine young healthy people, one night per condition, which is enough to show that suppression and sleep can separate and not enough to exclude a small effect. It also says nothing about what repeated nightly suppression does over months, which is the exposure most readers actually have.

Blume et al., melatonin suppression does not automatically alter sleepiness, vigilance, sensory processing or sleep · Sleep 2022;45(11):zsac199

With display content held constant and only melanopic irradiance varied, sleep onset latency lengthened in a dose-dependent way, melatonin concentration fell and melatonin onset was delayed, both dose-dependently. Subjective alertness differed between the low- and high-melanopic conditions but showed no dose-response across the intermediate levels.

Measured in: 72 healthy men exposed to display light in the evening before a laboratory sleep episode, across graded melanopic irradiance levels

Alertness was the one outcome without a dose-response: it did differ between the low and high melanopic conditions, but it did not scale across the intermediate levels the way melatonin and sleep latency did. Participants were 18 to 35.

Schöllhorn et al., melanopic irradiance defines the impact of evening display light on sleep latency, melatonin and alertness · Commun Biol 2023;6(1):228

Pooling only the double-blind randomized crossover trials, the lenses moved sleep onset latency by -4.86 minutes (95% CI -20.23 to 10.52, p=0.54) and total sleep time by +8.75 minutes (95% CI -35.31 to 52.82, p=0.70), with no significant effect on sleep efficiency or wake after sleep onset. Heterogeneity was I-squared 0%, so the trials agreed with each other.

Measured in: 3 double-blind randomized crossover trials, 49 participants in total, actigraphy outcomes only

Forty-nine people is a small pool and intervals that wide would not have detected a modest benefit, so this shows agreement between the trials that controlled for expectation rather than a demonstrated absence of any effect. Actigraphy estimates sleep rather than staging it. These two lens rows are not independent: all three trials in the pooled crossover set are included studies inside the Cochrane review, and two of them carry its sleep-quality outcome. Same trials, two analyzes.

Luna-Rangel et al., efficacy of blue-light blocking glasses on actigraphic sleep outcomes · Front Neurol 2025;16:1699303

Every extra 10 minutes of screen use in bed cost 3 minutes of total sleep (95% CI -6 to -1). Split by type, 10 minutes of interactive use cost 9 minutes (95% CI -16 to -2) against 4 minutes for passive use (95% CI -7 to 0), gaming cost 17 minutes (95% CI -28 to -7), and nights with multitasking cost 35 minutes against nights without (95% CI -67 to -4). Use in the 2 hours before getting into bed showed no association with most sleep measures that night.

Measured in: 79 New Zealand youths, 47 of them male (59.5%), mean age 12.9, with screen use captured objectively by video rather than recalled, over multiple nights each

The confidence interval on passive use touches zero, so the clean two-to-one ratio between interactive and passive is softer than it looks. This is nights compared within the same children rather than an intervention, so it describes what happened on nights they chose a given activity.

What could explain it instead: Nights when a child games or multitasks in bed are also nights they were more awake, more wound up or dealing with something on their mind, and that arousal would delay sleep whatever the screen was doing. Weekends, later bedtimes and social plans cluster with the heavier screen nights.

Brosnan et al., screen use at bedtime and sleep duration and quality among youths · JAMA Pediatr 2024;178(11):1147-1154

The gap between distal skin temperature (hands and feet) and trunk skin temperature was the strongest predictor of how quickly sleep began, ahead of core temperature itself and ahead of melatonin.

Measured in: Laboratory sleep-onset studies in healthy young adults at the Basel chronobiology unit

The finding is about the ordering of physiological events, not about a treatment. It establishes why warm feet in a cool room help; it does not tell you what any bedroom should be set to. The 1999 Nature paper is a one-page communication with no abstract; the ranking above is from the companion paper in 2000.

Kräuchi et al., functional link between distal vasodilation and sleep-onset latency · Am J Physiol Regul Integr Comp Physiol 2000;278(3):R741-8 Kräuchi et al., warm feet promote the rapid onset of sleep · Nature 1999;401(6748):36-7

Sleep was most efficient and least restless between 68°F (20 °C) and 77°F (25 °C), with a 5 to 10% fall in sleep efficiency as the bedroom warmed from 77°F (25 °C) to 86°F (30 °C). The relationship was nonlinear and the optimum varied widely between individuals.

Measured in: 50 community-dwelling older adults in Boston, 41 of them women, mean age 79, monitored in their own bedrooms over 12 months

Measured in a group whose thermoregulation is already less efficient than a younger adult's, so the range is theirs rather than everyone's. The authors emphasize the between-person spread and call for individual adjustment rather than a single target.

What could explain it instead: Observational and in real homes: people who set warmer rooms differ from people who set cooler ones in health, activity and housing quality. Poor sleepers may also change the thermostat because they are sleeping badly, which reverses the arrow.

Baniassadi et al., nighttime ambient temperature and sleep in community-dwelling older adults · Sci Total Environ 2023;899:165623

The warmest nights, above 86°F (30 °C), cost about 14 minutes of sleep against the nights with the least temperature-attributed loss. The loss is driven jointly by delayed sleep onset and by waking earlier, with sleep offset advancing above about 59°F (15 °C). The effect is more than twice as large in older adults and roughly three times as large in lower-income countries, and slightly larger in women than men.

Measured in: 47,628 adults across 68 countries, more than 7 million nights of wearable sleep tracking linked to local daily weather

Consumer wearables estimate sleep rather than stage it, and the exposure is outdoor temperature rather than the actual bedroom. The comparator is the least-affected nights, not an average night. Both push toward underestimating the effect for people without cooling and overestimating it for people with it.

What could explain it instead: Warm nights arrive with longer daylight, more evening activity and more alcohol, all of which delay sleep independently. Air conditioning access tracks income, so the geographic pattern partly measures wealth.

Minor et al., rising temperatures erode human sleep globally · One Earth 2022;5(5):534-549

Water at 104 to 108.5°F (40 to 42.5 °C), one to two hours before bed, for as little as 10 minutes, shortened sleep onset latency and improved sleep efficiency and self-rated sleep quality.

Measured in: 17 studies met inclusion, 13 of them pooled quantitatively; adult sleepers across a range of ages

The pooled trials are small and heterogeneous in timing, water temperature and outcome measure. The timing is the active part: warming immediately before bed works against the same mechanism.

Haghayegh et al., before-bedtime passive body heating by warm shower or bath to improve sleep · Sleep Med Rev 2019;46:124-135

Pooled across three trials, stage N3 rose by 7.5 (SD 21.6) minutes per 7.5 hours and heart rate fell by 2.36 (SD 1.08) beats per minute on a passive high heat capacity mattress. REM did not change.

Measured in: A pooled re-analysis of three separate trials totalling 72 healthy adults: 15 men in Torino (mean 26.5), 33 men in Berlin (mean 46.2) and 24 post-menopausal women in Chicago (mean 62.5)

The two figures above come from one pooled dataset, not from independent replications; the Berlin trial is published separately and is counted inside the total. Its own report gives 32 men where the pooled paper gives 33, unexplained. The wide standard deviation on the N3 change means many individuals gained nothing. A co-author is affiliated to Technogel Italia, which makes the mattress, on a paper that declares no competing interests. This tests a passive heat-drawing surface, not an actively refrigerated bed.

Herberger et al., enhanced conductive body heat loss during sleep increases slow-wave sleep and calms the heart (pooled re-analysis of three trials) · Sci Rep 2024;14:4669 Herberger et al., sleep on a high-heat capacity mattress in healthy middle-aged men (Berlin trial, included in the pooled total above) · Sleep 2020;43(5):zsz271

Per 10 dB rise in night-time noise level, the odds of being highly sleep-disturbed rose 2.52 for road, 2.97 for rail and 2.18 for aircraft when noise was named as the source. Where noise was not named, road came in at 1.14 (still significant) and rail at 1.17 (not significant).

Measured in: 36 studies, 11 new plus 25 from the original WHO review, in residential populations near roads, railways and airports

The gap between the two sets of numbers is the point: asking people about noise inflates how much they attribute to it. The lower figures are the more conservative estimate and they still show an effect. Evidence graded moderate for the noise-named surveys and low to very low otherwise.

Smith et al., environmental noise and effects on sleep, an update to the WHO systematic review and meta-analysis · Environ Health Perspect 2022;130(7):076001

One night under 100 lux of overhead room light raised HOMA-IR by about 15%, against a 4% fall in the dim-light arm (p=0.018), and lowered the Matsuda index by about 16% against a 3% rise (p=0.048). Heart rate stayed elevated through the night and heart rate variability fell.

Measured in: 20 healthy adults, 14 women and 6 men, mean age about 27, in a laboratory sleep suite; 10 per arm

Twenty people and a single night, so this shows the mechanism is live in humans rather than the size of any long-term risk. The arms were parallel rather than crossed over, so the two groups were not the same people.

Mason et al., light exposure during sleep impairs cardiometabolic function · PNAS 2022;119(12):e2113290119

At about 660 to 835 ppm CO2 against 2,400 to 2,600 ppm, measured sleep quality improved, next-day sleepiness fell, and performance on a logical-thinking test improved.

Measured in: Two balanced crossover field experiments in single-occupancy student dormitory rooms, 14 and 16 participants, one week per condition

Thirty people in total, in student rooms, with ventilation changed by opening a window in the first experiment and by a fan in the second. Balanced in order rather than randomized. CO2 is a marker for stale air generally rather than necessarily the agent, and the two cannot be separated in this design. Wargocki is a co-author on this and on the ventilation study below, so this section rests largely on one group.

Strøm-Tejsen et al., the effects of bedroom air quality on sleep and next-day performance · Indoor Air 2016;26(5):679-686

Deep sleep was lower at the low ventilation rate across 23 bedrooms. In a subgroup of 12, light sleep and awakenings also rose. Next-day cognitive performance did not differ between conditions.

Measured in: 29 bedrooms with mechanical extract ventilation enrolled, 23 contributing the deep-sleep result and 12 the subgroup result, four weeks each, fan speed altered covertly in balanced order

Sleep was measured with wrist-worn trackers rather than polysomnography, which stage sleep only approximately. Balanced in order rather than randomized. The stronger part of the result rests on 12 bedrooms, so it is the weaker half of this claim and should not be overstated. Wargocki is a co-author here and on the CO2 study above.

Fan et al., a single-blind field intervention study of whether increased bedroom ventilation improves sleep quality · Sci Total Environ 2023;884:163805

After four weeks, response (a 50% or greater fall in Insomnia Severity Index) came with an odds ratio of 25.8 (95% CI 6.8 to 85.7) and remission an odds ratio of 19.7 (4.4 to 87.9) against a light control blanket, Cohen’s d 1.90. Daytime symptoms improved alongside.

Measured in: 120 adults with clinical insomnia alongside major depression, bipolar disorder, ADHD or generalized anxiety; 68.3% women, mean age 39.6 (18 to 77)

A clinical population with psychiatric diagnoses, so this does not automatically describe an otherwise healthy poor sleeper. The control was 1,535 g of plastic chains sewn into the same shape and size as the 17.6 lb (8 kg) blanket, weights were never disclosed to participants, and the blind broke once, at an airport metal detector. The confidence intervals are very wide, which is what an odds ratio of 25.8 in 120 people looks like. A heavy blanket also traps heat, which works against the cooling logic elsewhere on this page.

Ekholm et al., a randomized controlled study of weighted chain blankets for insomnia in psychiatric disorders · J Clin Sleep Med 2020;16(9):1567-1577

At 62.6°F (17 °C) against 71.6°F (22 °C), REM increased and the proportion of N2 fell, while N1 rose; deep sleep rose but not significantly. Total sleep time, sleep efficiency and wake after sleep onset did not differ. Wool sleepwear shortened sleep onset latency against cotton. Skin and core temperature variables together explained about 68% of the variance in sleep onset.

Measured in: 17 healthy young adults, 10 of them men, nine nights each of polysomnography, randomized across sleepwear, bedding and ambient temperature

Small, young, and funded by Australian Wool Innovation, which also funded the senior author and the first author’s scholarship, and two of the five authors are executives of that organization. That is the specific reason the fiber result needs independent replication. The temperature result is less exposed to it, since both fabrics were tested at both temperatures, but the measures that moved were sleep architecture rather than how long or how well anyone slept.

Shin et al., the effects of fabric for sleepwear and bedding on sleep at 17 °C and 22 °C · Nat Sci Sleep 2016;8:121-31

Across the whole sample the device was associated with a 2% lower sleeping heart rate and 7% higher heart rate variability. The sleep-stage gains were sex-split and conditional on cooling the first half of the night: about 14 more minutes of deep sleep in men and about 9 more minutes of REM in women, rather than both in everyone.

Measured in: 54 adults in their own homes, 27 women and 27 men, mean age 36 (range 21 to 74), eight nights of home sleep testing in a fixed device-off then device-on sequence, with no randomization and no blinding

Designed, run, analyzed and written by the company that sells the device. Nine of the ten authors give Eight Sleep as their affiliation, six hold shares, and the rest are paid consultants or former employees. The order was fixed rather than randomized, so ordinary week-to-week variation, seasonal drift and expectation all sit inside the effect. The sleep-stage results also depend on a specific cooling schedule, not on ownership.

Moyen et al., sleeping for one week on a temperature-controlled mattress cover improves sleep and cardiovascular recovery · Bioengineering (Basel) 2024;11(4):352

Over 9.5 years, the brightest nights against the darker half of the sample carried hazard ratios of 1.47 for myocardial infarction, 1.56 for heart failure, 1.32 for coronary artery disease, 1.32 for atrial fibrillation and 1.28 for stroke. The stronger associations in women were for heart failure and coronary artery disease, and in younger participants for heart failure and atrial fibrillation, rather than across every outcome.

Measured in: 88,905 UK Biobank participants, mean age 62.4, 56.9% women, one week of wrist light sensing covering about 13 million hours

Light was sensor-measured rather than recalled, which is better than most work in this area, and it was one week of measurement used to represent a decade. The comparator is the darker half of the sample, not a dark room. This is the same UK Biobank light-sensor substudy behind our mortality and psychiatric-risk findings on the morning-light page, which the paper states itself, so those are separate analyzes of one dataset rather than independent corroboration.

What could explain it instead: Bright nights track shift work, urban housing, alcohol and untreated sleep apnea, all of which carry cardiovascular risk of their own. Early undiagnosed illness also disturbs sleep and increases night-time waking with lights on, which reverses the arrow.

Windred et al., light exposure at night and cardiovascular disease incidence · JAMA Netw Open 2025;8(10):e2539031

Bedrooms measured at 5 lux or more during sleep carried an incidence rate ratio for new diabetes of 3.74 (95% CI 1.55 to 9.05) against bedrooms under 5 lux, and 3.19 after propensity adjustment. At a 3 lux threshold it was 2.74 (1.19 to 6.33).

Measured in: 678 older Japanese adults in the HEIJO-KYO cohort, 55.5% women, mean age 70.6, 128 of them above the 5 lux threshold, followed a median 42 months

Nineteen incident cases produced that ratio, so the confidence interval is very wide and the point estimate should not be read as a size. The exposure is a two-night measurement taken as standing for the next 42 months, so it describes those two nights more securely than it describes a habit. Its value is that it is a separate population from the UK Biobank work, measured with bedroom sensors rather than wrist sensors.

What could explain it instead: People who leave a light on differ in housing, night-time waking, urinary frequency and existing metabolic risk. Nocturia is both an early diabetes sign and a reason to keep a light on, which makes reverse causation plausible here.

Obayashi et al., bedroom lighting environment and incident diabetes mellitus, the HEIJO-KYO cohort · Sleep Med 2020;65:1-3

A week of sleeping in an eye mask improved paired-associate learning and reaction time on a psychomotor vigilance test against a week without one. Two of the four reported outcomes were null. A reanalysis disputed the learning result, placing it just short of statistical significance, while the vigilance result of about 6 ms stood, and argued the practical size is small.

Measured in: 89 adults aged 18 to 35 in a within-subject crossover, one week per condition at home

Blinding is impossible with a mask, so expectation is inside the result. The dispute that followed is about how much the effect is worth rather than whether it exists, and the authors have replied to it. Participants were young, mean age 21.

Greco et al., wearing an eye mask during overnight sleep improves episodic learning and alertness · Sleep 2023;46(3):zsac305 Rhodes, reanalysis of Greco et al. 2023 · Sleep 2023;46(8):zsad105 Greco et al., reply to Rhodes · Sleep 2023;46(8):zsad148

Across 38 studies the results split in both directions, with some showing improved sleep and others showing disrupted sleep. The reviewers graded the quality of evidence for continuous noise improving sleep as very low and noted potential effects on hearing.

Measured in: 38 studies in adults and infants, spanning laboratory and home settings, with wide variation in the noise used and how sleep was measured

This is a null in the sense that the pooled picture does not settle the question, not in the sense that masking never helps a specific person in a specific loud bedroom. The heterogeneity in what counted as white noise is severe enough that the studies are hard to compare at all.

Riedy et al., noise as a sleep aid, a systematic review · Sleep Med Rev 2021;55:101385

Two independent systematic reviews, one of 24 controlled trials and one of 39 articles, neither pooling their results, both conclude that a medium-firm surface is the best general choice for comfort, sleep quality and spinal alignment, with self-adjustable inflation performing well in the first.

Measured in: Adults with and without non-specific low back pain, across trials mostly running two to twelve weeks

Neither review pools its trials, so there is no combined effect size behind this, only two independent readings of a literature arriving at the same place. The underlying trials are small and short. The 2021 review states it received no grant funding and that its authors report no financial conflicts. Neither review identifies a firmness that suits everyone, and both note that body weight, sleeping position and the type of back problem all move the answer.

Radwan et al., effect of different mattress designs on sleep quality, pain and spinal alignment · Sleep Health 2015;1(4):257-267 Caggiari et al., what type of mattress should be chosen to avoid back pain and improve sleep quality · J Orthop Traumatol 2021;22(1):51

Caffeine is an antagonist at adenosine A1 and A2A receptors. It occupies them without activating them, so the sleep pressure that adenosine signals stops being felt while it carries on accumulating. The A2A receptor is the subtype the human sleep-EEG and pharmacogenetic evidence points to.

Measured in: A review synthesizing human sleep-EEG and pharmacogenetic studies with receptor-knockout and transgenic rodent work

The receptor-subtype attribution is assembled across species rather than measured in one human experiment, and the same review states that no human study has yet tested functional A1 receptor variants against the sleep EEG. Blocking a signal is also not the same as removing it, which is what matters here, and what no receptor study measures.

Reichert, Deboer & Landolt, adenosine, caffeine, and sleep-wake regulation: state of the science and perspectives · J Sleep Res 2022;31(4):e13597

The largest curated dataset covers 141 publications, 500 subject groups and 4,714 individuals, and its authors decline to publish a single canonical half-life because the between-person variation is too large for an average to describe anyone.

Measured in: Adults in 141 published caffeine pharmacokinetic studies, curated and re-analyzed rather than newly measured

Any single half-life figure you are given, including the familiar five hours, is a central point pulled out of a distribution this dataset deliberately refuses to collapse. What it does establish is which modifiers move it, which is the useful part.

Grzegorzewski et al., pharmacokinetics of caffeine: a systematic analysis of reported data · Front Pharmacol 2021;12:752826 (online 2022)

Pooling 24 controlled crossover studies: total sleep time down 45.3 minutes, sleep efficiency down 7.0%, sleep onset latency up 9.1 minutes, wake after sleep onset up 11.8 minutes, light N1 sleep up 6.1 minutes, and combined deep N3 and N4 sleep down 11.4 minutes. The pattern is that sleep gets lighter and shorter while the time to fall asleep barely moves.

Measured in: Healthy adults, low-to-moderate habitual caffeine consumers, across 24 controlled crossover studies

Twenty-one of the 24 studies measured sleep with polysomnography and three used activity monitors, which overestimate sleep duration, so the pooled estimate is not purely electrode-based. The authors state the results may not generalize to adolescents, adults over 65, caffeine-naive people or heavy habitual consumers, and the per-outcome pools are small: the deep-sleep figure rests on 127 participants. Of the 24 studies, 18 used polysomnography, 3 used activity monitors and 3 measured only self-reported sleep, so 21 measured sleep objectively in some form. Activity monitors overestimate sleep duration.

Gardiner et al., the effect of caffeine on subsequent sleep: a systematic review and meta-analysis · Sleep Med Rev 2023;69:101764

No effect on REM duration (mean difference -4.4 minutes, 95% CI -10.5 to 1.6, p = 0.127) or on REM as a proportion of sleep (-0.02%, p = 0.980), and no effect on REM onset latency (-1.5 minutes, p = 0.581). A second independent meta-analysis found REM proportion unaffected as well.

Measured in: Healthy adults across two independently conducted meta-analyzes of controlled crossover trials

This is an absence of a detected effect in modestly sized pools, not a demonstration that none exists: the REM duration analysis rests on 163 participants and its confidence interval still admits a loss of up to 10 minutes. This is what makes late caffeine feel survivable while it is costing deep sleep.

Gardiner et al., the effect of caffeine on subsequent sleep: a systematic review and meta-analysis · Sleep Med Rev 2023;69:101764 Chang, Cheng & Cheng, age- and dose-specific effects of caffeine on sleep: a meta-analysis of controlled crossover trials · Sleep Med 2025;136:106874

Across 22 controlled crossover trials and 956 participants, all measured by nighttime polysomnography: total sleep time down 34.67 minutes, sleep efficiency down 4.74%, slow-wave sleep proportion down 1.01%, sleep onset latency up 8.35 minutes, REM proportion unchanged.

Measured in: 956 healthy adults across 22 controlled crossover trials, polysomnography only, searched to March 2025

The subgroup contrasts by age and dose did not reach statistical significance between groups and the meta-regressions found no linear relationship with either, so the age and dose story in the title is weaker than it first appears. Its value here is convergence: two teams pooling an overlapping literature landed 10 minutes apart, which bounds the number better than either does alone.

Chang, Cheng & Cheng, age- and dose-specific effects of caffeine on sleep: a meta-analysis of controlled crossover trials · Sleep Med 2025;136:106874

Pooling 16 randomized trials of at least seven days, 25 strata and 1,010 people: systolic blood pressure rose 2.04 mmHg and diastolic 0.73 mmHg overall. Caffeine trials alone gave a larger rise, 4.16 systolic and 2.41 diastolic, than coffee trials, 1.22 and 0.49. Effects on heart rate were negligible.

Measured in: 1,010 adults across 16 randomized controlled trials of at least a week, median caffeine dose 410 mg a day in the caffeine arms

This measures blood pressure with regular intake over a week or more, not the acute response to a single cup, and the effect is small enough that the authors conclude the blood pressure effect of caffeine taken as coffee is minor. The literature searched ended in 2003. It does not describe what happens in people with an arrhythmia or poorly controlled hypertension, who were not the trial population.

Noordzij et al., blood pressure response to chronic intake of coffee and caffeine: a meta-analysis of randomized controlled trials · J Hypertens 2005;23(5):921-8

In a systematic review and meta-analysis of 45 publications yielding 327 effect estimates in sleep-deprived or sleep-restricted adults, caffeine taken after sleep loss improved simple reaction time with a large pooled effect (Hedges g about 1.11, where g of roughly 0.8 counts as a large effect), improved sustained-attention (vigilance) response time (g about 0.86), and improved vigilance accuracy (g about 0.68), all versus placebo. Doses in this research were modest, on the order of the caffeine in one to a few cups of coffee. The authors concluded caffeine is an effective countermeasure to the cognitive impairments of sleep loss.

Measured in: Adults tested during sleep deprivation or sleep restriction across 45 controlled studies; most designs were placebo-controlled crossover trials.

This is an acute effect measured against placebo, not a cure for lost sleep: the underlying sleep debt remains. Regular daily use builds tolerance, so the lift is largest in people who are not habituated or who are acutely sleep-deprived. Effect sizes vary by task, and taking caffeine too close to bedtime disrupts the sleep that follows, which is the timing caution documented elsewhere on this page.

Irwin et al., Effects of acute caffeine consumption following sleep loss on cognitive, physical, occupational and driving performance: a systematic review and meta-analysis · Neurosci Biobehav Rev 2020;108:877-888

A meta-analysis of 46 randomized placebo-controlled studies found that moderate caffeine doses of 3 to 6 mg per kilogram of body weight improved endurance time-trial performance: mean power output rose by about 3.0% (3.03 plus or minus 3.07%, small effect size about 0.23) and time-trial completion time improved by about 2.2% (2.22 plus or minus 2.59%, effect size about 0.41) versus placebo. The International Society of Sports Nutrition position stand concurs that 3 to 6 mg per kilogram is consistently ergogenic, with a minimal effective dose possibly as low as 2 mg per kilogram. For a 154 lb (70 kg) adult, 3 mg per kilogram is about 210 mg.

Measured in: Meta-analysis of 46 randomized placebo-controlled trials of endurance exercise; participants were predominantly trained young adults. The ISSN position stand reviews the broader exercise-performance literature.

The average gain is small in size and varies between people: within the same meta-analysis a few studies recorded slower times or lower power in some participants, so response is individual and partly genetic. Trial populations skew toward trained young men, so the exact numbers transfer less certainly to older, female, or untrained people. Very high doses raise side effects such as jitteriness and disrupted later sleep without adding benefit.

Southward et al., The effect of acute caffeine ingestion on endurance performance: a systematic review and meta-analysis · Sports Med 2018;48(8):1913-1928 Guest et al., International Society of Sports Nutrition position stand: caffeine and exercise performance · J Int Soc Sports Nutr 2021;18(1):1

Mean salivary caffeine half-life was 3.5 hours in smokers against 6.0 hours in non-smokers, with body clearance 155 against 94 mL/kg/h (p < 0.05) and no difference in volume of distribution. The direction is corroborated across the curated dataset four decades later.

Measured in: 13 healthy smokers and 13 healthy non-smokers, compared as intact groups rather than randomized

Twenty-six people in 1978, measured in saliva, and the two groups were compared as they came rather than assigned. The effect is large, consistent with the enzyme-induction mechanism and reproduced in the modern pooled dataset, which is why it holds at moderate rather than at preliminary.

What could explain it instead: Smokers differ from non-smokers in far more than tobacco: alcohol intake, habitual caffeine dose, diet, body composition and occupation all track smoking and several of them also influence CYP1A2. With intact groups and no randomization, the design cannot separate the smoke from the smoker.

Parsons & Neims, effect of smoking on caffeine clearance · Clin Pharmacol Ther 1978;24(1):40-5 Grzegorzewski et al., pharmacokinetics of caffeine: a systematic analysis of reported data · Front Pharmacol 2021;12:752826 (online 2022)

Mean elimination half-life was 7.88 hours in users of low-dose estrogen oral contraceptives against 5.37 hours in matched controls, a rise of about 47%, driven by a fall in plasma clearance from 1.75 to 1.05 mL/min/kg with no change in volume of distribution.

Measured in: Nine women on low-dose estrogen oral contraceptives for more than three months, matched to nine non-smoking, drug-free women of similar age, weight and ethnic origin

Eighteen women, one 162 mg dose, and formulations from 1985. Modern preparations carry lower estrogen doses still, so the size of the effect on a contemporary pill is an extrapolation even though the direction is corroborated in the pooled pharmacokinetic dataset.

What could explain it instead: Matched groups, not randomized. Women who take the pill differ from women who do not in ways that touch hepatic enzyme activity, including body composition and alcohol intake, and the matching covered age, weight and ethnicity but could not cover those.

Abernethy & Todd, impairment of caffeine clearance by chronic use of low-dose oestrogen-containing oral contraceptives · Eur J Clin Pharmacol 1985;28(4):425-8 Grzegorzewski et al., pharmacokinetics of caffeine: a systematic analysis of reported data · Front Pharmacol 2021;12:752826 (online 2022)

Mean salivary half-life was 8.3 hours in 57 pregnant women, range 3 to 16, against 3.4 hours in a reference group of 25 adult men and non-pregnant women. Values returned to normal within about a month of delivery. The same group's follow-up reports an average of 10.5 hours across the last four weeks of pregnancy against about 3 hours in non-pregnant women.

Measured in: 57 pregnant women followed through pregnancy and after delivery, against 25 adult men and non-pregnant women as the reference

Saliva rather than plasma, 1981 methods, and no correlation was found with age, weight, coffee intake or smoking, which means the study identified a large effect without explaining the variation inside it. The top of the range, 16 hours, is one woman's value across the whole pregnant group and not a figure attached to any particular week.

What could explain it instead: Pregnant and non-pregnant participants were compared as intact groups. Pregnancy arrives with changes in diet, nausea, smoking cessation, body water and drug use that move caffeine clearance on their own, so the effect measured is of being pregnant rather than of any single mechanism.

Knutti, Rothweiler & Schlatter, effect of pregnancy on the pharmacokinetics of caffeine · Eur J Clin Pharmacol 1981;21(2):121-6 Knutti, Rothweiler & Schlatter, the effect of pregnancy on the pharmacokinetics of caffeine (follow-up reporting the final four weeks) · Arch Toxicol Suppl 1982;5:187-92

In this dataset, liver impairment had moderate to strong effects on caffeine clearance, with large variability between individuals. Obesity and malaria showed no clear effect.

Measured in: Adults with liver disease and other conditions, pooled from the published pharmacokinetic literature

The stronger statement often quoted from this paper, that the reduction scales with the degree of hepatic impairment, is its introduction summarizing earlier work rather than a result of its own analysis.

Grzegorzewski et al., pharmacokinetics of caffeine: a systematic analysis of reported data · Front Pharmacol 2021;12:752826 (online 2022)

A meta-regression of dose and timing gives no cut-off for a cup of black tea (47 mg), 8.8 hours before bed for a cup of coffee (107 mg), and 13.2 hours for a standard pre-workout serving (217.5 mg). Total sleep time recovered 2.8 minutes for every extra hour before bed and worsened 0.2 minutes for every extra milligram.

Measured in: Modelled from the crossover trials contributing timing and dose data, 30 effect sizes across 262 participants

These are group means projected onto a 10 p.m. bedtime, and they define the cut-off by statistical significance while assuming a statistically significant loss of sleep is a clinically meaningful one, which the authors say plainly. Individual half-life runs from about 2 to 10 hours, so any single hour is a starting point for a personal test rather than a rule. The model also uses only the final dose of the day where a study gave several.

Gardiner et al., the effect of caffeine on subsequent sleep: a systematic review and meta-analysis · Sleep Med Rev 2023;69:101764

100 mg of caffeine produced no significant effect on any objective or subjective sleep outcome at any timepoint tested, including 4 hours before bedtime (p > 0.05 throughout), against placebo in the same participants.

Measured in: 23 men, mean age 25.3, moderate habitual caffeine intake under 300 mg a day, seven conditions each with a 48-hour washout, measured by in-home partial polysomnography

A null in 23 people is a small trial failing to find an effect, not a demonstration that none exists, and the participants were young, healthy and already caffeine-tolerant. The measurement was in-home partial polysomnography rather than a full sleep laboratory montage.

Gardiner et al., dose and timing effects of caffeine on subsequent sleep: a randomized clinical crossover trial · Sleep 2025;48(4):zsae230

400 mg taken 4 hours before bed reduced total sleep time by an estimated 50.6 minutes and deep N3 sleep by 29.7 minutes (both p < 0.001), and cut perceived sleep quality by 34.0% (p = 0.006). At 8 and 12 hours before bed, 400 mg still altered objective sleep while perceived quality was unaffected.

Measured in: 23 men, mean age 25.3, placebo-controlled double-blind crossover, in-home partial polysomnography

400 mg is roughly four brewed coffees at once, which is a deliberately high test dose rather than a normal evening. The gap between the objective loss at 8 and 12 hours and the intact perceived quality at those times is the finding with the most practical weight, and it comes from 23 young men.

Gardiner et al., dose and timing effects of caffeine on subsequent sleep: a randomized clinical crossover trial · Sleep 2025;48(4):zsae230

Using inherited variants to approximate randomized exposure, faster caffeine metabolism showed no clear effect on sleep duration (p = 0.603 univariable, 0.578 multivariable) and none on insomnia (OR 1.040, p = 0.248; OR 1.010, p = 0.890). It did reduce the likelihood of daytime napping. Higher caffeine consumption reduced daytime sleepiness.

Measured in: Genome-wide summary data on caffeine metabolism (9,876 people) and on sleep traits in up to 453,379 people, predominantly UK Biobank, all European ancestry

Mendelian randomization estimates lifelong average exposure, so it is close to silent on what a particular cup does at four in the afternoon; a null here is compatible with the polysomnography trials finding real acute effects. The sleep outcomes are self-reported UK Biobank measures rather than recorded sleep, and the samples are European-ancestry only.

What could explain it instead: The instrument for caffeine metabolism is drawn from genes that also govern the metabolism of other substances, so horizontal pleiotropy is the standing threat; the authors run multivariable models partly to address it. UK Biobank is also a self-selected, healthier-than-average, largely white British cohort, which distorts any behavior-linked association including how much coffee people report drinking.

Das et al., exploring the relationship between caffeine consumption, caffeine metabolism, and sleep behaviours: a Mendelian randomisation study · J Sleep Res 2026;35(1):e70147 (published online 2025 Jul 14)

In 185 healthy non-smokers, CYP1A2 activity measured by caffeine metabolite ratio did not differ significantly between the three genotypes of the intron-1 C/A polymorphism reported as rs762551. In 51 smokers it did (p = 0.008). Meanwhile genome-wide work implicates rs2472297 between CYP1A1 and CYP1A2 and rs4410790 near AHR, and the GWAS of caffeine metabolites itself hits those loci plus CYP2A6 and CD83, not rs762551.

Measured in: 236 healthy Caucasian adults for the genotype-phenotype test; 47,341 and 2,680 to 4,300-person European samples for the genome-wide work; 9,876 for the metabolite GWAS

The distinction being drawn is real but should be stated precisely: rs2472297 and rs4410790 are the lead variants from genome-wide studies of coffee and caffeine consumption, at loci that govern caffeine metabolism, rather than from a genome-wide study of metabolic rate directly. The metabolite GWAS reports loci rather than those two names. What the evidence supports is that rs762551 is a poor stand-alone predictor in non-smokers, not that a better consumer test exists.

What could explain it instead: Genotype-phenotype and GWAS designs are both observational. Habitual caffeine intake, smoking, diet and medication all induce or inhibit CYP1A2 and all track with each other, so an activity difference attributed to genotype can carry the induction environment with it. The consumption GWAS additionally measure a behavior that is shaped by culture and price, not only by physiology.

Sachse et al., functional significance of a C to A polymorphism in intron 1 of the CYP1A2 gene tested with caffeine · Br J Clin Pharmacol 1999;47(4):445-9 Sulem et al., sequence variants at CYP1A1-CYP1A2 and AHR associate with coffee consumption (rs2472297) · Hum Mol Genet 2011;20(10):2071-7 Cornelis et al., genome-wide meta-analysis identifies 7p21 (AHR) and 15q24 as determinants of habitual caffeine consumption (rs4410790) · PLoS Genet 2011;7(4):e1002033 Cornelis et al., genome-wide association study of caffeine metabolites · Hum Mol Genet 2016;25(24):5472-5482

A meta-analysis of 13 studies found that regular caffeine consumers had a lower risk of developing Parkinson's disease, with a pooled hazard ratio of 0.797 (95% CI 0.748 to 0.849) in the healthy cohorts, meaning about a 20% lower risk, and a lower rate of disease progression among people who already had Parkinson's (hazard ratio 0.834, 95% CI 0.707 to 0.984). A separate prospective cohort of 184,024 European adults (EPIC4PD) found the highest coffee intake carried a hazard ratio of 0.63 (95% CI 0.46 to 0.88) versus non-consumers, and measured caffeine and its metabolites in stored blood years before diagnosis were also inversely associated with later Parkinson's, which points to caffeine itself rather than to something else in coffee.

Measured in: Meta-analysis of 13 prospective cohorts, plus the EPIC4PD population cohort of 184,024 adults across six European countries with a nested biomarker analysis of 351 matched case-control pairs.

This is an association from observational studies, so it cannot prove that caffeine prevents Parkinson's; correlation is not causation. Two confounders matter most. Smoking is independently linked to lower Parkinson's rates and travels with coffee-drinking, which can inflate the apparent effect. Reverse causation is also possible: the disease process can begin years before diagnosis and may itself reduce a person's taste for coffee, making pre-illness intake look artificially protective. The blood-biomarker data measured years before diagnosis helps, but does not fully settle the causal question. No trial has tested caffeine to prevent Parkinson's.

Hong et al., The effect of caffeine on the risk and progression of Parkinson's disease: a meta-analysis · Nutrients 2020;12(6):1860 Zhao et al., Association of coffee consumption and prediagnostic caffeine metabolites with incident Parkinson disease in a population-based cohort · Neurology 2024;102(8):e209201

A dose-response meta-analysis of 21 prospective studies (997,464 participants, 121,915 deaths) found coffee intake inversely and non-linearly associated with death from any cause, with the largest reduction at about 4 cups per day: 16% lower all-cause mortality (95% CI 13% to 18%) versus non-drinkers. A separate dose-response meta-analysis of 28 prospective studies (1,109,272 participants, 45,335 cases) found each additional cup per day was associated with about 9% lower type 2 diabetes risk for caffeinated coffee (relative risk 0.91) and about 6% lower for decaffeinated (relative risk 0.94), reaching a relative risk near 0.67 at about 6 cups per day. Because decaffeinated coffee shows much of the diabetes benefit, these effects reflect coffee as a whole drink, not caffeine alone.

Measured in: Two dose-response meta-analyzes of prospective cohorts: over 997,000 adults for mortality and over 1.1 million adults for type 2 diabetes, drawn largely from general adult populations in the US, Europe, and Asia.

These are observational associations, so they cannot prove coffee lengthens life or prevents diabetes; correlation is not causation. Healthy-user bias is a real concern, since coffee drinkers may differ in diet, activity, and income. Smoking is a major confounder in older cohorts because smokers drink more coffee and die earlier, which can mask or distort the association if not fully adjusted. Critically, decaffeinated coffee carries much of the diabetes benefit, so this is about coffee the beverage and its many compounds, not isolated caffeine. No randomized trial has tested coffee against these hard endpoints.

Crippa et al., Coffee consumption and mortality from all causes, cardiovascular disease, and cancer: a dose-response meta-analysis · Am J Epidemiol 2014;180(8):763-775 Ding et al., Caffeinated and decaffeinated coffee consumption and risk of type 2 diabetes: a systematic review and a dose-response meta-analysis · Diabetes Care 2014;37(2):569-586

A meta-analysis of 16 studies found coffee drinkers had lower odds of advanced liver disease: any coffee versus none carried an odds ratio of 0.61 (95% CI 0.45 to 0.84) for cirrhosis and 0.73 (95% CI 0.58 to 0.92) for advanced fibrosis, with high intake associated with an odds ratio of 0.53 for cirrhosis, and the protection held in subgroups with alcoholic liver disease and hepatitis C. A separate updated meta-analysis of 16 studies found the risk of hepatocellular carcinoma, the main form of liver cancer, was about 40% lower for any coffee versus none (relative risk 0.60), with each additional cup per day associated with a relative risk of about 0.80. These findings concern coffee as a drink rather than isolated caffeine.

Measured in: Two meta-analyzes each pooling 16 observational studies (cohort and case-control), covering general adults and subgroups with existing liver disease such as chronic hepatitis C and alcoholic liver disease.

This evidence is observational and cannot prove coffee protects the liver; correlation is not causation. Reverse causation is a particular risk here, because people with early liver or digestive disease often lose their appetite for coffee, which can make coffee drinkers look healthier than they would otherwise. The authors of the liver-cancer analysis flagged exactly this possibility. Healthy-user and smoking confounding also apply, and the protection is attributed to coffee the beverage and its many compounds, not to caffeine alone. No randomized trial has tested coffee against these liver outcomes.

Liu et al., Coffee consumption decreases risks for hepatic fibrosis and cirrhosis: a meta-analysis · PLoS One 2015;10(11):e0142457 Bravi et al., Coffee reduces risk for hepatocellular carcinoma: an updated meta-analysis · Clin Gastroenterol Hepatol 2013;11(11):1413-1421.e1

In the UK Biobank (172,315 participants free of type 2 diabetes, coronary heart disease, and stroke at baseline), moderate coffee intake near 3 drinks/day was associated with a hazard ratio of 0.519 (95% CI 0.417 to 0.647) for new-onset cardiometabolic multimorbidity, about a 48 percent lower risk, and caffeine intake of 200 to 300 mg/day with a hazard ratio of 0.593 (95% CI 0.499 to 0.704), about 41 percent lower, versus non-consumers or under 100 mg/day. A separate UK Biobank multi-state analysis (185,112 participants, median 11.4 years) found coffee drinkers had lower risk of first transitions to type 2 diabetes (HR as low as 0.79), coronary heart disease (0.91), and stroke (0.87), and of progression from coronary heart disease or stroke to multimorbidity (HR as low as 0.56 and 0.60). Associations held for unsweetened coffee.

Measured in: UK middle-aged and older adults (UK Biobank), free of cardiometabolic disease at baseline

Observational, so association is not proof of cause; both analyzes draw on the same UK Biobank population, so this is not yet replicated across cohorts. Coffee drinkers may share other protective habits (healthy-user effect), and early undiagnosed illness can lower intake (reverse causation). Benefit was clearest for unsweetened coffee; added sugar or heavy cream would change the picture. Caffeine here comes mainly from coffee, not energy drinks.

What could explain it instead: Healthy-user bias and smoking, and both analyzes draw on the same UK Biobank population, so the association is not yet replicated in an independent cohort.

Lu X, et al. Habitual Coffee, Tea, and Caffeine Consumption, Circulating Metabolites, and the Risk of Cardiometabolic Multimorbidity. · The Journal of Clinical Endocrinology & Metabolism 2025;110(6):e1845-e1855 Sun D, et al. Association of coffee consumption with cardiometabolic multimorbidity: A prospective cohort study in the UK biobank. · Nutrition, Metabolism & Cardiovascular Diseases 2024;34(12):2779-2788

A dose-response meta-analysis of 10 prospective studies across 11 cohorts (723,825 participants, 30,169 atrial fibrillation events) found no increased atrial fibrillation risk with coffee. Each additional cup per day was associated with a relative risk of 0.98 (95% CI 0.97 to 1.00, P = 0.02), a roughly 2 percent lower risk per cup, though the confidence interval reaches 1.00 and the dose-response across 1 to 7 cups was not clearly linear (P for non-linearity = 0.25). This overturns the older assumption that caffeine provokes atrial fibrillation. Separately, high caffeine intake above about 400 mg/day can transiently raise blood pressure, so the reassurance applies to habitual moderate coffee rather than large single doses.

Measured in: General adult populations across prospective cohorts (North America, Europe, Asia)

Observational, so association is not proof of cause; the per-cup association is small and its confidence interval reaches 1.00, so the robust conclusion is no increased risk rather than a proven reduction. The safety signal covers habitual coffee at moderate intake; very high caffeine can briefly raise blood pressure and may trigger palpitations in caffeine-sensitive individuals.

Cao Y, et al. Association of Coffee Consumption With Atrial Fibrillation Risk: An Updated Dose-Response Meta-Analysis of Prospective Studies. · Frontiers in Cardiovascular Medicine 2022;9:894664

In two pooled US prospective cohorts (Nurses' Health Study, n=86,606 women; Health Professionals Follow-up Study, n=45,215 men; 131,821 total, 11,033 incident dementia cases over up to 43 years, median 36.8), the highest quartile of caffeinated-coffee intake carried a hazard ratio for dementia of 0.82 (95% CI 0.76 to 0.89) versus the lowest quartile, about 18 percent lower risk (141 vs 330 cases per 100,000 person-years). The same top quartile showed less subjective cognitive decline (prevalence ratio 0.85, 95% CI 0.78 to 0.93) and a slightly higher cognitive-test score (TICS mean difference 0.11, 95% CI 0.01 to 0.21). The difference was most pronounced at roughly 2 to 3 cups per day, and decaffeinated coffee showed no association. An earlier dose-response meta-analysis of 8 prospective studies (328,885 participants, 7,486 dementia cases) found no linear association per 1 cup/day for dementia (RR 1.01, 95% CI 0.98 to 1.05) or Alzheimer's disease (RR 1.01, 95% CI 0.95 to 1.07), so the strongest signal comes from high-quality large cohorts at moderate intake rather than from a uniform dose gradient.

Measured in: US health-professional adults free of dementia, Parkinson's, and cancer at baseline; mean baseline age 46 to 54 years

Observational, so association is not proof of cause; coffee drinkers differ in smoking, diet, and health behavior (healthy-user and reverse-causation concerns since early cognitive change can alter coffee habits). The benefit tracked caffeinated coffee and tea, not decaffeinated coffee. A large dose-response meta-analysis found no overall linear coffee-dementia association, so treat moderate intake as associated with lower risk rather than dose-dependent protection.

What could explain it instead: Coffee drinkers differ in smoking, diet, education and health behavior (healthy-user bias), and reverse causation is possible because early dementia can change coffee habits years before diagnosis.

Zhang Y, et al. Coffee and Tea Intake, Dementia Risk, and Cognitive Function. · JAMA 2026;335(11):961-974 Larsson SC, Orsini N. Coffee Consumption and Risk of Dementia and Alzheimer's Disease: A Dose-Response Meta-Analysis of Prospective Studies. · Nutrients 2018;10(10):1501

A systematic review and dose-response meta-analysis of observational studies (12 studies, 346,913 individuals, 8,146 depression cases) found higher coffee intake associated with a pooled relative risk of depression of 0.76 (95% CI 0.64 to 0.91) versus lower intake. The dose-response curve was J-shaped, with the strongest protective association near 400 mL/day (about 1.5 to 2 cups). Caffeine intake in prospective studies was also associated with lower risk (RR 0.84, 95% CI 0.75 to 0.93). Tea showed a weaker, borderline signal (RR 0.70, 95% CI 0.48 to 1.01).

Measured in: General adult populations across observational studies (mixed regions)

Observational, so association is not proof of cause; low mood itself can reduce interest in coffee (reverse causation), and coffee drinkers may differ in social activity and other habits. The J-shape means benefit flattened past about 2 cups a day rather than climbing with more. This addresses depression risk in general populations, not treatment of diagnosed depression.

Grosso G, et al. Coffee, tea, caffeine and risk of depression: A systematic review and dose-response meta-analysis of observational studies. · Molecular Nutrition & Food Research 2016;60(1):223-234

400 mg taken 0, 3 or 6 hours before bed cut objectively measured total sleep time by 1.1 to 1.2 hours, significant at every one of the three timepoints. By sleep diary the six-hour dose cost 41 minutes and did not reach significance, so the sleepers substantially under-reported a real loss.

Measured in: 12 healthy adults, 6 women and 6 men, mean age 29.3, sleeping at home

The gap between the measured and the reported result is the point: people do not notice most of what late caffeine costs them, which is why self-assessment is a poor guide here. Funded by a sleep-technology company, and one of its co-authors was that company's vice president for scientific affairs.

Drake et al., caffeine effects on sleep taken 0, 3, or 6 hours before going to bed · J Clin Sleep Med 2013;9(11):1195-200

In the only modern polysomnography dose-and-timing trial that genotyped its participants, there was no significant main effect of CYP1A2 rs762551 or of the adenosine-receptor variant ADORA2A rs5751876 on any objective or subjective sleep outcome (p > 0.05).

Measured in: 23 men, genotyped for the two variants consumer caffeine panels most often report

The authors state directly that the trial was not powered to detect genotype effects. This is a small study failing to find a difference, which is not the same as showing there is none, and it is the reason this sits at preliminary rather than being read as a refutation.

Gardiner et al., dose and timing effects of caffeine on subsequent sleep: a randomized clinical crossover trial · Sleep 2025;48(4):zsae230

Both concern the same ADORA2A variant, c.1083T>C (rs5751876). The 2007 study found caffeine induced insomnia-like sleep EEG exclusively in C-allele carriers. The 2017 population study found the association between caffeine intake and objective sleep only in T-allele carriers and not in the CC genotype. Both results stand and they point in opposite directions.

Measured in: 2007: a self-rated caffeine-sensitivity survey of more than 4,300 respondents plus a double-blind sleep-EEG arm. 2017: the São Paulo EPISONO population sample, adults aged 20 to 80, stratified by sex, age and socioeconomic status, with full polysomnography

The two designs are not equivalent, which is part of why they disagree: one gave a controlled 2 x 200 mg dose and recorded the sleep EEG, the other correlated self-reported habitual intake with one night of polysomnography in the general population. Neither has been replicated in a design capable of settling the direction, which is why a consumer report of an ADORA2A result cannot yet be acted on.

What could explain it instead: The 2017 arm is observational: caffeine intake was self-reported and people choose their own dose, so genotype may be shaping consumption rather than the sleep response, and the association could run through anything that tracks coffee drinking in São Paulo, including smoking, shift work and socioeconomic status. The 2007 sensitivity classification was self-rated before it was genotyped.

Rétey et al., a genetic variation in the adenosine A2A receptor gene (ADORA2A) contributes to individual sensitivity to caffeine effects on sleep · Clin Pharmacol Ther 2007;81(5):692-8 Nunes et al., the association between caffeine consumption and objective sleep variables is dependent on ADORA2A c.1083T>C genotypes · Sleep Med 2017;30:210-215 Reichert, Deboer & Landolt, adenosine, caffeine, and sleep-wake regulation (confirms the 2007 result was confined to C-allele carriers) · J Sleep Res 2022;31(4):e13597

In cultured mammalian skeletal-muscle cells, caffeine increased autophagy by promoting calcium-dependent activation of AMP-activated protein kinase (AMPK). The AMPK inhibitor Compound C reduced the autophagy marker LC3b-II and autophagic vesicle accumulation, and blocking the calcium-sensing kinases CaMKKbeta and CaMKII cut the response, in a caffeine dose-dependent manner. This is a cell-culture mechanism study. It maps a plausible pathway by which caffeine could support cellular housekeeping, and it does not measure any longevity, disease, or clinical outcome in animals or people.

Measured in: Cultured mammalian skeletal-muscle cells (in vitro); no human or whole-animal outcomes

Mechanistic cell-culture work only; caffeine concentrations and exposure in a dish do not translate to doses, tissue levels, or outcomes in living animals or people. Read this as a candidate pathway, not evidence that coffee extends lifespan or clears disease. Human longevity and autophagy claims for caffeine remain at the mechanistic stage.

Mathew TS, et al. Caffeine promotes autophagy in skeletal muscle cells by increasing the calcium-dependent activation of AMP-activated protein kinase. · Biochemical and Biophysical Research Communications 2014;453(3):411-418

Ten minutes of walking after each main meal lowered the three-hour incremental glucose area under the curve to 0.88 of the same total walking taken as one 30-minute daily bout (95% CI 0.78 to 0.99), a 12% reduction. After the evening meal the ratio was 0.78 (0.67 to 0.91), a 22% reduction.

Measured in: 41 adults with type 2 diabetes in New Zealand, mean age 60, mean diabetes duration 10 years, each completing both two-week conditions in randomized order under continuous glucose monitoring

The outcome is two weeks of continuous glucose monitoring, not HbA1c and not any clinical event. The evening advantage is entangled with what was eaten, since dinner carried the most carbohydrate in this group and was followed by the most sitting.

Reynolds et al., advice to walk after meals is more effective for lowering postprandial glycaemia in type 2 diabetes mellitus than advice that does not specify timing, a randomised crossover study · Diabetologia 2016;59(12):2572-2578

Pooling eight three-armed randomized trials, exercise after eating lowered the postprandial glucose excursion more than the same exercise before eating (SMD 0.47, 95% CI 0.23 to 0.70) and more than no exercise (SMD 0.55, 0.34 to 0.75). Exercise taken before the meal was not distinguishable from sitting still (SMD -0.13, -0.42 to 0.17). Meta-regression found the effect shrinking as the gap between the meal and the start of activity widened (estimate -0.0151 per minute, p = 0.001).

Measured in: 116 adults across eight trials, 40 women and 76 men, 47 of them with type 2 diabetes. Activity was mostly treadmill walking, 7 to 60 minutes, light to vigorous

Eight small crossover trials, 116 people in total, each measuring one meal in a laboratory. The timing advantage was clear in participants without diabetes (SMD 0.57, 0.30 to 0.83) and did not reach significance in those with type 2 diabetes (0.24, -0.14 to 0.62), which is the reverse of what the popular version of this finding assumes.

Engeroff et al., after dinner rest a while, after supper walk a mile? A systematic review with meta-analysis on the acute postprandial glycemic response to exercise before and after meal ingestion · Sports Med 2023;53(4):849-869

Across 28 intervention studies that directly compared physical activity before a meal against activity after it, 21 measuring a single bout and 7 measuring multiple weeks, no significant difference in glycemia was found either acutely or after weeks of training.

Measured in: Adults across 28 studies spanning normal glucose tolerance, overweight and obesity, prediabetes and type 2 diabetes

The authors graded the evidence as low to moderate certainty, with wide variation in design, glucose measure and population. This is a failure to detect a difference across heterogeneous studies, which is a weaker statement than a demonstration that timing does not matter. It also pools trials that were not restricted to the three-armed design used by the review that did find a timing effect, and the two disagree. The review carries a 2026 corrigendum.

Slebe et al., the effect of preprandial versus postprandial physical activity on glycaemia, meta-analysis of human intervention studies · Diabetes Res Clin Pract 2024;210:111638

Breaking up sitting with light-intensity walking, in the included trials either 2 minutes every 20 minutes or 5 minutes every 30, lowered postprandial glucose against uninterrupted sitting (SMD -0.72, 95% CI -1.03 to -0.41) and postprandial insulin (-0.83, -1.18 to -0.48). Walking speeds ran from 1.0 to 2.7 mph (1.6 to 4.4 km/h) or a self-selected comfortable pace.

Measured in: 461 participant-conditions across 166 unique people, walking at 1.0 to 2.7 mph (1.6 to 4.4 km/h)

The count is of conditions rather than of people, which is a smaller evidence base than it first reads as. Walking speeds were slow, which is the point: this is not exercise.

Buffey et al., the acute effects of interrupting prolonged sitting time in adults with standing and light-intensity walking on biomarkers of cardiometabolic health, a systematic review and meta-analysis · Sports Med 2022;52(8):1765-1787

Standing instead of sitting lowered postprandial glucose (SMD -0.31, 95% CI -0.60 to -0.03) with no measurable change in insulin or systolic blood pressure. Light-intensity walking beat standing on both glucose (-0.30, -0.52 to -0.08) and insulin (-0.54, -0.75 to -0.33).

Measured in: The same seven randomized crossover trials, 461 adults for glucose and 358 for insulin, aged 18 to 79, mostly overweight or obese and sedentary

The standing effect is small and its confidence interval nearly touches zero. One of the constituent crossover trials, ten non-obese adults given 2-minute standing breaks every 20 minutes, found standing no better than remaining seated while walking breaks lowered the five-hour glucose area under the curve from 396 to 333 mg/dL (22.0 to 18.5 mmol/L) per 5 h.

Buffey et al., the acute effects of interrupting prolonged sitting time in adults with standing and light-intensity walking on biomarkers of cardiometabolic health, a systematic review and meta-analysis · Sports Med 2022;52(8):1765-1787 Bailey and Locke, breaking up prolonged sitting with light-intensity walking improves postprandial glycemia, but breaking up sitting with standing does not · J Sci Med Sport 2015;18(3):294-298

In 19 overweight adults, 2 minutes of light walking every 20 minutes cut the glucose area under the curve by roughly 25% and insulin by roughly 23% against uninterrupted sitting; moderate-intensity breaks gave about 29% and 23%. In 24 adults with type 2 diabetes, 3 minutes of light walking every 30 minutes lowered glucose, insulin and C-peptide responses across a full day of sitting, with a mean glucose difference of about 169 mg/h/dL (9.4 mmol/h/L).

Measured in: 19 overweight or obese adults aged 45 to 65, mixed sex (Dunstan); 24 inactive overweight or obese adults with type 2 diabetes, 14 of them men, mean age 62 (Dempsey)

Both are single-day laboratory crossovers with standardized test meals and enforced sitting between bouts. They measure what the breaks do to one day's glucose curve, not what the habit does to a person over months.

Dunstan et al., breaking up prolonged sitting reduces postprandial glucose and insulin responses · Diabetes Care 2012;35(5):976-983 Dempsey et al., benefits for type 2 diabetes of interrupting prolonged sitting with brief bouts of light walking or simple resistance activities · Diabetes Care 2016;39(6):964-972

Acarbose, a drug whose action is to blunt the post-meal glucose rise specifically, produced no reduction in major adverse cardiovascular events over a median of five years: 470 events (14%) on acarbose against 479 (15%) on placebo, HR 0.98, 95% CI 0.86 to 1.11. It did reduce progression to diabetes, 13% against 16%, rate ratio 0.82 (0.71 to 0.94).

Measured in: 6,522 Chinese adults with coronary heart disease and impaired glucose tolerance, randomized to acarbose or placebo and followed a median of 5.0 years

One trial, in one country, in people who already had coronary disease. It is also a drug trial rather than a walking trial: slowing carbohydrate absorption at the gut and having leg muscle take glucose up are different interventions with different effects on everything else. What it constrains is how much a lower post-meal number by itself can be read to promise. That trial was funded by the manufacturer of the drug it tested, and it still returned a null on cardiovascular events, which makes the null harder to dismiss.

Holman et al., effects of acarbose on cardiovascular and diabetes outcomes in patients with coronary heart disease and impaired glucose tolerance (ACE), a randomised, double-blind, placebo-controlled trial · Lancet Diabetes Endocrinol 2017;5(11):877-886

Three 15-minute walks at 3 METs, one after each meal, improved 24-hour glycemic control against a no-walking day (129 ± 24 vs 116 ± 13 mg/dL) and beat a single sustained 45-minute morning or afternoon walk over the three hours after dinner (p < 0.01).

Measured in: 10 inactive adults aged 60 and over with fasting glucose between 105 and 125 mg/dL and BMI under 35, each studied across three 48-hour stays in a whole-room calorimeter

Ten people. All meals were provided and every movement was measured in a metabolic chamber, so the control over the conditions is what this has instead of size, and the number should not be carried further than that.

DiPietro et al., three 15-min bouts of moderate postmeal walking significantly improves 24-h glycemic control in older people at risk for impaired glucose tolerance · Diabetes Care 2013;36(10):3262-3268

Fifteen minutes of slow walking after a carbohydrate-rich meal lowered glucose during the walk and delayed the peak (p = 0.003). Forty minutes also cut the two-hour incremental area under the curve (p = 0.014). The size of each woman's reduction tracked the size of her own seated glucose response (p < 0.001).

Measured in: 14 healthy women over 50, each completing all three conditions in randomized crossover order

Fourteen women, one meal each, capillary glucose. The correlation between benefit and the size of the seated response comes from within a single small sample and has not been reproduced as a formal moderator analysis.

Nygaard et al., slow postmeal walking reduces postprandial glycemia in middle-aged women · Appl Physiol Nutr Metab 2009;34(6):1087-1092

Sudden cold-water entry sets off an involuntary gasp and uncontrollable fast breathing in the first minute. In water deep enough to submerge, that is the moment people inhale water. Arrhythmias have been recorded in around 60% of people during head-out immersion following a maximal breath hold, so the breath hold is the multiplier rather than depth on its own.

The danger is highest with sudden entry into open or very cold water and lower in a shallow tub you sit upright in, but it is not zero. The practical reading is firm: sit at a depth you can stand up out of, do not go in alone at first, do not hold your breath, and keep breathing exercises away from the water.

Datta & Tipton, respiratory responses to cold water immersion, neural pathways, interactions and clinical consequences · J Appl Physiol 2006;100(6):2057-2064 Shattock & Tipton, autonomic conflict, a different way to die during cold water immersion · J Physiol 2012;590(14):3219-3230

Cold entry drives a sympathetic surge at the same time that breath-holding and face immersion drive the opposing parasympathetic response. This autonomic conflict can provoke a cardiac arrhythmia, and the risk rises with pre-existing heart disease and with breath-holding.

This is the physiology of why cold water loads the heart, drawn from experimental and drowning data rather than a controlled trial. Anyone with a cardiac history, uncontrolled blood pressure, or who is pregnant is the group for whom getting cleared first matters most.

Tipton et al., cold water immersion, kill or cure? · Exp Physiol 2017;102(11):1335-1355

In eight people cooled in 46.4°F (8°C) water to a core temperature of about 96.6°F (35.9°C), core temperature kept falling by 0.4 degrees Celsius for roughly 21 minutes after they got out. When shivering was blocked, the afterdrop grew to 1.1 degrees Celsius and lasted 89 minutes, and rewarming slowed by 37%.

This measured deliberate cooling to mild hypothermia, which is colder than a two-minute recreational dip, so the afterdrop after a short plunge is smaller. The mechanism is the same, and the practical point stands: anything that suppresses shivering, including alcohol, sedatives and rushing straight into a hot shower, makes the afterdrop deeper and longer, so rewarm gradually.

Giesbrecht et al., inhibition of shivering increases core temperature afterdrop and attenuates rewarming in hypothermic humans · J Appl Physiol 1997;83(5):1630-4 Giesbrecht, cold stress, near drowning and accidental hypothermia, a review · Aviat Space Environ Med 2000;71(7):733-52

Cold-water immersion reduced delayed-onset soreness and creatine kinase and modestly improved jump height, with ten to fifteen minutes at 41–59°F (5–15°C) the best-supported dose.

This is a dose-comparison analysis, so its finding is that particular dose windows work rather than that cold works generally, and it reduces the feeling of soreness rather than proving faster tissue repair. The benefit is covered in full on the cold exposure page; it is repeated here only to show what a build has to deliver.

Wang et al., dose of cold water immersion for recovery from exercise-induced muscle damage, network meta-analysis · Front Physiol 2025;16:1525726

Pseudomonas aeruginosa folliculitis is a skin infection commonly associated with pool and hot-tub water, often appearing as outbreaks among people who share the same inadequately disinfected water. It shows up as itchy red bumps a day or two after exposure.

Measured in: A single case in a 50-year-old woman, with a review of host risk factors and hot-tub outbreak literature

The cited source is a single case in a woman plus a review of who is susceptible, so it shows the infection is real and water-linked rather than how often a well-kept home plunge causes it. Regular water changes, filtration or ozone, and adequate disinfection lower the risk.

Jacob & Tschen, hot tub-associated Pseudomonas folliculitis, a case report and review of host risk factors · Cureus 2020;12(9):e10623

Pooled across randomized trials, stress fell sharply about 12 hours after cold-water immersion and at no other timepoint measured, while mood scores themselves did not change. Single studies reported better sleep and quality of life.

A benefit absent at four timepoints and present at one is a fragile pattern, and ten of the eleven trials were entirely male. This same-day lift is why people build a plunge; the full account, including the acute noradrenaline surge behind the alertness, is on the cold exposure page.

Cain et al., effects of cold-water immersion on health and wellbeing, systematic review and meta-analysis · PLoS One 2025;20(1):e0317615

A threat activates two arms on different timescales. The sympathetic nervous system fires within seconds, releasing adrenaline and noradrenaline that raise heart rate, blood pressure and available fuel. The slower hypothalamic-pituitary-adrenal (HPA) arm follows over minutes: the hypothalamus releases CRH, the pituitary releases ACTH, and the adrenal cortex releases cortisol, which mobilizes glucose and restrains the immune and inflammatory response. Cortisol then feeds back to switch the axis off. In a healthy system the whole loop is self-limiting and returns to baseline.

Measured in: Established human and animal neuroendocrine physiology described in major reviews.

That the acute response is adaptive is not in doubt. What varies between people, and what the rest of this page is about, is how reliably the system switches back off.

Chrousos, Stress and disorders of the stress system · Nat Rev Endocrinol 2009;5(7):374-381 Ulrich-Lai & Herman, Neural regulation of endocrine and autonomic stress responses · Nat Rev Neurosci 2009;10(6):397-409

The same mediators that protect during an acute challenge cause cumulative wear when they are called on too often, kept high for too long, or fail to shut off. McEwen named this allostatic load: the price the body pays for repeated adaptation. The pattern includes cortisol that stays elevated or loses its normal daily rhythm, a sympathetic system that stays active, and downstream effects on metabolism, immunity, the cardiovascular system and the brain. Cortisol itself is not harmful; the problem is chronic activation and the loss of self-limiting feedback.

Measured in: A conceptual framework grounded in human and animal neuroendocrine and cardiovascular data.

Allostatic load is a framework that organizes many findings well; it is not a single number you can measure cleanly in one person, and composite allostatic-load indices vary in how they are built.

McEwen, Protective and damaging effects of stress mediators · N Engl J Med 1998;338(3):171-179

Slowing the breath toward roughly six breaths a minute raises vagally mediated heart rate variability during and after the practice, a direct read on parasympathetic (rest-and-digest) engagement. A 2022 systematic review and meta-analysis of voluntary slow breathing found consistent increases in HRV, and a broader systematic review linked slow breathing to greater parasympathetic and reduced sympathetic activity alongside reports of calm and reduced arousal.

Measured in: Pooled controlled studies, mostly in healthy adults, sexes mixed and not always reported.

This measures an autonomic shift, not a health outcome. That slow breathing raises HRV is well supported; what that buys a person over months is a separate and less settled question.

Laborde et al., Effects of voluntary slow breathing on heart rate and heart rate variability: a systematic review and meta-analysis · Neurosci Biobehav Rev 2022;138:104711 Zaccaro et al., How breath-control can change your life: a systematic review on psycho-physiological correlates of slow breathing · Front Hum Neurosci 2018;12:353

Training people to raise their heart rate variability, usually through paced slow breathing with feedback, reduced self-reported stress and anxiety. A 2017 meta-analysis of 24 studies in 484 people found a large pre-to-post reduction, with a standardized effect around 0.8.

Measured in: 24 studies, 484 participants; sexes mixed and inconsistently reported.

The trials were small and often without an active comparison, so part of the effect may be expectancy. The self-directed version, slow breathing, is the free and low-risk part of it.

Goessl et al., The effect of heart rate variability biofeedback training on stress and anxiety: a meta-analysis · Psychol Med 2017;47(15):2578-2586

"Adrenal fatigue", the idea that ordinary chronic stress exhausts the adrenal glands so they underproduce cortisol and cause tiredness, is not a recognized medical diagnosis. A 2016 systematic review of 58 studies found the tests used to support it were inconsistent and the concept was not substantiated. Adrenal insufficiency (Addison's disease) is a real, separately diagnosed condition and is not what the term describes.

Measured in: Systematic review of 58 studies; participants across studies were mixed sex.

Saying the label is not validated is not saying the tiredness is imaginary. Fatigue is real and worth investigating; it is the specific adrenal-exhaustion explanation that the evidence does not support.

Cadegiani & Kater, Adrenal fatigue does not exist: a systematic review · BMC Endocr Disord 2016;16(1):48

Chronic work stress tracks with more heart disease. A collaborative meta-analysis of individual data from 197,473 people found that job strain (high demand with low control) was associated with about a 23% higher risk of coronary heart disease, hazard ratio 1.23 (95% CI 1.10 to 1.37).

Measured in: 197,473 adults across 13 European cohorts; both sexes.

This is observational. The effect is modest and stress is entangled with income, health behaviors and working conditions, so the number is an association, not a demonstrated cause.

What could explain it instead: Job strain clusters with lower socioeconomic position, smoking, physical inactivity and shift work; adjustment reduces but cannot remove these, and reverse causation (early ill health limiting job control) is possible.

Kivimaki et al., Job strain as a risk factor for coronary heart disease: a collaborative meta-analysis of individual participant data · Lancet 2012;380(9852):1491-1497

Sustained high cortisol and chronic stress act on brain regions dense in glucocorticoid receptors. The hippocampus (memory) and prefrontal cortex (regulation) tend to lose volume and function while the amygdala (threat detection) becomes more reactive, a pattern linked to depression, anxiety and impaired memory. Some of this is reversible when the stress lifts, and it varies across the lifespan.

Measured in: Human and animal neuroendocrine and imaging evidence synthesized in a major review.

The mechanism is well described but much of the causal detail is from animal models, and in people the brain changes and the mood disorder can drive each other, so the arrow is not one-directional.

Lupien et al., Effects of stress throughout the lifespan on the brain, behaviour and cognition · Nat Rev Neurosci 2009;10(6):434-445

In the INTERHEART study, psychosocial stress was one of the factors associated with first heart attack across 52 countries. People reporting permanent stress at work or home had roughly double the odds, and psychosocial factors together carried a population-attributable risk around 33%, of a similar order to smoking or high blood pressure in that analysis.

Measured in: 11,119 first-MI cases and 13,648 controls across 52 countries; both sexes.

The stress was reported after the event, so people who had just had a heart attack may recall or reframe their stress differently. The size of the study does not remove that bias.

What could explain it instead: Retrospective self-report of stress after a heart attack invites recall bias (patients reinterpret their prior stress) and reverse causation (early cardiac symptoms raising perceived stress); psychosocial stress also clusters with smoking and low income.

Rosengren et al., Association of psychosocial risk factors with risk of acute myocardial infarction in 11119 cases and 13648 controls from 52 countries (the INTERHEART study) · Lancet 2004;364(9438):953-962

Time in natural settings is associated with lower measured cortisol. A 2019 systematic review and meta-analysis of forest bathing found lower salivary cortisol after time in forest environments compared with urban ones, and a separate field study found the greatest reduction in salivary cortisol per minute at about 20 to 30 minutes spent in nature.

Measured in: Small short-term studies pooled in a meta-analysis, plus a field study, mostly younger healthy adults; sexes mixed.

These measure a stress biomarker over minutes to hours, not mood or health over time, and the studies are small and varied. The direction is consistent; the magnitude and durability are not settled.

Antonelli et al., Effects of forest bathing (shinrin-yoku) on levels of cortisol as a stress biomarker: a systematic review and meta-analysis · Int J Biometeorol 2019;63(8):1117-1134 Hunter et al., Urban nature experiences reduce stress in the context of daily life based on salivary biomarkers · Front Psychol 2019;10:722

Endurance training reliably increases mitochondrial content in skeletal muscle. Training volume is the main driver of how much content rises, while higher intensity drives respiratory quality, and the two do not always move together.

Measured in: Human training studies pooled in a systematic review; the underlying trials skew heavily male, as most exercise physiology does.

That training raises mitochondrial content is bedrock. The finer claims, how much volume versus intensity each contributes, rest on a literature that is small per study and male-dominated, and the review itself flags that short-term changes may not predict long-term training response.

Granata et al., training-induced changes in mitochondrial content and respiratory function in human skeletal muscle · Sports Med 2018

In 10 men, six sessions of single-leg interval training versus work-matched continuous training on the other leg raised citrate synthase activity (10.2 versus 8.4 mmol per kg protein per minute) and oxidative phosphorylation capacity more on the interval leg.

Measured in: 10 young active men, single-leg within-subject design.

A small, elegant within-person trial: the design is strong but the sample is ten men over two weeks, so it shows the direction of the intensity effect rather than its long-term size.

MacInnis et al., superior mitochondrial adaptations after interval compared to continuous single-leg cycling matched for total work · J Physiol 2017

Exercise transiently raises reactive oxygen species inside working muscle, and those molecules act as a signal to the nucleus that switches on the genes for mitochondrial biogenesis and the cell's own antioxidant defenses. In this account the stress is the message, not only damage.

That ROS carry an adaptive signal is well supported across cell, animal and short human studies. What is not pinned down is the precise dose in a person: too little stress gives no signal, too much is damage, and where the useful window sits is not settled.

Merry and Ristow, mitohormesis in exercise training · Free Radic Biol Med 2016

Muscle with more and better mitochondria switches more readily between burning fat and burning glucose as supply changes. This capacity, metabolic flexibility, is blunted in obesity and type 2 diabetes and tracks with insulin resistance.

The link between mitochondrial capacity and metabolic flexibility is consistent and mechanistically grounded, but much of the human evidence is cross-sectional, so impaired flexibility is at least as much a marker of metabolic disease as a proven cause of it.

Goodpaster and Sparks, metabolic flexibility in health and disease · Cell Metab 2017

In 39 young men doing a 4-week training program, exercise improved insulin sensitivity and induced PGC-1 alpha and the muscle's own antioxidant enzymes only in the group NOT taking vitamin C (1000 mg) and vitamin E (400 IU) daily. The supplements abolished those gains.

Measured in: 39 healthy young men.

One small mechanistic trial in young men, with a surrogate outcome (insulin sensitivity and gene expression) rather than a long-term health endpoint. It is enough to caution against megadose antioxidants around training, not to make a claim about disease.

Ristow et al., antioxidants prevent health-promoting effects of physical exercise in humans · Proc Natl Acad Sci USA 2009

In about 24 healthy adults aged 55 to 79, six weeks of nicotinamide riboside raised blood NAD+ and was well tolerated, but did not move most metabolic measures. A drop in blood pressure appeared only in a subgroup and needs confirming.

Measured in: About 24 healthy middle-aged and older adults.

One small crossover trial. It establishes safety and that the supplement does raise NAD+, which is the mechanism people buy it for, but not that the mechanism translates into a health outcome.

Martens et al., chronic nicotinamide riboside supplementation is well-tolerated and elevates NAD+ in healthy middle-aged and older adults · Nat Commun 2018

A meta-analysis of 28 trials and 830 people found coenzyme Q10 lowered biomarkers of exercise-induced muscle damage (creatine kinase down about 51 IU/L, lactate dehydrogenase, myoglobin) and oxidative stress (malondialdehyde), but it did not establish a meaningful gain in performance, and most trials were small and run in Asia.

Measured in: 830 participants across 28 trials, most conducted in Asia.

Lower damage markers are not the same as better performance or health, and blunting exercise-induced oxidative stress could even work against adaptation, echoing the antioxidant finding above. The trials are mostly small and geographically narrow, which is why this sits at emerging.

Talebi et al., effects of coenzyme Q10 supplementation on biomarkers of exercise-induced muscle damage, physical performance and oxidative stress: a systematic review and meta-analysis · Clin Nutr ESPEN 2024

The original free-radical, or mitochondrial, theory of aging held that accumulating oxidative damage from mitochondria is the primary driver of aging. Better evidence has revised it: antioxidant supplements do not extend healthy lifespan, and higher ROS can accompany longer life, so ROS are now read as signals as well as damage.

This is a reframing of a theory, not a finished answer. Oxidative damage is real and does accumulate; what changed is the claim that removing it is straightforwardly good. How much mitochondrial aging is cause versus consequence is still open.

Vina et al., the free radical theory of aging revisited: the cell signaling disruption theory of aging · Antioxid Redox Signal 2013

In a first-in-human trial in sedentary elderly adults, four weeks of urolithin A at 500 to 1000 mg a day was safe and shifted plasma acylcarnitines and muscle mitochondrial gene expression toward a molecular signature of improved mitochondrial health. No clinical or performance outcome was measured.

Measured in: Healthy sedentary elderly adults.

A biomarker signature is a promissory note, not a result a person can feel. Whether urolithin A improves function, strength or healthspan in humans is untested, so this sits at preliminary.

Andreux et al., the mitophagy activator urolithin A is safe and induces a molecular signature of improved mitochondrial and cellular health in humans · Nat Metab 2019

Autophagy captures worn-out proteins and damaged organelles inside a double membrane, delivers them to the lysosome for breakdown, and recycles the parts. A low level runs constantly as housekeeping and ramps up sharply under starvation to free amino acids, and the core machinery is conserved from yeast to humans.

That autophagy exists and does this job is settled cell biology. The open question is not whether it happens but how much a given human behavior changes it and whether that change affects health.

Mizushima, Autophagy: process and function · Genes Dev 2007

When energy is plentiful, the growth sensor mTOR phosphorylates the autophagy-initiating enzyme ULK1 and keeps it off. When energy runs low, the fuel-gauge enzyme AMPK is activated and mTOR releases ULK1, and AMPK switches it on directly. Low insulin, low energy availability and mTOR-inhibiting drugs all converge on this switch.

This is molecular cell biology, robust and repeatedly confirmed. It explains how autophagy is regulated; it does not by itself show that any particular fast in a person produces a health benefit.

Kim et al., AMPK and mTOR regulate autophagy through direct phosphorylation of Ulk1 · Nat Cell Biol 2011

The field's consensus guidelines stress that autophagy is a flux, a rate of capture and degradation, not a static amount, so a single marker such as LC3-II can rise either because autophagy sped up or because it stalled at the final step. Reliably measuring it requires flux assays that are largely impractical in living humans.

This is a statement about measurement, not about whether fasting does anything. It is the reason the human column of this topic is mostly empty: not that autophagy fails to respond in people, but that we largely cannot yet watch it do so.

Klionsky et al., Guidelines for the use and interpretation of assays for monitoring autophagy (4th edition) · Autophagy 2021

Rapamycin, which inhibits mTOR and thereby releases autophagy, extended median and maximal lifespan in genetically diverse mice even when started late in life, at 600 days of age. On the basis of age at 90 percent mortality, survival rose by 9 percent in males and 14 percent in females.

Mice, not people, and rapamycin does many things besides inducing autophagy, so the lifespan gain cannot be attributed to autophagy alone. Its side-effect profile makes off-label use for longevity risky and not backed by human evidence.

Harrison et al., Rapamycin fed late in life extends lifespan in genetically heterogeneous mice · Nature 2009

In mice, 24 hours of food withdrawal produced a dramatic rise in autophagy, including in neurons, a tissue that had been thought resistant to starvation-induced autophagy, with a further increase at 48 hours.

A mouse result. Mice have a far faster metabolism than people, so the fasting durations do not translate directly, and the study measured autophagy markers in tissue you cannot sample in a healthy person.

Alirezaei et al., Short-term fasting induces profound neuronal autophagy · Autophagy 2010

In the worm C. elegans, dietary restriction extended lifespan, and knocking down essential autophagy genes abolished that extension. Autophagy was necessary for the longevity effect of eating less in this model, not merely switched on by it.

C. elegans is a worm that lives about three weeks. This establishes a causal mechanism in a model organism; it is a long way from a claim about human lifespan, which has never been tested this way and could not be.

Hansen et al., A role for autophagy in the extension of lifespan by dietary restriction in C. elegans · PLoS Genet 2008

The natural polyamine spermidine induced autophagy and extended lifespan in yeast, flies and worms, and reduced markers of age-related cell death in cultured human immune cells. The lifespan effect was lost when autophagy genes were deleted, pointing to autophagy as the route.

Model organisms plus human cells in a dish. This shows autophagy induction can extend life in simple animals; whether dietary or supplemental spermidine meaningfully extends human life is not established.

Eisenberg et al., Induction of autophagy by spermidine promotes longevity · Nat Cell Biol 2009

Mice engineered to keep normal baseline autophagy but unable to increase it in response to exercise had reduced endurance and lost the protection that exercise normally gives against high-fat-diet glucose intolerance.

An elegant mouse-genetics result showing autophagy is necessary for one exercise benefit in mice. It does not measure exercise-induced autophagy in humans, and it does not mean more autophagy is always better.

He et al., Exercise-induced BCL2-regulated autophagy is required for muscle glucose homeostasis · Nature 2012

In a population cohort, higher dietary spermidine intake was associated with lower blood pressure and a lower incidence of cardiovascular disease. The same study showed spermidine reduced cardiac ageing and extended lifespan in mice.

Measured in: A community-based population cohort including both sexes; the mechanistic arm was in mice.

The human data are observational and the mechanism rests on mice. A correlation between a food compound and better heart outcomes is a reason to study it, not evidence that supplementing it works.

What could explain it instead: Spermidine comes mainly from plant foods, whole grains and legumes, so a higher intake marks an overall plant-rich, higher-quality diet. The association may reflect the whole dietary pattern and the kind of people who eat that way rather than spermidine itself.

Eisenberg et al., Cardioprotection and lifespan extension by the natural polyamine spermidine · Nat Med 2016

In 11 adults with overweight, eating within an early 6-hour window versus a 12-hour window raised expression of the autophagy gene LC3A and lowered 24-hour average glucose by 4 mg/dl, over a 4-day randomized crossover.

Measured in: 11 adults with overweight, 7 men and 4 women.

Eleven people, four days, and a single gene-expression marker rather than a measure of autophagy actually running. It is the kind of early human signal the fasting-autophagy story rests on, and it is a long way from demonstrating a health benefit driven by autophagy.

Jamshed et al., Early Time-Restricted Feeding Improves 24-Hour Glucose Levels and Affects Markers of the Circadian Clock, Aging, and Autophagy in Humans · Nutrients 2019

The vagus is the principal direct neural cable between the gastrointestinal tract and the brainstem, and the majority of its fibres are afferent, carrying sensory information from gut to brain rather than commands the other way, which makes the brain a continuous recipient of reports on the state of the gut.

Direction is left open because this is descriptive anatomy of a communication channel, not a benefit or a harm. That the vagus carries the signal is established; the therapeutic weight of deliberately stimulating it for mood or gut disorders is a separate and less settled question.

Breit et al., Vagus Nerve as Modulator of the Brain-Gut Axis in Psychiatric and Inflammatory Disorders · Front Psychiatry 2018;9:44

The wall of the digestive tract contains the enteric nervous system, a network of hundreds of millions of neurons intrinsic to the gut that senses gut contents and controls motility and secretion largely autonomously, and which is evolutionarily ancient, present even in animals that have no central nervous system.

Direction is descriptive, not a benefit or harm. The often-quoted precise neuron counts vary between sources and species; the durable, well-supported point is that the ENS is a large, semi-autonomous nervous system, not that it holds an exact number of cells.

Furness & Stebbing, The first brain: species comparisons and evolutionary implications for the enteric and central nervous systems · Neurogastroenterol Motil 2018;30(2):e13234

The gastrointestinal tract holds the large majority of the body's serotonin (5-HT), on the order of ninety percent, synthesised by enterochromaffin cells with input from the microbiota, where it regulates motility and secretion and is carried by platelets. This peripheral serotonin does not cross the blood-brain barrier, so it is a separate pool from the serotonin the brain makes for itself.

The popular framing that the gut produces most of your happiness chemical is accurate about where serotonin is made and misleading about what it does, since gut serotonin acts peripherally and does not become the brain's serotonin. The precise percentage is an oft-cited estimate rather than a single measured constant.

O'Mahony et al., Serotonin, tryptophan metabolism and the brain-gut-microbiome axis · Behav Brain Res 2015;277:32-48 Yano et al., Indigenous bacteria from the gut microbiota regulate host serotonin biosynthesis · Cell 2015;161(2):264-76

Under the Rome IV framework, irritable bowel syndrome and related conditions are classified as disorders of gut-brain interaction: the gut is structurally normal but shows disturbed motility, visceral hypersensitivity, altered mucosal and immune function, altered gut microbiota, and altered central nervous system processing of gut signals, occurring in combination.

This is a classification and an explanatory framework rather than a single experimental result, so it is presented without a benefit/harm direction. It describes how the disorder is understood; the size of any particular treatment effect is a separate question, handled on the IBS condition page.

Drossman, Functional Gastrointestinal Disorders: History, Pathophysiology, Clinical Features and Rome IV · Gastroenterology 2016;150(6):1262-1279

Germ-free mice, raised with no microbes, mount an exaggerated hypothalamic-pituitary-adrenal response to stress (higher ACTH and corticosterone) and show reduced brain-derived neurotrophic factor in the cortex and hippocampus compared with normally colonised mice; reconstituting them with bacteria early in life corrects the exaggerated response, and later colonisation does not.

This is powerful causal evidence in mice, not in people. Germ-free animals are an extreme, artificial model, and a critical early-life window in a rodent does not translate directly to an adult human swallowing a supplement. It establishes that the microbiome can shape the developing stress system, which is a different claim from a treatment effect.

Sudo et al., Postnatal microbial colonization programs the hypothalamic-pituitary-adrenal system for stress response in mice · J Physiol 2004;558(Pt 1):263-75 Cryan & Dinan, Mind-altering microorganisms: the impact of the gut microbiota on brain and behaviour · Nat Rev Neurosci 2012;13(10):701-12

Feeding mice Lactobacillus rhamnosus (JB-1) reduced stress-induced corticosterone and anxiety- and depression-related behavior and altered brain GABA receptor expression; none of these effects appeared in vagotomized mice, identifying the vagus nerve as the route by which the gut bacterium reached the brain.

The direction is a benefit in mice only. It demonstrates a mechanism and a route; it is not evidence that this or any strain lifts mood in humans, where the trial evidence is far weaker and mixed.

Bravo et al., Ingestion of Lactobacillus strain regulates emotional behavior and central GABA receptor expression in a mouse via the vagus nerve · Proc Natl Acad Sci USA 2011;108(38):16050-5

In a network meta-analysis of psychological therapies for IBS, several treatments were more effective than control, with cognitive behavioral therapy and gut-directed hypnotherapy holding the largest evidence base and the most durable long-term effect (gut-directed hypnotherapy relative risk of remaining symptomatic 0.67, 95% CI 0.49 to 0.91). The authors note the trials were at high risk of bias with funnel-plot asymmetry, so the true effect is likely smaller than the pooled estimate.

The benefit is real but its size should be read with the authors' own warning: the trials are difficult to blind, risk of bias is high, and the pooled effect probably overstates what an individual can expect. None of the therapies was clearly superior to the others.

Black et al., Efficacy of psychological therapies for irritable bowel syndrome: systematic review and network meta-analysis · Gut 2020;69(8):1441-1451

Short-chain fatty acids (SCFAs), produced when gut bacteria ferment dietary fiber, are proposed mediators of microbiota-gut-brain communication, acting through immune, endocrine, neural and humoral routes, but the pathways by which they influence mood and cognition are not fully worked out and direct human evidence is sparse.

Graded emerging and left without a direction because the mediating role is supported by mechanism and animal work rather than by human outcome data. The molecules are real and the fermentation is real; whether manipulating them changes mood or cognition in people has not been established.

Dalile et al., The role of short-chain fatty acids in microbiota-gut-brain communication · Nat Rev Gastroenterol Hepatol 2019;16(8):461-478

Pooled trials give a weak and inconsistent signal. A meta-analysis of 34 controlled trials found probiotics produced a small effect on depression (d = -0.24) and a smaller effect on anxiety (d = -0.10), concentrated in clinical and psychiatric samples (d = -0.45, rising to -0.73 in psychiatric patients) with little in community samples, while prebiotics showed no effect. A separate meta-analysis of 10 trials in 1349 people found no significant mood benefit overall (SMD -0.128, 95% CI -0.261 to 0.005, p = 0.059), with any effect confined to already-depressed subgroups.

Graded emerging with no single direction because the pooled evidence points different ways and the effects, where present, are small and concentrated in clinical populations. Strains, doses and outcome measures vary widely between trials, so the aggregate hides real heterogeneity. This is the gap between the strong animal mechanism and the modest human reality.

Liu et al., Prebiotics and probiotics for depression and anxiety: a systematic review and meta-analysis of controlled clinical trials · Neurosci Biobehav Rev 2019;102:13-23 Ng et al., A meta-analysis of the use of probiotics to alleviate depressive symptoms · J Affect Disord 2018;228:13-19

Piezo1 and Piezo2 are ion channels that open within milliseconds when a cell membrane is stretched or pushed, letting positive ions in and turning a physical force directly into an electrical and chemical signal.

This is molecular and cellular work, much of it in cell lines and mice; it establishes the machinery exists and how it behaves, not how any one clinical outcome follows from it.

Coste et al., Piezo1 and Piezo2 are essential components of distinct mechanically activated cation channels · Science 2010

Cells are physically tethered to the matrix around them through integrin proteins linked to an internal cytoskeleton, so a force applied outside is transmitted through the cell to the nucleus and translated into biochemical signals that change how the cell behaves.

A review of mechanism across many tissues rather than a single experiment; it frames how force sensing works and why it matters, and leans on disease genetics rather than controlled human trials.

Jaalouk and Lammerding, Mechanotransduction gone awry · Nat Rev Mol Cell Biol 2009

The proteins YAP and TAZ move into the nucleus and switch genes on or off according to how stiff the surrounding matrix is and how much a cell is stretched, linking a physical property of the tissue to which genes a cell reads.

The pathway is well mapped in cultured cells and stem-cell models; how precisely it governs a given tissue's response to exercise or loading in a living person is still being worked out.

Dupont et al., Role of YAP/TAZ in mechanotransduction · Nature 2011

Osteocytes, the cells buried inside mineralized bone, make up 90 to 95% of all bone cells and act as the resident controllers of remodeling, signaling to the bone-building and bone-removing cells on the surface.

A review of cell biology, largely from animal and in-vitro work; it establishes the osteocyte as the coordinating cell, while the exact way it reads a mechanical load in a person remains an active question.

Bonewald, The amazing osteocyte · J Bone Miner Res 2011

The mechanostat model holds that a bone adjusts its own strength to the everyday mechanical loads placed on it: habitual loading adds and maintains bone, and the absence of load lets it thin, so voluntary mechanical usage sets most of a healthy bone's postnatal strength.

A hypothesis and organizing framework rather than a single trial; it is widely used because it fits the loading and disuse data, but as a model it summarises evidence rather than being a test of it.

Frost, Bone's mechanostat: a 2003 update · Anat Rec A Discov Mol Cell Evol Biol 2003

In postmenopausal women with low bone mass, 8 months of twice-weekly heavy resistance and impact training raised lumbar spine bone mineral density by 2.9% versus a 1.2% fall in the control group, and raised femoral neck density, with only one minor adverse event.

Measured in: 101 postmenopausal women aged around 65 with osteopenia or osteoporosis, otherwise healthy and screened.

A single well-run trial in one supervised setting: it shows heavy loading can add bone in a group long told to avoid it, but it is 101 women over 8 months, and the safety record depended on close supervision.

Watson et al., High-Intensity Resistance and Impact Training (LIFTMOR RCT) · J Bone Miner Res 2018

Astronauts on 4 to 6 month space station missions lost bone at roughly 0.9% per month at the spine and 1.4 to 1.5% per month at the hip, with the fastest loss (2.2 to 2.7% per month) in the trabecular bone of the hip, when near-weightlessness removed the usual mechanical load.

Measured in: 14 astronauts (13 men, 1 woman), aged 40 to 55, on long-duration missions.

A small before-and-after series in 14 highly selected, fit adults, and near-weightlessness is an extreme form of unloading; the rate of loss on Earth from bed rest or immobilisation is slower, though the direction is the same.

Lang et al., Cortical and trabecular bone mineral loss from the spine and hip in long-duration spaceflight · J Bone Miner Res 2004

Mechanical loading raises collagen synthesis and turnover in tendon and muscle connective tissue, while inactivity markedly lowers it; sustained training shifts the balance toward net collagen synthesis, which stiffens and strengthens the tissue against future load.

A mechanism and physiology review synthesising human and animal tissue studies; it establishes that loading drives collagen adaptation, which underpins but does not by itself prove any specific rehabilitation protocol.

Kjaer, Role of extracellular matrix in adaptation of tendon and skeletal muscle to mechanical loading · Physiol Rev 2004

Prolonged inactivity and the loss of mechanical stimulus drive a marked fall in skeletal muscle mass and strength (disuse atrophy), seen with limb immobilization, bed rest and spinal injury; resistance training is the most effective way to reverse it.

A mechanism-focused review drawing heavily on rodent models for the molecular detail; that disuse shrinks muscle and loading rebuilds it is well established in humans, while the finer molecular pathways are mostly animal work.

Yeo, Muscle Disuse Atrophy · Adv Exp Med Biol 2025

Mechanical loading sharply lowers sclerostin, a protein made by osteocytes that normally holds bone formation back, with the largest fall in the bone regions that felt the most strain; removing load raised it. Since sclerostin blocks the Wnt pathway that drives bone building, suppressing it under load frees that pathway, which is the molecular route from a load an osteocyte senses to new bone.

The direct loading-and-unloading experiments are in mice, so the strain-to-sclerostin detail is animal work; the pathway is supported in humans indirectly, through the bone-building effect of the anti-sclerostin drug romosozumab, rather than by the same controlled loading study.

Robling et al., Mechanical stimulation of bone in vivo reduces osteocyte expression of Sost/sclerostin · J Biol Chem 2008;283(9):5866-75

Strengthening exercise is the mainstay treatment for patellar tendinopathy, but the pooled randomized evidence is thin and low-certainty: it may make little or no difference to function versus no treatment and performs about the same as glucocorticoid injection or surgery on pain, function and return to sport.

Measured in: 211 people with chronic patellar tendinopathy across 7 trials; 88% male, all athletes.

The certainty of evidence is low to very low, the trials are small and mostly in young male athletes, and none measured harms, so this grades the clinical trial base, not the underlying biology, which is stronger.

Lopes et al., Exercise for patellar tendinopathy (Cochrane Review) · Cochrane Database Syst Rev 2025

Pooling studies that measured BDNF in blood around exercise, a single session produced a moderate rise, and regular training raised the size of that post-session response and, to a smaller degree, resting levels.

Measured in: Adults across exercise-physiology studies pooled in a meta-analysis; sample sizes per study are generally small and the literature skews male, as exercise research does.

That exercise raises BDNF measured in blood is well replicated. What that blood number means for the brain, and whether a bigger rise translates into any outcome a person can feel, is a separate question this measurement cannot answer.

Szuhany, Bugatti and Otto, a meta-analytic review of the effects of exercise on brain-derived neurotrophic factor · J Psychiatr Res 2015

In rats, blocking BDNF action in the hippocampus removed the learning and plasticity gains that exercise otherwise produced, which is the strongest evidence that BDNF is not a bystander but part of the causal path.

This establishes necessity in rodents, not in people. You cannot block BDNF in a human hippocampus to test the same thing, so the causal step that is proven here stays inferred for humans.

Vaynman, Ying and Gomez-Pinilla, hippocampal BDNF mediates the efficacy of exercise on synaptic plasticity and cognition · Eur J Neurosci 2004

In a one-year trial in older adults, moderate aerobic exercise increased the volume of the anterior hippocampus by about 2% and improved spatial memory, and within the exercise group the volume gain tracked with a rise in serum BDNF.

Measured in: 120 older adults, roughly ages 55 to 80, community-dwelling and free of dementia; the sample was majority women.

One well-run trial in healthy older adults. The BDNF-to-volume link is a correlation inside the exercise group, so it supports the story rather than proving BDNF caused the growth, and the result is specific to an ageing population.

Erickson et al., exercise training increases size of hippocampus and improves memory · PNAS 2011