Condition: Anxiety

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, not 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%).

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.

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.

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, not an equally engaging activity.

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.

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.

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.

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, not less. Liver function abnormalities were significantly more frequent on kava, in an aqueous extract.

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.

The value here is access. The same CBT that leads the evidence for anxiety can be delivered over the internet with a clinician supporting it remotely, and across 38 trials and more than 3,000 people it lowered anxiety symptoms with a large effect against inactive controls and was statistically indistinguishable from in-person therapy where the two were compared head to head. The guided form, with a therapist checking in, is the one carrying the evidence; fully unguided apps are less well supported. The Cochrane reviewers rated the body of evidence low to moderate quality, mostly because psychotherapy trials cannot blind their participants.

Measured against non-mindful exercise, not 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.

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.

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. 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, not the treatment of a disorder.. 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.

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.

Compared against anxiolytic drugs, not 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.

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, not 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.

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.

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. 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, not its cause.. The pattern is subtle and the studies are small and varied. This supports checking thyroid function once, not treating a borderline number as the explanation. An overactive thyroid produces anxiety symptoms directly; a marginal TSH in an anxious person usually does not.