Condition: Headache & Migraine

41% of the acupuncture group halved their headache frequency against 17% of controls (RR 2.40, 95% CI 2.08 to 2.76; NNT 4). Headache frequency SMD -0.56 (95% CI -0.65 to -0.48). Moderate certainty. Measured in: 4 trials, 2,199 to 2,519 participants with episodic migraine, compared against acute treatment or usual care alone. Neither participants nor practitioners were blinded in this comparison, so it measures the whole package of attention, ritual, expectation and needling against nothing added. It says what a course of acupuncture delivers, not what the needle placement contributes.

About 1.2 fewer attacks per 28 days than placebo (MD -1.20, 95% CI -1.59 to -0.80), and roughly double the proportion halving their attack frequency (RR 2.02, 95% CI 1.57 to 2.60; NNT 4). 200 mg was no more effective than 100 mg. No significant difference against amitriptyline, flunarizine or propranolol. Measured in: 17 unique trials; 1,737 participants for attack frequency and 1,190 for the responder rate. Adverse events at 100 and 200 mg were significantly more frequent than placebo, with numbers needed to harm from 2 to 25, driven by paresthesia, taste change, weight loss and cognitive slowing. Topiramate is teratogenic and reduces the effectiveness of some hormonal contraception, which constrains its use in exactly the group most affected by migraine.

Headache days halved in 52% on amitriptyline, 55% on topiramate and 61% on placebo. The trial was stopped early for futility. Fatigue 30% and dry mouth 25% on amitriptyline; paresthesia 31% and weight loss 8% on topiramate. Three mood disturbances on amitriptyline and one suicide attempt on topiramate. Measured in: 328 children and adolescents aged 8 to 17 with migraine, analyzed of 361 randomized. A 61% placebo response is high enough that a small true drug effect could hide inside it, and stopping early for futility widens the confidence intervals. This does not rule out benefit in a subgroup, and it does say that neither drug should be started in a child on the strength of the adult data.

50% of the true acupuncture group against 41% of the sham group halved their headache frequency immediately after treatment (RR 1.23, 95% CI 1.11 to 1.36; NNT 11), and 53% against 42% at follow-up (NNT 10). Headache frequency SMD -0.18 (95% CI -0.28 to -0.08). Moderate certainty. Measured in: 12 to 14 trials, 1,646 to 1,825 participants with episodic migraine. Sham acupuncture in these trials involves a practitioner, a private room, half an hour of attention and shallow needling or skin contact at non-points, so it is an active comparator. The 9 percentage point margin is what remains after all of that is subtracted, and adverse effects did not differ between the arms.

57% of the acupuncture group against 46% of the drug group halved their headache frequency after treatment (SMD -0.25, 95% CI -0.39 to -0.10). By six months the gap had narrowed to 59% against 54% and was no longer significant (SMD -0.13, 95% CI -0.28 to 0.01). Adverse effects were far fewer with acupuncture (OR 0.25, 95% CI 0.10 to 0.62). Moderate certainty. Measured in: 3 to 5 trials, 451 to 931 participants; comparators were the standard oral prophylactics in use when the trials ran. A daily tablet with side effects cannot be blinded against a course of needling, and participants knew which they were getting. The efficacy advantage had largely disappeared by six months, so the durable finding is the adverse-effect difference, not the efficacy difference. NICE (CG150, amended 2025) recommends acupuncture for migraine prophylaxis only as third-line, after propranolol, topiramate and amitriptyline have failed, which is a narrower endorsement than the guideline gave before. It still recommends it for chronic tension-type headache, and still recommends against it for low back pain, so the split across conditions is real. The comparison against drugs rests on 3 trials and 739 participants; the wider participant range quoted elsewhere belongs to the adverse-event analysis, not the efficacy one. Acupuncture had lower odds of adverse events, not literally a quarter of the count.

Against routine care, 48% responded to acupuncture against 19% in a trial of 1,265 people (RR 2.5, 95% CI 2.1 to 3.0) and 45% against 4% in a trial of 207 (RR 11). Against sham, 51% (205 of 391) against 43% (133 of 312) halved their headache frequency (RR 1.3, 95% CI 1.09 to 1.5). Against physiotherapy, massage or exercise, acupuncture showed no advantage and some outcomes slightly favored the comparator. Measured in: 12 trials, 2,349 participants with frequent episodic or chronic tension-type headache. The routine-care comparisons come from two unblinded trials, and the 8 percentage point margin over sham rests on five trials providing data. Adverse effects were reported by 17% on acupuncture and 12% on sham, a difference that did not reach significance.

Migraine days fell 3.9 a month in the manual acupuncture group against 2.2 with sham needling at non-acupoints at weeks 17 to 20, and attacks fell 2.3 against 1.6. No severe adverse events. Measured in: 147 adults with episodic migraine without aura at seven Chinese centers, mean age 36.5, 82% women, 20 sessions over 8 weeks. This is one of the better-designed trials in the area: the sham was non-penetrating, not shallow needling, participants were acupuncture-naive by design, and blinding was measured and held, with 79% of the real group and 75% of the sham group believing they had been penetrated (P=0.891).

Monthly headache days fell 12.3 with withdrawal plus a preventive started at the same time, 9.9 with a preventive and no withdrawal, and 8.5 with withdrawal alone. Reversion from chronic to episodic headache was 74.2%, 60.0% and 41.7% respectively at six months. Measured in: 120 patients with medication overuse headache at a Danish tertiary headache center, mean age 43.9, 79.4% women; 102 completed follow-up. Open-label at a single specialist center over six months, and every arm received the center's structured patient education. The numbers describe what happens with expert support, not what happens attempting this alone. The requirement that the headache improve after withdrawal was part of the older ICHD-2 definition, not ICHD-3, which no longer conditions the diagnosis on getting better.

Monthly migraine days fell 3.7 on erenumab 140 mg and 3.2 on 70 mg against 1.8 on placebo, from a baseline of 8.3. Migraine days halved in 50.0% and 43.3% against 26.6% on placebo. Adverse event rates matched placebo. Measured in: 955 adults with episodic migraine randomized to 70 mg, 140 mg or placebo for six months. Six months of follow-up in episodic migraine only, so it says nothing about chronic migraine or about years of continuous use. Constipation, occasionally severe, emerged as a class problem after launch, not in the trial. The gain over placebo is under two migraine days a month.

Monthly migraine days fell 4.2 on atogepant 60 mg, 3.9 on 30 mg and 3.7 on 10 mg against 2.5 on placebo, from a baseline of 7.5 to 7.9. The gain over placebo was 1.2 to 1.7 days. Constipation 6.9 to 7.7%, nausea 4.4 to 6.1%. Measured in: 873 adults with episodic migraine analyzed over 12 weeks. Twelve weeks, episodic migraine only. The advantage over placebo sits in the same one to two day range the older oral preventives occupy; what has changed is tolerability, not effect size.

Migraine days a month: candesartan 16 mg 2.95 (95% CI 2.35 to 3.55), propranolol slow release 160 mg 2.91 (2.36 to 3.45), placebo 3.53 (2.98 to 4.08). Attacks halved in 43% on candesartan and 40% on propranolol against 23% on placebo. Adverse events 133, 143 and 90 respectively. Measured in: 72 adults with episodic or chronic migraine, triple-blind double crossover, three 12-week treatment periods. 72 people at one Norwegian center. Candesartan is off-label for migraine in most countries and is contraindicated in pregnancy. Propranolol's own Cochrane review was withdrawn in 2017, not updated, so this crossover carries more evidential weight for it than a first-line drug should need. The trial was supported by the maker of candesartan, which supplied the drug, contributed funding, and reviewed and approved the manuscript, on a comparison of its own drug against the standard first-line preventive.

Against placebo, standardized mean difference -1.29 for tension-type headache and -0.70 for migraine. Relative risk of halving headache intensity 1.80 for migraine and 1.41 for tension-type headache. Effectiveness increased with longer duration of treatment. No significant difference against SSRIs. Measured in: Randomized trials of tricyclic antidepressants in adults with migraine, tension-type headache or both. Adverse effects were more likely than placebo (RR 1.53), predominantly dry mouth, drowsiness and weight gain, though dropouts were not raised. Many pooled trials are old and small, and an effect size of -1.29 for tension-type headache is large enough that publication bias should be assumed to contribute. Against SSRIs the two did not differ on headache frequency, but tricyclics were significantly better at achieving a 50% reduction in intensity, so "no difference" understates them.

Aerobic exercise reduced migraine days by 0.6 plus or minus 0.3 a month. Unpooled, attack duration fell 20 to 27% and pain intensity 20 to 54%. Moderate quality evidence for the migraine-days finding. Measured in: 6 studies of aerobic exercise programs in adults with migraine. Six studies, with outcome measures too heterogeneous to pool for duration or intensity, so the reviewers drew no conclusion on either. Exercise cannot be blinded, and 0.6 days a month is roughly a third of what a preventive drug delivers.

Petasites hybridus root extract 75 mg twice daily cut attack frequency 48% against 26% on placebo over four months (p = 0.0012), with 68% halving their attacks against 49%. The 50 mg dose reduced attacks 36% and did not separate from placebo. The commonest adverse effects were mild gastrointestinal, predominantly burping. Measured in: 245 adults aged 18 to 65 with migraine, three parallel arms over four months. The trial numbers are per-protocol, not intention-to-treat. Germany and Switzerland acted against butterbur products (Swissmedic revoked its authorization in 2004); in the UK the products were unlicensed to begin with, so there was no authorization to lose, and the regulator requested a voluntary withdrawal in 2012. The hepatotoxicity concern is why the product-quality question, whether the liver-toxic alkaloids are actually removed, is the whole decision.

Attacks were significantly more frequent during the late luteal and early follicular phases, when estrogen is falling, and significantly less frequent during phases of rising estrogen. Measured in: 38 women with regular menstrual cycles and one to four migraines a month, tracked with fertility monitors and daily urinary hormone metabolites. What could explain it instead: Cycle phase carries everything else that varies across the month with it: sleep, mood, analgesic use, and the prostaglandin release around menstruation. Timing alone cannot isolate estrogen withdrawal as the operative cause.. 38 women with regular cycles and relatively infrequent migraine, which excludes exactly the people whose cycles are irregular and whose attacks are frequent. It supports the estrogen withdrawal hypothesis, not establishing it.

400 mg daily for three months: 59% halved their attacks against 15% on placebo (NNT 2.3), with improvement in attack frequency (p = 0.005) and headache days (p = 0.012). Adverse events were diarrhea and polyuria. Measured in: 55 adults with migraine at a single Belgian center, three months. One small single-center trial from 1998, with a responder gap unusually large for that sample size, and it has not been replicated at scale in adults. The separation appeared only after the first month, so a four-week trial of it proves nothing.

600 mg daily of trimagnesium dicitrate cut attack frequency 41.6% against 15.8% on placebo in weeks 9 to 12, with diarrhea in 18.6% and gastric irritation in 4.7%. A second trial the same year using magnesium aspartate found 28.6% responders against 29.4% on placebo and was stopped at interim analysis for futility. Measured in: Two randomized placebo-controlled trials: 81 adults aged 18 to 65 averaging 3.6 attacks a month, and 69 adults (64 women, 5 men) in a trial planned for 150. The two trials used different magnesium salts with different absorption, which is the most likely reason they disagree, and neither has been repeated. 600 mg a day of supplemental magnesium exceeds the tolerable upper intake level, and roughly one person in five got diarrhea.

100 mg three times daily: 47.6% halved their attack frequency against 14.4% on placebo at three months (NNT 3), with fewer headache days and fewer days with nausea. Well tolerated. Measured in: 42 adults with migraine, double-blind placebo-controlled, three months. 42 people at one center, and the separation appeared only in the third treatment month. A responder gap this wide in a trial this small is the pattern that most often fails to replicate.

Results across trials are inconsistent. The one larger rigorous trial found a difference of 0.6 attacks a month between feverfew and placebo. The reviewers rated the overall evidence low quality and found no major safety concerns, with mild and reversible adverse effects. Measured in: Randomized placebo-controlled trials of feverfew for migraine prevention in adults. Preparations differ widely in parthenolide content, which is a plausible reason the trials disagree, and no trial has tested a standardized stable extract at scale. Professional bodies split on it, with one recommending against offering it and others rating it probably effective.

Oral sumatriptan 100 mg against placebo: number needed to treat 4.7 for pain freedom at two hours, 3.5 for headache relief at two hours, and 6.5 for sustained pain freedom through 24 hours without rescue medication (24% against 8%). Adverse events were transient and mild with a clear dose response from 25 mg to 100 mg. Measured in: 61 trials, 37,250 participants treating acute migraine attacks. Most trials treated a single attack, so nothing here describes repeated use over months, which is where the medication overuse threshold bites. Trial populations exclude the cardiovascular disease that contraindicates the drug, so the safety profile does not describe the people who need to be excluded.

Ibuprofen 400 mg left 26% pain-free at two hours against 12% on placebo (NNT 7.2), and gave headache relief at two hours in 57% against 25% (NNT 3.2), with 24-hour sustained relief in 45% against 19% (NNT 4.0). 200 mg was weaker: 20% against 10% pain-free (NNT 9.7). Adverse events matched placebo. Measured in: 9 trials, 4,373 participants, 5,223 migraine attacks. Trials treated one or a few attacks, so the gastrointestinal and renal risk of repeated use over months does not appear in these numbers. No included trial combined ibuprofen with an antiemetic, despite that being routine practice.

Paracetamol 1000 mg gave pain freedom at two hours in 19% against 10% on placebo (NNT 12) and headache relief at two hours in 56% against 36% (NNT 5.0). Combined with metoclopramide 10 mg it was not significantly different from oral sumatriptan 100 mg for two-hour headache relief. Adverse event rates matched placebo. Measured in: 11 studies, 2,942 participants, 5,109 migraine attacks. A number needed to treat of 12 for pain freedom is poor beside the alternatives on this page, and trials recruited people willing to treat migraine with paracetamol, who plausibly have milder attacks. The combination with metoclopramide is where the useful result sits.

Pain freedom at two hours in 32.2% on lasmiditan 200 mg (OR 2.6, 95% CI 2.0 to 3.6) and 28.2% on 100 mg (OR 2.2, 1.6 to 3.0) against 15.3% on placebo. Relief of the most bothersome symptom in 40.7% and 40.9% against 29.5%. Measured in: 1,856 adults who treated a migraine attack; 77.9% had one or more cardiovascular risk factors in addition to migraine. Dizziness and sedation are frequent enough that driving is restricted for at least eight hours after a dose. There is no head-to-head trial against a triptan, so its place is set by who cannot take one, not by comparative efficacy.

Probable medication-overuse headache had a population prevalence of about 5.6% in women and 2.3% in men. Measured in: 41,614 adults aged 18 to 65 across 17 countries, 53.5% female, assessed by structured diagnostic interview. What could explain it instead: Probable medication overuse headache is assigned from self-reported medication days, and the people with the most headache days both take the most medication and recall it differently. A cross-sectional survey also cannot separate medication causing the headache from headache driving the medication.. Population survey figures for PROBABLE medication-overuse headache, since a survey cannot establish the pre-existing headache pattern or the three-month medication duration that a full diagnosis needs. That is also why the diagnosis is "probable", not confirmed: it is a limitation of the survey, not of the person.

Chocolate was no more likely to provoke a headache than carob in any headache diagnostic group. Prior belief that chocolate was a trigger did not change the result. Measured in: 63 women with chronic headache (50% migraine, 37.5% tension-type, 12.5% both), given two chocolate and two carob samples in random order under double-blind conditions after two weeks on a diet low in vasoactive amines. 63 people, all women, all pre-treated with an elimination diet that may itself change reactivity, and carob differs enough in taste and texture that blinding is imperfect. It cannot exclude an effect in the minority who report a very consistent chocolate reaction.

81% of people with migraine or tension-type headache endorsed at least one trigger (95% CI 0.75 to 0.86). Stress was endorsed by 58% and sleep by 41%. Endorsement rates rose with the number of trigger categories the questionnaire asked about, and with year of publication. Measured in: 85 articles published 1958 to 2015, 27,122 participants, 420 unique triggers grouped into 15 categories. This measures what people believe triggers their headaches, not what provokes one under test. The rise in endorsement with the number of categories asked about is a property of the questionnaire, not of the person answering it, and the literature is described by its own reviewers as extremely heterogeneous.

Premonitory symptoms appear hours before the headache and include tiredness, difficulty concentrating and a stiff neck. Food cravings are widely described in this phase but the study cited here does not list them among the symptoms it measured. Measured in: 97 people with usable data of 120 recruited, using a prospective electronic diary. What could explain it instead: Expectancy. Participants knew the study was about symptoms preceding attacks, so both attention to early symptoms and the labeling of a subsequent headache as migraine are inflated by taking part.. The point that matters for triggers still holds: because a warning phase precedes the pain by hours, a food eaten in that window can be an early symptom of the attack, not its cause. The specific attribution of craving to this citation is what does not hold.

Probable medication-overuse headache ran about 5.6% in women against 2.3% in men, part of the broader pattern of migraine being roughly two to three times more common in women. Measured in: 41,614 adults aged 18 to 65 across 17 countries, 53.5% female; and 162,576 US respondents aged 12 and over. What could explain it instead: Case ascertainment method. Structured interviews find far more migraine than mailed questionnaires do, so the two surveys differ in the size of the sex gap partly because they differ in how they ask. Men are also less likely to report headache to a clinician or to describe it as migraine.. The two surveys disagree on the size of the sex difference by roughly a factor of two, so the direction is settled and the ratio is not. The 2025 pooled sample over-represents lower-middle income countries, and the US survey is now nearly two decades old.

Adults with migraine scored worse on the Pittsburgh Sleep Quality Index than controls (g = 0.75, 95% CI 0.54 to 0.96), with a larger gap in chronic migraine (g = 1.03) than episodic (g = 0.63). On polysomnography, both adults (g = -0.22) and children (g = -0.71) had a lower percentage of REM sleep. Children also had less total sleep time (g = -1.37) and more wakefulness (g = 0.52). Measured in: 32 studies; 21 measuring subjective sleep quality in adults, 6 polysomnography in adults, 5 polysomnography in children. What could explain it instead: Reverse causation and shared drivers. Anxiety, depression and preventive medication all affect sleep architecture and headache frequency together, and the pooled studies did not consistently adjust for them.. Every comparison is cross-sectional, so it cannot say whether poor sleep drives migraine, migraine wrecks sleep, or both. The adult polysomnography effect is small and rests on six studies.

Migraine of any kind carried a pooled relative risk of ischemic stroke of about 1.73, rising to 2.16 for migraine with aura. The association in men (1.37, 95% CI 0.89 to 2.11) did not reach significance. Measured in: Pooled observational studies of migraine and cardiovascular disease; nine studies contributed the ischemic stroke estimate. What could explain it instead: Smoking and oral contraceptive use both travel with migraine and with stroke, and aura status in most contributing studies came from questionnaires, not neurological assessment, so misclassification could push the aura and no-aura estimates in either direction.. Pooled observational data, not trials. The absolute risk of ischemic stroke in a young woman is low, so a doubling of it remains a small number, and the subgroup findings by aura, sex, age, smoking and contraceptive use are secondary analyzes, not the primary question.

One case-control analysis found migraine WITHOUT aura raised stroke odds in combined-contraceptive users while migraine WITH aura did not, contradicting the older work the prescribing rules rest on. Any migraine raised the odds (OR 2.00), and a pill containing 30 µg or more of estrogen carried higher odds (OR 1.52) than lower-dose pills. Measured in: 127 stroke cases matched to 635 controls, drawn from 203,853 combined hormonal contraception users at one US tertiary center between 2010 and 2019. What could explain it instead: Channeling by indication. Clinicians already avoid combined hormonal contraception in women with known aura, so the women with aura who were nonetheless prescribed it are a selected, lower-risk group, which drags the aura estimate down. Aura status was taken from chart review, not prospective classification.. 127 events at a single center. The aura finding runs against the larger pooled literature and against the eligibility criteria that guide prescribing worldwide, and it is a reason to keep watching the question, not to change what is prescribed. This is the less reliable of the two conflicting sources: the aura result comes from a group of aura sufferers who were prescribed the pill anyway, which selects for low baseline risk and makes a null nearly uninterpretable. The safety advice on this page follows the more reliable meta-analysis and is not softened. The estrogen-dose finding is the actionable part.