Migraine is a neurological disorder. The headache is a symptom of it, not the whole of it, and the distinction is not pedantry — it is the reason migraine was under-treated for so long. Attacks bring nausea, and a sensitivity to light, sound and movement severe enough that the standard response is to lie in a dark room and wait. A third of people with migraine also get aura: transient neurological symptoms, most often visual, preceding or accompanying the headache [1].
It is extraordinarily common and extraordinarily disabling. Global burden analysis ranks migraine and tension-type headache among the leading causes of years lived with disability [2], and the disability is concentrated in young and middle adult life, disproportionately in women [3] [4].
Two things have changed. The first is conceptual: migraine is now understood as a disorder of sensory processing — an abnormally excitable brain that responds to ordinary stimuli with a cascade of pain and sensory amplification — rather than a vascular problem [5] [6]. The second follows from it: the identification of a specific signalling molecule, CGRP, produced the first drugs designed for migraine rather than borrowed from epilepsy, cardiology or psychiatry.
Start here: what happens in an attack
The trigeminovascular system is the pain pathway. Trigeminal sensory fibres innervating the meninges and their blood vessels, projecting to the brainstem and onward to the thalamus and cortex, are what generate migraine pain [7] [8] [9].
CGRP — calcitonin gene-related peptide — is a neuropeptide released from trigeminal nerve endings, and it is not merely a marker of attacks. In a double-blind crossover study, infusing CGRP into people with migraine caused headache in all of them against one on placebo, and in three the delayed headache met formal criteria for migraine without aura [10]. That experiment moved CGRP from correlate to cause, and everything therapeutic follows from it [11] [12].
Aura has a different substrate: cortical spreading depression, a slow wave of neuronal and glial depolarisation followed by suppression, propagating across the cortex at a few millimetres per minute. Functional MRI in human visual cortex captured exactly that during visual aura [13] [14]. And spreading depression is not merely parallel to the headache — it can trigger it, by activating central trigeminovascular neurons [15] and through pannexin-1 channel opening [16] [17].
The disorder is substantially genetic in the polygenic sense: a meta-analysis of 375,000 individuals identified 38 susceptibility loci [18].
Pillar 1: measurement and diagnosis
There is no test
Migraine is diagnosed clinically, against explicit criteria. The International Classification of Headache Disorders specifies attack duration, headache character, associated symptoms and the number of prior attacks required [19] [20]. There is no biomarker, no scan finding, no blood test. A normal MRI does not argue against migraine; it argues against something else.
The practical diagnostic distinctions are three:
With or without aura. Aura matters beyond classification because migraine with aura carries an elevated ischaemic stroke risk, established in systematic review and meta-analysis [21].
Episodic or chronic. Chronic migraine — headache on 15 or more days a month, with migraine features on at least 8 — is a distinct entity with its own criteria [22], its own prevalence [23], a heavier comorbidity burden [24], and different treatments.
Primary or secondary. Most headache is primary. Red flags — sudden maximal onset, fever, neurological deficit, new headache in later life, headache changing character — warrant imaging. The rest do not [25] [26].
The diary is the instrument
Because there is no test, monthly migraine days is the measurement. It is the primary endpoint in essentially every preventive trial [27], and it is what a patient's headache diary produces. Recording days rather than impressions matters: it detects medication overuse, distinguishes episodic from chronic, and provides the only reliable way to tell whether a preventive is working.
Centerpiece: a simple simulatable model of what prevention delivers
The CGRP antibodies were greeted as transformative. They are a genuine advance, and the honest account of what they deliver is more modest and more interesting than the framing suggests.
A trial of 875 patients with episodic migraine reported baseline and 12-week monthly migraine days for every arm. Monthly fremanezumab fell from 8.9 days to 4.9; a single higher dose from 9.2 to 5.3; and placebo from 9.1 to 6.5 [28].
Those numbers permit the reductions to be computed rather than quoted, and then checked against the differences the trial reports separately: (8.9 − 4.9) − (9.1 − 6.5) = 1.4 days against a published 1.5, and (9.2 − 5.3) − (9.1 − 6.5) = 1.3 days against a published 1.3. Both reproduce.
Two facts sit uncomfortably together, and both are true.
The placebo arm improved by 2.6 days a month — 29 percent of its baseline. Some of that is regression to the mean, since patients enrol during bad periods and migraine frequency fluctuates; some is expectation. Whatever the mixture, it means the raw fall on treatment is not the treatment effect, and the honest number is the difference: about 1.4 days a month. A second trial of a different antibody, in 955 patients, found a comparable net effect of 1.9 days (3.7 against 1.8 on placebo) [29]. Two antibodies, two trials, the same order of magnitude.
And treatment does not abolish migraine. The best-performing arm still averaged 4.9 migraine days a month at the end [28]. Prevention shifts the frequency distribution downward; it does not move anyone to zero.
The responder framing is the more useful one, and it is why the right panel exists. A mean fall of one and a half days could describe everyone improving slightly or a quarter of people improving enormously — and it is closer to the latter. In the erenumab trial, 50.0 percent of patients on the higher dose achieved at least a 50 percent reduction in migraine days, against 26.6 percent on placebo [29]. That is a 23.4-point gain, a number needed to treat of about 4.3.
That is the teaching point. Half of treated patients halve their migraine days — an outcome that genuinely changes a life — but a quarter of them would have done so anyway, so roughly one in four owes it to the drug. Both halves of that sentence are needed: the first is why these drugs matter, the second is why a trial of one is a trial and not a guarantee.
Two honest limits. Everything above is episodic migraine over 12 to 24 weeks; chronic migraine trials show larger absolute falls from a higher baseline [30] [31], and long-term durability is a separate question. And treating the placebo-arm fall as regression to the mean plus expectation is an interpretation, not a measurement — the trials cannot separate those components.
Pillar 2: treatment
Migraine treatment has two arms that are chosen independently: stopping the attack you have, and having fewer of them.
Acute treatment
NSAIDs and simple analgesics work for many attacks and should not be dismissed.
Triptans were the first migraine-specific drugs — serotonin 5-HT1B/1D agonists, whose efficacy across the class was established in a meta-analysis of 53 trials [32] [33]. Two practical points shape their use. They work best taken early: once central sensitisation has developed and the scalp is tender, response falls off sharply [34]. And their vasoconstrictor action makes them cautious ground in people with cardiovascular disease — which for decades left those patients with nothing migraine-specific.
Gepants — small-molecule CGRP receptor antagonists — fill exactly that gap. The concept was proven with intravenous BIBN 4096 BS [35], progressed through telcagepant [36], and arrived as orally usable agents such as rimegepant [37]. They do not constrict blood vessels, which is the point.
Ditans are 5-HT1F agonists. The receptor subtype was pursued precisely because agonists at it inhibit neurogenic dural inflammation [38] without contracting cerebral or coronary arteries [39], and an early agent showed efficacy in acute migraine [40]. Lasmiditan followed, from preclinical characterisation [41] to acute-treatment trials [42] — a second route to migraine-specific acute treatment for people who cannot take triptans [43].
Preventive treatment
Prevention is offered when attacks are frequent or disabling enough that treating each one is insufficient.
The older drugs all came from somewhere else: beta-blockers from cardiology, topiramate and valproate from epilepsy, amitriptyline from psychiatry, candesartan from hypertension. They work — the evidence has been assembled in guidelines [44] [45] — and they are limited by side effects incurred for an indication the molecule was not designed for.
The CGRP monoclonal antibodies were designed for migraine. Early proof-of-concept trials of antibodies against CGRP and its receptor [46] [47] led to phase 3 programmes across the class — erenumab against the receptor [29] [48], fremanezumab [28] and galcanezumab [49] against the peptide — with parallel trials in chronic migraine [30] [31]. They are given monthly or quarterly by injection, and their tolerability is their main advantage over the older preventives: in the erenumab trial, adverse event rates were similar to placebo [29].
OnabotulinumtoxinA is specifically a chronic-migraine treatment, established in the PREEMPT programme [50], with no role in episodic migraine. Topiramate also has chronic-migraine trial evidence [51]. In children the picture is humbling: a trial of amitriptyline, topiramate and placebo found no benefit of either drug over placebo [52].
Medication-overuse headache
Frequent use of acute medication can drive episodic migraine into chronic migraine — an association shown longitudinally, with the risk differing by drug class [53]. This is one of the few genuinely iatrogenic mechanisms in headache medicine, and it is why "take a painkiller as needed" is inadequate advice for someone having headaches most days. Recognising it changes the treatment from more analgesia to less.
Comorbidity is part of the disease
Migraine travels with depression and anxiety, and the relationship is bidirectional — each raises the risk of the other, which argues against a simple causal story in either direction [54] [55]. Comorbid conditions also predict progression to more frequent headache [56], and sleep disorders are both trigger and consequence [57]. Treating migraine without addressing these is treating half the problem.
Pillar 3: what is unresolved
Who responds, and why. Roughly half of patients respond to a CGRP antibody and half do not, and nothing predicts which in advance [29] [49]. Given the cost and the injection burden, a response biomarker would be worth a great deal.
Why the placebo response is so large. A 2.6-day fall on placebo is not noise; understanding it would improve both trial design and clinical practice [27].
What aura is for. Cortical spreading depression is well characterised as the aura substrate and can trigger headache [13] [15] [16], but why some people have aura, why it raises stroke risk [21], and whether it can be interrupted therapeutically are open.
Chronification. The transition from episodic to chronic migraine involves medication overuse [53], comorbidity [56] and central sensitisation [34], and preventing it is more valuable than treating it — with no established intervention that specifically does so.
Getting the drugs to people. Migraine remains under-diagnosed and under-treated, and integrating the new agents into practice has required explicit position statements about who should receive them [58] [45] — an access and recognition problem rather than a scientific one.
Dig deeper in lmmol
Migraine and depression are bidirectionally comorbid, and the shared genetic and environmental contributions argue against reading either as simply causing the other [54] [55] — the same interpretive care the depression review applies to its own contested effect sizes. Migraine with aura carries an elevated ischaemic stroke risk [21], connecting to stroke. Epilepsy is the closest mechanistic neighbour: both are episodic disorders of neuronal excitability, cortical spreading depression and seizure share conceptual territory, and several drugs — topiramate, valproate — are used in both, which is exactly why migraine's purpose-built agents were such a departure. Sleep disorders are both trigger and consequence [57] — see obstructive sleep apnea. And for the contrast between a disorder with no biomarker and one defined almost entirely by a number, see thyroid disease. The full collection is at health.