How Long Do People Stay on Anti-CGRP Migraine Drugs? What 53 Real-World Studies Show

Why Persistence Matters

Clinical trials tell you whether a drug works over 12 to 24 weeks in people who were screened, monitored, and reminded to show up. They do not tell you what happens in an ordinary neurology clinic a year later, when the patient has to pay a copay, remember an injection, deal with a side effect, and decide whether the improvement is worth it. Persistence, the share of people still on a drug at a given time, captures all of that at once. It is the closest thing to a single number for "did this work out in practice."

For the anti-CGRP class, which now includes four monoclonal antibodies (erenumab or Aimovig, fremanezumab or Ajovy, galcanezumab or Emgality, eptinezumab or Vyepti) and the daily oral gepant atogepant (Qulipta), dozens of real-world cohorts have been published since 2020, but each covers one country, one clinic, or one drug. A new systematic review and meta-analysis in Pain and Therapy pulled 53 of them together to answer two questions: how many people are still on treatment at 3, 6, and 12 months, and what reasons were recorded when they stopped.

What the Review Pooled

The authors searched Embase and PubMed from inception through November 30, 2025, for real-world studies of adults receiving an anti-CGRP therapy for migraine prevention with longitudinal clinical follow-up. Claims and insurance databases were excluded on purpose, because they can show that a prescription stopped being filled but not why a physician documented the stop. The protocol was registered in PROSPERO.

Persistence was defined as the proportion of patients still on treatment from the day they started, and was pooled with a random-effects meta-analysis of proportions. Reasons for stopping were pooled descriptively. Subgroups looked at migraine type (episodic plus chronic versus chronic only), drug class (antibody versus gepant), and each individual antibody.

Persistence at 3, 6, and 12 Months

Random-effects pooled proportions with 95% confidence intervals. I squared measures how much of the variation between studies is real rather than chance; values above 75% are considered substantial.

The shape of the curve is the useful part. The drop from 3 to 6 months (about 9 points) is larger than the drop from 6 to 12 months (about 8 points over twice the time). In other words, attrition is front-loaded. People who make it past the six-month mark tend to stay. That matches how these drugs are prescribed: most guidelines and payers expect a response by 3 months, and clinicians typically make a continue-or-switch decision somewhere between months 3 and 6.

The heterogeneity is high at every timepoint, climbing from 82% to 93%, and the authors flag the wide prediction intervals that come with it. In practice that means the pooled 76% at one year is a reasonable central estimate across settings, but any individual clinic could sit well above or below it depending on its patients, its reimbursement rules, and how it defines a discontinuation.

Why People Stop

Thirty-nine cohorts recorded a reason for each discontinuation, giving 1,242 events to work with.

Two things stand out. First, side effects account for a small share of stops. For a drug class that patients often worry about because it is new and injectable, roughly one in seven discontinuations being for tolerability is a low figure, and it fits the safety data from both the trials and the long-term extension studies. Second, cost shows up at under 6%, but that number is shaped by where the studies were done. Most came from European countries with national reimbursement for anti-CGRP therapy once criteria are met. A US cohort with prior authorization hurdles and step-therapy requirements would likely report a larger cost share.

The "lack of efficacy" category deserves care. In most of these cohorts it means the physician recorded that the patient did not meet a response threshold, usually a 30% or 50% reduction in monthly migraine days by a set checkpoint, or that the patient reported no meaningful change. It does not mean the drug did nothing for everyone in that group, and it does not say whether the patient went on to try a different anti-CGRP agent and respond to it, which happens.

Drug by Drug

The review compared the individual monoclonal antibodies at each timepoint and found no statistically significant differences (p = 0.71 at 3 months, 0.33 at 6 months, 0.14 at 12 months). The point estimates still spread out, which is expected when some drugs have far more data than others.

Pooled 3-month and 12-month persistence where the review reported it. Eptinezumab and the gepants had few cohorts, so their estimates are less stable.

The one comparison that did reach significance was atogepant against the antibodies as a class at 3 months: 83.5% versus 94.2% (p = 0.013). The authors are careful here, and they should be. Atogepant contributed only four cohorts at 3 months, one at 6, and none at 12. A daily pill is also easier to stop than an injection that was already given, and early-stage gepant cohorts may include more people trying it after an antibody failed. Whether the gap holds with more data, or shrinks, is open. Rimegepant (Nurtec ODT) for prevention had too little real-world persistence data to pool; the authors note a Danish registry showing high early discontinuation but could not analyze it.

Too few gepant cohorts recorded reasons for discontinuation to compare the classes on why people stopped.

Chronic vs Episodic

Cohorts that enrolled only chronic migraine were compared with cohorts that mixed episodic and chronic. Persistence was nearly identical.

This is a somewhat reassuring finding for people with chronic migraine, who often arrive at anti-CGRP therapy after several failed preventives and with medication overuse in the picture. Their likelihood of still being on treatment at one year was the same as the broader population's. Lack of efficacy remained the leading reason for stopping in both groups.

What This Cannot Tell You

A meta-analysis of proportions has real limits, and the authors list most of them.

• Who the patients were. The analysis cannot adjust for how many preventives each patient had already failed, whether they had medication overuse, or how severe their migraine was at baseline. Several cohorts were treatment-resistant populations by design.
• Where they were treated. Nineteen countries contributed, but no between-country comparison was run, and reimbursement rules differ enormously. Persistence in a system where the drug is free once approved will not look like persistence in one where a patient pays several hundred dollars a month.
• Double counting. Italy and Spain contributed 27 of the 53 studies, several from overlapping headache centers. The authors could not rule out that some patients appear in more than one cohort.
• How persistence was reconstructed. Studies reported follow-up in different ways, and the authors had to make assumptions to put them on a common timeline.
• What happened next. A discontinuation is the end of the record. The review does not track switching to a second anti-CGRP agent, re-starting after a pause, or outcomes after stopping.
• Predictors. Factors that predict who will stop (age, prior failures, baseline frequency, early response) were not systematically extracted.

The heterogeneity statistic of 93% at 12 months is the single number to remember when quoting the 76% figure. It is a pooled average over very different settings, not a prediction for any one clinic or any one patient.

Key Takeaways

For someone starting an anti-CGRP therapy, the practical reading is this: the odds of still being on it a year later are good, the first six months are where the decision usually gets made, and if it does not work the most likely reason will be that it did not reduce attacks enough, not that it caused a problem. That is also an argument for giving the drug its full trial period. Guidelines generally recommend at least 3 months for the monthly antibodies and up to 6 months for the quarterly ones before calling a non-response, and the persistence curve here suggests that is where real-world practice already sits.

For clinicians, the review is a useful benchmark. A clinic whose one-year persistence sits far below 76% is worth examining for access barriers, follow-up gaps, or expectations that were set too high at the start. One sitting well above it may have a more selected population. And the 2.7% who stopped because they got better is a reminder that planned discontinuation after sustained remission is part of the picture, and should be recorded as its own category rather than folded into failure.