The First Migraine Prevention Guideline of the CGRP Era: What the 2026 AAN/AHS Update Changes at the Point of Care

Bottom Line: The August 31, 2026 AAN/AHS practice guideline — the first update since 2012 — places CGRP monoclonal antibodies, gepants, onabotulinumtoxinA, and legacy oral preventives on a single GRADE-based evidence hierarchy and lowers the operational bar for offering prevention to ≥4 migraine days per month or any functionally impairing migraine. The guideline itself is deliberately agnostic on comparative superiority — its systematic review, searched through June 6, 2024, found active-comparator evidence "generally of low or very low confidence." But head-to-head and special-population trials published after that search window (TEMPLE, RESOLUTION, UNITE) and the earlier HER-MES and APPRAISE data consistently favor CGRP-targeted therapy on tolerability and persistence, creating a growing mismatch between the guideline's even-handed framework and fail-first step-therapy architecture at the point of care.

Fourteen Years Between Guidelines: Why 2026 Is Different

On August 31, 2026, the American Academy of Neurology and the American Headache Society released their first joint migraine-prevention guideline since 2012, published simultaneously in Neurology and Headache and endorsed by the American Academy of Family Physicians. The document arrives into a therapeutic landscape transformed: in 2012 there was no migraine-specific preventive medication; in 2026 there are six US preventive agents targeting the CGRP pathway — four monoclonal antibodies (erenumab, fremanezumab, galcanezumab, eptinezumab) plus the gepants atogepant and rimegepant.

The guideline's threshold for offering prevention is operationally lower than many clinicians use: ≥4 migraine days per month, OR ≥4 moderate-to-severe headache days per month, OR migraine that impairs work or daily functioning. Chronic migraine is defined as ≥15 headache days per month with ≥8 having migraine features; episodic migraine falls below that line. Guideline author Tamara Pringsheim — also first author of the companion systematic review — framed the intent plainly: "For people experiencing frequent migraine attacks or attacks that affect the ability to function normally, this guideline can help clinicians determine which preventive medications may be able to help." Guideline author Rebecca Burch added: "The research shows many different types of medications may be effective for preventing migraine attacks and reducing symptoms."

That last quote deserves emphasis. This is not a CGRP-first manifesto. It is a selection-oriented document built on a systematic review of 217 randomized trials, graded with a modified GRADE process — and its most consequential feature is that old and new drugs now sit on one evidence scale.

What the Hierarchy Actually Says

The companion systematic review (searches through June 6, 2024) assigns confidence-in-evidence levels per drug. For episodic migraine: high-confidence evidence that galcanezumab and erenumab reduce headache frequency versus placebo; moderate-confidence evidence for atogepant, eptinezumab, fremanezumab, propranolol, topiramate, and valproate; and low-confidence evidence suggesting possible benefit for amitriptyline, bisoprolol, flunarizine, fluoxetine, levetiracetam, metoprolol, nifedipine, pizotifen, telmisartan — and rimegepant. For chronic migraine: high-confidence evidence for fremanezumab, galcanezumab, and onabotulinumtoxinA; moderate-confidence evidence for atogepant, eptinezumab, erenumab, topiramate, and valproate.

Three features of this hierarchy matter at the point of care. First, amitriptyline — a de facto first-line agent for fifty years — lands in the low-confidence tier for episodic migraine, while the four CGRP monoclonals and atogepant sit at moderate or high. Second, the hierarchy discriminates within the CGRP class: rimegepant, despite its migraine-specific mechanism, is grouped among the low-confidence episodic options — evidence that the panel graded drugs, not drug classes, and that "CGRP-pathway" is not itself an evidence tier. Third, on comparative effectiveness the review is explicit: "Evidence comparing active treatments was limited and generally of low or very low confidence, restricting conclusions about comparative effectiveness." The guideline does not declare CGRP agents superior to anything. What follows below — the head-to-head story — must therefore be read as evidence that intensifies the clinical and payer tension around the guideline, not as the guideline's own conclusion.

What the Legacy Orals Actually Rest On

Topiramate's foundation is genuinely randomized — a pair of pivotal phase 3 trials. MIGR-001 (Silberstein et al., Arch Neurol 2004) randomized 487 patients across placebo and three doses: monthly migraine frequency fell by 2.1 (100 mg/d) and 2.4 (200 mg/d), both significant versus placebo (P<0.001), while 50 mg/d (−1.3) was not; its companion MIGR-002 (Brandes et al., JAMA 2004; 483 randomized) was concordant at 100 and 200 mg/d, and together they secured FDA approval, with 100 mg/d emerging as the standard efficacy–tolerability target. But note the dose-dependent adverse-event discontinuations even in the pivotal program: 8% on placebo rising to 14%, 21%, and 29% across the ascending doses.

Amitriptyline's most-cited foundation is thinner by modern standards. Couch and colleagues (Neurology 1976) reported an open-label, single-center series of 110 analyzed patients, mostly on 50–75 mg/day, with ≥50% migraine-score reduction in 72% and ≥80% reduction in 57% — impressive figures, but with no placebo arm. Controlled amitriptyline data do exist (including a subsequent placebo-controlled trial by the same group and a later 2011 report), so the drug does not rest on a single uncontrolled study; still, this evidence base predates modern endpoints such as monthly-migraine-day change and ≥50% responder rates, which is precisely why cross-era comparisons are limited and why the guideline's per-drug confidence ratings — amitriptyline: low confidence — are the fairest available summary. Propranolol and valproate, by contrast, earned moderate-confidence ratings, a reminder that "legacy orals" are not a monolith.

The CGRP Monoclonals: The Placebo-Controlled Core

The pivotal-trial core the guideline weighed is consistent across the class: roughly 1.5–2.5 placebo-adjusted monthly migraine days of benefit, onset within the first month, and near-placebo tolerability. Erenumab in STRIVE (NEJM 2017; 955 patients) reduced monthly migraine days by 3.2 (70 mg) and 3.7 (140 mg) versus 1.8 with placebo, with ≥50% responder rates of 43.3% and 50.0% versus 26.6%. Galcanezumab in EVOLVE-1 (JAMA Neurol 2018; 858 patients) achieved −4.7 and −4.6 days versus −2.8 (differences −1.9 and −1.8, both P<0.001), with responder rates of 62.3% and 60.9% versus 38.6%; EVOLVE-2 was concordant (~−4.3 vs −2.3 days; 59%/57% vs 36%), and REGAIN extended the result to chronic migraine (1,117 randomized, baseline 19.4 migraine headache days: −4.8 and −4.6 vs −2.7, P<0.001). Fremanezumab was positive in both HALO CM (NEJM 2017; 1,130 patients; headache days −4.6 monthly and −4.3 quarterly vs −2.5, both P<0.001) and HALO EM (JAMA 2018; 875 patients; placebo-adjusted −1.5 and −1.3 days, both P<0.001). Eptinezumab added a rapid-onset IV option: in PROMISE-2 (chronic migraine), monthly migraine days fell −7.7 (100 mg) and −8.2 (300 mg) versus −5.6 (P<0.0001), with ≥50% response in 57.6% and 61.4% versus 39.3% — and migraine prevalence on day 1 after infusion was 28.6% and 27.8% versus 42.3%. In PROMISE-1 (episodic), 100 and 300 mg were significant while 30 mg failed the prespecified testing hierarchy.

Gepants and OnabotulinumtoxinA: The Rest of the Modern Hierarchy

Atogepant dissolved the old oral-versus-injectable tradeoff. In ADVANCE (NEJM 2021; 910 randomized, 873 in the modified ITT analysis), monthly migraine days fell −3.7, −3.9, and −4.2 (10/30/60 mg) versus −2.5 with placebo — placebo-adjusted −1.2, −1.4, and −1.7 days (all P<0.001) — with ≥50% responder rates of 55.6%, 58.7%, and 60.8% versus 29.0%. In PROGRESS (Neurology 2024, chronic migraine), atogepant worked with or without acute-medication overuse: in the overuse subgroup, migraine days fell versus placebo by −2.7 (30 mg BID; 95% CI −4.0 to −1.4) and −1.9 (60 mg QD; 95% CI −3.2 to −0.6), with ≥50% response in 44.7% (OR 2.5) and 41.8% (OR 2.3) versus 24.9%, and 52.1%–61.9% of atogepant-treated participants no longer met overuse criteria over 12 weeks (Class II evidence). Rimegepant's every-other-day preventive indication rests on a smaller evidence base, which is why it sits in the low-confidence tier — within the gepant class, atogepant carries the preventive evidence.

OnabotulinumtoxinA deserves more prominence than it usually gets in CGRP-era discussions: it is one of only three agents with high-confidence chronic-migraine evidence in the new hierarchy. The pooled PREEMPT program (Headache 2010; 1,384 patients) showed headache-day reduction of −8.4 versus −6.6 for placebo at week 24 (difference −1.8 days, 95% CI −2.52 to −1.13, P<0.001), ≥50% response in 47.1% versus 35.1% (P<0.001), and adverse-event discontinuation of only 3.8% versus 1.2% — a tolerability profile that keeps it firmly in the chronic-migraine algorithm, including in patients with medication overuse (about two-thirds of the PREEMPT population).

The Head-to-Head Turn: Read the Dates Carefully

Three randomized comparisons now pit CGRP-targeted therapy against oral standard of care — but only some of this evidence was available to the guideline panel, and none of it changed the panel's comparative-effectiveness verdict.

HER-MES (Cephalalgia 2022), a double-dummy trial of erenumab versus topiramate, made tolerability the primary endpoint: adverse-event discontinuation 10.6% versus 38.9% (RR 0.27, 95% CI 0.20–0.37), with ≥50% response in 55.4% versus 31.2% (OR 2.76) and migraine-day change −5.86 versus −4.02 (difference −1.84). By week 6 — the end of up-titration — 26.6% had already stopped topiramate versus 8.3% erenumab. APPRAISE (JAMA Neurol 2024), an open-label pragmatic 12-month trial of erenumab versus nonspecific oral preventives in episodic migraine after one or two oral failures, found its composite endpoint (≥50% reduction plus persistence) in 56.2% versus 16.8% (OR 6.48, 95% CI 4.28–9.82); completion on the assigned drug was 86.9% versus 37.5% (OR 11.27, 95% CI 7.53–16.87), switching 2.2% versus 34.6%, and patient-reported global improvement 76.0% versus 18.8% (OR 13.75, 95% CI 9.08–20.83). TEMPLE (Lancet Neurol 2026; 545 adults) — published after the June 6, 2024 search cutoff and therefore outside the guideline evidence base — replicated the pattern for atogepant 60 mg versus topiramate: treatment-related discontinuation 12% versus 30% (RR 0.4, 95% CI 0.3–0.6, p<0.0001), ≥50% response at months 4–6 in 64% versus 39% (RR 1.6, 1.4–2.0), migraine-day change −6.3 versus −4.5 (Δ −1.8, −2.5 to −1.0), and PGIC much/very-much improved 69% versus 37%. Cognitive tolerability also separated: PROMIS cognitive-function scores favored atogepant by +5.0 points (95% CI 3.3–6.7), and paresthesia occurred in 5% versus 41%.

The synthesis is consistent — and consistently misread. The superiority signal is largest not on migraine-day counts (Δ ~1.8 days) but on persistence and tolerability: patients simply stay on CGRP-targeted therapy and abandon topiramate. But each trial tests one drug in one population — erenumab in HER-MES and APPRAISE (the latter episodic-only and open-label), atogepant in TEMPLE — and extending their conclusions to the whole CGRP class, or to chronic migraine, is extrapolation. Because the within-window designs carried limitations (open-label pragmatic design in APPRAISE; tolerability-weighted endpoints), the systematic review graded active-comparator evidence low to very low confidence, and TEMPLE arrived too late to be weighed at all. The honest formulation: the guideline built the unified scale; the head-to-head trials are post-hoc and post-window pressure on it.

Populations Where the Update Bites

Medication overuse. The guideline recommends offering prevention to patients with medication overuse or medication-overuse headache, preferentially using agents with evidence in that population — CGRP monoclonals, atogepant, onabotulinumtoxinA, and topiramate. That moderate, treat-forward stance is supported by within-window data: the erenumab chronic-migraine overuse subgroup (Neurology 2019: migraine days −6.6 vs −3.5; Δ −3.1, 95% CI −4.8 to −1.4), galcanezumab's efficacy regardless of overuse status (REGAIN/EVOLVE subgroup analyses), and MOTS (Neurology 2022), where prevention without forced symptomatic-medication switching was noninferior. Post-window evidence pushes further: in RESOLUTION (Neurology 2026), a single eptinezumab 100 mg infusion plus a brief educational intervention reduced migraine days by −6.9 versus −3.7 (Δ −3.2; 95% CI −4.2 to −2.2; p<0.0001), and 37.8% versus 18.1% no longer met chronic-migraine or MOH criteria within a month (OR 3.3). A 2025 network meta-analysis (J Headache Pain) found abrupt withdrawal alone was not statistically better than control for monthly headache-day reduction (MD −2.77; 95% CI −5.74 to 0.20) — though withdrawal combined with prevention performed well, and withdrawal may still serve selected patients. The fair summary: the evidence increasingly favors leading with prevention, with or without structured withdrawal, over relying on detoxification alone.

Psychiatric comorbidity. UNITE (JAMA Neurol 2025, post-window) tested fremanezumab in migraine with active major depression: migraine days −5.1 versus −2.9 (p<0.001) and HAM-D-17 −6.0 versus −4.6 at week 8 (p=0.02; downstream endpoints nominal after the hierarchy stopped). The counterweight is the guideline's own comorbidity-driven logic: antidepressants and antihypertensives offer dual benefit in depression and hypertension, and legacy agents retain defined roles in pregnancy planning (where the guideline emphasizes maximizing nonpharmacologic approaches and careful risk–benefit discussion — valproate and topiramate teratogenicity remain disqualifying considerations), obesity, and cost-constrained care.

Safety: The Unresolved CGRP Question

A claims-based analysis (Neurology 2026; 900,370 MarketScan beneficiaries — also post-window) associated CGRP-inhibitor initiation with a composite cardiovascular aHR of 1.26 (95% CI 1.10–1.45; 8.77 vs 6.76 events per 1,000 person-years), driven by ischemic stroke (aHR 1.26, 1.07–1.49), with MI, revascularization, CRAO, and intracranial hemorrhage not elevated. This is a hypothesis-generating observational signal — initiators had more baseline cardiovascular morbidity, the absolute increase is small, and no CV signal emerged in STRIVE — not a demonstration that CGRP blockade causes stroke. The practical translation is vascular risk assessment before initiation, particularly in aura plus vascular risk factors, not class avoidance. Class-specific tolerability is otherwise well characterized: constipation with atogepant (6.9–7.7% vs 0.5% in ADVANCE), injection-site pain (~9% in EVOLVE-2), and topiramate's cognitive burden, now quantified head-to-head in TEMPLE.

Evidence Hierarchy Meets Step Therapy

What follows is editorial analysis — neither the guideline nor its press releases address payer policy. Most commercial step edits require failure of two generic oral preventives before CGRP-pathway access. Set that against the numbers: APPRAISE suggests that, at least for episodic migraine after prior failure, the oral-preventive path yields a 16.8% composite success rate and 34.6% switching at one year; HER-MES and TEMPLE quantify the toll of a mandated topiramate trial at 30–39% adverse-event discontinuation. The guideline will not adjudicate this — its own review found comparative evidence insufficient, and generics still cost dollars a month while CGRP agents cost hundreds. But the guideline's per-drug confidence ratings are now citable, society-endorsed, AAFP-co-signed text: amitriptyline and rimegepant low confidence; atogepant and the monoclonals moderate to high. That the hierarchy demotes one CGRP agent alongside the tricyclic is, paradoxically, its greatest rhetorical strength in appeals — it demonstrates the ratings are evidence-driven, not class advocacy. A conservative prediction: those ratings, quoted verbatim, become the working lever in prior-authorization appeals — less because the guideline demands CGRP-first therapy than because it removes the pretense that all "first-line" orals rest on equivalent evidence.

What the General Neurologist Needs to Know

• Offer prevention at ≥4 migraine days/month, ≥4 moderate-to-severe headache days/month, or any functionally impairing migraine — a lower bar than common practice.
• The guideline's hierarchy is placebo-based, not comparative: active-comparator evidence through June 6, 2024 was graded low/very low confidence. TEMPLE, RESOLUTION, UNITE, and the CV claims analysis all post-date the review.
• The ratings are per-drug, not per-class: the CGRP monoclonals and atogepant sit at moderate-to-high confidence, but rimegepant lands in the low-confidence episodic tier alongside amitriptyline.
• Head-to-head, topiramate's record is consistently inferior on tolerability (AE discontinuation 30–39% vs 10–12%) and response (31–39% vs 55–64%) — post-window context, not guideline text.
• After 1–2 oral failures in episodic migraine, switching to erenumab is directly evidence-backed (APPRAISE composite OR 6.48); applying that result to other CGRP agents or chronic migraine is reasonable extrapolation, not trial-proven.
• OnabotulinumtoxinA holds high-confidence chronic-migraine evidence with low AE discontinuation (3.8% in pooled PREEMPT) — do not let the CGRP narrative crowd it out.
• In medication overuse, lead with prevention using agents proven in that population (CGRP mAbs, atogepant, onabotulinumtoxinA, topiramate); abrupt withdrawal alone was not statistically better than control in the 2025 network meta-analysis, so favor adding prevention over detox-only strategies.
• Screen vascular risk before CGRP initiation — the claims-based stroke signal (aHR 1.26) warrants awareness, not avoidance.
• Legacy orals keep defined roles: cost, dual-benefit comorbidities (hypertension, depression), and — with caution and counseling — pregnancy planning.

Conclusion

The 2026 AAN/AHS guideline is best understood not as a verdict on the CGRP era but as its constitution: for the first time, a 50-year-old tricyclic, the 2004 topiramate program, botulinum toxin, and six migraine-specific CGRP-pathway agents are graded on one scale, by one panel, with modern endpoints — and the scale cuts across class lines, elevating atogepant and the monoclonals while placing rimegepant beside amitriptyline. The guideline itself remains scrupulously agnostic about which drug beats which — its comparative evidence base, frozen in June 2024, would not support more. Yet the trials accumulating just outside that window keep pointing the same direction: when patients are randomized between a CGRP-targeted agent and topiramate or usual oral care, they stay on the former and abandon the latter. General neurologists should use the guideline for what it firmly establishes — a lower threshold for offering prevention, per-drug confidence ratings, and treat-forward guidance in medication overuse — while recognizing that the collision between this evidence framework and fail-two-generics-first payer architecture is now the live policy question of the next several years. The next guideline should not take fourteen years, and when it comes, comparative effectiveness — not placebo superiority — will be the question it must answer.