Between late 2025 and April 2026, two major trials of GLP-1 peptides in Alzheimer's disease published contrasting results. Semaglutide missed its Phase 3 primary endpoint. Liraglutide hit Phase 2b secondary endpoints with a striking MRI signal. If you've been following the "GLP-1s might cure dementia" headlines from 2024, this is the reality check — and it changes how buyers should think about GLP-1 peptides for brain health.
Research-context information only. Peptides discussed below are research compounds. Protocols, doses, and reactions reported come from published research and self-reported community sources. This article reports what has been documented, not what should be done. Consult a licensed physician for personal medical decisions.
The Two Trials That Changed the Story
Two datasets matter more than the rest. Everything else published in this window either supports one of these or reflects observational data with known limitations.
EVOKE and EVOKE+ (semaglutide, Phase 3) — negative. Novo Nordisk ran two parallel Phase 3 trials enrolling 3,808 participants aged 55 to 85 with mild cognitive impairment or mild Alzheimer's dementia confirmed by amyloid PET or CSF. Patients received oral semaglutide 14 mg or placebo for 104 weeks. The primary endpoint was change in CDR-SB (Clinical Dementia Rating — Sum of Boxes). Results published in The Lancet March 2026: mean CDR-SB change was 2.3 on semaglutide vs 2.3 on placebo in EVOKE, and 2.2 vs 2.1 in EVOKE+. Secondary cognitive and functional outcomes also showed no meaningful difference. The 52-week blinded extension was discontinued.
The one positive signal: amyloid and tau biomarkers dropped modestly in the semaglutide arm (roughly 10% reductions in some markers), and inflammation markers moved in the right direction. Biological activity was real. It just didn't translate into cognitive or functional benefit over two years.
ELAD (liraglutide, Phase 2b) — mixed but interesting. Published in Nature Medicine in late 2025, this Imperial College-led UK trial randomized 204 patients with mild to moderate Alzheimer's (without diabetes) to daily injected liraglutide or placebo for 52 weeks. The primary endpoint — change in cerebral glucose metabolism on FDG-PET — was not met (p = 0.14). But secondary endpoints told a different story:
- 18% slower cognitive decline on the ADAS-Exec (executive function) measure
- Nearly 50% less gray matter volume loss on MRI, specifically in frontal, temporal, parietal, and total gray matter regions
- Safety and tolerability: well tolerated in a non-diabetic AD population
The effect sizes are large enough that they shouldn't be dismissed. They're also Phase 2b, underpowered for clinical endpoints, and hinge on surrogate measures that don't always predict Phase 3 success.
Why the Results Diverged
Nobody knows for certain, but four hypotheses are on the table:
1. Blood-brain barrier penetration. Liraglutide is a smaller peptide with properties that appear to cross the blood-brain barrier more readily than semaglutide, which was engineered for albumin binding and long half-life at the cost of CNS penetration. If the target is neuronal GLP-1 receptors, liraglutide may simply reach more of them.
2. Oral vs injectable formulation. EVOKE used oral semaglutide (oral tablet formulation, roughly 1% bioavailability). ELAD used daily subcutaneous injections. Peak CNS exposure differs substantially.
3. Disease stage. ELAD enrolled mild to moderate dementia. EVOKE specifically targeted early-stage symptomatic AD (MCI or mild dementia). Counter-intuitively, the less-advanced population in EVOKE showed no benefit. This may reflect floor effects in early patients — less to "save" in 104 weeks — or something about amyloid-confirmed populations that differs from the broader ELAD cohort.
4. Trial duration and power. ELAD ran 52 weeks in 204 patients. EVOKE ran 104 weeks in nearly 20x the sample. A real but small effect could have been missed in ELAD power-wise; in EVOKE, a real small effect should have been detected. The most parsimonious read: semaglutide's CNS effect is too weak to matter at scale.
What the Observational Data Still Shows
The negative EVOKE readout does not erase the observational evidence, which is large and consistent:
- A 295,000-patient propensity-matched cohort found 70% reduced dementia risk in GLP-1 users vs non-users
- A separate retrospective cohort showed HR 0.58 for dementia in GLP-1 users vs DPP-4 inhibitor users (95% CI 0.55–0.61, p < 0.0001)
- Multiple 2025 reviews of diabetic medical records reported 40–70% lower dementia incidence in GLP-1 users across different drug classes
Observational data can't establish causality. Healthier patients get put on GLP-1 drugs. Better glycemic control independently reduces dementia risk. Socioeconomic confounders are hard to fully adjust for. But the signal has been reproduced across independent datasets, drug classes, and countries. It's real; it's just not necessarily the drug doing the work.
For vendor comparisons and current pricing on the GLP-1 peptides referenced in these trials:
- Best semaglutide vendors — ranked by price, COA, reputation
- Best liraglutide vendors — daily injection GLP-1
- Best retatrutide vendors — triple-agonist GLP-1
- All vendor discount codes — current coupon round
