Semaglutide and Alzheimer's after EVOKE

A commentary published in the Journal of Alzheimer's Disease on 10 September 2026 takes on a question that has been simmering since late 2025: what should researchers do next after semaglutide failed to slow Alzheimer's disease in two large phase 3 trials^[1]? The trials, EVOKE and EVOKE+, missed their primary endpoint and every cognitive or functional secondary endpoint^[1][2]. That result has not changed. The new paper's authors, Pereira Da Silva and Haddad Santos, are arguing about how the field should read a more recent reanalysis of that same negative data, and what kind of study should come after it.

What EVOKE actually found

EVOKE and EVOKE+ were randomised, placebo-controlled phase 3 trials of oral semaglutide in people aged 55 to 85 with mild cognitive impairment or mild dementia and amyloid confirmed by PET scan or spinal fluid test^[2]. Across the two trials, more than 3,800 people were randomised to a daily semaglutide tablet titrated up to 14 mg, or to placebo, for up to 156 weeks^[2]. The primary outcome was change in the Clinical Dementia Rating Sum of Boxes score at week 104, a clinician-rated measure of memory, judgement and daily function.

On that measure, the drug did not separate from placebo in either trial, and neither did any of the cognitive or functional secondary measures^[1][2]. Both trials were stopped on that basis. This is the fact that has to stay in view through everything below: EVOKE was a negative result, not a mixed or borderline one.

The reanalysis, and the pushback against it

A separate paper published in the same journal in August 2026 went back into the EVOKE data past week 104, into the extension period out to weeks 130 and 156^[3]. It reported a statistically significant difference favouring semaglutide on a daily-function scale at week 130, plus a trend on a cognitive scale at week 156, and proposed that semaglutide's limited ability to cross the blood-brain barrier explains why the primary-endpoint result came back flat^[3]. That paper is the one the new commentary responds to.

Pereira Da Silva and Haddad Santos argue that reanalysis leans on evidence weaker than it appears^[1]. Their first point is attrition: by weeks 130 and 156, a large share of the original trial population had already dropped out, so the group still being measured is not the group that was randomised. A late-timepoint comparison of survivors is a weaker kind of evidence than the prespecified week-104 comparison of everyone enrolled.

Their second point concerns the biomarker substudy run alongside the trials. Rather than a clean, supportive signal for disease modification, the authors say the biomarker data complicate that story rather than confirm it^[1]. Movement in a blood or spinal-fluid marker is not the same thing as a drug changing the course of a disease, and here even the marker picture does not line up neatly.

From a single hypothesis to a stratified program

The commentary's stated conclusion is that EVOKE does not eliminate the broader idea connecting metabolic drugs like semaglutide to brain health, sometimes called the cardiometabolic hypothesis, but it does change what the next study needs to look like^[1]. In the authors' own words, the open questions are now "which pathways, which patients, and which disease stages are most likely to benefit"^[1].

That is a call to stop repeating a single, all-comers trial design (one drug, one broad population of people who already have symptoms, one hoped-for outcome) and instead design smaller, better-targeted studies. Candidates researchers have floated elsewhere include enrolling people earlier in the disease process, before symptoms appear, or selecting participants by a specific biomarker or metabolic profile rather than an Alzheimer's diagnosis alone^[3]. None of that has been tested yet. It is a proposed research direction, not a result.

What this is not

This is not evidence that semaglutide or any other GLP-1 receptor agonist treats or prevents Alzheimer's disease. The only completed phase 3 test of that idea, EVOKE and EVOKE+, was negative on its primary endpoint and on every secondary endpoint^[1][2]. The commentary covered here is an argument about how to interpret a reanalysis of that negative data and what to study next, not a new trial and not a reversal of the original result.

It is also not a reason to use semaglutide off-label for memory protection. In the United States, oral semaglutide is approved only for glycaemic control in type-2 diabetes and to reduce cardiovascular risk in adults with type-2 diabetes at high risk^[4]. The label does not mention Alzheimer's disease, dementia or cognition, and no regulator has authorised any GLP-1 drug for a cognitive indication.

What remains genuinely uncertain

A few questions stay open. Whether a more brain-penetrant GLP-1 drug would perform differently has not been tested in a phase 3 trial. Whether enrolling people earlier, before symptoms appear, would change the outcome is also untested, and would take years to answer. And whether the biomarker movements inside EVOKE reflect anything a patient or family would notice is a question the commentary itself says the data cannot settle.

What this means for readers

If you or someone in your family has an Alzheimer's diagnosis, this commentary is not a reason to add or seek out a GLP-1 drug for cognitive protection. It is a methods debate among researchers about how to design the next study, running on top of a trial result that stayed negative. If you already take semaglutide for diabetes or weight management, nothing here changes that drug's approved use or safety profile^[4].

None of this is medical advice. Decisions about an Alzheimer's diagnosis or a prescription medicine belong with a clinician who can see the full picture. For more, see our explainer on reading the EVOKE trial result and the semaglutide overview.