Semaglutide was tested against Alzheimer’s in 3,808 people. It did not slow decline.

Why this note exists

"GLP-1 drugs protect the brain" has become one of the most repeated secondary claims around semaglutide, and it had real reasons behind it. Until this year, those reasons were observational. The evoke program was the first time the idea was put through large randomized trials with a clinical endpoint, and the answer is now published in full, in The Lancet, with the numbers attached.[1]

In pepmg's price index generated on September 8, 2026, 30 of 108 vendors carried a semaglutide listing, 86 listings in all.[15] None of those 86 names a tablet or any other oral form, and the sizes cluster at 5, 10, 20 and 30 mg, which together account for 63 of them.[15] The product tested in evoke was an oral tablet.[1]

A note on what kind of evidence this is: every efficacy and safety figure below is human data, reported with its design and participant count. Where animal work is mentioned, it is only as the reason the trials were run, and it is labeled as animal work. Observational results are called observational in the same sentence as their number. Doses appear only as their sources published them.

The trials

Two trials, 566 sites, and the highest approved oral dose

Evoke and evoke+ were multicentre, randomised, double-blind, placebo-controlled phase 3 trials run across 566 sites in 40 countries.[1] They enrolled people aged 55 to 85 with amyloid-confirmed mild cognitive impairment or mild dementia due to Alzheimer's disease; evoke+ also admitted people with significant small vessel pathology, although in practice only 54 of its participants (2.8%) had it.[1] Participants were randomly assigned 1:1 to once-daily semaglutide 14 mg (flexible dose) or placebo for up to 156 weeks, on top of standard care.[1][2]

The dose matters for reading the result. 14 mg is the maximum recommended dose of oral semaglutide on its US label for type 2 diabetes.[13] The trials did not test a token dose. They tested oral semaglutide up to 14 mg daily.

Figures as published in the primary report.[1] The CDR-SB runs from 0 to 18, with higher scores meaning greater impairment across memory, orientation, judgement, community affairs, home life and personal care.[3] Both registry records still show no posted results as of September 10, 2026, and list enrollment as 1,840 each, which does not match the published totals; this note uses the published figures.[3][4]

The result

What "not efficacious" looks like in numbers

The primary endpoint was the change in Clinical Dementia Rating–Sum of Boxes from baseline to week 104, assessed in everyone randomized.[1] In evoke, both groups worsened by a mean of 2.3 points; the estimated difference was −0.08 (95% CI −0.35 to 0.20, p = 0.57).[1] In evoke+, the semaglutide group worsened by 2.2 and the placebo group by 2.1; the estimated difference was 0.10 (95% CI −0.17 to 0.38, p = 0.46).[1]

The confidence intervals are the useful part. Both are narrow, and both sit across zero. That is what a well-powered null result looks like: the trial did not fail to see an effect because it was too small; it saw no difference in a large sample.[1] Conference coverage of the December 2025 presentation reported that the secondary clinical measures (ADCS-ADL-MCI, MoCA, ADAS-Cog13, MMSE and ADCOMS) showed the same picture, as did time to progression from mild cognitive impairment to dementia.[5]

Both trials were stopped after the primary analysis. The published report states that "both trials have been discontinued due to negative clinical outcome," and Novo Nordisk announced in November 2025 that "[t]he 1-year extension period in the evoke and evoke+ trials will be discontinued based on the efficacy results."[1][2]

The biomarkers

Some spinal-fluid markers moved. The patients did not.

This is the part of the story most likely to be quoted selectively. Novo Nordisk's own announcement says that "treatment with semaglutide resulted in improvement of Alzheimer's disease-related biomarkers in both trials," and in the same sentence that "this did not translate into a delay of disease progression."[2]

The detail, as reported by Alzforum from the December 2025 presentation, is modest. The cerebrospinal fluid substudy included roughly 100 participants per group at baseline and about 60 per group at week 78. In that subset, seven CSF markers (including p-tau181, p-tau217, total tau, neurogranin and the inflammation marker YKL-40) showed nominally significant reductions of 10 percent or less.[5] The same markers did not shift in blood, where the report instead noted a fall of about 30 percent in high-sensitivity C-reactive protein, a general inflammation marker.[5] The Alzheimer's Drug Discovery Foundation described reductions "of up to 10%" as statistically significant but "not large enough to have a clinical impact."[12]

Two cautions follow from the design. "Nominally significant" usually means significant before correcting for the number of markers tested, and the week-78 comparison rests on about 60 people per group, out of 3,808 randomized.[5] A biomarker shift that arrives without any clinical difference is a clue about biology. It is not evidence of benefit.

The same coverage reported one pharmacology finding that bears on the mechanism question: a CSF substudy presented as a poster reported that after 12 weeks of treatment, median semaglutide concentration in spinal fluid was 0.14 nmol/L against 34.8 nmol/L in plasma, a ratio of about 0.4 percent, cited by a Novo Nordisk investigator answering whether the drug reaches the brain at all.[5]

Safety

The safety profile was the familiar one

Treatment-emergent adverse events were reported in 1,729 of 1,896 people receiving semaglutide (91.2%) and 1,613 of 1,902 receiving placebo (84.8%).[1] Investigators judged five deaths to be treatment-related: one in the semaglutide groups and four in the placebo groups.[1] The authors describe safety and tolerability as "consistent with studies in other indications," and a Novo-funded plain-language summary states that "no new adverse effects were identified."[1][8]

Alzforum's report of the safety presentation gives the texture: nearly a quarter of semaglutide participants reported nausea, 14 percent diarrhea and 12 percent vomiting, more semaglutide participants reduced their dose or stopped because of gastrointestinal effects, and average body weight fell 5.8 percent over two years against a 0.6 percent gain on placebo.[5] The drug did what semaglutide does to body weight. It did not change the course of the disease.[1]

The backstory

Where the hope came from, and why it was never the same question

The trial report's own background names the sources: "animal, clinical, and real-world studies in individuals with type 2 diabetes and/or obesity" had suggested a reduced risk of dementia after GLP-1 receptor agonist exposure.[1] Each of those pieces answered a different question from the one evoke asked.

The most-cited human signal is a 2022 analysis that pooled three cardiovascular outcome trials of GLP-1 drugs in people with type 2 diabetes, 15,820 patients in all, and reported a dementia hazard ratio of 0.47 (95% CI 0.25–0.86) for drug versus placebo; the same paper's Danish registry cohort of 120,054 patients reported a hazard ratio of 0.89 per additional year of exposure.[9] Those were trials designed to measure heart attacks and strokes, and the dementia comparison was drawn from them afterwards; several of the authors were Novo Nordisk employees.[9] A diagnosis of dementia recorded during a diabetes trial is not the same outcome as the rate of decline in people who already have Alzheimer's disease.

The observational literature has also turned out to depend heavily on what the drug is compared against. A 2026 target trial emulation using University of Pennsylvania health records, an observational design, not a randomized trial, found GLP-1 drug use in older adults with type 2 diabetes associated with lower dementia risk than DPP-4 inhibitors (HR 0.76, 95% CI 0.59–0.97) but higher risk than SGLT2 inhibitors (HR 1.53, 95% CI 1.13–2.07), and its external validation confirmed the first comparison but not the second.[10]

The one earlier randomized treatment trial of a GLP-1 drug in Alzheimer's disease pointed the same way evoke did. ELAD, a phase 2b trial of daily injected liraglutide in 204 people with mild to moderate Alzheimer's disease and no diabetes, published in Nature Medicine in December 2025, missed its primary endpoint of cerebral glucose metabolism (difference −0.17, 95% CI −0.39 to 0.06, P = 0.14) and found no difference on the CDR-SB (−0.06, 95% CI −0.57 to 0.44).[11] One secondary executive-function score favored liraglutide, with an unadjusted P of 0.01.[11] Its authors cite neuroprotective effects "in animal models" as the rationale, and that animal work is the base the whole hypothesis stands on.[11]

Where the field is going, briefly

The argument has moved from treatment to prevention. The Lancet published an accompanying comment with the trial report, and in August 2026 a letter titled "Semaglutide treatment fails but might prevent Alzheimer's disease."[6][7] The Alzheimer's Drug Discovery Foundation, which disclosed that it funded the liraglutide phase 2 from 2011, called the data "disappointing" and suggested exploring GLP-1 drugs "as a preventive therapy, which may still hold promise."[12] In Alzforum's coverage, one researcher argued that starting in people who are already amyloid-positive may be too late, and Lon Schneider suggested a phase 2 trial to settle pharmacokinetics and biomarker response should have come before two phase 3s.[5]

Those are hypotheses, not findings. As of this note, no randomized trial designed to test whether semaglutide prevents dementia has reported.

Three things this note is not saying

First, it is not saying the trials were badly run or too small. They were large, double-blind, placebo-controlled, published in full, and their confidence intervals are tight enough to be informative.[1] Novo Nordisk funded them, and five of the eleven authors of the primary report were current or former Novo Nordisk employees; the result is negative for the sponsor's own drug, which is the direction least likely to reflect sponsor bias.[1]

Second, it is not saying semaglutide does nothing. Its approved uses rest on their own trial programs: glycemic control and cardiovascular risk reduction in type 2 diabetes for the oral tablets, and chronic weight management, cardiovascular risk reduction and MASH with fibrosis for Wegovy.[13][14] None of those labels lists a cognitive or neurological indication.[13][14]

Third, it is not saying GLP-1 drugs cannot prevent dementia. That question was not tested here. The observational signal was about new diagnoses in people with diabetes; evoke tested slowing in people who already had symptomatic Alzheimer's disease.[1][9]

What it is saying is narrower: for treating early Alzheimer's disease, the question has now been asked in 3,808 people at oral semaglutide up to 14 mg daily, and the answer was no.[1] Any claim that semaglutide protects memory or slows cognitive decline is running ahead of the only large randomized evidence there is.

Questions people are asking