Semaglutide and Alzheimer's: What the EVOKE Trials Found

For a few years, semaglutide looked like it might do something no drug had managed: meaningfully slow Alzheimer’s disease through a route that had nothing to do with amyloid plaques.

The hope was built on a stack of observational studies. People with type 2 diabetes who took semaglutide seemed to develop dementia less often. The drug reduced inflammation. GLP-1 receptors exist in the brain. It was a coherent story.

Then the trials read out, and the story ended.

This article walks through what EVOKE and EVOKE+ tested, what they found, why some numbers moved while the ones that mattered did not, and what it means for the separate question of whether these drugs might prevent dementia in people who do not have it.

What were the EVOKE and EVOKE+ trials?

They were two matched phase 3 trials run by Novo Nordisk, and they were enormous by Alzheimer’s standards [1].

The basics:

• Who: adults aged 55 to 85 with mild cognitive impairment or mild dementia due to Alzheimer’s disease, with amyloid confirmed by brain scan or spinal fluid
• How many: nearly 10,000 people screened, 3,808 enrolled across 40 countries. EVOKE enrolled 1,855; EVOKE+ enrolled 1,953
• What they took: oral semaglutide once a day, or a matching placebo. Doses went from 3 mg to 7 mg at four weeks, then 14 mg at eight weeks, mirroring the type 2 diabetes regimen. Participants with side effects could stay on a lower dose
• How long: a 104-week main phase with a 52-week extension planned on top, for up to 156 weeks of treatment in total
• Main measure: change from the start to week 104 on the Clinical Dementia Rating-Sum of Boxes, a standard scale covering memory, orientation, judgment, community affairs, home life and personal care

The only difference between the two trials was that EVOKE excluded people with signs of small vessel disease on brain imaging while EVOKE+ allowed them. In practice that difference nearly disappeared: only 54 participants with small vessel pathology ended up enrolled, so the two study populations were essentially identical [2].

Participants averaged 72 years old. About half were overweight or obese, 47 percent were a healthy weight and 2.5 percent were underweight. Nearly 14 percent had type 2 diabetes. They were allowed to keep taking cholinesterase inhibitors, and some started anti-amyloid antibodies during the trial.

What did the trials find?

Nothing.

Novo Nordisk announced on November 24, 2025 that semaglutide treatment improved some Alzheimer’s-related biomarkers but that this did not translate into any delay in disease progression [3]. The planned one-year extension was discontinued on the basis of the efficacy results. When the trials were published in The Lancet on March 19, 2026, the authors’ conclusion was that the results “do not support the efficacy of 14 mg/day of semaglutide given for up to 156 weeks in participants with biomarker-confirmed Alzheimer’s disease in the MCI or mild dementia stage” [10]. The trials ran across 566 sites in 40 countries.

At the Clinical Trials on Alzheimer’s Disease conference in San Diego on December 3, 2025, Jeffrey Cummings of the University of Nevada, Las Vegas presented the detail. For both trials, the placebo and semaglutide groups declined at the same rate on the primary measure throughout the study, tracking exactly what you would expect for untreated early Alzheimer’s disease [2].

Every secondary outcome matched. Participants declined the same amount on daily-function scales. They progressed from mild cognitive impairment to dementia at the same rate. The same was true on the MoCA, ADAS-Cog13, MMSE and ADCOMS cognitive measures.

Asked whether any subgroup benefited, Novo Nordisk’s Peter Johannsen told the audience: “When you look at the curves, they are exactly on top of each other.” If a benefiting subgroup existed, he said, it would be a very small one [2].

The Alzheimer’s Association said it was disappointed and stressed that the results underscore the need for a diverse research pipeline [4].

If nothing worked, why do people mention the biomarkers?

Because a few measurements did move, and it is worth understanding why that was not enough.

About 100 participants per group had spinal fluid collected at the start, and roughly 60 per group at week 78. In that small substudy, seven markers showed nominally significant reductions of 10 percent or less with semaglutide [2]:

• p-tau181 and p-tau217 (tau phosphorylation markers)
• non-phosphorylated tau181 and tau205
• total tau and neurogranin (markers of neurodegeneration and synaptic loss)
• YKL-40 (a neuroinflammation marker)

In the blood, none of those moved. Instead, plasma GFAP rose about 4 percent on semaglutide in both trials, and plasma NfL rose about 5 percent in EVOKE+. Cummings said he had no explanation for that.

One blood marker moved a lot in the expected direction: high-sensitivity C-reactive protein, a general marker of inflammation, dropped roughly 30 percent. That suggests semaglutide was doing something real to inflammation in the body. It just did not help the brain.

Johannsen’s own assessment was that a 10 percent shift in spinal fluid markers is in the same range seen with drugs that later failed, while the anti-amyloid antibodies that did win approval, lecanemab and donanemab, produced much larger biomarker responses [2].

Did the drug even reach the brain?

Partly, and this is one of the open questions.

A substudy presented as a poster at the conference measured semaglutide concentrations after 12 weeks. Median cerebrospinal fluid concentration was 0.14 nmol/L, against 34.8 nmol/L in plasma, a ratio of about 0.4 percent [2]. Concentrations in fluid and blood were not correlated across participants.

Whether being present in spinal fluid means the drug actually penetrated brain tissue is a separate question, and whether brain penetration was even necessary is another. Johannsen argued it was plausible that peripheral effects, like reduced inflammation, could have benefited the brain without much direct entry. That argument did not survive the result.

Why did observational studies get it so wrong?

They did not exactly get it wrong. They answered a different question.

Multiple large studies have tied GLP-1 drugs to lower dementia risk. A real-world target trial emulation in people with type 2 diabetes found semaglutide associated with hazard ratios ranging from 0.54 versus insulin to 0.80 versus older GLP-1 drugs for all Alzheimer’s disease and related dementias, with the strongest effect for vascular dementia at 0.48 versus insulin [5][6]. A pooled analysis of Danish registry data and cardiovascular trial data found people on semaglutide or liraglutide were less than half as likely to develop dementia [2].

Those are studies of who develops dementia. EVOKE studied whether the drug slows dementia that has already started. Those are not the same thing, and a drug can plausibly do one and not the other.

There is also a confounding problem that does not go away with statistics. Johannsen acknowledged it directly: people who control their type 2 diabetes with GLP-1 drugs differ from people who do not, in health behaviors, access to care and more. And reverse causality runs the other way too. Someone with plaques and tangles quietly accumulating may be less organized about managing their diabetes, which would make them less likely to be on a newer drug and more likely to get a dementia diagnosis later.

Lon Schneider of the University of Southern California put it bluntly to Alzforum: the negative result is the latest lesson in the risk of using epidemiological findings as the basis for an intervention trial. Statins, NSAIDs and fish oil all followed the same arc [2].

Schneider also noted that Novo Nordisk skipped a phase 2 dose-finding trial and went straight into two large phase 3 studies, using real-world evidence “to talk themselves into Phase 3.”

Is the GLP-1 brain hypothesis dead?

Not completely, but it has been narrowed considerably.

Several researchers argued after the readout that different questions remain open:

Prevention rather than treatment. Malú Tansey of Indiana University suggested that if the diabetes-dementia link is driven by chronic inflammation, then metabolic dysregulation of immune cells likely happens long before amyloid and symptoms appear. Starting a drug in someone who is already amyloid-positive may simply be too late. “A prevention strategy at stage 0 could turn out to be more effective,” she wrote [2].

After amyloid removal. Paul Aisen of the University of Southern California argued the best way to evaluate non-amyloid therapies is after amyloid has been cleared, since amyloid is the primary driver early on. A GLP-1 drug might be worth testing in early Alzheimer’s disease after anti-amyloid treatment, or in non-Alzheimer’s conditions like vascular cognitive impairment [2].

A proteomic hint. A separate poster applied a machine-learning plasma protein score to participants in the SELECT cardiovascular trial. Two years of semaglutide was associated with a 26 percent lower predicted five-year dementia risk and a 9 percent lower predicted 20-year risk [2]. A predicted risk score is not a diagnosis, and this was a conference poster, but it keeps the prevention question alive.

The Alzheimer’s Drug Discovery Foundation framed the completion of two rigorous phase 3 trials targeting a non-amyloid pathway as real progress toward combination approaches, noting that existing anti-amyloid drugs slow decline by around 30 percent and something has to address the other 70 percent [7].

What about Parkinson’s disease?

The GLP-1 story in Parkinson’s has followed the same shape, and it got there first.

Exenatide-PD3 was the largest and longest trial of a GLP-1 drug in Parkinson’s disease: 194 people at six UK hospitals, randomized to weekly exenatide or placebo for 96 weeks. Published in The Lancet on February 4, 2025, it found no advantage at any point, on movement examination or on patients’ own reports [8].

The Michael J. Fox Foundation said the results “effectively close the door on the idea of exenatide as a treatment for Parkinson’s,” while noting that other GLP-1 drugs act differently and more research could still be warranted [9]. As of May 2025, no phase 3 study had shown disease-modifying efficacy for any GLP-1 therapy in Parkinson’s disease.

A phase 2b trial of liraglutide in mild to moderate Alzheimer’s disease also missed its primary endpoint, and was not powered to detect cognitive change [2].

What this means if you or a family member takes semaglutide

A few practical points.

Semaglutide is not a dementia treatment and should not be taken as one. No regulator anywhere has approved it for that, and the largest trials ever run found no effect.

If you are already taking it for diabetes, weight, heart or kidney reasons, nothing about this changes that. The safety profile in EVOKE was consistent with what is known: more total adverse events than placebo, driven by mild to moderate gastrointestinal symptoms, with no difference in serious, severe or fatal events [2]. Participants lost an average of 5.8 percent of body weight over two years while the placebo group gained 0.6 percent.

Do not read the observational dementia studies as a reason to start. That is precisely the inference EVOKE tested and failed to support.

If someone in your family has Alzheimer’s disease, treatment decisions belong with their care team. Anti-amyloid antibodies, symptomatic medications and non-drug care approaches all exist and all have their own trade-offs.

Sources

1. evoke and evoke+: design of two large-scale phase 3 studies evaluating semaglutide in early-stage symptomatic Alzheimer’s disease — Alzheimer’s and Dementia, January 2025
2. Semaglutide Does Not Treat Alzheimer’s. Could It Prevent Dementia? — Alzforum, December 16, 2025
3. Evoke phase 3 trials did not demonstrate a statistically significant reduction in Alzheimer’s disease progression — Novo Nordisk, November 24, 2025
4. Alzheimer’s Association Statement on Oral Semaglutide Phase 3 Topline Data Release — Alzheimer’s Association, November 24, 2025
5. Associations of semaglutide with Alzheimer’s disease-related dementias in patients with type 2 diabetes: a real-world target trial emulation study — 2025
6. Semaglutide Use Tied to Lower Dementia Risk in Patients With Type 2 Diabetes — Neurology Advisor
7. Readout of Phase 3 Semaglutide Trials Marks Critical Moment in Alzheimer’s Research — Alzheimer’s Drug Discovery Foundation, November 24, 2025
8. Exenatide once a week versus placebo as a potential disease-modifying treatment for Parkinson’s disease — The Lancet, February 4, 2025
9. Exenatide, a GLP-1 Drug, Shows No Impact on Parkinson’s Symptoms — Michael J. Fox Foundation, February 2025
10. Efficacy and safety of oral semaglutide 14 mg (flexible dose) in early-stage symptomatic Alzheimer’s disease (evoke and evoke+): two phase 3, randomised, placebo-controlled trials — Cummings JL, Atri A, Sano M, et al., The Lancet, March 19, 2026;407(10544):2167-2179
11. GLP-1 Semaglutide Fails to Outperform Placebo in Phase 3 EVOKE Trial of Alzheimer Disease — NeurologyLive
12. Can Semaglutide Prevent Dementia? What the Research Says — GoodRx