Semaglutide and Alzheimer's: The Trials Came Back Negative

For three years the most exciting claim about GLP-1 drugs was that they might slow Alzheimer’s disease. Two large randomized trials set out to test it directly, enrolled 3,808 people, ran for two years — and found nothing. Both were discontinued for negative clinical outcome.^[1] This is what the answer looks like when the question is finally asked properly.

What the trials did

The evoke and evoke+ trials randomized adults with early-stage symptomatic Alzheimer’s disease to oral semaglutide 14 mg or placebo, and followed them for 104 weeks. Participants had a mean age of 72.2 and a mean CDR-SB score of 3.7 at baseline — a standard measure of dementia severity where higher is worse and change over time is the outcome that matters.^[1]

Both confidence intervals sit squarely across zero and the point estimates go in opposite directions. This is not a trial that narrowly missed, or one that would have succeeded with more participants. It is a flat result in nearly four thousand people.

Two point three against two point three. In a field where every fraction of a point is fought over, that is as clear as evidence gets.

Why the earlier signals looked so promising

This is the useful part, because the earlier evidence was not fabricated and was widely reported. We covered it in three analyses, three answers: a meta-analysis of 26 trials found a 45% reduction in dementia outcomes, a cohort of 396,963 patients found 33%, and a continuously updated review found nothing significant and graded its own certainty as low.

The pattern is familiar and worth learning. Dementia outcomes in those trials were incidental — recorded as adverse events in studies designed around diabetes and cardiovascular disease, in people who were not selected for cognitive symptoms. Cohort studies compared people prescribed a newer drug against people who were not, and the difference between those groups is never only the drug. We take that apart in the cancer cohorts, where the same mechanism produced a 91% mortality difference that evaporated against a fair comparator.

A second drug, a third negative primary endpoint

A different GLP-1 drug has now failed the same question in a separate trial. ELAD randomized 204 people with mild to moderate Alzheimer’s disease and no diabetes to daily liraglutide or placebo for 52 weeks, with brain imaging and detailed cognitive testing.^[2]

Its primary outcome was the rate of glucose metabolism in the brain, measured by PET. There was no significant difference — a difference of −0.17 with a 95% confidence interval from −0.39 to 0.06, P = 0.14.

The pattern across the two programs is now consistent, and it is the pattern that should be believed: three primary endpoints, three failures, two different drugs. Positive findings survive in secondary and exploratory measures, which is where positive findings usually survive after a primary endpoint has gone the other way.

It is worth adding that a smaller trial in a different population reported a comparable shape — three cognitive domains improving, one surviving adjustment, at exactly the significance threshold — in three domains improved, then one. Nothing here rules out a real cognitive effect. It does mean that every properly powered test of it has so far come back negative on the question it set out to ask.

What the trials do not settle

Being precise here matters, because overclaiming in the skeptical direction is the same error wearing a different coat.

• This tested treatment, not prevention. Participants already had symptomatic Alzheimer’s. Whether taking a GLP-1 in midlife changes the odds of developing dementia decades later is a different question and remains open.
• It tested oral semaglutide at 14 mg. Not injectable semaglutide, not tirzepatide, and not the higher doses used for weight management.
• Early-stage symptomatic disease is a specific window. Most Alzheimer’s drug trials fail in it, and that is a hard place to show benefit.
• It says nothing about cognition in people without dementia, which is a separate and much murkier literature.

On safety, the numbers went the other way

Adverse events were more common on semaglutide — 91.2% against 84.8% — which is consistent with everything known about this drug class and mostly reflects gastrointestinal effects.^[1]

But of five deaths that investigators judged related to treatment, one was in the semaglutide group and four were in the placebo group.^[1] Nothing should be concluded from five events. It is worth stating plainly because a failed trial in an elderly population is exactly the setting where a safety scare tends to be assembled from selective reporting, and the actual numbers do not support one.