Low-Dose GLP-1 and the Brain: What the Alzheimer's Trials Found

"Neuroprotection" has become one of the standard selling points for low-dose GLP-1 programs. We found it on the marketing pages of providers we track, stated as established fact — one of them tells readers that GLP-1 microdosing "has been proven to improve" cognitive function and neuroprotection.

That claim has now been tested at scale, and it failed.

This page is about what the trials found, why the earlier signal looked so convincing, and what any of it has to do with a deliberately low dose. The short version of the last part: nothing, because nobody has studied it.

What EVOKE and EVOKE+ were

Two phase 3 randomized, double-blind, placebo-controlled trials of oral semaglutide 14 mg in early-stage symptomatic Alzheimer's disease, designed together and reported together ¹.

The scale is the point. 3,808 people randomized across 566 sites in 40 countries, all with amyloid-confirmed early Alzheimer's disease, with change on the Clinical Dementia Rating Sum of Boxes at week 104 as the primary endpoint ². This was not a substudy or an incidental finding. It was the definitive test of the hypothesis, designed by people who believed it, and it is by a wide margin the largest and best-controlled examination of a GLP-1 and cognition that exists.

What they found

Nothing.

The difference on the primary endpoint was −0.08 in EVOKE (95% CI −0.35 to 0.20, p=0.57) and +0.10 in EVOKE+ (p=0.46) ². Both confidence intervals sit across zero. Neither trial found a signal in either direction.

The published report states that both trials were discontinued because of the negative clinical outcome, and both registry records show status completed as of 30 January 2026 ^3,4. The program is over. This is not a result awaiting a longer follow-up.

Why the earlier signal looked so strong

It is worth understanding, because the same reasoning is still being used on marketing pages that have not caught up.

The randomized signal was a subgroup inside a null result. A systematic review and meta-analysis of 26 randomized trials of cardioprotective glucose-lowering agents and dementia risk reported an overall null finding — odds ratio 0.83, with a confidence interval crossing one. Only the GLP-1 subgroup reached significance, at 0.55 ⁶. A significant subgroup inside a null overall result is a hypothesis, and it was correctly treated as one: it is part of what EVOKE was built to test ⁵.

And those dementia diagnoses were not adjudicated cognitive endpoints. They were diagnoses captured incidentally inside cardiovascular outcome trials that were measuring something else entirely ⁵. That is a legitimate way to generate a hypothesis and a poor way to settle one.

The observational signal is almost certainly confounding. The widely circulated real-world claim comes from an electronic-health-record study across 116 million US patients, reporting semaglutide associated with a large reduction in first-time Alzheimer's diagnoses in people with type 2 diabetes ⁹. It is an enormous dataset and it is not a trial.

The most useful check on it is a study that ran the same method across drug classes. In a target-trial emulation, GLP-1 receptor agonists were associated with lower dementia risk against other glucose-lowering drugs — and SGLT2 inhibitors showed the same or a larger association, with no significant difference between the two classes head to head ⁷.

That is the finding that should change your mind. If two mechanistically unrelated drug classes produce the same apparent protection, the most parsimonious explanation is not that both are neuroprotective. It is that the people who get prescribed either one differ from the people who do not, in ways no adjustment fully captures — a confounding pattern that a randomized trial removes, and which the randomized trial duly did.

The bit that matters here: dose was never tested

Everything above is about full doses in people with established early Alzheimer's disease. Now the part that applies to this site.

There is no randomized or observational evidence on low-dose or microdose GLP-1 and cognitive outcomes. Not weak evidence. None.

EVOKE used oral semaglutide 14 mg — the maximum approved dose of that formulation. No study identified anywhere stratifies cognitive or dementia outcomes by dose. The dose-response for cognition is entirely untested, in both directions: nobody has shown a lower dose works, and nobody has shown a higher one does either.

So a provider claiming that microdosing delivers cognitive benefit is making a claim with three separate gaps in it. The benefit itself failed its definitive randomized test. The observational support behind it is not class-specific. And even if all of that had gone the other way, none of it was measured at anything like a microdose.

What EVOKE does not rule out

Being fair to the hypothesis, because the honest reading of a negative trial is bounded too.

EVOKE tested treatment of established early Alzheimer's disease over 104 weeks. It did not test primary prevention in cognitively healthy people, and a null result in symptomatic disease does not formally exclude an effect over a much longer horizon in people who are well. Amyloid pathology accumulates for many years before symptoms; a two-year trial in people who already have symptoms is a different question from a twenty-year one in people who do not.

That is a real limitation and it is worth stating. It is also not a rescue. No trial of that longer question has reported, and until one does, "it might work over decades in people without the disease" is a hypothesis with no supporting randomized evidence — which is precisely what "GLP-1s protect the brain" was before EVOKE, and precisely what EVOKE was built to resolve.

One more caution if you go reading around this yourself. The Lancet published a linked commentary alongside the trial report, and at least one bibliographic record for commentary in this area is indexed under a trial-shaped publication type despite being a single-author opinion piece. A database label is not a study design. Check what a paper actually is before treating it as evidence — the same rule we apply to every citation on this site.

What to do with a provider that claims this

Treat it as a marketing claim that has been tested and did not hold, and let it inform how you read everything else on the page.

We record exactly this kind of overreach on the rows where we find it — one provider we track claims on its own microdosing page that the practice has been proven to improve longevity, cognitive function, neuroprotection and inflammation, which is a considerable distance beyond what any of those literatures support. It is not a reason on its own to avoid a provider. It is a reason to check the rest of their claims against a source rather than taking them.

For what the evidence does support about low-dose GLP-1 — which is real, narrower, and mostly about metabolism rather than cognition — start with what the evidence actually shows and GLP-1 for longevity.

The honest summary

The largest, best-designed test of a GLP-1 on cognition ever run enrolled 3,808 people with confirmed early Alzheimer's disease and found nothing, on both trials, and the program was stopped.

The observational data that made the idea popular shows the same association for a completely different drug class, which is what confounding looks like.

And no study of any design has ever examined a low dose. Anyone selling you neuroprotection at a microdose is three separate steps ahead of the evidence.