Ozempic and dementia risk: what the GLP-1 research actually shows

Short answer: The hopeful headlines came from observational data. The rigorous test has now been run — and it failed. Two large phase-3 trials (EVOKE and EVOKE+) randomised 3,808 people with early Alzheimer's to oral semaglutide or placebo, and after two years found no clinical difference between the groups. That doesn't mean GLP-1 drugs are bad for the brain; it means we now know they don't slow Alzheimer's once it has started. Whether they lower risk in people without cognitive symptoms years earlier is still an open question no trial has been built to answer. If you take Ozempic, Wegovy or Mounjaro for diabetes or weight, keep taking it as prescribed — and take any change, in either direction, to your prescriber rather than deciding off an article. Poorly controlled blood sugar and midlife obesity are established modifiable risk factors for dementia, so they are worth managing well on their own merits. Just don't take a GLP-1 drug for your brain.

This is educational, not medical advice. Never start, stop, or change a prescription medication based on an article. Decisions about GLP-1 drugs — including whether they're right for you — belong with your clinician.

Where the excitement came from

For a few years, the observational evidence looked genuinely exciting.

The most-quoted study analysed the health records of 1,094,761 US adults with type 2 diabetes and found that those prescribed semaglutide had a 40% to 70% lower rate of a first Alzheimer's diagnosis than those on other diabetes medications — a hazard ratio of 0.33 compared with insulin, and 0.59 compared with other GLP-1 drugs (Wang & Xu et al., Alzheimer's & Dementia, 2024).

It's worth knowing what that percentage sits on top of. First Alzheimer's diagnoses were uncommon over the study's three-year window: among adults aged 60 and over, the risk of a first-time Alzheimer's diagnosis was 0.33% — about twice the 0.16% seen in the general population. Large relative differences built on small absolute numbers are easy to generate and easy to over-read.

The biology made sense too. GLP-1 receptors exist in the brain. In animal studies these drugs reduce neuroinflammation, calm activated microglia, and lower amyloid and tau markers. And they improve nearly every cardiometabolic risk factor that matters for the brain — weight, blood sugar, blood pressure.

So: a plausible mechanism, a million-patient signal, and a drug millions of people were already taking. It's easy to see why "Ozempic may prevent Alzheimer's" became a headline.

What the trials found

Novo Nordisk ran the definitive test. EVOKE and EVOKE+ enrolled 3,808 participants across 40 countries, average age 72, all with mild cognitive impairment or mild dementia due to Alzheimer's. They took oral semaglutide titrated up to 14 mg daily, or placebo, for up to three years, with the primary result measured at two years.

The result, published in The Lancet in 2026: nothing. There was no difference between semaglutide and placebo on the primary measure of dementia severity (CDR-SB) at week 104, and none on the secondary measure of daily functioning. The two groups' decline curves sat on top of each other for the whole trial (The Lancet, 2026 · NeurologyLive summary).

The biomarker results are the most interesting part, and they cut both ways. A handful of spinal-fluid markers dropped slightly — 10% or less, and only nominally significant. Inflammation in the body fell substantially: C-reactive protein dropped around 30%. But in the blood, two markers of brain injury went in the wrong direction — GFAP rose about 4% in both trials, and neurofilament light rose about 5% in EVOKE+ only (Alzforum's conference coverage).

In plain terms: the drug did what it does — it reduced body-wide inflammation — and that simply didn't translate into a brain benefit in people who already had Alzheimer's pathology.

Why the observational studies looked so much better

This gap between "huge effect in health records" and "no effect in a trial" is one of the most useful lessons in all of health research, and it's worth understanding because you'll meet it again.

Confounding by indication. Doctors don't prescribe randomly. The people who get put on semaglutide, stay on it, tolerate it, and can afford it are systematically healthier, better-engaged with healthcare, and better off than the people put on insulin. Insulin is often the drug of last resort in long-standing, poorly controlled diabetes — which is itself strongly linked to dementia. Comparing the two compares two different kinds of patient, not just two drugs.

Reverse causation. Very early, undiagnosed cognitive decline changes behaviour years before diagnosis: people manage their medications less consistently, attend fewer appointments, lose weight. That can make the people already drifting toward dementia look like non-users.

The comparator decides the answer. This is the clearest evidence that the observational signal isn't solid. A 2026 study emulating a randomised trial in health-system records — the same method as the semaglutide study above, with different comparison drugs — found that GLP-1 drugs were associated with lower dementia risk than DPP-4 inhibitors (hazard ratio 0.76, 95% CI 0.59–0.97) — but higher dementia risk than SGLT2 inhibitors (hazard ratio 1.53, 95% CI 1.13–2.07) (target trial emulation, 2026). The same drug looks protective or harmful depending on which medication you compare it to. When that happens, you're usually looking at differences between patients, not effects of the drug.

None of this makes the observational researchers wrong to have looked. It's why we run trials.

Is the prevention question still open? Honestly, yes

EVOKE tested treatment in people with established Alzheimer's. It did not test risk reduction in healthy middle-aged people, and the researchers were careful to say so.

Alzheimer's pathology builds for a decade or two before symptoms appear. By the time someone qualifies for a trial like EVOKE, amyloid plaques and tau tangles are already extensive — and the argument from several researchers is that a metabolic drug might matter much earlier, before that damage accumulates. Malú Tansey of Indiana University, commenting to Alzforum on the trial results, put it this way: "a prevention strategy at stage 0 AD could turn out to be more effective, so I hope this is not the end of the road for semaglutide" (Alzforum).

That's a reasonable hypothesis. It is also, right now, only a hypothesis. No trial has been designed to test whether GLP-1 drugs lower dementia risk in people who don't yet have cognitive symptoms. The closest we have is REWIND, which randomised 9,901 people aged 50 and over with type 2 diabetes to weekly dulaglutide or placebo for a median of 5.4 years. In an exploratory analysis of the 8,828 participants with cognitive testing, the hazard ratio for substantive cognitive impairment — defined as a drop on a cognitive test, not a dementia diagnosis — was 0.93 (95% CI 0.85–1.02), which was not statistically significant. A 14% reduction appeared only after a post-hoc adjustment for baseline scores (HR 0.86, 95% CI 0.79–0.95) (Cukierman-Yaffe et al., Lancet Neurology, 2020). That is a hint, not an answer, and a trial built to answer the question properly would need to run for many years. Anyone telling you today that Ozempic protects your brain is describing hope, not evidence.

What this means if you take a GLP-1 drug

If you're prescribed one for diabetes or obesity: keep taking it as prescribed, and take any change to your prescriber first. On the clinical measures, EVOKE gives no signal that semaglutide harms cognition — the two groups declined at the same rate, and tolerability was consistent with what's known about the drug in its other uses. The small rises in GFAP and neurofilament light noted above remain unexplained; they are one reason the biomarker picture is better described as mixed than as reassuring. And here's the point that gets lost: poorly controlled blood sugar and midlife obesity are established modifiable risk factors for dementia. Trials haven't yet shown that treating them lowers dementia rates specifically, but they are worth managing well on their own merits, and doing so is consistent with the risk-factor evidence — see our guides to blood sugar and dementia and the 14 modifiable risk factors.

If you're considering starting one specifically to protect your brain: that's not a supported reason. GLP-1 drugs are approved for diabetes and weight management. They aren't approved for dementia, and after EVOKE there's no evidence they treat it. The decision should rest entirely on your metabolic health, with your clinician.

If you're stopping or switching: talk to your prescriber first, not to the internet. Weight regain and blood-sugar rebound after stopping have their own consequences for cardiovascular and brain health.

The bigger picture

There's a pattern here worth naming, because it will keep repeating: a drug or supplement shows a dramatic effect in observational data, the headlines run, and the trial comes back flat. It happened with hormone therapy, with vitamin E, with ginkgo, and now with semaglutide for Alzheimer's. The next case waiting to be settled the same way is the shingles vaccine, which is associated with lower dementia risk across an unusually consistent set of studies — including two natural experiments that handle confounding far better than the semaglutide records did — but whose first randomised trial won't report until around 2029.

Meanwhile the boring stuff keeps holding up. In July 2026 the World Health Organization published its updated guidance on reducing dementia risk, and the recommendations were the unglamorous ones: physical activity, managing blood pressure and diabetes and cholesterol, hearing aids where needed, not smoking, limiting alcohol, cognitive and social engagement, and — newly — reducing air-pollution exposure. At population level, WHO estimates that up to 45% of dementia risk can be attributed to modifiable factors like these (WHO, July 2026). That is a figure about whole populations. It is not a promise that any one person can cut their own risk by 45%.

How we research and review our content →

This article is educational and is not medical advice, diagnosis, or treatment. Studies cited — including the EVOKE and EVOKE+ semaglutide trials in The Lancet, the 2024 Alzheimer's & Dementia observational health-records analysis, the 2026 target trial emulation of GLP-1, DPP-4 and SGLT2 drugs, the REWIND dulaglutide trial and the 2026 WHO guidance — describe trial results, risk factors and associations in general, not Solenna specifically; the observational GLP-1 comparisons cannot establish cause and effect. Solenna does not diagnose, prevent, treat, or cure Alzheimer's disease or any form of dementia; risk reduction means lowering probability, not eliminating it, and no outcome is guaranteed. Never start, stop, or change semaglutide, tirzepatide or any other prescription medication without speaking to a qualified healthcare professional.