GLP-1 and Alzheimer's: Why the EVOKE 2025 Trial Failed

In late 2025, Novo Nordisk and a network of international study sites delivered the most important negative result in GLP-1 research of the year. The Phase 3 EVOKE and EVOKE+ trials tested oral semaglutide in patients with early Alzheimer disease. Neither trial showed a significant benefit over placebo on the primary cognitive endpoint. The programme was discontinued.

This article breaks with the GLP-1 Alzheimer narrative that has circulated through specialist press and tabloid headlines since 2023. It places the EVOKE result in the context of other failed neuroprotection trials with incretins, explains why the underlying hypothesis was plausible, and shows what a genuine positive cognition trial looks like with a very different peptide. For the European peptide research community, EVOKE is a case study in honest evidence evaluation, not schadenfreude.

Why the Hype Emerged in the First Place

Speculation that GLP-1 agonists could slow cognitive decline did not come from nowhere. It rested on three pillars, each plausible in its own right.

First, the epidemiological signal. A retrospective cohort study by Wang et al., published in 2024 in Alzheimer's & Dementia, analysed claims data from over one million patients and found significantly lower risk of newly diagnosed Alzheimer disease in diabetics on GLP-1 therapy compared to other antidiabetic regimens. Tang et al. replicated similar signals in a JAMA Neurology paper the same year. Both studies were methodologically sound, but both were observational.

Second, the mechanistic hypothesis. Type 2 diabetes and obesity are well-established risk factors for Alzheimer disease. Brain insulin resistance, chronic neuroinflammation, and cerebrovascular damage overlap with the pathophysiological signature of Alzheimer pathology. A drug that addresses all three seemed biologically natural. Preclinical studies showed partial blood-brain barrier passage of GLP-1 agonists, GLP-1 receptors in the hippocampus, and anti-inflammatory effects in microglia.

Third, the success in other indications. SELECT for cardiovascular events, FLOW for kidney protection, STEP-HFpEF for heart failure. Semaglutide had an almost unbroken streak of positive Phase 3 results in new indications between 2023 and 2025. Expectations of another multiplier effect in Alzheimer were correspondingly high.

EVOKE and EVOKE+ Trial Design

Novo Nordisk launched two parallel Phase 3 programmes with nearly identical design in 2021. Together, EVOKE (NCT04777396) and EVOKE+ (NCT04777409) enrolled approximately 3,840 patients.

The choice of the oral format was strategic. Injectable semaglutide has stronger weight loss data, but the oral version allows continuous daily exposure, which theoretically could be more favourable for neuroprotective effects. The population was deliberately restricted to early disease stage, an approach that has become standard after years of disappointment with anti-amyloid therapies in advanced disease.

The Results: No Significant Benefit

In late 2025, Novo Nordisk published topline results in a press release. Science covered the context extensively (Science, 2025).

The primary endpoint, change on the CDR-SB scale at week 104, did not differ significantly between oral semaglutide and placebo. CDR-SB scores increased in both arms at the expected rate; the difference between groups was too small to support clinical relevance. Secondary cognitive endpoints such as ADAS-Cog and functional endpoints such as ADCS-ADL also showed no clinically meaningful separation.

The biomarker data were interesting. Individual signals on CSF p-Tau, amyloid ratio, and volumetric MRI parameters suggested biological activity, though without translation into cognitive or functional benefit. This is a familiar pattern in the history of Alzheimer trials: biomarker shifts without clinical consequence. The programme was terminated, and semaglutide will not be developed further for Alzheimer disease.

Why It Likely Failed

Several factors could explain the result. None is conclusive alone. Together they sketch a more consistent picture.

The blood-brain barrier passage of semaglutide is minimal. PET studies with radiolabelled semaglutide show signal in circumventricular organs and select brainstem regions, but the penetrance into cortical and hippocampal tissue, where Alzheimer pathology concentrates, is low. Any neuroprotective effect must therefore be mediated either via indirect routes (vascular, inflammatory, metabolic) or the minimal direct exposure must suffice.

The dosing was optimised for metabolic effect, not a hypothetical CNS action. 14 mg oral semaglutide corresponds in systemic exposure to roughly the lower subcutaneous dose. Whether higher doses would have shifted CNS exposure clinically relevantly is an open question, but the tolerance corridor for semaglutide is constrained by gastrointestinal side effects.

The disease stage may still be too late. Amyloid pathology begins decades before clinical symptoms. By the time patients receive the MCI diagnosis, the neuronal network is already structurally altered. Mechanistic interventions that rely on metabolic tailwinds likely require much earlier intervention, for which no practicable screening model currently exists.

The mechanism itself may have been misaligned. Insulin resistance and neuroinflammation are risk factors, but the causal chain from peripheral GLP-1 signalling to synaptic function preservation is multi-step and fragile. Each step can only suffer minimal efficiency loss for the end benefit to accumulate. EVOKE shows that in reality this is not enough.

Context: Other Failed GLP-1 and Metabolic Neuro Trials

EVOKE is not the first negative trial of this hypothesis. The list is surprisingly long and has been largely overlooked by the public.

ELAD with liraglutide. The ELAD trial (Evaluating Liraglutide in Alzheimer's Disease) randomised 204 patients with mild Alzheimer dementia to liraglutide or placebo over 12 months. Published in 2024 in Alzheimer's & Dementia, ELAD showed no significant difference in the primary cognitive endpoint. Some secondary MRI markers suggested less volume loss, but clinical translation was absent. ELAD was the first larger GLP-1 cognition Phase 2 trial with negative result and should have been a warning signal.

TOMMORROW with pioglitazone. The TOMMORROW trial tested pioglitazone, a PPAR-γ agonist with robust metabolic effects, in patients with genetically elevated Alzheimer risk. The trial was terminated early for futility in 2018. Pioglitazone had shown protective signals in observational data that did not replicate in the RCT.

Intranasal insulin. The SNIFF trial with intranasal insulin in MCI and early Alzheimer disease ran over eight years and published a disappointing primary result in 2020. Post-hoc analyses showed signals in subgroups, but the primary endpoint was not met.

The pattern is clear. Metabolic interventions with plausible mechanisms have repeatedly missed primary endpoints in Alzheimer RCTs. The epidemiological signals that fed each of these hypotheses did not withstand randomised testing.

What the Evidence Actually Supports

The GLP-1 Alzheimer hypothesis has failed. What remains is the evidence that treating metabolic risk factors across the lifespan lowers Alzheimer risk. That is a different claim.

Lancet commissions on modifiable dementia risk factors list type 2 diabetes, midlife obesity, and hypertension as established risk factors. Aggressive treatment of these risk factors reduces Alzheimer incidence, as FINGER, LookAHEAD, and other lifestyle intervention trials have shown. Semaglutide, as a potent antidiabetic and weight-reducing agent, lowers these risk factors and can therefore plausibly be part of a prevention strategy.

The step from prevention to treatment, however, is qualitatively different. Someone with poorly controlled diabetes and progressive obesity for twenty years accumulates cognitive risk over decades. A therapeutic intervention that straightens this curve is not equivalent to an intervention in a brain that is already showing neurodegenerative changes. EVOKE tested the latter and showed that semaglutide cannot make that transition.

Implications for Observational Studies

The most striking learning from EVOKE is the discrepancy between observational evidence and RCT result. Wang 2024 and Tang 2024 had shown clear risk reductions. EVOKE shows no therapeutic effect. How does this reconcile?

The answer lies in confounding. GLP-1 therapy users differ systematically from non-users, even after statistical adjustment. They are likely more health-conscious, more adherent to other medications, have better access to care, and better eating behaviour. These are exactly the variables that also influence Alzheimer risk. Part of the observed protective effect in claims data was likely a signal for the patient type, not for the drug.

This trap has been set before: for hormone replacement therapy and cardiovascular prevention, for vitamin E in dementia, and for numerous other hypotheses. Epidemiology plus plausible mechanism often yields a hypothesis, but almost never a causal conclusion. Only RCTs clarify the effect. EVOKE is a textbook case of this discrepancy.

Parallel: Tesamorelin Baker 2012 - What a Positive Cognition Trial Looks Like

For contrast, it is worth looking at a peptide that actually showed cognitive benefit in a controlled study. Tesamorelin is a GHRH analogue, mechanistically entirely different from GLP-1.

Baker et al. published a randomised study with tesamorelin in older adults with MCI and cognitively healthy controls in Archives of Neurology in 2012. Participants received 20 weeks of subcutaneous tesamorelin or placebo. The primary endpoint was change in cognitive test batteries, with secondary endpoints including IGF-1, body composition, and MR spectroscopy. Results showed significant improvements in executive function and verbal memory with tesamorelin versus placebo, with effect sizes between 0.03 and 0.002 on standardised scales. IGF-1 rose by 117 percent. A follow-up study by Friedman et al. in 2013 in JAMA Neurology confirmed MR spectroscopic changes consistent with GABAergic modulation.

The Baker study is small (n approximately 137), not conclusively replicated, and the effect sizes are moderate. But it achieves what EVOKE did not: a positive cognitive signal on a prespecified endpoint. Crucially, the mechanism is different. Tesamorelin acts via the GH-IGF-1 axis and has direct effects on hippocampal growth and synaptogenesis in preclinical models. The causal chain is shorter and more robust, and the intervention was applied in patients without amyloid pathology.

This does not mean tesamorelin is an Alzheimer drug. It means what a positive peptide cognition trial looks like and what questions it still leaves open. For researchers interested in the stabilised GHRH analogue used in Baker 2012, see Tesamorelin.

What This Means for the Peptide Research Community

The GLP-1 Alzheimer hypothesis is dead. That is good news, not because the result was negative, but because it was tested cleanly. For the European peptide research community, this means two things.

First, honesty in peptide research is a genuine differentiator. Many vendors in this market work with citable but uncontextualised references, with observational data sold as evidence, with preclinical results presented as clinically relevant. The RCT map is more complicated. Those who differentiate demonstrate more expertise, not less.

Second, the semaglutide dossier remains impressive despite EVOKE. The cardiovascular, renal, and heart failure data are robust. What disappears is the Alzheimer hope, which was speculative from the start. For a comprehensive overview of the semaglutide evidence base, see Semaglutide Science 2026.

EVOKE is not a stain on semaglutide. It is a boundary, clearly drawn by a well-designed RCT. That is exactly how science works. For researchers working with the currently most potent triple agonist (GLP-1/GIP/glucagon) backed by the TRIUMPH-4 Phase 3 data, see Retatrutide.