Semaglutide changed a dementia-risk blood test—here’s what that does and doesn’t mean

What the source reports

A post-hoc analysis of the SELECT cardiovascular trial examined whether semaglutide changed a blood-protein pattern associated with the probability of a future dementia diagnosis. The work was presented at the 2026 American Academy of Neurology meeting.

Researchers analysed non-fasting serum samples from 2,970 SELECT participants aged 65 or older. The participants had overweight or obesity and established cardiovascular disease, but no diabetes. Samples taken at the start, week 20 and week 104 were assessed with the Dementia SomaSignal Test, known as dSST.

After 104 weeks, the semaglutide group had a more favourable change in the protein-based prediction than the placebo group. This is a biomarker finding—not evidence that fewer participants developed dementia.

What the blood test measures

The dSST combines measurements from 25 blood proteins using a machine-learning model. It estimates the probability of receiving an all-cause dementia diagnosis over five- and 20-year periods.

The model was developed using 11,277 participants in the Atherosclerosis Risk in Communities study and evaluated in other cohorts. Its validation study also examined associations with cognitive decline and brain-volume changes.

A prediction model can identify a statistical pattern associated with later diagnosis. It is not a direct measurement of memory, thinking, brain pathology or whether dementia will occur in a particular person. Changing its score does not automatically mean that the underlying clinical outcome has changed.

What changed after two years

The investigators reported that the increase in predicted five-year dementia risk was 2.5 times smaller with semaglutide than with placebo at week 104. Their model translated this into a 26.0% lower predicted five-year event rate, with an odds ratio of 0.74 and a 95% confidence interval from 0.64 to 0.85.

For the 20-year prediction, the increase was 1.7 times smaller with semaglutide. This corresponded to an 8.8% lower predicted event rate, with an odds ratio of 0.91 and a 95% confidence interval from 0.88 to 0.94.

A separate category analysis found 36% lower odds of being placed in a higher predicted-risk group with semaglutide. All three results come from the same protein-based model and should not be counted as three independent clinical benefits.

What the analysis did not show

SELECT was designed to test cardiovascular outcomes, not dementia prevention. The proteomic work was a later analysis of a subgroup with stored serum samples, rather than a prespecified trial whose primary outcome was diagnosed dementia or cognitive decline.

Only 18 participants in the analysed older subgroup had a dementia-related adverse event recorded during SELECT. The poster reported no statistically significant difference in those events between semaglutide and placebo.

The researchers noted that dementia events may have been underreported because the trial collected them through adverse-event reporting rather than a dedicated programme of cognitive testing and diagnostic assessment. The analysis therefore cannot tell us whether semaglutide prevented dementia, delayed symptoms or maintained memory.

Why SELECT still provides useful evidence

SELECT randomly assigned 17,604 adults with overweight or obesity and established cardiovascular disease, but without diabetes, to weekly semaglutide 2.4 mg or placebo. Randomisation makes the treatment groups more comparable than an ordinary observational study.

The main trial found fewer major cardiovascular events with semaglutide. That matters to dementia research because vascular and metabolic health can influence future cognitive risk, although a cardiovascular benefit does not prove a dementia benefit.

The proteomic subgroup is reasonably large and includes measurements at several time points. However, restricting the analysis to participants aged 65 or older with available samples can reduce the balance created by the original randomisation, and the findings may not generalise beyond this cardiovascular-risk population.

The essential counterpoint: the Alzheimer’s trials were negative

The evoke and evoke+ Phase 3 trials directly tested oral semaglutide in 3,808 people with amyloid-confirmed early symptomatic Alzheimer’s disease. Participants received semaglutide or placebo for up to 156 weeks, with the primary analysis at 104 weeks.

Semaglutide did not significantly slow worsening on the Clinical Dementia Rating–Sum of Boxes measure in either trial. Those negative clinical results are stronger evidence about treating established early Alzheimer’s disease than a protein-risk model can provide.

The SELECT biomarker finding does not reverse the evoke results. The studies involved different formulations, doses, populations and questions: one explored predicted future all-cause dementia risk in people without diabetes, while the others tested clinical progression in people who already had Alzheimer’s disease.

Could prevention and treatment produce different results?

It is biologically possible for an intervention to influence risk factors before symptoms develop yet fail to slow an established neurodegenerative disease. Cardiovascular health, inflammation, glucose regulation and body weight could all affect protein patterns connected with long-term risk.

That remains a hypothesis here. The SELECT analysis did not establish which biological changes drove the score, whether they reflect brain-specific effects, or whether the difference was independent of weight loss and cardiovascular improvement.

A prevention claim would require a long, prospectively designed trial that repeatedly measures cognition and uses independently assessed dementia diagnoses as clinical outcomes. Biomarkers can help design such a trial, but they cannot replace it.

Funding and publication context

SELECT was funded by Novo Nordisk, the manufacturer of semaglutide. Five of the seven authors listed on the proteomic poster were affiliated with Novo Nordisk; another was affiliated with Standard BioTools, which offers SomaSignal testing, and one with Amsterdam University Medical Center.

The current report is a conference presentation hosted on Novo Nordisk’s scientific website rather than a full peer-reviewed article describing the analysis in detail. Conference findings can be valuable and timely, but methods and conclusions may change during journal review.

Company funding and author affiliations do not make a result incorrect. They are important context, particularly when a surrogate biomarker could be presented more strongly than the clinical evidence permits.

What remains uncertain

We do not know whether the change in the dSST score will translate into fewer dementia diagnoses, later onset, slower cognitive decline or no measurable clinical difference. We also do not know whether any effect would apply to people without cardiovascular disease, people with diabetes, younger adults or those taking a different semaglutide product or dose.

The model predicts all-cause dementia rather than one specific disease. Alzheimer’s disease, vascular dementia and other dementias have overlapping but different biological causes, so one composite score cannot show which pathway may have changed.

Independent replication, a full peer-reviewed report and clinical-outcome trials are needed before this signal can support a prevention claim.

What this means for the individual

The result is not a reason to start semaglutide for brain health or to change an existing prescription. Semaglutide has defined approved uses and risks; dementia prevention is not an established indication.

A calculated protein-risk score also should not be interpreted as an individual diagnosis or guarantee. Personal concerns about memory, thinking or dementia risk need appropriate clinical assessment rather than conclusions drawn from a research biomarker.

For someone already using a GLP-1 medicine, this study is useful context about ongoing research. It does not provide a new treatment target, dose or monitoring requirement.

Bottom line

Semaglutide produced a more favourable change in a validated 25-protein model of future dementia risk among 2,970 older SELECT participants. That is an intriguing biological signal worth testing further.

It did not demonstrate better memory, less cognitive decline or fewer dementia diagnoses. Combined with the negative evoke trials in established Alzheimer’s disease, the evidence supports a careful research question—not a claim that semaglutide prevents or treats dementia.