Post Hoc SELECT Trial Analysis Finds Semaglutide Slowed a Blood-Based Dementia Risk Score

Alzheimer's & Dementia: Diagnosis, Assessment & Disease Monitoring, a journal of the Alzheimer's Association, published a post hoc analysis on August 8, 2026 examining whether semaglutide affects a blood-based proteomic score that predicts future dementia risk. The study was led by Marti Jimenez-Mausbach and Jan Christian Refsgaard of Novo Nordisk A/S in Denmark, with co-authors Betty M. Tijms of the Alzheimer Center Amsterdam at Amsterdam UMC and Clare Paterson of Illumina in Boulder, Colorado.

The analysis drew on the SELECT trial (ClinicalTrials.gov identifier NCT03574597), a completed phase 3, randomized, double-blind, placebo-controlled cardiovascular outcomes trial that enrolled 17,604 adults with overweight or obesity and established cardiovascular disease but no diabetes, comparing once-weekly semaglutide 2.4 mg with placebo. This post hoc analysis was restricted to the subset of participants aged 65 or older who had non-fasted serum samples available at both baseline and week 104, totaling 2,970 people: 1,382 assigned to semaglutide and 1,588 assigned to placebo. Baseline characteristics, including mean age of about 69.8 years, were balanced between the two groups.

Samples were analyzed using the Dementia SomaSignal Test (dSST), an aptamer-based 25-protein blood score that was previously validated in external cohorts, including the Atherosclerosis Risk in Communities (ARIC) study, to predict an individual's 5-year and 20-year probability of all-cause dementia, and to sort people into ordered risk categories of low, medium-low, medium-high, or high. The prespecified outcomes for this analysis were the change in predicted 5-year and 20-year dementia risk from baseline to week 104, and any shift in dSST risk category over that period.

Compared with placebo, semaglutide was associated with a 2.5-fold smaller increase in predicted 5-year dementia risk, corresponding to a 26.0% lower predicted event rate (odds ratio 0.74, 95% confidence interval 0.65-0.85), and a 1.67-fold smaller increase in predicted 20-year risk, an 8.8% lower predicted event rate (odds ratio 0.91, 95% CI 0.88-0.94). Ordinal regression on the risk-category shift found semaglutide was associated with 36% lower odds of moving into a higher dSST risk category at week 104 compared with placebo (beta -0.44, standard error 0.078, p<0.001; equivalent odds ratio 0.64, 95% CI 0.55-0.75).

In a sensitivity analysis that adjusted for change in body mass index between baseline and week 104, the treatment effect on 20-year risk attenuated by approximately 28%, which the authors interpret as meaning about 72% of the semaglutide-associated difference in predicted 20-year dementia risk persisted after accounting for weight change, a signal they say is consistent with a partial contribution from mechanisms beyond weight loss. As an exploratory, non-adjudicated analysis, the authors also examined baseline dSST risk categories among the 18 trial participants aged 65 or older who experienced a dementia-related adverse event during SELECT, identified using a narrow-scope MedDRA standardized query rather than clinical diagnosis. None of these 18 participants were classified as low risk at baseline; 7 (39%) were medium-high and 11 (61%) were high risk.

The authors describe several limitations. The analysis is post hoc and exploratory, without prespecified correction for multiple statistical comparisons across the proteomic endpoints. It relies on a predicted risk score rather than adjudicated dementia diagnoses or systematic cognitive assessment within SELECT, and the dementia-related adverse events used in the enrichment analysis were captured through MedDRA reporting, which the authors say likely under-ascertains true cases. The dSST was developed and validated in EDTA plasma cohorts, while this analysis measured serum, introducing a possible matrix-related difference the authors flag as a limitation. The study was funded by Novo Nordisk A/S, which designed the study in collaboration with academic co-authors and contributed to data collection, analysis, and manuscript preparation. Jimenez-Mausbach and Refsgaard are Novo Nordisk employees and minor shareholders; Paterson is an Illumina employee and shareholder; Tijms has received consulting and lecture fees from Novo Nordisk, Roche, and Sanofi, paid to her institution.

Semaglutide is already approved by the FDA, the European Medicines Agency, and Australia's Therapeutic Goods Administration, marketed as Ozempic and Rybelsus for type 2 diabetes and Wegovy for chronic weight management. This post hoc analysis does not change that regulatory status. It does not establish that semaglutide prevents, slows, or treats dementia or Alzheimer's disease in humans, since the outcome measured is a predicted risk score rather than a clinical diagnosis, and the authors themselves frame the findings as hypothesis-generating rather than conclusive. Nothing in this analysis constitutes treatment or dosing guidance; cognitive and dementia risk should be assessed by a qualified clinician.