Clinical Proteomics: Semaglutide Slows Dementia Risk
Dementia trials stall at the readout. Cognitive decline is slow to measure over the decade a cardiometabolic drug needs to prove itself. A post hoc analysis of the SELECT phase 3 trial took another route, using clinical proteomics to ask whether semaglutide 2.4 mg bends a validated 25-protein blood risk score across 104 weeks in 2,970 non-diabetic adults aged 65 and older. It did.
Key Takeaways
Semaglutide 2.4 mg slowed the two-year rise of a validated 25-protein blood dementia risk score in 2,970 older adults.
Clinical proteomics can supply an interim readout when the real endpoint, adjudicated dementia, is a decade away.
Adjusting for weight change absorbed only 28% of the effect, so adiposity reduction is not the mechanism.
Predicted risk is not diagnosed disease, so the result stays hypothesis-generating until cognitive outcomes confirm it.
Clinical Proteomics as a Two-Year Trial Endpoint
The Dementia SomaSignal Test is a 25-protein machine learning score read out on the SomaScan aptamer platform, which quantifies roughly 7,000 circulating proteins per sample. Duggan and colleagues (2025) validated it against 5-year and 20-year dementia risk. SELECT applied that algorithm without retraining, and the distinction carries the study: a locked model read at two timepoints is an endpoint, while a model refit at week 104 only describes the data. One wrinkle deserves attention. The score was built in EDTA plasma and SELECT banked serum, so the authors ran a concordance check first.
Key Findings
Five-year predicted risk rose 2.5-fold less on semaglutide than placebo (OR 0.74, 95% CI 0.65 to 0.85), a 26.0% lower modeled event rate.
The 20-year projection moved less, a 1.67-fold smaller increase (OR 0.91, 95% CI 0.88 to 0.94).
Treated participants had 36% lower odds of shifting into a higher risk category (OR 0.64; P < 0.001).
Weight loss explained part of it, not most. Adjusting for BMI change moved the coefficient from -0.092 to -0.066.
Serum held up against plasma, median analyte concordance 0.808 and 0.959 for the score itself.
Figure 1. Design of the SELECT proteomics substudy. Non-diabetic adults aged 65 and older with overweight or obesity and established cardiovascular disease received weekly subcutaneous semaglutide 2.4 mg or placebo, with serum drawn at weeks 0 and 104. Adapted from Jimenez-Mausbach M. et al. (2026). Alzheimer's & Dementia.
Why the Serum Question Is Not Trivial
Coagulation releases platelet and clotting proteins that plasma never sees. A plasma-calibrated score can drift on the absolute scale yet stay usable between arms, because the bias falls on both groups equally. Quoting one participant's 20-year probability would still be a mistake.
Predicted Risk Versus Observed Disease
No dementia was adjudicated here. A projection moved, and projections inherit their training cohorts; ARIC, BLSA, and NILS-LSA were community samples that included diabetics. Read it as a strong pharmacodynamic signal and a weak clinical one. The same gap turns up in omics models for early Parkinson's.
Implications for Drug-Discovery Pipelines
A pre-locked protein score can carry an interim decision at 24 months, provided the algorithm is frozen before the first sample is assayed. We hit the same constraint in client work at Dalton, where the hard question is rarely whether a signature exists but whether it survives a matrix switch or a second collection site, which is why bridging samples belong in the plasma and serum proteomics design rather than the analysis. Anyone scoping a longitudinal multi-omics biomarker discovery program should budget that concordance work up front.
Frequently Asked Questions
What is clinical proteomics used for?
Clinical proteomics measures thousands of proteins in patient blood or tissue to find markers of disease, risk, or treatment response. In drug development it supports patient stratification and pharmacodynamic readouts. Platforms range from aptamer panels to MS-based proteomics.
Can clinical proteomics show whether a drug lowers dementia risk?
It can show movement in a validated protein risk score, as SELECT reported for semaglutide. That is a pharmacodynamic signal, not proof of prevention, which needs adjudicated cognitive outcomes over longer follow-up.
How many samples does a proteomics biomarker study need?
It depends on effect size and whether the score is already locked. Applying a validated signature inside a randomized trial worked with roughly 1,500 participants per arm. De novo discovery needs more, plus a replication set.
Conclusion
Two years of semaglutide 2.4 mg shifts a locked proteomic dementia risk score, and weight loss doesn't account for most of it. Whether that means fewer diagnoses is unproven. Sponsors should pair such a score with a real cognitive endpoint in one protocol, so the interim signal has something to be checked against.
Related Reading
See how we run these analyses in one lab: Dalton's multi-omics CRO services.
Citation
Jimenez-Mausbach, M., Tijms, B. M., Paterson, C., & Refsgaard, J. C. (2026). Semaglutide attenuates a proteomics-based dementia risk signature in older adults with overweight or obesity and cardiovascular disease without diabetes: A post hoc analysis of the SELECT phase 3 trial. Alzheimer's & Dementia: Diagnosis, Assessment & Disease Monitoring, 18(3), e70432. https://doi.org/10.1002/dad2.70432
Note
This blog post summarizes findings from the above-cited research. Figures are adapted from the original publication. For full details, please refer to the source article.
By Seungjun Yeo, CEO at Dalton Bioanalytics. Specializing in multi-omics mass spectrometry for drug discovery and biomarker research.