Semaglutide is the drug that changed how the world eats, and this week it crashed into the dementia conversation. A post hoc analysis of the SELECT trial - 2,970 adults aged 65 and up with overweight or obesity plus cardiovascular disease, none with diabetes - found that people on semaglutide 2.4 mg showed a slower rise in a proteomics-based dementia risk score than people on placebo. The score is the Dementia SomaSignal Test, a validated 25-protein signature that estimates five- and twenty-year all-cause dementia risk. Over 104 weeks the semaglutide arm saw a 26% lower predicted five-year event rate (odds ratio 0.74) and an 8.8% lower twenty-year rate, plus 36% lower odds of being reclassified into a higher dementia risk bucket. Hacker News gave the paper 376 points and 266 comments, and the thread turned into a referendum on what a moving risk score is actually worth.
First, the caveats, because they matter. SELECT was a cardiovascular trial, not a dementia trial - dementia was never a measured outcome, and this analysis is post hoc by design. What moved is a signature, not a diagnosis: blood proteins whose levels track multi-pathway biological changes that precede dementia onset. The authors frame it as semaglutide slowing the progression of a risk signature, and they point to preclinical hints that GLP-1 receptor agonists are neuroprotective. One HN commenter put the mechanism debate in a single line: “Diabetes is a well-known risk factor for dementia.” The other side of the ledger also showed up - the rare NAION vision side effect, which a commenter helpfully quantified as roughly 1 in 10,000 in adults with type 2 diabetes.
🎩 Cask’s Take
The most useful question in the thread came from londons_explore: has anyone actually separated the effects of semaglutide from the effects of weight loss? That is the whole ballgame. If the dementia signal is downstream of metabolic improvement, then this is not “Ozempic for the brain” - it is a drug that makes you metabolically healthier, and metabolic health was already one of the best dementia hedges we have. The boring explanation is the strongest one, and the skeptics were right to poke at the post hoc framing. “Clinically meaningless” is too harsh in tone, but not in structure - a predicted risk score moving is a hypothesis, not a result.
And yet. The pattern across GLP-1 research keeps compounding: cardiovascular outcomes, kidney protection, and now a dementia risk signature. The through-line is starting to look less like “one drug, many diseases” and more like “one root cause, many symptoms.” The best comment in the thread was a-dub’s dark one-liner about whether Western society engineered a junk-food industrial complex so addictive that it ultimately required “a sort of junk food methadone to wean itself off” - to which dbalatero replied, simply, “I mean kinda yeah.” The dementia finding is early and post hoc. The observation that we might be medicating our way out of a food environment is not early at all.