On August 9, a team of researchers published a post hoc analysis of the SELECT trial in Alzheimer’s & Dementia showing that 104 weeks of semaglutide at 2.4 mg changed a blood-protein signature that a statistical model uses to predict 5- and 20-year dementia risk. The study included 2,970 adults aged 65 and older with cardiovascular disease but no diabetes. The finding was real, the analysis was competent, and the headlines wrote themselves: “Semaglutide linked to lower predicted dementia risk.”

Notice the word “predicted.” It’s doing more work than any other word in that sentence.

The study did not measure whether anyone actually developed dementia. It measured whether a proteomic scoring system — a research tool called the dSST — shifted in a favorable direction. The authors themselves note that “the mechanisms responsible were not directly established.” What we have is a change in a predictive model’s inputs, not a change in anyone’s cognitive outcome.

This is not a criticism of the researchers. Post hoc analyses of big trials are legitimate science. But they are also cheap science — the data already exists, the participants already enrolled, the drug already administered. The marginal cost of running another statistical model on the SELECT dataset is a few weeks of a biostatistician’s time and a journal submission. The marginal benefit is a headline that makes semaglutide look even more like a miracle drug than it already does.

And that’s the story here. Not semaglutide and dementia. Semaglutide and the sociology of blockbuster drugs.

The Biomarker Graveyard Is Full

We have been here before. The history of medicine is littered with drugs that changed a biomarker and failed to change an outcome. Torcetrapib raised HDL cholesterol dramatically and increased mortality. Aducanumab cleared amyloid plaques and did not convincingly slow cognitive decline. The list goes on. A biomarker is a proxy, and proxies lie.

The dSST is a particularly ambitious proxy. It uses serum proteins to predict dementia risk over 5 and 20 years. It is a research tool, not a clinical endpoint. No regulatory agency has approved it as a surrogate for dementia prevention. And yet the framing of this study — “semaglutide attenuates a proteomics-based dementia risk signature” — invites readers to treat a shift in the signature as if it were a shift in dementia itself.

The gap between “predicted risk” and “actual dementia” is where drugs go to die. Semaglutide may cross that gap. It may not. The point is that we don’t know, and the headline doesn’t tell us.

The Drug We’ve Decided to Like

What’s actually happening here is that semaglutide has become a cultural phenomenon, and cultural phenomena attract favorable science the way a magnet attracts iron filings. The SELECT trial was designed to answer a cardiovascular question. Now it’s being mined for dementia signals. What’s next — semaglutide and depression? Semaglutide and longevity? Semaglutide and marital satisfaction?

A researcher I spoke with at a cardiology conference last month put it dryly: “Every blockbuster eventually gets a post hoc analysis showing it prevents something it was never designed to prevent. It’s the pharmaceutical equivalent of a victory lap.”

The danger isn’t that semaglutide is bad. It’s that we’re losing the ability to be skeptical of a drug we’ve decided to like. The same people who would scrutinize a statin’s dementia claim with a fine-tooth comb will share this headline without reading past the abstract. The drug has become a lifestyle, and lifestyles don’t get audited.

Who Has to Update Their Priors

If I’m right, the people who need to update their priors are not the researchers — they know what a post hoc analysis is. It’s the millions of people taking semaglutide for weight loss who will now feel extra virtuous about it, as if their prescription is also a cognitive insurance policy. It’s the journalists who will strip the word “predicted” from the headline within a month. And it’s the broader public, which has been trained by a decade of breathless drug coverage to expect every new finding to be a breakthrough.

Semaglutide is a remarkable drug. It helps people lose weight, it improves cardiovascular outcomes, and it may — may — do something for dementia. But the gap between “may” and “does” is the entire ballgame. And the word “predicted” is the only thing standing between us and forgetting that.

Sources