Two analyses of the SELECT trial, published a day apart, reach the same uncomfortable conclusion from different directions. The first, in the European Heart Journal, tried to work out how much of semaglutide’s 20% reduction in heart attacks, strokes and cardiovascular deaths runs through weight loss and the other risk factors it improves, and could not pin it down. The second, in Diabetes Care, found that nothing a clinician can measure before treatment predicts who will lose the most weight. Both come from the Novo Nordisk-funded trial and list Novo Nordisk employees among their authors.
What the two papers measured
SELECT randomised 17,604 adults with established cardiovascular disease, a body mass index of 27 or more and no diabetes to weekly semaglutide (target dose 2.4 mg) or placebo. The mediation analysis took the changes each person had after 24 months in weight, waist circumference, high-sensitivity C-reactive protein (hsCRP, a marker of inflammation), HbA1c, cholesterol, blood pressure and two kidney measures, and asked how much of the drop in major adverse cardiovascular events each change could account for.
Semaglutide improved every one of those markers. But the estimates of how much of the benefit each explained were imprecise to the point of being uninformative. Waist circumference had the largest point estimate at 64.0%, with a 95% confidence interval of 27.5% to 179.6% — an interval that runs past 100%, which is not a real-world possibility and signals an unstable estimate. hsCRP came in at 42.1% (17.9% to 110.5%). Body weight on its own was 19.5%, with an interval from −33.0% to 110.7%, meaning the data are compatible with weight loss explaining none of the benefit or all of it. All the markers together explained an estimated 31.4% (−30.1% to 143.6%). The authors themselves flag the weight and waist estimates as unreliable, because the link between weight change and heart events appeared to differ between the two arms.
The companion paper looked at the 7,594 people who stayed on semaglutide for at least a year — 86.3% of those randomised to it. Their median peak weight loss was 12.5% (interquartile range 7.9% to 17.9%), reached after a median of 19 months. A model built from demographics, medical history, blood tests and clinical measurements explained under 10% of the variation in that peak, and sex did almost all of that work: women reached a median peak of 16.6% against 12.4% for men.
What it changes
A common reading of SELECT has been that semaglutide protects the heart because people lose weight. These papers do not refute that, but they do not support the simple version of it either. Weight change could not be shown to carry most of the benefit, and the point estimate for inflammation was at least as large as the one for weight. For someone weighing the drug, the practical upshot is that the cardiovascular result in the trial was not confined to the people who lost the most weight — though this analysis cannot say how much of it someone with a small weight response would keep.
The predictor paper matters for a different reason. It means a doctor cannot look at a new patient and forecast their response; the spread from a quarter of people losing under 8% to a quarter losing over 18% is mostly unexplained. The authors point to unmeasured biological, behavioural or genetic differences. Maximum dose tracked with results — a median peak of 6.0% among people whose highest dose was 0.24 mg against 13.3% at 2.4 mg — but dose was not randomised. People who stayed on a low dose may have done so because of side effects or early discontinuation, so this is not evidence of what a dose change would do for an individual.
What is not yet known
Mediation analysis is an attempt to infer mechanism from a trial that was not designed to test one, and it rests on assumptions that cannot be checked from the data. The confidence intervals here are wide enough that the honest summary is “we cannot tell”, which is what the authors conclude. SELECT’s population was 72% male, averaged 62 years old and already had heart disease, so neither result transfers directly to younger people without cardiovascular disease or to people with diabetes. The predictor analysis covers only people who stayed on treatment for a year, leaving out those who stopped early.
Sources: Colhoun HM, Lincoff AM, Ryan D, Lingvay I, Kahn SE, et al. “Semaglutide and cardiovascular risk reduction: a mediation analysis of the SELECT trial.” European Heart Journal, published online 23 September 2026. doi:10.1093/eurheartj/ehag524. Lingvay I, Kushner RF, Kaul U, Kokkinos A, Lincoff AM, et al. “Predictors of peak body weight loss with semaglutide in the SELECT trial.” Diabetes Care, published online 24 September 2026. doi:10.2337/dc26-0784.
