A Benefit Nobody Observed

A study in a leading cardiology journal reports that GLP-1 therapy could cut ten-year cardiovascular risk by 22% in people who have never had heart disease.^[1] Nobody in it took the drug. Researchers applied the risk-factor changes from a completed trial to 200,012 people in health surveys and ran a risk equation — and the assumption underneath that arithmetic is one the source trial partly contradicts.

What was actually done

First, a model was fitted in 610,789 people without cardiovascular disease, predicting ten-year cardiovascular events and death from five inputs: BMI, HbA1c, systolic blood pressure, C-reactive protein, and non-HDL cholesterol.

Then that model was applied to 200,012 people from two health surveys. For each, the researchers adjusted the five inputs by the amounts semaglutide changed them in the SELECT trial, and recalculated the predicted risk. The difference between the two predictions is the reported benefit.

Absolute reductions were larger in men than women, 3.14 against 2.7 percentage points, with similar relative effects. Both shrank in people who were not obese, in people at lower baseline risk, and further still once imperfect adherence was assumed.

The assumption the source trial contradicts

SELECT itself examined that assumption, in a prespecified analysis of its own 17,604 participants, and found the cardiovascular benefit largely independent of weight loss. There was no linear relationship between weight lost by week 20 and later events, and only about a third of the benefit was mediated through waist circumference. We covered it in losing weight predicted more heart attacks.

The model assumes the benefit travels by a route the trial found it largely does not take.

That does not make the projection too high. It could as easily be too low — if most of the benefit arrives by some pathway the five factors do not capture, a model built only on those factors would miss it. The honest statement is that the calculation is not measuring the thing it is presented as measuring, and its direction of error is unknown.

There is a second, more ordinary problem. SELECT enrolled people who already had cardiovascular disease. This study applies its risk-factor changes to people who do not. Whether a drug moves risk factors by the same amount in a lower-risk population is an assumption, and whether the same movement buys the same protection is another.

The number worth carrying, in a usable form

Set the mechanism question aside for a moment and take the projection at face value. A 2.99 percentage point absolute reduction over ten years corresponds to a number needed to treat of roughly 33 — treat 33 people for a decade to prevent one cardiovascular event.

That is a real and defensible figure for a preventive therapy, and it is also ten years of an expensive injection with the side effects documented throughout this register. Whether it is worth it is a genuine question involving cost, tolerability and preference, and no risk model answers it. The general case for reading absolute numbers rather than relative ones is in how many people to treat to help one.

What the authors actually claim

The value of work like this is in deciding what to test, not in deciding what to prescribe. It says a trial of these drugs for primary prevention would be worth running, which nobody has done, and it says roughly how large an effect such a trial would need to be powered to detect.