Composite endpoints: when a trial bundles outcomes together

When you read that a drug cut the risk of the primary endpoint by a fifth, it is worth checking what the endpoint actually was. Many trials, especially in heart disease, do not measure a single outcome. They use a composite endpoint, also called a combined endpoint, that counts a patient as having an event if any one of several things happens, say death, or a heart attack, or a hospitalisation.

There is a good reason to do this. Serious events like death are, thankfully, fairly rare, so a trial measuring death alone might need enormous numbers of people or many years to see a difference. Bundling several outcomes together produces more events, which lets the trial be smaller or shorter and still detect an effect. Used well, a composite can be a sensible, efficient design.

The catch is that a composite treats very different events as if they were equal. A death and an overnight hospital stay both count as one event, but they are not remotely the same to a patient. So a drug can move the composite mostly by reducing the mildest, most common component, hospitalisations, say, while barely touching death at all. The headline number then sounds more impressive than what the drug really does for the outcomes people care about most.

So when a trial reports a composite endpoint, look inside it. Ask what the components were, and which ones actually drove the result. A composite is trustworthy when its parts are of similar importance and the effect is spread across them. It is misleading when a serious-sounding result turns out to rest almost entirely on the least serious component. The bundle can hide as much as it reveals.