Three of today's findings turn on the composition of a composite endpoint, and this is the single most transferable skill in reading a trial outside your own field.
A composite counts several different events as one. It is used because it gives more events and therefore a smaller trial. The problem is that the events are almost never equally important, and the commonest ones drive the result. A composite of death, myocardial infarction and hospitalisation will be dominated by hospitalisation, because hospitalisation is ten times more frequent — so a trial can report a difference in its primary endpoint that is entirely a difference in admission thresholds.
Three questions cover it. What are the components? Which one supplied most of the events? And did the direction hold for the component you actually care about? If the answer to the third is no, the headline result is weaker than it looks.
The reverse trap also exists. A composite that mixes a fatal event with the event that causes it will misbehave when one arm prevents the death: the survivor lives to have the non-fatal event counted. Any trial reporting both sudden death and arrhythmia in one endpoint needs reading with that in mind.
The rule of thumb: read the components before the hazard ratio, and if the paper does not report them separately, treat the headline with suspicion.
- Ask which component supplied most of the events — that component is what the trial actually measured
- Check whether the direction of effect held for mortality, or only for softer components
- Be sceptical of composites containing hospitalisation when trials ran across different health systems
- Watch for composites that mix a fatal event with its non-fatal precursor; preventing the death inflates the precursor
- A noninferiority margin is a judgement, not a fact — look at what width the investigators allowed themselves
Read the rest in the app
You have read your two free briefings this month. The app carries all 27 specialties, every morning, free — and this finding is waiting in it.
Scan to keep reading on your phone. No account needed to start.
Tomorrow morning, before your first patient
One edition a day for top clinical updates, written by the desk, every claim tied to its paper. Six minutes.
Get the app — free