Composite outcomes
Why they are used
More events means more power, so a smaller or shorter trial. They also sidestep multiplicity by nominating one primary outcome instead of several, and they can capture a treatment whose benefit spreads across related events rather than concentrating in one.
The three conditions
For a composite to mean anything, its components should be of comparable importance to patients, occur with comparable frequency, and be affected by the treatment in the same direction and to a similar degree. In practice all three are frequently violated, and the violation is visible in the component table if one is published.
What to report
- The effect on each component separately, with counts, alongside the composite.
- How many patients contributed each type of first event, which shows what is driving the result.
- Whether any component moved in the opposite direction, which a favourable composite can conceal.
- The definition of each component, at protocol level, including how it was adjudicated.
- Total events, not only first events, where recurrent events matter clinically.
Better constructions
Hierarchical approaches rank the components by importance and compare patients pairwise on the most serious event first, which is what the win ratio does. Recurrent-event models use all events rather than the first. Both keep the power advantage while restoring the clinical ordering that a plain composite discards.
Why Mallard treats it as an estimand question
What the composite counts IS the outcome the study is about, so it belongs in the estimand rather than in the analysis section, and the component structure must be stated there.
What to watch out for
A composite is usually driven by its most frequent component, which is usually its least serious one, so a headline benefit in a death-or-hospitalisation composite is often entirely hospitalisation. Standard time-to-first-event analysis compounds this by counting whichever event happens first, so a hospitalisation at week 2 outranks a death at week 30. A composite is only interpretable if the components are of similar importance, occur at similar frequency, and move in the same direction under treatment.
Clinical example
A trial reports a reduction in death, myocardial infarction or revascularisation. The component table shows revascularisation, the least serious and most operator-dependent component, accounts for nearly all of the difference, and mortality is unchanged.
- Composite endpoints (opens in a new tab) · J Clin Epidemiol · doi 10.1016/j.jclinepi.2020.07.017
- The win ratio: a new approach to the analysis of composite endpoints in clinical trials based on clinical priorities (opens in a new tab) · Eur Heart J · doi 10.1093/eurheartj/ehr352