What went into the bundle?
Imagine a fictional trial whose endpoint includes either an urgent clinic visit or a hospitalization. If urgent visits are much more common, they can account for most of the events in the combined outcome.
Suppose the comparison group has 100 participants with a qualifying event and the intervention group has 80. That difference alone does not tell us whether hospitalizations fell. They might have remained similar while urgent visits decreased.
We also cannot add component counts blindly. One person may experience both events. A composite defined as “at least one event” counts that person once; a table of individual event types may include them twice.
A useful order for reading the results
- Find the exact definition of the composite, including the time window.
- Check whether components differ substantially in importance to patients.
- Look at component-specific estimates and their uncertainty.
- Check whether the analysis counts people, first events, or all events.
Cochrane's guidance on interpreting outcomes emphasizes understanding what was measured before drawing conclusions about benefit. In a composite, the label can conceal several different experiences.
Avoid the opposite overreaction
A composite is not inherently a trick. Combining events can address a coherent question and increase the number of observable outcomes. The problem arises when a broad phrase such as “serious health events improved” suggests that every component improved, or that the most important one drove the result.
In the fictional example, “fewer people had an urgent visit or hospitalization” may faithfully describe the primary endpoint. “The program prevented hospitalizations” needs its own supporting analysis. Reading the components protects that distinction without dismissing the combined result.
