Subgroup analysis

A clinical trial reports its main result for everyone who took part. A subgroup analysis asks a narrower question: how did the treatment do in one slice of those people, such as the women, the over 70s, or the patients who were sickest at the start. The European Medicines Agency defines a subgroup as a subset of the trial population marked out by one or more characteristics of the patients themselves, usually measured before treatment began.

These findings carry less weight than the trial's main result, for reasons that compound. A subgroup is smaller than the whole trial, so its numbers are less precise and the uncertainty around them is wider. The balance that randomisation creates between the treatment and the control group is not fully preserved once you look inside it, so a difference can reflect who happened to land in each group rather than what the drug did. And the more slices a trial examines, the higher the chance that one of them shows a difference by chance alone.

The agency's guideline says the credibility of a subgroup finding rests on three things: whether the subgroup was defined in advance rather than picked out once the results were in, whether there is a biological reason to expect the treatment to behave differently in it, and whether other trials show the same pattern. A result that comes from a subgroup is a signal worth following rather than a settled answer, and regulators treat it as supporting evidence rather than proof.