Evidence-appraisal glossary
Subgroup analysis
Subgroup analysis splits a study's participants into groups (by age, sex, disease severity, and so on) and estimates the treatment effect within each. It asks whether the intervention works differently for different kinds of people, rather than reporting a single average effect for everyone in the trial.
Also called: subgroup effect, effect modification analysis, subset analysis.
What it is
A trial reports one overall (average) treatment effect. A subgroup analysis re-examines that effect inside slices of the population, for example men versus women, or mild versus severe disease, to see whether benefit or harm varies.
How to read it
Subgroups are notoriously unreliable and generate false leads. When you see a subgroup claim, check:
- Pre-specified or post-hoc? Groups named in the protocol before unblinding are more credible than ones found by dredging the data afterward.
- Multiplicity. Testing many subgroups makes at least one "significant" finding likely by chance alone. Ten independent tests give a false-positive risk over 40%.
- Interaction test, not eyeballing. A within-subgroup p-value only shows the effect held up in a smaller sample. The right question is a test for interaction: does the effect genuinely differ between subgroups?
- Biological plausibility and replication in other trials.
Treat subgroup findings as hypothesis-generating, not confirmatory, unless the trial was designed to test them.
This is a plain-language methodology definition for reading research. It is general education, not medical advice.