Box 94 Guidance for the analysis of subgroups

♦ Treat all subgroup analyses with a degree of scepticism. In the protocol pre-specify a few important subgroups which are to be analysed, providing a rationale for why the effect of the experimental treatment is likely to be different in the different groups. Regard all other subgroup analyses as exploratory.

♦ Report all subgroup analyses (whether pre-specified or not), even if this is only in summary form (see examples below of how this might be done).

♦ Subgroup analyses should not be performed by comparing the experimental group against the control group and reporting a p-value separately for each subgroup, for example performing and emphasizing separate analyses for older and younger patients.

♦ Subgroup analyses should be performed by performing an analysis looking at the difference between the results in subgroups by performing a test for interaction (see below).

♦ Different results in different subgroups are probably more plausible when the overall result comparing experimental treatment against the control treatment is itself positive and statistically significant. This is because then we are looking for quantitative differences between subgroups, for example is the effect in younger patients larger than the effect in older patients?

♦ Different results in different subgroups are less plausible when there is no evidence of a difference between the experimental treatment against the control treatment. This is because then we are looking for qualitative differences between subgroups, for example, is the effect in younger patients in favour of the control treatment, while the effect in older patients in favour of the experimental treatment? Such qualitative differences are usually inherently biologically less plausible.

0 0

Post a comment