ABSTRACT
Evaluation of subgroups is a routine part of the analysis of nearly every large clinical trial. The purpose is to determine whether the effects of the treatment are consistent across the study population or whether there are patient characteristics that can be used to predict which patients will experience a particularly large benefit or a particularly large harm (i.e., treatment-by-factor interactions). Unfortunately, the power to detect interactions is typically low, and there is a belief that subgroup analyses are far more likely to lead to false-positive findings than to identify true interactions. However, it is important to note that the presence of an interaction depends on the scale on which the treatment effect is measured, and this has important implications for the assessment of benefit–risk in subgroups. Notably, when there is reason to believe that the relative effects of a treatment (e.g., the hazard ratio for a time-to-event endpoint) are consistent across subgroups, the absolute effects (e.g., the absolute risk reductions) are often highly variable. This is because there are often subgroups with greater and lesser disease severity, where event rates vary considerably; in this case, a constant hazard ratio across subgroups will lead to highly variable absolute risk reductions. In this chapter, we argue that benefit–risk conclusions should be based on the absolute effects for benefits and harms and, therefore, that subgroup analyses play a particularly important role.
