Five theses concerning the clinical consequences of pathology and prognostic factors

We discuss possible justifications to split study populations from a biometrical point of view. The existence of prognostic differences between subgroups are neither a sufficient nor a necessary reason to justify a splitting decision. There are essentially two separate types of relevant arguments to justify a split of patient study populations: a) Different toxicity/benefit trade-offs concerning the acceptability of a particularly aggressive treatment, b) Evidence for strong treatment by subgroup interactions, i.e. Treatment differences differ markedly by biologically defined subgroups. The latter is what the research ideal of biologically specific treatment asks for. Subgroup analysis is notoriously difficult. Formal statistical analysis must be complemented by specific evidence from basic sciences. Meta-subgroup analyses may be an option if a biologically specific hypothesis on which treatment component interacts with what biological feature allows to operationally identify all those randomised trials in which the effect should be present. In this conceptual and methodological paper we defend five theses concerning the clinical consequences of pathological and biological differences from a somewhat unorthodox biometrical point of view.