ABSTRACT
Multiple comparisons are commonly encountered in clinical trials. Multiple comparisons may involve comparisons of multiple treatments (dose groups), multiple endpoints multiple time points, interim analyses, multiple tests of the sample hypothesis, variable/model selection, and subgroup analyses in a study. In this chapter, statistical methods for controlling error rates for multiple comparisons will be reviewed. From a statistical reviewer’s point of view, Fritsch (2012) suggested that when dealing with the issue of multiplicity, one should carefully select the most appropriate hypotheses, i.e., choose “need to have” endpoints, but don’t pile on “nice to have” endpoints; put the endpoints in the right families; carefully consider which hypotheses represent distinct claims; and ensure all “claims” are covered under the multiplicity control structure. In addition, one should ensure a good match between the study objectives and the multiplicity control methods by utilizing natural hierarchies (but avoid arbitrary ones) and taking the time to understand complex structures to ensure overall control.
