Responder Analysis 1 (Bad) and 2 (Good)
*Janet Turk Wittes, Statistics Collaborative, Inc. 

Keywords: responder, genomics, adherer

Responder analyses come in at least two flavors. Both require a definition of a responder. Analyses of the distasteful flavor, long justly maligned but still performed and even often reported in journals, compare responders to non-responders. (A variant compares adherers to non-adherers.) The typical analysis of this flavor presents a Kaplan-Meier plot showing that responders live longer than non-responders (surprise!). We statisticians should refuse to produce such analyses even if our clinician colleagues plea they are “clinically relevant”. The more tasty flavor consists of analyses that compare two or more groups with respect to the proportion of responders. The goal is to identify the more effective treatment. Another important goal may be to identify covariates that predict response. The advent of genomics makes such analyses especially important.