The Two Sides of Dichotomization, and Their Synthesis?
*Heng Li, FDA/CDRH 

Keywords: responder analyses, dichotomization, patient-reported outcomes, measurement error, stochastic ordering, non-parametric statistics

An essential part of clinical trial design is the pre-specification of endpoints, or outcomes, by which to assess the safety and effectiveness of the medical product or therapy to be studied in the trial. Sometimes transformations of endpoints are considered and study hypotheses are formulated in terms of some transformed outcome measures. In this presentation we discuss the transformation to a binary variable, otherwise known as dichotomization. We try to reconcile some opposing views on this subject from more general perspectives.