American Statistical Association New York City Metropolitan Area Chapter Chapter Seminar May 23, 2012

Ideally, a clinical study should be able to demonstrate not only a statistically significant improvement in a key endpoint, but also that the magnitude of the effect is clinically relevant. One approach to address the question, often proposed by regulatory guidances, is a responder analysis, whereby a continuous efficacy measure is dichotomized into "responders" and "nonresponders." This presentation is taken directly from a recently published paper on responder analyses (Uryniak et al. Responder Analyses-A PhRMA Position Paper. Statistics in Biopharmaceutical Research 2011; 3:476-487) and represents a Pharmaceutical Research and Manufacturers of America (PhRMA) position on responder analyses. After providing a literature review, this presentation covers the results of simulations and the logic of conceptual underpinnings to highlight the consequences of dichotomization on statistical power and the implication of responder analyses in the assessment of a clinically relevant treatment effect. Any real or perceived benefit from a responder analysis is found to be outweighed by its loss of statistical power on the existence of a treatment effect. However, between-group comparisons on the percentages of "responders" can play a role in the assessment and reporting of the clinical meaningfulness of the treatment effect.