Dichotomization of Continuous Endpoints: Examination of Efficiency Under Departures from Normality

• Activity Number: 394
• Type: Contributed
• Date/Time: Tuesday, August 5, 2014 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Davis Gates*+
• Companies: Merck
• Keywords: dichotomizing ; responders ; efficiency

Abstract

Dichotomizing continuous endpoints is a useful practice in clinical trials design when it is desirable to offer a simple risk assessment or responder's outcome in lieu of a continuous outcome. It is known dichotomizing a continuous endpoint is accompanied with reduced efficiency under the normal distribution assumption. However, observed departures from normality can generate treatment comparisons in which testing the binomial outcome is more powerful than testing the continuous outcome. Such cases can be observed even under formal acceptances of normality. A review of loss of efficiency of dichotomization under the normality assumption is presented. The presentation continues with an exploration of the probability of observing a stronger result under dichotomization versus a continuous outcome from the same data set. The nature and degree of departure from the normality assumption required for this finding is explored using simulations, but was motivated by actual results from a randomized clinical trial.