On a Primary Efficacy Endpoint *

Abstract

In some confirmatory clinical trials, data on the number of episodes of a particular type during baseline and treatment phase periods are collected. The primary efficacy variable proposed by the sponsor to compare the test drug with placebo is Y = (B — T)/(B + T), where B and T are x-day rates of episodes for baseline and the treatment phase, respectively. In this article, it is argued that this efficacy variable does not have clinical interpretation as required by the International Conference on Harmonization (ICH) E9 guidelines (1). It is claimed that the difference D = B-T would be a natural and legitimate efficacy endpoint as it represents a reduction in the frequency of episodes. The relative difference RD = D/B is also examined for its feasibility. It is indicated that using Y as the dependent variable instead of D (or RD) lowers the p-values for the pooled t-test and the Wilcoxon test. It is concluded that the proposed efficacy variable Y is obscurant in nature and introduces statistical bias in favor of the test drug. The arguments against the use of the response variable Y are also valid for clinical trials designed to increase the rate of an outcome.