TESTING FOR TREATMENT DIFFERENCES WITH DROPOUTS PRESENT IN CLINICAL-TRIALS - A COMPOSITE APPROACH

A major problem in the analysis of clinical trials is missing data fro m patients who drop out of the study before the predetermined schedule . In this paper we consider the situation where the outcome measure is a continuous variable and the final outcome at the end of the study i s the main interest. We argue that the hypothetical complete-data marg inal mean averaged over the dropout patterns is not as relevant clinic ally as the conditional mean of the completers together with the proba bility of completion or dropping out of the trial. We first take the p attern-mixture modelling approach to factoring the likelihood function , then direct the analysis to the multiple testings of a composite of hypotheses that involves the probability of dropouts and the condition al mean of the completers. We review three types of closed step-down m ultiple-testing procedures for this application. Data from several cli nical trials are used to illustrate the proposed approach. (C) 1997 by John Wiley & Sons, Ltd.