Randomization-based analysis of covariance for inference in the sequential parallel comparison design

Abstract

The sequential parallel comparison design (SPCD), with sequence groups P:P, P:T, and T:T, together with the exclusion of the second-period information from placebo responders in the first period, can serve usefully for studies with highly favorable placebo response, for example, psychiatric clinical trials. This paper presents a methodology for the first-period treatment difference in the overall population and the second-period treatment difference in the placebo nonresponders for the first period, as well as other available sources of information that could be of potential interest. Without any assumptions, a hypothesis testing method is proposed based on the randomization distribution of comparisons using the covariance structure for the randomized population under the null hypothesis to control type I error. Randomization-based analysis of covariance (ANCOVA) is introduced to adjust for baseline and for the observations that serve as baselines for the second period. Related methods are proposed for the study population as a simple random sample of an almost infinite population. The statistical properties of the proposed methods are described with simulation studies; and the use of the methods is illustrated for an example based on the data from the ADAPT-A clinical trial.