A rank-based test for clinical trials with multivariate longitudinal outcomes

Abstract

AbstractClinical trials of many chronic diseases such as Parkinson’s disease often collect multiple health outcomes to monitor the disease severity and progression. It is of scientific interest to test whether the experimental treatment has an overall efficacy on the multiple outcomes across time, as compared to placebo or an active control. To compare the multivariate longitudinal outcomes between two groups, the rank-sum test1and the variance-adjusted rank-sum test2can be used to test the treatment efficacy. But these two rank-based tests, by utilizing only the change from baseline to the last time point, do not fully take advantage of the multivariate longitudinal outcome data, and thus may not objectively evaluate the global treatment effect over the entire therapeutic period. In this paper, we develop rank-based test procedures to detect global treatment efficacy in clinical trials with multiple longitudinal outcomes. We first conduct an interaction test to determine whether treatment effect varies over time, and then propose a longitudinal rank-sum test to assess the main treatment effect either with or without the interaction. Asymptotic properties of the proposed test procedures are derived and thoroughly examined. Simulation studies under various scenarios are performed. The test statistic is motivated by and applied to a recently-completed randomized controlled trial of Parkinson’s disease.