Bayesian decision-making with multiple correlated binary outcomes in clinical trials
share
Clinical trials often evaluate multiple outcome variables to form a comprehensive picture of the effects of a new treatment, and use this multidimensional insight to make a decision about treatment superiority. Common statistical procedures to make these superiority decisions with multiple outcomes have three important shortcomings: 1.) Outcomes are often modelled individually, and consequentially fail to consider the relation between outcomes; 2.) Superiority is often defined as a relevant difference on a single, any, or all outcomes(s); and lacks a compensatory mechanism that allows positive effects on some outcomes to outweigh negative effects on other outcomes; 3.) A priori sample size computation relies on multiple pieces of information that are often not or only partially available at the time of study design, thereby potentially undermining accuracy. These shortcomings may result in trials including too many patients or leading to erroneous decisions. In this paper, we suggest solutions to each of the introduced shortcomings by proposing 1.) a Bayesian multivariate model for the analysis of correlated binary outcomes; 2.) a decision criterion with a compensatory mechanism; and 3.) an adaptive stopping rule that relies on interim monitoring to terminate data collection timely.
READ FULL TEXT