A Bayesian Group Sequential Approach for Multiple Endpoints

Abstract

ABSTRACT Making decisions regarding overall treatment effectiveness can be problematic when one considers multiple outcomes, especially if the treatment effects are discordant across the profile of outcomes. A typical example involves a case where a treatment has both increased efficacy in one endpoint (e.g., improved disease-free survival) and increased side effects (e.g., more acute toxicities). Often a study is designed to test one primary hypothesis while other outcomes vital to the decision-making process are not formally incorporated into the study design. We describe a Bayesian approach that provides a mechanism for combining information from two normally distributed endpoints and accounts for the magnitude of those effects. This procedure is implemented for the case of comparing two different treatments to each other and allows for multiple looks at the data. Information from more than one endpoint is combined through the use of utility functions. Our group sequential procedure is demonstrated for the design of a cancer clinical trial that involves two looks at the data. The example shows the effect of different utility functions applied to the same data. Because the selection of the utility function is crucial to the interpretation of the two endpoints, the results are not invariant to the utility function, and great care must be exercised in choosing an appropriate function. Additionally, since results are not robust to the choice of prior, we select a noninformative prior for our example. Despite some limitations in the specification of the utility structure and prior distribution, our procedure provides a mechanism that is useful for simultaneously monitoring multiple endpoints in a clinical trial.