PS5E - Innovation with Hierarchical Composite Endpoints in Complex Trial Design

Two complex innovative trial designs using a hierarchy of endpoints will be presented. The first is an adaptive platform trial for the treatment of ALS. The second is a confirmatory study in heart failure that borrows information from a previous trial. In both ALS and heart failure the standard primary endpoint is an integration of survival and/or hospitalization events with function. Traditional analyses of these endpoints are non-parametric joint rank analyses. However, in innovative trial designs such as platform trials or trials with historical borrowing, a parametric analysis is often needed to accommodate complexities within the setting. In both examples, sharing of information from multiple data sources is utilized and analysis models must account for potential differences across these sources. The Bayesian paradigm lends a natural framework for synthesis of these data through hierarchical modeling. 

There are significant challenges for designing a cardiovascular device trial appropriate for FDA regulatory approval. Some of these issues are more specific to device trials as opposed to drug trials. The lecturer will review some of the current FDA Center for Devices and Radiological Health (CDRH) thinking on this topic and highlight considerations pertinent to the use of hierarchical composite enpoints and innovative trial designs in this setting. 

The primary design questions for clinical trials are sample size, early stopping for efficacy or futility and adaptive designs such as sample size adjustment. The purpose of this talk is to discuss how to address these issues for trials that use Hierarchical Composite Endoints.

Sample size calculation is complicated by the fact that if followup times are different the ordering is not transitive so that the simple Mann-Whitney variance formula cannot be used, I will discuss simulation approaches to this issue. Most large clinical trials use group sequential designs, where there are several interim analyses in which the trial can stop for efficacy or futility. It is necessary to determine boundaries so that the probability that efficacy is declared is controlled and the trial has adequate power. The calculations to do this usually assume that the multiple test statistics that are calculated have independent increments. Unfortunately this isn't usual true with test statistics based on Hierarchical Composite Endpoints. We show how these calculations can be modified in this case when a trial is being analyzed, and discuss how such a trial might be designed. Finally some trials are designed with adaptive designs. Basically and interim analysis is conducted in which there are three possible decisions. The first being stop and declare efficacy the next being stop and declare futility and the third being, continue but recalculate the sample size. We show how this could be done for a Hierarchical Composite Endpoints.