Abstract:
|
Adaptive enrichment trial designs (AETDs) can be an attractive option when there is prior uncertainty in treatment heterogeneity across subpopulations. Unfortunately, such designs often require several tuning parameters. This makes it difficult to ascertain the value of adopting an AETD, or the relative benefits of one design versus another. Here we present a Simulated Annealing approach for optimizing the tuning parameters of an AETD, in order to find a best case implementation of a given AETD method in a given scientific application. Optimization is done with respect to either expected sample size, or expected trial duration, and subject to constraints on power. We use this optimization framework compare approximate best-case implementations of AETD methods based Type I error rate reallocation and on the covariance of the test statistics. We also compare against conventional choices for tuning parameters that approximate O'Brien Fleming boundaries and Pocock boundaries. We find empirical evidence that optimized designs can be substantially more efficient than either standard Pocock or O'Brien Fleming boundaries.
|