Abstract:
|
In non-standard settings of confirmatory clinical trials (e.g., trials incorporating historical data), the associated statistics for hypothesis testing are not pivotal quantities, in the sense that their distributions depend on unknown parameters. A common practice is to choose the critical value as a constant by a grid-search method to control type I error rates in testing within a certain range of the null space. However, potential power loss may be observed under scenarios with conservative error rate. We introduce an alternative computational framework to perform hypothesis testing for such trials. The proposed method is extended to accommodate studies with multiple comparisons. Some case studies illustrate that our approach properly controls error rates with power preservation.
|