Abstract:
|
Many new experimental treatments outperform the current standard only on a subset of the population. Current methods for analyzing confirmatory clinical trials focus on establishing an average benefit across the population and then screen for evidence of subgroups which deviate from that benefit. However, such an approach may incorrectly conclude benefit in non-benefiting subgroups due to lack of power in the screening procedure, or fail to identify benefiting subgroups due to a lack of average benefit in the broad population. We propose an approach that directly identifies the subsets of patients for which there is sufficient evidence of benefit, or lack of benefit, fully controlling for multiplicity. Our method constructs two bounding subgroups for the benefiting subgroup: one which likely contains only patient types who benefit, and one which likely contains all who do. We illustrate the method on a data set from an Alzheimer's disease treatment clinical trial.
|