Data-Driven and Confirmatory Subgroup Analysis in Clinical Trials

Abstract

In this chapter we provide an overview of the principles and practice of subgroup analysis in late-stage clinical trials. For convenience, we classify different subgroup analyses into two broad categories: data-driven and confirmatory. The two settings are different from each other primarily by the scope and extent of pre-specification of patient subgroups. First, we review key considerations in confirmatory subgroup analysis based on one or more pre-specified patient populations. This includes a survey of multiplicity adjustment methods recommended in multi-population Phase III clinical trials and decision-making considerations that ensure clinically meaningful inferences across the pre-defined populations. Secondly, we consider key principles for data-driven subgroup analysis and contrast it with that for a guideline-driven approach. Methods that emerged in the area of principled data-driven subgroup analysis in the last 10 years as a result of cross-pollination of machine learning, causal inference and multiple testing are reviewed. We provide examples of recommended approaches to data-driven and confirmatory subgroup analysis illustrated with data from Phase III clinical trials. We also illustrate common errors, pitfalls and misuse of subgroup analysis approaches in clinical trials often resulting from employing overly simplistic or naive methods. Overview of available statistical software and extensive bibliographical references are provided.