Biomarker Evaluation and Subgroup Identification in a Pneumonia Development Program Using SIDES

This chapter discusses the general problem of exploratory subgroup analysis in the context of late-stage clinical development. In this context, exploratory subgroup analysis focuses on biomarker discovery and identification of subgroups with enhanced treatment effect in large clinical trial databases. A case study based on a Phase III development program in patients with nosocomial pneumonia is used to compare traditional approaches to subgroup search, based on univariate assessments of individual biomarkers, and a novel subgroup exploration method, which utilizes a recursive partitioning algorithm with a local treatment effect modeling approach. The SIDES (Subgroup Identification based on Differential Effect Search) method and its extensions (SIDEScreen method) have been used in multiple Phase II and Phase III programs to perform a comprehensive evaluation of candidate biomarkers and identify biomarker-based subgroup of patients with desirable characteristics (improved efficacy or acceptable safety). The chapter provides a detailed summary of key features of the SIDES method, including complexity control (subgroup pruning), biomarker screening to prevent data overfitting and application of resampling-based techniques to account for Type I error rate inflation inherent in subgroup exploration.