Testing for a treatment effect in a selected subgroup.

Area of Science:

• Biostatistics
• Clinical Trial Design
• Pharmacogenomics

Background:

• Increasing interest in personalized medicine using continuous biomarkers to predict treatment response.
• Statistical challenges arise when using the same data for threshold selection and subgroup analysis.
• Need for robust methods to control Type I error rates in biomarker-guided clinical trials.

Purpose of the Study:

• To propose a hierarchical testing framework for familywise Type I error rate control.
• To introduce two novel statistical tests for identifying treatment effects in biomarker-defined subpopulations.
• To address the statistical complexities of threshold-based subgroup analysis in clinical trials.

Main Methods:

• Development of a hierarchical testing procedure.
• Proposal of two distinct statistical tests: one based on linear regression with interaction, and a more robust alternative.
• Evaluation of familywise Type I error rate control under different assumptions.

Main Results:

• The proposed framework provides control over the familywise Type I error rate.
• A linear regression-based test is powerful but sensitive to model assumption violations.
• A more robust test offers better performance when linear model assumptions are not met, albeit with slightly reduced power.

Conclusions:

• The hierarchical testing framework is effective for managing Type I errors in biomarker-stratified treatment effect analyses.
• The choice between the two proposed tests depends on the specific clinical trial context and data characteristics.
• These methods enhance the statistical rigor of identifying treatment benefits in targeted patient populations based on continuous biomarkers.