Robustness of testing procedures for confirmatory subpopulation analyses based on a continuous biomarker

Area of Science:

• Biostatistics
• Clinical Trials
• Personalized Medicine

Background:

• Personalized medicine increasingly focuses on treatment effects within specific patient subpopulations defined by biomarkers.
• Continuous biomarkers are often dichotomized, creating nested subpopulations (low/high levels) requiring careful statistical analysis.
• Group sequential trial designs offer a framework for analyzing such nested subpopulations.

Purpose of the Study:

• To investigate the inflation of family-wise type 1 error rates when analyzing biomarker-defined subpopulations in clinical trials.
• To propose and evaluate alternative hypothesis tests that control family-wise type 1 error rates under minimal assumptions.
• To illustrate the proposed methods using a depression clinical trial example.

Main Methods:

• Simulations were used to quantify the inflation of type 1 error rates in testing biomarker subpopulations using group sequential designs.
• The impact of outcome variability across subpopulations, particularly with unadjusted prognostic biomarker effects, was examined.
• Alternative statistical tests were developed and assessed for their error rate control.

Main Results:

• Analysis of biomarker subpopulations using standard group sequential methods can lead to inflated family-wise type 1 error rates.
• This inflation is particularly pronounced when biomarker prognostic effects are not accounted for in the statistical model, causing outcome variability differences.
• The proposed alternative tests effectively control the family-wise type 1 error rate under minimal assumptions.

Conclusions:

• Standard group sequential approaches are not always appropriate for testing biomarker-defined subpopulations due to potential type 1 error inflation.
• Careful consideration of biomarker effects and outcome variability is crucial for accurate subgroup analysis in personalized medicine.
• The developed hypothesis tests offer a robust solution for controlling error rates in biomarker-based subpopulation analyses.