Resampling-based stepwise multiple testing procedures with applications to clinical trial data.

Area of Science:

• Biostatistics
• Clinical Trial Methodology

Background:

• Clinical trial endpoints are frequently correlated, complicating standard multiple testing procedures.
• Existing methods often ignore correlations or require known correlation structures.
• Resampling-based methods offer a powerful alternative for handling unknown correlations.

Purpose of the Study:

• To advocate for and evaluate resampling-based multiple testing procedures in clinical trial data analysis.
• To extend stepdown resampling methods to stepup procedures.
• To assess the performance of these methods across various correlation structures and data types.

Main Methods:

• Simulation studies were conducted to compare resampling-based stepdown and stepup methods.
• Methods were evaluated under different correlation structures and distribution types, including binary data and small samples.
• The Family-wise type I error rate (FWER) was controlled strongly.

Main Results:

• Resampling-based methods demonstrated strong FWER control across various scenarios.
• These methods were more powerful than Holm and Hochberg methods under positive dependence and independence for binary data.
• The practical advantages were illustrated using cardiovascular and oncology trial data.

Conclusions:

• Resampling-based stepdown and stepup methods are effective and powerful tools for clinical trial analysis.
• These methods provide superior performance compared to traditional procedures, particularly with correlated endpoints.
• The application of these advanced statistical techniques is encouraged for clinical trial data analysis.