Evaluation of inferential methods for the net benefit and win ratio statistics

Area of Science:

• Biostatistics
• Clinical Trial Methodology
• Statistical Inference

Background:

• General Pairwise Comparison (GPC) statistics like net benefit and win ratio are crucial in clinical trial analysis and design.
• Existing inferential methods for GPC statistics (re-sampling, asymptotic, exact) lack comparative evaluation.
• Performance characteristics such as bias, Type I error, and confidence interval coverage require thorough assessment.

Purpose of the Study:

• To evaluate and compare the performance of different inferential methods for General Pairwise Comparison (GPC) statistics.
• To assess small sample bias in variance estimation for GPC statistics.
• To determine the Type I error control and 95% confidence interval coverage of various GPC inferential approaches.

Main Methods:

• Simulation studies were conducted to evaluate GPC inferential methods.
• Performance metrics included small sample bias, Type I error rate, and confidence interval coverage.
• Methods compared encompass re-sampling, asymptotic, and exact statistical approaches.

Main Results:

• Exact permutation and bootstrap tests demonstrated optimal performance across all evaluated aspects for net benefit.
• The exact bootstrap test exhibited the best performance specifically for the win ratio.
• Simulations revealed differences in bias, Type I error control, and CI coverage among methods.

Conclusions:

• Exact permutation and bootstrap tests are recommended for robust analysis of net benefit in clinical trials.
• The exact bootstrap test is particularly suitable for analyzing win ratio data.
• These findings provide guidance for selecting appropriate inferential methods in GPC analysis, especially with limited sample sizes.