Robust inference from multiple test statistics via permutations: a better alternative to the single test statistic...

Area of Science:

• Biostatistics
• Clinical Trials
• Statistical Inference

Background:

• Formal inference in randomized clinical trials typically relies on a single pre-specified test statistic.
• This approach is limited by its dependence on potentially unmet assumptions.
• A single statistic may not adequately capture treatment effects under diverse data distributions.

Purpose of the Study:

• To propose a robust inference method using multiple test statistics in clinical trials.
• To enhance the reliability of hypothesis testing for treatment effects.
• To provide an alternative to single-statistic reliance, especially the logrank test in survival analysis.

Main Methods:

• Pre-specifying multiple test statistics for analysis.
• Utilizing the minimum p-value across these statistics for hypothesis testing.
• Determining critical values from permutation distributions for error rate control.
• Testing the null hypothesis that treatment groups are indistinguishable.

Main Results:

• The minimum p-value approach effectively controls the Type I error rate.
• This method maintains good power even if some candidate statistics have low power.
• It offers a more reliable alternative to relying on a single test statistic.

Conclusions:

• Relying on the minimum p-value from multiple statistics enhances robustness in clinical trial inference.
• This method is particularly advantageous for complex survival trials where the logrank test may be insufficient.
• The proposed approach improves the reliability and validity of statistical conclusions in randomized trials.