Wild bootstrap logrank tests with broader power functions for testing superiority

TL;DR

This paper presents new wild bootstrap logrank tests for survival data that combine classical weighted tests to achieve broader power, allowing for right censoring and demonstrating strong theoretical and practical performance.

Contribution

Introduction of novel wild bootstrap procedures combining classical weighted logrank tests for improved power in survival superiority testing.

Findings

Tests are asymptotically exact under the null hypothesis.

Tests are consistent for fixed alternatives.

Simulations show strong finite-sample performance.

Abstract

We introduce novel wild bootstrap procedures for testing superiority in unpaired two-sample survival data. By combining different classical weighted logrank test we obtain tests with broader power behavior. Right censoring within the data is allowed and may differ between the groups. The tests are shown to be asymptotically exact under the null, consistent for fixed alternatives and admissible for a larger set of local alternatives. Beside these asymptotic properties we also illustrate the procedures' strength in simulations for finite sample sizes. The tests are implemented in the novel R-package mdir.logrank and its application is demonstrated in an exemplary data analysis.