Saddlepoint inference for rank-based k-sample tests in clustered survival trials

Area of Science:

• Biostatistics
• Clinical Trials Methodology
• Statistical Inference

Background:

• Standard statistical methods struggle with complex derived endpoints in cluster randomized trials.
• Limited cluster numbers and high intra-cluster dependence cause conventional nonparametric tests to fail, inflating Type I errors and reducing confidence interval coverage.

Purpose of the Study:

• To develop a unified framework for robust statistical inference in cluster randomized trials with complex, clustered, right-censored survival data.
• To address the breakdown of standard asymptotic theory and nonparametric procedures in challenging trial designs.

Main Methods:

• A rank-based k-sample test statistic was developed for clustered, right-censored survival data.
• A multivariate saddlepoint approximation was derived for the exact permutation distribution, offering high-order precision.
• The approach enables nonparametric confidence interval construction via test inversion.

Main Results:

• The proposed saddlepoint approximation method accurately controls Type I error rates in small-cluster settings where standard methods fail.
• It achieves accuracy comparable to computationally intensive Monte Carlo resampling but at a significantly lower computational cost.
• Nonparametric confidence intervals for relative treatment effects were practically constructed.

Conclusions:

• The developed rank-based saddlepoint approximation offers a robust and efficient inferential tool for complex cluster randomized trials.
• This methodology provides reliable results in challenging scenarios, as demonstrated in analyses of leukemia, vision loss, and periodontitis clinical trials.
• It overcomes the limitations of conventional techniques, ensuring trustworthy conclusions in borderline cases.