Answers are grounded in this paper and cite their sources.
Treatment allocation in multi-center clinical trials frequently employs restricted randomization procedures such as Efron’s biased coin design to maintain balance between treatment groups. For this design, statistical inference is complicated if outcomes are observed as panel counts or current status data, in which there are potential sequential assignment dependencies, intra-cluster correlation, and discrete event counts. For small numbers of clusters or under finite sizes, traditional inferential procedures with first order central limit theorem approximations have skewed type I error rates and inadequate coverage of the confidence intervals caused by path dependent correlation and skewness. This paper proposes an inferential methodology based on saddlepoint approximation for two-sample non-parametric rank tests under Efron’s biased coin design. Our method is based on the Markovian property of the allocation sequence, which allows us to calculate the exact cumulant generating function of the test statistic without having to perform prohibitively many enumerations of permutations. The Lugannani–Rice formula is then used to approximate the discrete mid-p values with high asymptotic accuracy. In addition, we use the proportional mean model and then find accurate confidence intervals for the recurrent event rate ratio. Monte Carlo simulations confirm that the proposed saddlepoint method maintains near-nominal Type I error control and accurate coverage probabilities across finite-sample scenarios where standard asymptotic normal approximations are not adequate. The practical utility of the methodology is demonstrated through applications to four multi-center clinical trial datasets.
No takes yet. Share an insight, caveat, or question.
Haidy A. Newer (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: