Two-Sample Inference for Median Survival Times Based on One-Sample Procedures for Censored Survival Data

摘要

Nonparametric two-sample tests based on ranks, e.g. median and Wilcoxon tests, have been extended by Gehan, Efron, and Brookmeyer and Crowley among others to accommodate censored survival data. While there is a vast literature on general linear rank test procedures with censored observations, two-sample confidence intervals have received much less attention. Nonparametric confidence intervals for the ratio of two scale parameters were derived for randomly right censored data by Wei and Gail based on the idea of Hodges and Lehmann. Their method can also by applied to obtain confidence intervals for the difference of two location parameters. The confidence sets were computed by a numerical method based on a grid search and may not yield an interval. This paper focuses on the problem of testing and confidence intervals of the difference of two population medians with randomly right censored data. The authors show how to construct a two-sample test and confidence interval based on one-sample confidence intervals for the median of each population. (kr)