Using weighted Kaplan-Meier statistics in nonparametric paired censored survival outcomes

This research introduces methods for nonparametric testing of weighted inte grated survival differences in the context of paired censored survival desi gns, The current work extends work done by Pepe and Fleming (1989, Biometri cs 45, 497-507), which considered similar test statistics directed toward i ndependent treatment group comparisons. An asymptotic closed-form distribut ion of the proposed family of tests is presented, along with variance estim ates constructed under null and alternative hypotheses using nonparametric maximum likelihood estimates of the closed-form quantities. The described m ethod allows for additional information from individuals with no correspond ing matched pair member to he incorporated into the test statistic in sampl ing scenarios where singletons are not prone to selection bias. Simulations presented over a range of potential dependence in the paired censored surv ival data demonstrate substantial power gains associated with taking into a ccount the dependence structure. Consequences of ignoring the paired nature of the data include overly conservative tests in terms of power and size, In fact, simulation results using tests for independent samples in the pres ence of positive correlation consistently undershot both size and power tar gets that would have been attained in the absence of correlation. This addi tional worrisome effect on operating characteristics highlights the need fo r accounting for dependence in this popular family of tests.