Case influence diagnostics in the Kaplan-Meier estimator and the log-rank test (Q2488422)

The authors propose a version of Cook's distance \(CD\) to assess the influence of the \(i\)-th observation on the Kaplan-Meier estimator of the survival function \(S\) by censored data: \[ CD_i=\int_0^\infty (\log \widehat S(t)-\log \widehat S_{-i}(t))^2\widehat S(t)^2dt, \] where \(S\) and \(S_{-i}\) are the Kaplan-Meier estimates of \(S\) by all the data and by the data with the \(i\)-th observation deleted. An analogous diagnostics is proposed for the log-rank test for homogeneity of two censored samples. Results of simulations are presented.