scientific article; zbMATH DE number 124966

From MaRDI portal
Publication:4026602





[For the entire collection see Zbl 0745.00009.] Left-censored survival times occur when the failure event has already occurred before observation begins such that the exact failure time \(X\) is unknown but the fact of failure is known. Let denote \(X\) the true life time with survival function \(S_ X\) (assumed continuous for simplicity) and \(L\) the left-censoring time with survival function \(S_ L\). \(U=\max(X,L)\) is the observed failure time and \(\delta=\mathbf{1}(L\leq X)\) the censoring status. If \(L>X\), respectively \(\delta=0\), the survival time is left-censored. \(X\) and \(L\) are assumed to be independent non- negative random variables. A Left-Kaplan-Meier estimator (LKM) is derived using the basic integral equation \[ 1-S_ X(t)=1+\int^ \infty_ t(1- S_ X(u))(1-Q(u))^{-1} dQ_ X(u)\tag{*} \] with \(Q=1-(1-S_ X)(1-S_ L)\) and subsurvival function \[ Q_ X(t)=-\int^ \infty_ t (1-S_ L)dS_ X, \] where (*) is a backward Doléans equation of the form \[ Z_ t=Z_ \infty+\int^ \infty_ t Z_ s dA_ s,\quad t\geq 0, \] whose unique solution gives an explicit LKM estimate which can be viewed as a symmetrical version of the right-censored Kaplan Meier estimate. The authors prove that the LKM estimate is a self-consistent, generalized maximum likelihood estimator and can be derived by a redistribution to- the-left algorithm completely in analogy to the left-censored case. Strong consistency is obtained from Glivenko-Cantelli theorem and continuity of the integrals. The influence curve can be calculated and asymptotic normality is obtained by standard arguments.











This page was built for publication:

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4026602)