B0666
Title: Nonparametric estimation of quantiles of the conditional residual lifetime distribution
Authors: Steven Abrams - University of Antwerp (Belgium) [presenting]
Paul Janssen - Hasselt University (Belgium)
Noel Veraverbeke - Hasselt University (Belgium)
Abstract: In medical research, interest is often in studying either the association between an event time $T_1$ and a continuous covariate $T_2$ or between two non-negative event times $T_1$ and $T_2$, where event times are potentially right-censored. This implies that time-to-event data are incomplete for some subjects. More specifically, for right-censored observations, the true event time is unobserved and is only known to exceed the observation time. Such a censored nature of the data should be accounted for when studying the association between $T_1$ and $T_2$. Although the strength of dependence between such random variables $T_1$ and $T_2$ can be expressed in terms of global and local association measures, alternative quantities are useful to study as well. For example, the median residual time to the occurrence of a specific event, given that an individual belongs to a specific group based on his/her value of $T_2$, is often a useful quantity for clinicians to work with. Therefore, existing methods are extended to non-parametrically estimated quantiles of the conditional residual lifetime distribution to encompass a more flexible classification of subjects into subgroups based on their respective $T_2$-values. More specifically, two estimators under one component are proposed, respectively univariate censoring, and a detailed study of their finite-sample performance is provided. The use of these estimators for different medical datasets is demonstrated.
