Title: Observations on bivariate event-times subject to informative censoring
Authors: Dongdong Li - Simon Fraser University (Canada) [presenting]
Xiaoqiong Joan Hu - Simon Fraser University (Canada)
John Spinelli - University of British Columbia and British Columbia Cancer Agency (Canada)
Abstract: Studies on association between two event-times and how covariates affect the association are often of interest to researchers. Conventional statistical approaches usually assume non-informative censoring which could lead to biased inference when the assumption is violated. We conduct semi-parametric regression analysis with right-censored bivariate event-times in presence of informative censoring, using a motivating example which attempts to evaluate how clinical factors (e.g. treatment) affect the risk of getting cardiovascular disease among breast cancer patients who have experienced relapse/second cancer. We propose a pair-wise modeling approach, where the dependence structure between each event-time and the censoring time is modeled through copula functions. We develop a pseudo likelihood-based inference procedure. Simulation studies are conducted to examine the performance of the proposed modelling and inference procedure. Asymptotic properties of the proposed estimator are obtained. The proposed modeling and inference procedure is applied to a motivating research question using breast cancer data for illustration.