A0272
Title: Estimation and log-rank testing procedure via bivariate survival copula models under semi-competing risk
Authors: Tomoyuki Sugimoto - Osaka University (Japan) [presenting]
Abstract: In time-to-event data, two primary endpoints of interest are often non-fatal and fatal, and then the issue of semi-competing risks arises when designing trials or performing statistical analyses. If such non-fatal and fatal event-times are mutually uncorrelated, we can conduct the usual statistical analysis. However, when the two event times are correlated, the usual use of log-rank analysis or Cox regression models for non-fatal events will reduce the power of the analysis. The problem of estimating and testing the hazard ratio of the marginal distribution of the non-fatal event is difficult to handle not only in theory but also in practical use. Assuming a bivariate survival copula model in which two event times are possibly correlated under semi-competing risks, we propose an estimation method and algorithm for making an inference on the marginal distribution of the non-fatal events. We discuss the theoretical validity and theoretical properties of the estimated survival function based on the proposed estimation procedure. Furthermore, we discuss the extension to the two-sample problem via the estimation for survival functions and copula-related parameters. We discuss a modified version of the log-rank analysis and evaluate the performance of this estimation procedure using bivariate survival copulas.
