Quantile Residual Lifetime Regression for Multivariate Failure Time Data

The quantile residual lifetime (QRL) regression is an attractive tool for assessing covariate effects on the distribution of residual life expectancy, which is often of interest in clinical studies. When the study subjects are exposed to multiple events of interest, the failure times observed for the same subject are potentially correlated. To address such correlation in assessing the covariate effects on QRL, we propose a marginal semiparametric QRL regression model for multivariate failure time data. Our new proposal facilitates estimation of the model parameters using unbiased estimating equations and results in estimators, which are shown to be consistent and asymptotically normal. To overcome additional challenges in inference, we provide three methods for variance estimation based on resampling techniques and a sandwich estimator, and further develop a Wald-type test statistic for inference. The simulation studies and a real data analysis offer evidence of the satisfactory performance of the proposed method.