报告题目：Clustered Event Time with Covariates Missing Not at Random
Abstract：Motivated by a study of tuberculosis (TB) screening, this paper considers the regression analysis of right-censored event times from clustered individuals. Specifically, we are concerned with situations where some covariate entries are missing not at random (MNAR). We propose a likelihood-based estimation procedure under the Cox proportional hazards frailty model. The proposed estimator is in fact an adaption of the semiparametric maximum likelihood estimator with complete covariate information. Readily available supplementary information on the missing covariate is used to provide the variation of the estimator. We present algorithms to compute the estimator and its variation, establish their consistency and weak convergence, and examine their finite-sample performance via simulation. The proposed approach is illustrated with the TB study data.