2 p.m. - 3 p.m. Location: JO 4.614
Postdoctoral Research Associate
Department of Biostatistics and Medical Informatics
University of Wisconsin – Madison
Semiparametric Accelerated Failure Time Models with Missing Covariates Under Censoring-Ignorable Missingness at Random Mechanism
Incomplete covariate data commonly occurs in survival analysis, which complicates the estimation. Most existing methods for missing covariates in survival data rely on the missing at random (MAR) assumption, which requires that the missingness mechanism to be independent of the missed covariate values conditioning on the observed follow-up time and other observed covariates. In many cases the conditioning should be on the failure time itself, instead of the follow-up time, which is the minimum of a censoring event and the interested failure time. Therefore, MAR is not a reasonable assumption for such cases. We propose a set of estimating functions for semiparametric accelerated failure time (AFT) model with missing covariates, under censoring-ignorable missingness at random (CIMAR) mechanism (the missingness depends on survival time but not on censoring time). We first propose an estimating equation based on the inverse probability weighting technology using uncensored subjects. To improve efficiency, we further propose an estimating equation adopting information from censored subjects. Our estimating equations yield consistent and efficient estimators, with possibility to extend to other survival methods such as Cox proportional hazard model. Numerical studies demonstrate that our estimators work quite well in reasonably large samples.
Sponsored by the Department of Mathematical Sciences