Amy Herring, Harvard University

Missing Covariates in Survival Analysis

Problems associated with missing covariate data are well-known but often ignored. Intuitively, when subjects with missing covariate values differ systematically from those with complete data with respect to the outcome of interest, results from a traditional complete case data analysis omitting the missing cases may no longer be valid. We develop methods for handling missing covariates in the framework of the Cox regression model. We propose a model-based procedure with a Monte Carlo EM algorithm that allows us to make inferences based on the likelihood of the observed data. The estimation procedure is illustrated using cancer clinical trials in which categorical and continuous covariates are not completely observed. This methodology may be applied when covariates are ignorably or nonignorably missing and has been extended to allow the inclusion of random effects.