Reweighting Estimators for Cox Regression With Missing Covariates

Abstract

Missingness in covariates is a common problem in survival data. In this article we propose a reweighting method for estimating the regression parameters in the Cox model with missing covariates. We also consider the augmented reweighting method by subtracting the projection term onto the nuisance tangent space. The proposed method provides consistent and asymptotically normally distributed estimators when the missing-data mechanism depends on the outcome variables, as well as on the observed covariates with either monotone or arbitrary nonmonotone missingness patterns. Simulation results indicate that in most situations, the proposed reweighting estimators are more efficient than the inverse probability weighting estimators for the regression coefficients of the missing covariates and are as efficient as or more efficient than the inverse probability weighting estimators for the regression coefficients of the completely observed covariates. The simple reweighting estimators can be easily implemented in standard statistical packages.