American Statistical Association New York City Metropolitan Area Chapter New York State Psychiatric Institute at Columbia University Medical Center Biostatistics Seminar

When new covariates are collected in the middle of a follow-up study, these covariates may not be available for those who already died or dropped out of the study. In this talk, we propose a method handling this situation in the Cox proportional hazard model for survival data. We consider a weighting method and the associated augmented reweighting method for Cox model with missing covariates. 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 non-monotone missingness patterns. Simulation results indicate that, in common situations, the proposed reweighting estimators are equally or more efficient than the inverse probability weighting estimators. The simple reweighting estimators can be easily implemented in standard statistical packages such as SAS or SPLUS. We illustrate the method using Northern Manhattan Stroke Study.