Pseudo-observations under covariate-dependent censoring

A regression analysis using jack-knife pseudo-observations from the Kaplan-Meier estimator, or related estimators, can be biased when censoring times depend on event times or covariates. We study ways in which other, covariate-dependent, estimators can be used in place of the Kaplan-Meier related estimators to overcome the problem. These estimators are inverse probability weighted estimators, weighting with an estimate of the probability of observation based on a model of the censoring distribution. We study an additive hazard model and a proportional hazards model for the censoring distribution. We argue that, under certain assumptions, the pseudo-observation method with pseudo-observations from such estimators will produce consistent and asymptotically normal parameter estimates. (C) 2019 Elsevier B.V. All rights reserved.