Abstract:
|
In the presence of informative censoring even after stratification of baseline covariates, Kaplan-Meier method gives biased estimates of risk throughout the follow-up period. To account for informative censoring, time-dependent covariates can be used along with two statistical methods: inverse probability of censoring weighted (IPCW) Kaplan-Meier estimator and parametric g-formula estimator. The asymptotic unbiasedness of IPCW estimators depends on the correctness of model specification of censoring hazard, while that of g-formula depends on the models for event hazard and time-dependent covariates. We propose a doubly robust estimator that provides asymptotically unbiased estimates of risk if either the model for censoring or event hazard, but not necessarily both, is correctly specified under the assumption that covariates model is correctly specified. Simulation studies with time-dependent covariates that affect both time-to-event and censoring showed the theoretical property of our estimator, and that the proposed estimator gets slight efficiency recovery from IPCW method. The proposed method is illustrated in the large clinical trial with censoring before the end of follow-up.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.