Analysis of Competing Risks Data with Covariates Subject to Detection Limits

Competing risks data are commonly encountered in biomedical studies when subjects may experience multiple types of events and the occurrence of the primary event of interest can be precluded by a competing event. Challenges arise for regression analysis of such data with covariates subject to censoring due to detection limits. We propose a novel multiple imputation method for inference under Fine-Gray’s subdistribution hazard model with censored covariates subject to detection limits. Our proposal uses the information from the fully observed covariate values and the failure outcomes to impute the censored covariates iteratively using rejection sampling, which makes the imputation model compatible to the substantive model and the estimation efficiency improve significantly. We show the consistency and asymptotic normality of the resulting estimator and demonstrate its promising finite sample performance through simulation studies. Moreover, we extend this new proposal to assess the impacts of censored covariates on the predictive performance of the competing risks model. To illustrate its practical utility, we provide an application to the data from a study of community acquired pneumonia. Supplementary materials for this article are available online.