The expectation-maximization approach for Bayesian additive Cox regression with current status data

In this paper, we propose a Bayesian additive Cox model for analyzing current status data based on the expectation-maximization variable selection method. This model concurrently estimates unknown parameters and identifies risk factors, which efficiently improves model interpretability and predictive ability. To identify risk factors, we assign appropriate priors on the indicator variables which denote whether the risk factors are included. By assuming partially linear effects of the covariates, the proposed model offers flexibility to account for the relationship between risk factors and survival time. The baseline cumulative hazard function and nonlinear effects are approximated via penalized B-splines to reduce the dimension of parameters. An easy to implement expectation-maximization algorithm is developed using a two-stage data augmentation procedure involving latent Poisson variables. Finally, the performance of the proposed method is investigated by simulations and a real data analysis, which shows promising results of the proposed Bayesian variable selection method.