|
Abstract:
|
Survival data are referred to as the time to event data in the literature, and the main feature of such data is that the failure times are often incomplete due to censoring. In this article, we study regression analysis of arbitrarily censored survival data, which contain a mixture of exactly observed, left-censored, interval-censored, and right-censored observations. A novel Bayesian estimation method is proposed under the framework of semiparametric proportional odds models. The proposed approach adopts monotone splines to approximate the unknown cumulative baseline odds function, involving only a finite number of unknown parameters in modeling this nonparametric function. Posterior computation is carried out by an efficient Gibbs sampler, which is developed based on a novel data augmentation. All the unknown parameters and latent variables are sampled either from standard distributions or from an automatic Adaptive-reject sampling (ARS) in the proposed Gibbs sampler. Our method is evaluated by extensive simulations studies and illustrated by a real-life example from a sexually transmitted infection study.
|
