On Bayesian modeling of censored data in JAGS

Area of Science:

• Statistics
• Computational Statistics
• Bayesian Inference

Background:

• Markov Chain Monte Carlo (MCMC) methods are essential for Bayesian modeling.
• Just Another Gibbs Sampling (JAGS) is a popular software for MCMC analysis.
• The default deviance calculation in JAGS for censored data is inaccurate, hindering model comparison.

Purpose of the Study:

• To develop an automated approach for correct deviance function specification in JAGS for censored data.
• To provide a generic and simple alternative modeling strategy for analyzing censored outcomes.
• To enable accurate likelihood-based Bayesian model comparison.

Main Methods:

• Proposed an alternative modeling strategy for censored data analysis in JAGS.
• Implemented the strategy to automatically specify the correct deviance function.
• Applied the method to survival data and drug safety data examples.

Main Results:

• The alternative strategy correctly draws posterior samples in JAGS.
• The proposed method automatically yields the correct deviance for model assessment.
• Accurate mean deviance values were obtained for survival data using the exact likelihood.
• Deviance Information Criterion (DIC) and penalized expected deviance were computed for Bayesian models of censored drug safety data.

Conclusions:

• An effective strategy for modeling censored data in JAGS with correct deviance specification was proposed.
• The approach simplifies the computation of Kullback-Leibler based model selection measures.
• The method is applicable to various censored data types, including survival data, and diverse Bayesian model structures.