Semiparametric Bayesian accelerated failure time model with interval-censored data

Abstract

Many existing approaches to analysing interval-censored data lack flexibility or efficiency. In this paper, we propose an efficient, easy to implement approach on accelerated failure time model with a logarithm transformation of the failure time and flexible specifications on the error distribution. We use exact inference for the Dirichlet process without approximation in imputation. Our algorithm can be implemented with simple Gibbs sampling which produces exact posterior distributions on the features of interest. Simulation and real data analysis demonstrate the advantage of our method compared to some other methods.

Keywords:

• Dirichlet process
• accelerated failure time
• transformation
• nonparametric Bayes
• interval-censored time

Acknowledgements

The work of the first author was supported by Early Career Research grant of Central Michigan University, c62229. We thank Xiaoye Wang for assistance in producing some simulation results, the anonymous reviewers for helpful suggestions which led to significant improvement of the paper.