Nonparametric Bayesian inference for mean residual life functions in survival analysis

Area of Science:

• Statistics
• Biostatistics
• Reliability Engineering
• Survival Analysis

Background:

• Survival analysis models typically focus on survival distribution functions.
• The mean residual life (MRL) function, representing expected remaining lifetime, is vital in reliability, medical, and actuarial sciences.
• The MRL function offers a direct interpretation and characterizes the survival distribution.

Purpose of the Study:

• To develop a general Bayesian nonparametric inference framework for the mean residual life (MRL) function.
• To model the MRL function using a Dirichlet process mixture model for the underlying survival distribution.
• To explore the flexibility of the proposed model in capturing various MRL function shapes.

Main Methods:

• Developed a Bayesian nonparametric inference approach for MRL functions.
• Utilized a Dirichlet process mixture model for the survival distribution.
• Represented the MRL function model as a mixture of kernel MRL functions with time-dependent weights, emphasizing a gamma distribution kernel.

Main Results:

• The proposed model allows for a wide range of shapes for the MRL function.
• The inference method was successfully illustrated using two experimental group datasets.
• The method's performance was further validated with a dataset involving right-censored data.

Conclusions:

• The Bayesian nonparametric approach provides a flexible and effective method for MRL function inference.
• The model's structure, based on a Dirichlet process mixture, yields desirable properties for the MRL function.
• The method is applicable to various survival analysis scenarios, including those with right censoring.