Bayesian nonparametric analysis of restricted mean survival time

Area of Science:

• Biostatistics
• Survival Analysis
• Bayesian Statistics

Background:

• Restricted mean survival time (RMST) is crucial for comparing survival curves when mean survival is not estimable due to censoring.
• Frequentist methods for RMST are established, but Bayesian approaches, especially nonparametric ones, are less explored.
• Existing Bayesian methods for RMST are limited, particularly for right- and interval-censored data.

Purpose of the Study:

• To propose novel Bayesian nonparametric estimation and inference procedures for RMST.
• To extend RMST analysis to both right- and interval-censored data using Bayesian methods.
• To compare the performance of proposed Bayesian nonparametric methods with frequentist approaches.

Main Methods:

• Developed Bayesian nonparametric estimation using a mixture of Dirichlet processes (MDP) prior for the distribution function.
• Utilized Dirichlet process mixture models for an alternative Bayesian nonparametric approach.
• Conducted simulation studies to evaluate estimation robustness and prior incorporation.
• Applied methods to real clinical trial data with right- and interval-censored survival data.

Main Results:

• Bayesian nonparametric RMST with diffuse MDP priors demonstrated robust estimation.
• Informative priors in the Bayesian approach allowed for the incorporation of prior knowledge.
• The proposed methods showed flexibility and interpretability in analyzing real trial data.
• Bayesian nonparametric methods provided reliable estimates for both right- and interval-censored data.

Conclusions:

• Bayesian nonparametric methods provide a flexible and interpretable framework for RMST estimation with censored data.
• These methods effectively handle both right- and interval-censored data, expanding the utility of RMST.
• The ability to incorporate prior knowledge through informative priors is a key advantage of the Bayesian approach.
• The proposed Bayesian nonparametric RMST procedures offer a valuable alternative to existing frequentist methods.