Bayesian analysis of proportional hazards models built from monotone functions

Area of Science:

• Biostatistics
• Survival Analysis
• Statistical Modeling

Background:

• The standard proportional hazards model assumes known baseline hazard and covariate link functions.
• Unknown baseline hazard and covariate link functions present significant challenges in survival data analysis.
• Monotone functions are often used to characterize these unknown components.

Purpose of the Study:

• To develop and apply a Bayesian method for proportional hazards models with unknown, monotone baseline hazard and covariate link functions.
• To analyze lung cancer patient survival data using this novel statistical approach.
• To provide a robust framework for model criticism in survival analysis.

Main Methods:

• Utilizing a dense class of monotone functions, characterized as mixtures of Beta distributions.
• Employing a Bayesian framework with vague prior specifications, including Jeffreys's prior.
• Implementing sampling-based computational methods for model fitting and inference.

Main Results:

• The developed Bayesian approach effectively handles unknown baseline hazard and covariate link functions.
• The model was successfully applied to a real-world dataset of lung cancer patient survival.
• The study discusses and demonstrates methods for model criticism.

Conclusions:

• The proposed Bayesian proportional hazards model offers a flexible and powerful tool for survival data analysis.
• This approach enhances the ability to model complex relationships between covariates and survival outcomes.
• The methodology is validated through application to lung cancer survival data.