Bayesian cure rate models induced by frailty in survival analysis

Area of Science:

• Biostatistics
• Survival Analysis
• Statistical Modeling

Background:

• Frailty models are crucial for handling unobserved heterogeneity in survival data.
• Continuous frailty distributions may be inappropriate for data with a cure rate, leading to incorrect inference.
• Existing models often fail to adequately address the complexities of long-term survival and zero frailty.

Purpose of the Study:

• To propose a flexible probability distribution using discrete frailty for survival data with a cure rate.
• To introduce and analyze a special hyper-Poisson discrete frailty distribution.
• To develop Bayesian inference methods, including simulation and diagnostics, for the proposed model.

Main Methods:

• Development of a novel discrete frailty probability distribution.
• Focus on a specific hyper-Poisson distribution.
• Application of intensive Markov chain Monte Carlo (MCMC) algorithms for Bayesian simulation and influence diagnostics.
• Analysis of a real-world dataset to demonstrate model utility.

Main Results:

• The proposed discrete frailty model effectively handles unobserved dependence and heterogeneity in survival data with a cure rate.
• The hyper-Poisson distribution demonstrates flexibility and appropriateness for such data.
• Bayesian inference methods and MCMC algorithms provide robust parameter estimation and model assessment.
• The application to a real dataset confirms the practical utility and improved inferential accuracy of the proposed model.

Conclusions:

• Discrete frailty models, particularly the proposed hyper-Poisson distribution, offer a superior alternative to continuous models for survival data with a cure rate.
• The developed Bayesian inferential framework ensures reliable analysis and interpretation of complex survival data.
• The study highlights the importance of accounting for discrete frailty to avoid misinference in the presence of cure rates.