An accelerated failure time regression model for illness-death data: A frailty approach

Area of Science:

• Biostatistics
• Survival Analysis
• Statistical Modeling

Background:

• Illness-death models are crucial for analyzing competing risks and time-to-event data.
• Existing models may not fully capture unobserved dependencies between different failure types.
• Accelerated Failure Time (AFT) models offer interpretable covariate effects.

Purpose of the Study:

• To propose a novel illness-death survival model incorporating shared frailty and AFT models.
• To enhance the handling of unobserved dependencies in time-to-event data.
• To provide a statistically robust and interpretable framework for survival data analysis.

Main Methods:

• Development of a new illness-death model with shared frailty and AFT hazard functions.
• Implementation of a semiparametric maximum likelihood estimation via a kernel smoothed-aided expectation-maximization algorithm.
• Variance estimation using weighted bootstrap and a novel graphical goodness-of-fit procedure.

Main Results:

• The proposed model effectively accounts for unobserved heterogeneity and dependence between nonterminal and terminal events.
• The shared frailty variate enhances the interpretability of AFT models in the illness-death framework.
• Analysis of breast cancer data demonstrates the practical utility and performance of the new model.

Conclusions:

• The novel shared frailty AFT model provides a valuable tool for analyzing complex survival data.
• The estimation procedure is computationally feasible and statistically sound.
• The approach offers improved insights into disease progression and risk factors compared to existing methods.