A flexible bivariate cure model with shared random effect and associated inference and application to diabetic...

Area of Science:

• Biostatistics
• Survival Analysis
• Statistical Modeling

Background:

• Traditional cure models are limited to univariate data.
• Many real-world scenarios involve paired lifetimes with potential for cure and dependence.

Purpose of the Study:

• To propose a flexible bivariate cure model for paired lifetime data.
• To simultaneously model cure fractions and dependence structures in marginal survival distributions.

Main Methods:

• Utilized a shared frailty term (generalized gamma distribution) to capture dependence.
• Employed parametric modeling for marginal survival distributions.
• Developed likelihood-based inference using the expectation-maximization algorithm.

Main Results:

• The proposed model effectively accommodates cured individuals in one or both marginals.
• Simulation studies demonstrated accurate parameter estimation and model efficiency.
• A real-world application on diabetic retinopathy highlighted the model's ability to reveal insights not found in univariate analyses.

Conclusions:

• The bivariate cure model provides a comprehensive statistical tool for paired lifetime data.
• It enhances the understanding of cure dynamics and survival dependence in biomedical research.
• Broadens the applicability of cure rate methodology to complex paired data scenarios.