Fusion Learning Algorithm to Combine Partially Heterogeneous Cox Models.

Area of Science:

• Biostatistics
• Survival Analysis
• Machine Learning

Background:

• Cox proportional hazards models are widely used for survival analysis.
• Covariate effects can differ significantly across subgroups (e.g., age groups), leading to model heterogeneity.
• Existing methods struggle with efficiently learning these heterogeneous patterns, especially with many subgroups.

Purpose of the Study:

• To develop a computationally efficient fusion learning procedure for clustering regression coefficients in Cox models.
• To address the challenge of partially heterogeneous parameters across predefined subgroups.
• To accurately identify and learn differing covariate effects on survival time.

Main Methods:

• A novel fusion learning algorithm is proposed.
• The method incorporates estimated parameter ordering to enhance computational efficiency.
• The procedure is designed to mitigate computational burden while preserving statistical power.

Main Results:

• Extensive simulation studies demonstrate the method's effectiveness.
• The proposed approach shows strong performance in learning heterogeneous regression coefficient patterns.
• The method offers a significant reduction in computational cost compared to conventional techniques.

Conclusions:

• The fusion learning procedure provides an efficient and powerful tool for analyzing survival data with subgroup heterogeneity.
• The method successfully captures variations in covariate effects across different groups.
• The approach is validated through simulations and a real-world vehicle leasing contract renewal analysis.