IRIS: Interpretable Risk Clustering Intelligence for Survival Analysis

Area of Science:

• Biostatistics
• Machine Learning
• Medical Informatics

Background:

• Deep learning survival analysis models often lack interpretability and robust risk stratification.
• Existing methods typically perform risk clustering post-hoc, limiting direct data-driven insights.

Purpose of the Study:

• To introduce Interpretable Risk Clustering Intelligence for Survival Analysis (IRIS), a novel framework enhancing interpretability and risk clustering in survival analysis.
• To develop a model that directly learns patient risk groups from data while providing transparent feature importance.

Main Methods:

• Developed the IRIS framework integrating deep learning with interpretable risk clustering.
• Employed feature contribution functions for transparent feature importance estimation.
• Validated IRIS on benchmark, Alzheimer's disease, and electronic health record datasets.

Main Results:

• IRIS demonstrated superior performance in risk clustering and predictive reliability across diverse datasets.
• Achieved a successful balance between model interpretability and prediction accuracy.
• Showcased improved clinical utility for treatment planning and resource allocation.

Conclusions:

• IRIS offers a significant advancement in interpretable survival analysis, enabling meaningful risk stratification.
• The framework provides clinicians with actionable, data-driven insights for personalized medicine.
• IRIS successfully addresses the limitations of current deep learning survival models regarding interpretability and risk grouping.