Explainable machine learning framework to predict personalized physiological aging.

Area of Science:

• Biomedical data science
• Gerontology
• Machine learning applications

Background:

• Personalized healthy aging necessitates precise monitoring of physiological changes and identification of aging markers.
• Traditional biostatistical methods struggle with complex inter-parameter interactions and lack explainability.
• Machine learning (ML) offers potential but its 'black box' nature hinders clinical adoption.

Purpose of the Study:

• To develop an explainable ML framework for estimating Personalized Physiological Age (PPA).
• To identify key biological variables predicting physiological aging.
• To enable physician confidence and clinical utility in ML-based aging assessment.

Main Methods:

• Utilized the National Health and Nutrition Examination Survey (NHANES) dataset.
• Selected XGBoost as the optimal ML algorithm.
• Implemented SHapley Additive exPlanations (SHAP) for model interpretability.

Main Results:

• Developed a PPA metric predicting chronic disease and mortality independently of chronological age.
• Identified 26 key variables sufficient for PPA prediction.
• Quantified the contribution of each variable, with HbA1c showing significant weight.
• Clustering of SHAP values revealed distinct aging trajectories.

Conclusions:

• PPA is a robust, quantitative, and explainable ML-based metric for personalized health status monitoring.
• The framework provides a method for precision physiological age estimation.
• Identified aging trajectories offer opportunities for tailored clinical interventions.