Prediction of biological age using machine learning.

Area of Science:

• Gerontology and Bioinformatics
• Precision Medicine and Public Health

Background:

• Taiwan faces a rapidly aging population and increasing demand for personalized healthcare.
• Accurate assessment of individual physiological aging is crucial for targeted health interventions.

Purpose of the Study:

• To develop a machine learning model for predicting biological age using health examination data.
• To quantify physiological age through residual life estimation and identify key aging biomarkers.
• To provide data-driven health recommendations for life extension.

Main Methods:

• Utilized LightGBM for biological age prediction, comparing its performance against XGBoost.
• Employed MICE imputation to handle missing data, assessing its impact on prediction accuracy.
• Applied Kaplan-Meier (K-M) estimator survival analysis to validate biological age as a health indicator.
• Identified influential aging biomarkers using SHAP and PCC interpretations.

Main Results:

• The LightGBM model demonstrated an 11.40% improvement in predictive performance (R-squared) over XGBoost.
• MICE imputation enhanced prediction accuracy by 23.35%.
• The model effectively differentiated health status groups and identified key aging biomarkers with significant overlap with leading causes of death.
• Personalized health recommendations could potentially extend life expectancy by 64.58%.

Conclusions:

• The proposed machine learning model accurately predicts biological age and serves as a valid health status indicator.
• The identified biomarkers and personalized recommendations offer practical relevance for precision health and life extension strategies.
• This research provides methodological support and data for addressing aging populations and advancing personalized healthcare.