Development and validation of deep learning- and ensemble learning-based biological ages in the NHANES study.

Area of Science:

• Biomedical Informatics
• Machine Learning in Healthcare
• Aging Research

Background:

• Conventional biological age (BA) models often rely solely on blood markers, limiting their comprehensive assessment.
• There is a need for advanced machine learning (ML) models that integrate diverse data types for more accurate BA prediction.

Purpose of the Study:

• To develop and validate novel ML-based BA models using clinical, behavioral, and socioeconomic factors.
• To evaluate the predictive performance of these models for all-cause and cause-specific mortality.

Main Methods:

• Utilized data from 24,985 National Health and Nutrition Examination Survey (NHANES) participants (1999-2010).
• Selected 30 features using LASSO regression for training Deep Biological Age (DBA) and Ensemble Biological Age (EnBA) models.
• Assessed model performance using Mean Absolute Error (MAE) and predictive accuracy for mortality via Area Under the Curve (AUC) and Hazard Ratios (HR).

Main Results:

• DBA and EnBA accurately predicted chronological age (MAE ~3-3.6 years).
• Both models demonstrated strong predictive capability for all-cause mortality (AUC ~0.89).
• Accelerated BA predicted increased mortality risk and was linked to chronic diseases, with medication usage and physical activity being key predictors.

Conclusions:

• Developed and validated novel ML models (DBA, EnBA) for biological age prediction.
• These models accurately predict all-cause and cause-specific mortality.
• DBA and EnBA show potential for early identification of high-risk individuals and guiding preventive interventions.