Biological Age Estimation From the Age Gap Using Deep Learning Integrating Morbidity and Mortality: Model Development...

Area of Science:

• Gerontology and computational biology.
• Development of advanced machine learning models for health assessment.

Background:

• Biological age (BA) is a key health indicator, but current models lack comprehensive data integration.
• Existing BA models are limited by clinical parameters and do not fully incorporate morbidity and mortality data.

Purpose of the Study:

• To develop and validate a novel transformer-based model for biological age (BA) estimation.
• To improve BA predictive accuracy by incorporating morbidity and mortality information.
• To enhance early identification of age-related disease risk.

Main Methods:

• Retrospective analysis of 151,281 adults' health checkup data (2003-2020).
• Development of a custom transformer architecture for multi-objective learning (feature reconstruction, BA/CA alignment, health discrimination, mortality prediction).
• Comparison with conventional BA methods (Klemera and Doubal, CA cluster, DNN) using health status stratification and mortality prediction.

Main Results:

• The BA-CA gap model demonstrated superior health status reflection and mortality risk stratification compared to existing methods.
• The model effectively distinguished between normal, pre-disease, and disease groups, showing a clear BA gap gradient.
• Kaplan-Meier analyses indicated stronger future mortality discrimination in men.

Conclusions:

• The transformer-based BA-CA gap model offers a granular, clinically meaningful aging assessment by integrating morbidity and mortality data.
• This approach supports personalized health management and risk stratification.
• External validation in diverse populations is recommended to confirm generalizability.