Large language model-based biological age prediction in large-scale populations.

Area of Science:

• Biomedical Informatics
• Gerontology
• Artificial Intelligence in Healthcare

Background:

• Accurate individual aging assessment is vital for proactive health management and disease prevention.
• Existing aging proxies have limitations including methodological constraints, weak outcome associations, and poor generalizability.

Purpose of the Study:

• To introduce a novel framework utilizing large language models (LLMs) for estimating overall and organ-specific individual aging.
• To validate the LLM-based aging assessment framework across diverse, large-scale population cohorts.

Main Methods:

• Developed and validated an LLM-based framework using health examination reports from over 10 million participants across six cohorts.
• Compared LLM-predicted aging metrics against established aging proxies and machine-learning models for predicting mortality and disease outcomes.

Main Results:

• LLM-predicted overall age demonstrated superior predictive power for all-cause mortality (C-index 0.757) compared to telomere length, frailty index, epigenetic ages, and ML models.
• The LLM-based approach showed strong associations between age gap and adverse health outcomes (HR 1.055 for mortality).
• LLM-predicted organ-specific ages and age gaps outperformed ML models in predicting organ-specific diseases.

Conclusions:

• The LLM-based aging assessment framework provides a precise, reliable, and cost-effective method for estimating overall and organ-specific aging.
• This framework holds significant potential for personalized health assessment and management in large populations.
• LLMs can identify aging biomarkers and develop disease risk prediction models, enhancing our understanding of aging processes.