A NEW AI-powered transformer model integrating morbidity and mortality data has produced a more clinically relevant way to estimate biological age in adults, outperforming conventional methods at reflecting health status and predicting future risk. The tool clearly distinguished between normal, predisease and disease groups based on biological age gaps, improving early identification of age-related health risks.
Biological age, unlike chronological age, offers a measure of an individual’s physiological health and rate of ageing, making it a promising metric for predicting disease risk, personalising care and guiding interventions. Conventional models, however, have relied on limited clinical information and failed to fully exploit patient outcomes, such as disease progression and death, which are crucial for accurate health assessments.
Researchers retrospectively analysed data from over 151,000 adults who attended routine health checks between 2003 and 2020. Participants were classified into three groups—normal, predisease and disease—based on the presence of diabetes, hypertension or dyslipidaemia. The custom transformer model was trained using unsupervised and self-supervised learning, simultaneously optimising feature reconstruction, alignment of biological and chronological age, health status discrimination and mortality prediction. Performance was robustly compared with established methods, including the Klemera and Doubal model, cluster-based approaches and deep neural networks. The new model achieved superior stratification of mortality risk, especially among men, and demonstrated clear gradients of biological age gap values across health statuses. Results held steady in multiple sensitivity analyses.
For clinical practice, the biological age gap model represents a meaningful advance for identifying patients at risk of age-related diseases well before symptoms appear. By providing a more precise and granular assessment of ageing, this approach supports early, proactive management and personalisation of care. Nonetheless, further validation in diverse populations will be essential before routine adoption, ensuring that estimates remain robust and clinically actionable across different demographic groups.
Reference
Moon SE et al. Biological age estimation from the age gap using deep learning integrating morbidity and mortality: model development and validation study. J Med Internet Res. 2025;27:e71592.
