Aging-related diseases in the US population: a data-driven approach using All of Us electronic health record data

Aging is the strongest risk factor for most chronic diseases, yet aging-related diseases (ARDs) lack consistent, data-driven definitions. Prior work has proposed frameworks to classify ARDs based on age-specific onset patterns, but applications have largely relied on demographically homogeneous populations.

We extended a data-driven framework integrating unsupervised clustering and actuarial modeling to electronic health record data from 633 547 participants in the US All of Us Research Program. Age-specific onset rates for 274 high-burden diseases between ages 21 and 74 were analyzed. Standardized onset curves were grouped using hierarchical agglomerative clustering. Aging-related patterns were assessed with Gompertz and Gompertz-Makeham (GM) models to quantify age-dependent increases in disease onset.

Eleven clusters of disease onset trajectories were identified, with 4 main clusters encompassing 96.0% of diseases. Two clusters exhibited pronounced late-life increases in onset rate, while 1 showed earlier but moderate age-related increases, and another displayed largely age-independent patterns. Cardiovascular diseases and cancers were enriched in clusters with strong aging-related onset trajectories. Overall, 165 diseases demonstrated strong evidence of being aging-related, characterized by positive Gompertz age coefficients and good GM model fit. Median onset ages varied across clusters and disease categories and showed statistically significant discrepancies compared with UK populations.

In a large, racially diverse US cohort, we identified distinct ARD clusters with shared age-related onset patterns. These findings extend prior work to a more diverse population and support integrated clustering and actuarial approaches to systematically define ARDs and inform aging research and public health strategies.