Machine learning reveals hidden longevity pathways beyond chronic disease

This video summarises the paper “Longitudinal machine learning uncouples healthy aging factors from chronic disease risks,” published in Nature Aging (2024). Most aging research treats chronic diseases as the main lens for understanding longevity. But this study takes a different approach: using longitudinal machine learning on millions of electronic health records, the authors reconstruct entire lifespan trajectories from just a few years of data. The result is a quantitative “longevity score” that predicts survival to age 85, even in people who appear perfectly healthy. The surprising finding is that healthy aging is not just the absence of chronic disease. A sizable portion of longevity variation remains unexplained by classic risks like diabetes, kidney disease, cardiovascular disease or COPD. Instead, subtle patterns in routine lab tests, such as mildly low neutrophils or low alkaline phosphatase, emerge as independent indicators of long-term health potential. These signals appear early in adulthood and hold across populations in Israel, the UK and the US. The study also shows that these longevity signatures have a heritable component. Even after removing all patients with disease predisposition, parents of high-scoring individuals live at least one year longer than parents of matched controls. Genetic analyses further reveal variants that influence aging trajectories independently of disease risk.