Latent biochemical phenotypes delineate divergent health trajectories in older adults.

Ageing heterogeneity hampers prevention and care. We used routine biochemical panels and unsupervised learning to identify latent phenotypes in community-dwelling older adults. In 1491 participants from the Toledo Study for Healthy Ageing (TSHA) with ~10-11 years of follow-up, 39 blood biomarkers were dimension-reduced and clustered, yielding three phenotypes: Healthy, Metabolic (subclinical dysmetabolism), and Haematological (low erythroid/renal profile). Phenotypes differed in functional capacity, frailty, and independence at baseline (all p < 0.05 after age/sex adjustment) and predicted long-term mortality (Metabolic women HR = 1.49, p = 0.016). Sex-specific analyses revealed distinct disease-trajectory patterns (e.g., hypertension in Metabolic women HR = 1.30, p = 0.005; thrombosis in Haematological men HR = 7.20, p = 0.018; syncope in Haematological women HR = 1.88, p = 0.009). Findings are partially replicated in a cohort of physically active older adults (EXERNET), supporting the generalizability of the Metabolic phenotype. Standard laboratory data, integrated through machine learning, capture ageing-relevant biology and stratify future risk without specialised assays, enabling low-cost, scalable precision prevention.