Latent biochemical phenotypes delineate divergent health trajectories in older adults.
Standard blood work combined with AI reveals hidden health patterns that predict who'll develop frailty or disease years ahead, enabling cheaper, personalized prevention.
Using 39 routine blood biomarkers from 1,491 older adults followed for ~10 years, unsupervised ML identified three biochemical phenotypes that predicted long-term mortality, frailty, and disease trajectories with sex-specific patterns. The findings were partially replicated in an independent physically active older adults cohort, demonstrating that ML applied to standard laboratory data can enable scalable, low-cost precision prevention without specialized assays.
What the study was
- Study design
- Longitudinal cohort study with unsupervised ML clustering; replication cohort validation
- Population
- Community-dwelling older adults; Toledo Study for Healthy Ageing (TSHA) n=1491 + EXERNET replication cohort
- Sample size
- 1491
- Category
- Prevention
- Maturity
- Validated
- Journal
- npj aging
Why it surfaced
Demonstrates that ML on standard routine blood panels (CBC + biochemical panel) stratifies aging-related health trajectories and mortality in a large longitudinal cohort with replication. Near-term implementable because it uses universally available routine labs rather than specialized assays. Addresses the CBC/ML gap noted in recent run streak.
A plain-language summary of published research — not medical advice. Talk to a clinician about your own care.