Biological Age Tools Reveal Preclinical Health Risks

A review by Cheema et al. published in The Journal of Clinical Investigation (JCI 2026;136(12):e205777, doi:10.1172/JCI205777) surveys contemporary approaches to estimating biological age (BA). The paper documents a progression from simple functional tests such as gait speed and frailty measures to molecular and digital approaches, and explicitly categorizes epigenetic clocks into generations: first-generation clocks trained on chronological age (examples: Horvath, Hannum), second-generation clocks tied to phenotypic outcomes (examples: PhenoAge, GrimAge), and third-generation clocks measuring pace or rate of aging (example: DunedinPACE). The review lists data modalities now in active development for BA: DNA methylation, proteomics, blood biomarkers, medical imaging, wearables, and AI analysis of clinical text and scans. The authors state that prospective validation is required before routine clinical deployment and note that the role of these tools in standard care remains unresolved.

Per the review, methods differ by target and training objective: earlier models predict chronological age, later models predict mortality or morbidity, and newer approaches aim to quantify current aging rate or organ-specific decline. The review summarizes methodological challenges including cross-cohort calibration, confounding by adiposity, differential predictive performance across populations, and the need for longitudinal, prospective outcome data to establish clinical utility. The paper also highlights digital and AI-enabled clocks that repurpose imaging and clinical-text data but emphasizes variable evidence supporting each modality.

Editorial analysis: Reviews that map a measurement field typically accelerate multi-disciplinary uptake by clarifying terminology and performance gaps. For data scientists and ML engineers, the JCI review consolidates which label definitions and outcome targets are in use, and it underscores common data-quality and generalization challenges when models are trained on surrogate labels such as chronological age or cross-sectional biomarkers rather than long-term clinical endpoints.

Editorial analysis: Observers should track prospective cohort studies and randomized prevention trials that include BA measures as prespecified endpoints, cross-population benchmarking efforts, and open datasets that link multi-omic, imaging, wearable, and EHR-derived streams. For practitioners, improvements in covariate adjustment, calibration across ancestries, and standardized reporting of effect sizes versus established clinical risk scores will be key indicators of maturity.