A Systematic Review Of Phenotypic And Epigenetic Clocks Used For Aging And Mortality Quantification In Humans
Research Summary
A systematic review, designed after PRISMA, that surveys the human biological clocks of ageing proposed to date – epigenetic clocks built on DNA methylation and phenotypic clocks built on routine clinical biomarkers – excluding cancer-risk (mitotic) clocks and clocks modelled on non-human samples, and reporting the predictive performance of the 33 that qualified together with the statistical or machine-learning methods behind each.
Who Or What Was Studied?
Published primary studies of human biological clocks, identified by a PubMed search run on 8 June 2023 plus citation tracing and Google Scholar: 134 records were screened, 87 were excluded on abstract and 14 after full-text review, leaving 33 clocks. Epigenetic clocks are grouped into first-, second- and third-generation designs, and phenotypic clocks are separated by whether chronological age is used as an input.
What Did The Research Show?
- First-generation methylation clocks predict chronological age with correlations typically of 0.96 or higher; the best reported values were 0.990 (Horvath 2020, about 36,000 CpG sites, n = 850; Zhang 2019, 514 CpGs, n = 13,566), and the next best (r = 0.98) used artificial neural networks.
- Second-generation clocks – PhenoAge, GrimAge and MetaboHealth – are trained on time-to-event data and are reported to associate with all-cause mortality more strongly than first-generation clocks, while third-generation clocks such as DunedinPoAm use longitudinal data to estimate the rate of ageing itself.
- Phenotypic clocks built from clinical measurements are reported to predict mortality more accurately than chronological age; models that take chronological age as an input reach correlations as high as 0.98, whereas those that do not range from about 0.72 to 0.91.
- The clinical biomarkers that recur with the largest model coefficients across phenotypic clocks include systolic and diastolic blood pressure, forced expiratory volume, albumin, C-reactive protein, creatinine, alkaline phosphatase and blood glucose.
- The authors conclude that methylation arrays remain relatively costly for routine clinical or hospital use, so phenotypic clocks built from standard clinical variables may offer the most utility in the short term, while neural-network clocks need interpretability tools such as SHAP to be usable.
Limitations And Interpretation
- The authors state they could not objectively compare performance across the heterogeneous cohorts and environments involved, and this review is not a meta-analysis.
- Mitotic clocks, non-human clocks and non-English publications were excluded, so the survey is deliberately partial.
- This is a review of measurement tools: it evaluates no drug or intervention, including rapamycin, and gives no protocol for applying a clock in practice.
Summary prepared from the published paper (abstract and full text where accessible). This is not medical advice — always consult the original publication.