A study in The Lancet Digital Health describes ClinicalAge, a transformer-based deep learning model that estimates how fast a person is ageing from routine electronic health record data, developed in UK Biobank and tested in UK primary care records from CPRD Aurum1. The model was trained to track PhenoAge, an existing validated ageing biomarker, so that a similar signal can be read from records a health system already holds rather than from new blood tests1.
What the study found
The development cohort drew 320,891 UK Biobank participants, with a median chronological age of 58.0 years1. External validation used three CPRD Aurum cohorts covering 4,728,376 people with a median age of 51.0 years, taken from a primary care database that holds anonymised data on more than 30 million patients across the UK1.
ClinicalAge reached an R2 of 0.819 against PhenoAge1. It was significantly associated with all 17 functional measures and all ten biological ageing biomarkers the team tested1. It was also linked to the incidence of each of the 20 non-cancer diseases examined in both datasets, and to nine of 11 cancers in UK Biobank and ten of 11 in CPRD, with two of the cancer associations negative in each dataset1.
The mortality signal held after adjustment for clinical covariates and lifestyle factors. In CPRD, the adjusted hazard ratio for all-cause mortality was 1.36, and 1.36 again for cardiovascular deaths; in UK Biobank the figures were 1.23 and 1.331. For cancer mortality, the adjusted hazard ratios were 1.12 in UK Biobank and 1.17 in CPRD1. Tracking trajectories over time, the authors identified midlife as the point where ageing paths split, with earlier acceleration in people who went on to develop diabetes, hypertension, depression or lifestyle disorders1.
What it means for the market
The commercial interest lies in the input. ClinicalAge runs on data that GP and hospital systems already collect, so a health system or population health vendor could in principle score whole registered lists without new tests or appointments1. The authors frame it as a way to find people at risk of faster ageing early and target prevention such as lifestyle change, cardiometabolic control and mental health support1. That puts it alongside frailty scoring and risk stratification tools: the team compared it with the electronic frailty index1.
For buyers, the study does not yet show that acting on a ClinicalAge score changes outcomes. It reports associations, not the results of an intervention, and it gives no information on cost, deployment or regulatory status1.
Caveats
The evidence comes from UK data only, and UK Biobank recruited people aged 40 to 69 between 2006 and 20101. The model was built to mirror PhenoAge, so it inherits that biomarker's definition of ageing1. The authors describe the work as a model development and validation study, and they report no funding for it1.
Sources
This briefing summarises publicly available research and reporting for information only. It is not medical, investment or legal advice. Follow the sources listed for the primary research and reporting.