> ## Content Index
> Fetch the complete content index at: https://www.coios.me/llms.txt
> Use this file to discover other available public pages before exploring further.

# ClinicalAge, a deep learning ageing measure built from UK electronic health records
- URL: https://www.coios.me/clinicalage-a-deep-learning-ageing-measure-built-from-uk/
- Published: 2026-11-15T11:37:05.000Z
- Updated: 2026-11-15T12:17:08.000Z
- Description: ClinicalAge, trained to mimic PhenoAge, was associated with all 20 non-cancer diseases studied in UK Biobank and CPRD.
- Author: Daniel Ryan
- Tags: Data, models and systems, What linked data reveals, AI in medicine, United Kingdom, #peer-reviewed, #new, #item, #source-2026-11, #score-70

A transformer model was trained on health records of 320,891 UK Biobank participants to produce ClinicalAge, using PhenoAge as reference. External validation used 4,728,376 people in CPRD Aurum primary care records. ClinicalAge correlated strongly with PhenoAge and was associated with 17 functional measures and ten ageing biomarkers. It was associated with all 20 non-cancer diseases in both datasets, and with all-cause mortality. Of 11 cancers, nine were associated in UK Biobank and ten in CPRD, two inversely in each. The model was trained to reproduce PhenoAge rather than to predict disease or death directly.

*Why it is interesting: It derives a biological age measure from data already collected in routine care. Validation in nearly 5 million primary care patients tests whether it holds beyond the development cohort.*

Source

[The Lancet Digital Health, 12 November 2026](https://doi.org/10.1016/j.landig.2026.101060?ref=coios.me)

Paper

ClinicalAge, a framework for modelling ageing trajectories from longitudinal electronic health records in the UK: a model development and validation study

Authors

Lian J, Fan Z, Fan W, Rao S, Yang Q, Zeng G et al.

DOI

10.1016/j.landig.2026.101060

Type

Peer-reviewed article

Design

Model development and validation: transformer trained on UK Biobank linked EHR (n=320,891, recruited 2006-2010) with external validation in CPRD Aurum (n=4,728,376)

Verdict

New finding

Driver

[What linked data reveals](https://www.coios.me/d-linked-data/)

Driver

[AI in medicine](https://www.coios.me/d-ai-medicine/)

Abstract — the authors' own words, CC BY

Background

Ageing is a key determinant of morbidity and mortality. In this study, we aimed to develop and assess a new ageing biomarker, ClinicalAge, derived from routinely collected longitudinal electronic health record (EHR) data. Methods

In this model development and validation study, we used UK Biobank data for model development and UK Clinical Practice Research Datalink (CPRD) Aurum data for model validation. The UK Biobank is a longitudinal population-based cohort of approximately 500 000 individuals aged 40-69 years at recruitment (March 16, 2006, to Oct 1, 2010), with linked EHR from the November 2022 UK Biobank data release. CPRD Aurum is a primary care EHR database containing anonymised longitudinal data from over 30 million patients across the UK, including diagnoses, medication, procedures, and laboratory results. A transformer-based deep learning model was trained from longitudinal EHR data to generate ClinicalAge using the validated biomarker PhenoAge as reference. We used a UK Biobank cohort for model development and internal validation, and three CPRD Aurum cohorts for external validation, longitudinal trajectory analysis, and comparison with the electronic frailty index. ClinicalAge was evaluated for associations with 17 functional measures, ten biological ageing biomarkers, 31 ageing-related diseases (11 cancer diagnoses and 20 non-cancer diseases), and all-cause, cancer, and cardiovascular mortality. We further examined longitudinal ClinicalAge trajectories and investigated clinical and disease-specific factors associated with accelerated ageing. Findings

The UK Biobank cohort included 320 891 participants (median chronological age 58·0 years \[IQR 51·0-63·0\]; median PhenoAge 56·0 years \[48·6-62·0\]; 173 067 \[53·9%\] were female; and 147 824 \[46·1%\] were male) and the CPRD cohort included 4 728 376 individuals (51·0 years \[44·0-60·0\]; 2 293 243 \[48·5%\] were female and 2 435 133 \[51·5%\] were male). ClinicalAge had an R 2 score of 0·819 with PhenoAge and showed a significant association with the 17 functional measures and ten biological ageing biomarkers. It was also significantly associated with the incidence of all 20 non-cancer diseases in both datasets, as well as nine (82%) of 11 cancers in UK Biobank (seven positive associations and two negative) and ten (91%) of 11 cancers (eight positive associations and two negative) in CPRD. Moreover, ClinicalAge was significantly associated with all-cause mortality (UK Biobank adjusted hazard ratio \[HR\] 1·23 \[95% CI 1·21-1·24\]; CPRD adjusted HR 1·36 \[1·36-1·36\]), cancer mortality (UK Biobank adjusted HR 1·12 \[1·10-1·14\]; CPRD adjusted HR 1·17 \[1·16-1·17\]), and cardiovascular mortality (UK Biobank adjusted HR 1·33 \[1·29-1·36\]; CPRD adjusted HR 1·36 \[1·35-1·36\]), after adjustment for clinical covariates and lifestyle factors. Midlife was identified as a key window for divergence, with early acceleration in individuals developing diabetes, hypertension, depression, or lifestyle disorders. Interpretation

ClinicalAge is a non-invasive biomarker that leverages routinely collected EHR data to quantify ageing across the life course. It shows strong associations with ageing-related biomarkers, functional decline, disease risk, and mortality. Its interpretability and anticipatory signal offer a practical tool for the early identification of individuals at risk of accelerated ageing, enabling targeted preventive strategies such as lifestyle modification, cardiometabolic control, and mental health support. Funding

None.