Medical-record aging score linked to mortality in Hungarian study

A study published September 28 in npj Aging describes an aging score built from routine medical histories, offering a potential way to study age-related health differences without specialized laboratory testing. Researchers found that the score’s age-acceleration measure was associated with death from any cause.

The model, called EHR-AGE, draws on Hungarian diagnosis records covering approximately 9.5 million people between 2010 and 2021. Its immediate significance is methodological: information already collected during health care may contain useful signals about aging-related risk.

### What the researchers measured

In this retrospective modeling study, researchers trained machine-learning algorithms to predict chronological age from the preceding three years of ICD-10 diagnosis history. In independent test samples, EHR-AGE achieved a correlation of 0.89 with actual age and a mean absolute prediction error of 7.78 years.

The team also examined the relationship between its age-acceleration measure and all-cause mortality. The journal has released the work as an accepted, peer-reviewed article ahead of its final edited version.

### Prediction and clinical usefulness are different tests

A strong correlation across a population does not mean a model precisely measures each person’s biological condition. An average age-prediction error approaching eight years is particularly relevant when interpreting small differences between an estimated age and a birthdate-based age.

There are also separate questions about what a score captures and what clinicians could usefully do with it. A medical-history model may reflect accumulated disease, patterns of diagnosis, or contact with health services. Distinguishing those possibilities from underlying aging biology is an important validation task; the mortality association alone cannot resolve it.

For prevention research, the practical question is whether such a score adds useful information beyond existing assessments. That requires comparisons with simpler measures, testing in other health systems, and evidence that acting on the results improves outcomes. These are requirements for clinical translation, rather than benefits established by an age-prediction result.

### A broader challenge for aging clocks

Earlier research illustrates why those comparisons matter. In a February 2024 summary, the National Institute on Aging described a study of more than 3,500 participants in the U.S. Health and Retirement Study. Researchers compared DNA-methylation aging clocks with social, behavioral and other health measures.

Newer generations of those clocks were associated with outcomes including functional limitations and impaired cognition. However, demographic, socioeconomic, mental-health and behavioral factors were comparable—and often stronger—predictors of later-life health. That study concerned different measurements and a different population; it provides context for evaluating aging scores, rather than independent validation of EHR-AGE.

Electronic records also already support other forms of risk assessment. A 2024 study in the Journal of the American Geriatrics Society evaluated an Elders Risk Assessment score in 12,885 primary-care patients aged 65 or older. Its endpoint was death or intensive-care admission within one year. Such tools provide relevant comparators when considering whether an additional age-based score offers practical value.

### What remains unresolved

The new study’s observational design supports an association, not a conclusion that changing the score would extend life. Likewise, successfully estimating age does not establish that a tool can monitor the benefits of an intervention.

For now, EHR-AGE is a research candidate for extracting aging-related information from existing records. Its healthspan relevance will depend on whether that information improves predictions and decisions beyond what clinicians can already learn from a patient’s medical history.