Retinal Age and Systemic Aging: What AI Can Reveal from the Retina

1. AI estimated age accurately — including in an independent population

In UK Biobank, retinal age strongly correlated with chronological age:

r = 0.90, with a mean absolute error of 2.85 years.

The model was then tested in 4,757 participants from the independent Rotterdam Study. It remained accurate, with a correlation of r = 0.85 and a mean absolute error of 3.35 years.

This external validation strengthens confidence that the model’s ability to estimate age was not specific to UK Biobank.

However, it validates the model’s ability to predict chronological age from retinal images. It does not by itself validate retinal age as a clinical test of biological age.

2. An older retinal age was associated with metabolic and inflammatory health

A higher retinal age gap was associated with several less favorable health characteristics, including:

greater lifetime smoking exposure,

higher HbA1c and glucose,

higher body mass index and waist-to-hip ratio,

higher systolic blood pressure,

higher white blood cell count,

and higher C-reactive protein.

Higher estimated VO₂max, a measure of cardiorespiratory fitness, was associated with a lower retinal age gap.

Together, these findings suggest that an older-appearing retina is associated with several aspects of metabolic, vascular and inflammatory health.

They do not establish that these factors cause retinal aging or that improving them will necessarily reduce retinal age.

3. Retinal age gap was associated with future disease and mortality

The researchers also examined health outcomes occurring during long-term follow-up.

People who already had a particular disease were excluded from the analysis of new cases of that condition.

A higher retinal age gap was associated with subsequent all-cause mortality and with new cases of several conditions, including:

cardiovascular disease,

chronic obstructive pulmonary disease,

stroke,

dementia,

thrombosis,

ischemic heart disease,

and cancer.

These associations remained after accounting for established clinical risk factors.

However, adding retinal age gap to conventional risk factors generally produced only modest improvements in prediction, and improvement was not statistically significant for every outcome.

This distinction is important. Retinal age may add information about future health risk, but this study does not show that it can replace established clinical risk assessment.

4. Retinal aging patterns differed between women and men

Age prediction itself performed similarly in women and men, but some of the health associations with retinal age gap differed.

In men, retinal age gap showed stronger relationships with metabolic syndrome and related metabolic factors.

In women, several associations changed with menopausal status. Postmenopausal women had a higher retinal age gap than premenopausal women, and some of their health associations became more similar to those observed in men.

These findings suggest that biological differences between women and men — potentially including hormonal changes across menopause — may influence retinal aging patterns.

However, the study does not demonstrate that menopause or declining estrogen causes accelerated retinal aging.

5. Genetics and retinal imaging provided clues about the biological signal

The researchers identified 40 genome-wide significant genetic loci associated with retinal age gap.

The implicated biological pathways included metabolism, longevity, neurodegeneration, cellular senescence, retinal disease and retinal vascular characteristics.

The estimated genetic contribution to variation in retinal age gap was approximately 28%.

The AI model also appeared to obtain age-related information from several retinal regions, particularly the macula, optic disc and retinal blood vessels.

However, conventionally measured retinal characteristics explained only part of the signal.

This suggests that retinal age gap reflects biologically meaningful information, but the precise biological processes captured by the AI model are still not fully understood.

Retinal photography is rapid and non-invasive and provides direct images of neural and vascular tissue. This makes it an attractive potential source of additional information about aging.

The 2026 study strengthens the evidence for retinal age, but four limitations remain particularly important.

First, retinal age is not a direct measurement of whole-body biological age. The model predicts chronological age from retinal photographs and calculates how far an individual’s result differs from that prediction.

Second, association does not establish causation. A higher retinal age gap was associated with poorer health and future disease, but we do not yet know whether retinal changes contribute to these conditions, reflect underlying disease processes, or both.

Third, generalizability remains limited. UK Biobank and Rotterdam Study participants were predominantly of European ancestry and relatively healthy. The findings require confirmation in more diverse and clinical populations.

Fourth, clinical usefulness has not yet been established. Retinal age gap added some predictive information, but improvements over established risk factors were generally modest. The study did not show that using retinal age improves clinical decisions or health outcomes.

For longevity medicine, retinal age is therefore best considered an additional source of information about aging and systemic health, rather than a stand-alone biological-age score.

This study substantially extends earlier retinal-age research.

It shows that an AI model can estimate age from retinal photographs with good accuracy and that this ability remains strong in an independent population.

More importantly, retinal age gap was associated with multiple dimensions of health — including metabolic and inflammatory markers, physical fitness, future disease and mortality.

But the meaning of the number must remain clear.

The model was trained to predict chronological age from the retina. It was not trained to measure the biological age of the whole body.

Therefore, a retinal age gap of +5 years does not mean that a person’s body is biologically five years older.

It means that the retinal features detected by the model appear older than expected for that person’s chronological age. The study shows that this difference carries health-related information.