Can the Retina Reveal How We Are Aging? Retinal Age Gap and Mortality

1. Artificial intelligence could estimate age from retinal photographs

The model showed a strong correlation between retinal age and chronological age:

r = 0.81

The mean absolute error was 3.55 years. This means that the absolute difference between the age predicted from the retinal image and the participant’s actual age averaged approximately 3.5 years.

The model was not equally accurate across all ages. It tended to predict older ages for younger participants and younger ages for older participants.

Interestingly, when researchers examined which parts of the retinal images contributed to age prediction, the model frequently focused on regions around the retinal blood vessels.

This suggests that vascular features may contribute to the age-related information detected by the algorithm.

⸻

2. A larger retinal age gap was associated with higher all-cause mortality

During the median 11-year follow-up, 1,871 participants (5.21%) died.

After adjustment for age, sex, ethnicity, socioeconomic factors, education, smoking, physical activity, general health and major medical conditions, each additional year of retinal age gap was associated with approximately 2% higher relative hazard of all-cause mortality.

Adjusted HR: 1.02 per additional retinal year

This does not mean that one additional year of retinal age directly causes a 2% increase in mortality.

It means that people whose retinas appeared older relative to their chronological age tended, on average, to have a higher subsequent mortality risk.

⸻

3. The relationship was stronger at larger positive retinal age gaps

The researchers also divided participants into four groups according to their retinal age gap.

Compared with participants in the lowest group:

the second group did not have significantly higher mortality,

the third group had approximately 21% higher adjusted mortality hazard,

the highest group had approximately 35% higher adjusted mortality hazard.

The relationship was not simply linear.

When researchers examined the full range of retinal age gaps, the association followed a J-shaped pattern, with mortality risk becoming more apparent as retinal age increasingly exceeded chronological age.

This is important because retinal age gap should not be interpreted as a simple scale where every additional retinal year carries exactly the same meaning.

⸻

4. The association differed according to cause of death

Among the 1,871 deaths:

321 were attributed to cardiovascular disease,

1,018 to cancer,

532 to other causes.

After full adjustment, retinal age gap was not significantly associated with cardiovascular mortality or cancer mortality when these were examined separately.

The association was stronger for deaths from causes other than cardiovascular disease or cancer.

Each additional retinal year was associated with approximately 3% higher relative hazard of non-cardiovascular, non-cancer mortality.

Participants in the third and highest retinal age-gap groups had approximately 49% and 67% higher adjusted hazards, respectively, for mortality from these other causes compared with the lowest group.

The study was not designed to establish why this association occurred.

⸻

Many proposed measures of biological aging require blood tests, molecular analysis or advanced imaging.

Retinal photography is rapid, non-invasive and already widely used in clinical medicine.

This makes the possibility of extracting additional aging-related information from a retinal photograph particularly interesting.

However, there are several important reasons why retinal age was not yet ready to function as a stand-alone biological-age test in this study.

First, each participant’s retina was assessed at one point in time. The researchers could therefore compare retinal age with chronological age, but they could not determine whether an individual’s retina was actually aging faster or slower over time. Measuring the rate of retinal aging would require repeated retinal photographs over several years.

Second, the algorithm was developed using UK Biobank participants and was not externally validated in a completely independent population. A model that performs well in one large dataset may perform differently in people from other populations, age groups or clinical settings.

Third, UK Biobank participants are volunteers and tend to be healthier than the general population. The results therefore may not translate perfectly to every population.

Finally, this was an observational study. A larger retinal age gap was associated with higher mortality, but the study cannot establish that an older-appearing retina itself causes poorer health or shorter survival. Both could reflect other underlying biological or health processes.

These limitations do not remove the importance of the finding. They define what the study can — and cannot — tell us.

The study provided an important proof of concept that age-related information extracted from retinal images may contribute to the assessment of aging and health risk.

Further studies, including external validation and repeated measurements over time, are needed before retinal age can be interpreted as an established clinical measure of biological aging.

⸻

This study introduced an important idea: a retinal photograph may contain information about aging that extends beyond the eye itself.

The retina is particularly interesting because it allows direct, non-invasive visualization of blood vessels and neural tissue.

Artificial intelligence can analyse complex patterns within these images that may not be apparent during routine visual examination.

The study therefore provides evidence that retinal imaging can generate an age-related signal associated with future health outcomes.

But retinal age needs careful interpretation.

A retinal age of 65 in a 60-year-old does not mean that the person’s entire body is biologically 65 years old.

Retinal age is an estimate generated by a particular artificial-intelligence model from a retinal photograph. The retinal age gap simply describes the difference between this estimate and chronological age.

The study showed that this difference was associated with mortality.

It did not demonstrate that retinal age is a direct measurement of whole-body biological age.

⸻