Retinal Biological Age and Future Health Risks — A 2024 Cohort Study

1. Higher RetiPhenoAge Was Associated With Greater Mortality Risk

In UK Biobank, 1,310 participants died during follow-up.

Participants in the highest quartile of RetiPhenoAge scores had a higher mortality hazard than those in the lowest quartile.

After adjustment for PhenoAge, chronological age, sex, body mass index, blood pressure, cholesterol, hypertension, diabetes, smoking and ethnicity, the associations remained statistically significant.

Compared with the lowest quartile, the highest quartile had:

• All-cause mortality: HR 1.92 (95% CI 1.42–2.61).
• Cardiovascular mortality: HR 1.97 (95% CI 1.02–3.82).
• Cancer mortality: HR 2.07 (95% CI 1.29–3.33).

These estimates compare groups defined by their RetiPhenoAge model scores, not by the difference between retinal age and chronological age.

They represent relative hazards between groups, not individual probabilities of death.

The cardiovascular mortality estimate had a relatively wide confidence interval, indicating uncertainty about the precise effect size.

2. Retinal Aging Patterns Were Associated With Cardiovascular Events

Higher RetiPhenoAge was also associated with subsequent cardiovascular events.

In the fully adjusted UK Biobank analysis, participants in the highest quartile had a 70% higher relative hazard of cardiovascular events compared with those in the lowest quartile.

Hazard ratio: 1.70 (95% CI 1.17–2.47).

However, the corresponding association with new cancer events was not statistically significant after full adjustment.

For cancer events, the highest-versus-lowest-quartile estimate was:

Hazard ratio: 1.02 (95% CI 0.89–1.15).

This distinction is important. The retinal marker was associated with cancer mortality but did not independently predict new cancer events in the fully adjusted UK Biobank analysis.

3. Findings Were Partly Replicated in Independent Populations

The researchers evaluated RetiPhenoAge in cohorts from Singapore and the United States.

In Singapore, higher RetiPhenoAge was associated with several outcomes, including cardiovascular mortality, myocardial infarction and end-stage kidney disease.

After full adjustment, associations remained statistically significant for myocardial infarction and end-stage kidney disease.

In the American AREDS cohort, the highest quartile of RetiPhenoAge was associated with cardiovascular mortality.

Hazard ratio: 1.92 (95% CI 1.01–3.64).

These independent cohorts support the reproducibility of selected prognostic associations across different populations.

However, the associations were not uniformly significant for every outcome, and their magnitudes varied.

Importantly, replication of these associations does not establish that the model is fully validated for individual clinical risk prediction.

Further evaluation of predictive discrimination, calibration and clinical usefulness is required.

4. RetiPhenoAge Showed Stronger Associations Than Several Other Aging-Related Measures

Researchers compared RetiPhenoAge with handgrip strength, leukocyte telomere length and physical activity.

For all-cause mortality, the age- and sex-adjusted hazard ratios per standard deviation in the risk-associated direction were:

• RetiPhenoAge: HR 1.49 (95% CI 1.34–1.67).
• Handgrip strength: HR 1.21 (95% CI 1.11–1.33).
• Telomere length: HR 1.08 (95% CI 1.01–1.15).
• Physical activity: HR 1.28 (95% CI 1.19–1.38).

RetiPhenoAge showed the strongest association in this comparison.

However, the comparison concerned standardised association strengths rather than comprehensive predictive performance.

A stronger hazard ratio does not automatically establish that a biomarker is more accurate, clinically useful or superior for individual risk assessment.

The study did not demonstrate that RetiPhenoAge outperforms validated cardiovascular risk calculators or comprehensive clinical assessments.

5. Retinal Age Was Associated With Metabolic Health and Smoking

Higher RetiPhenoAge scores were associated with several established health risk factors.

These included:

• Higher body mass index.
• Smoking.
• Diabetes.
• Hypertension.

The researchers also identified associations with selected blood metabolites, including glucose, lactate and citrate.

These findings suggest that retinal images contain information related to systemic metabolic and vascular health.

However, the associations do not establish that these factors directly cause changes in RetiPhenoAge or that changing them will necessarily reduce the score.

6. Retinal Photography Offers a Potentially Accessible Assessment Method

Retinal photographs can be obtained quickly and without blood sampling.

This makes retinal imaging attractive for future population-level health assessment.

The study demonstrated that an artificial intelligence model trained on retinal images could estimate patterns associated with PhenoAge.

In UK Biobank, the correlation between predicted RetiPhenoAge and measured PhenoAge was 0.71.

The mean absolute prediction error was approximately 5.5 years.

This error describes how closely the model estimated the blood-based PhenoAge measure. It does not establish the accuracy of an individual’s true biological age.

The model also tended to overestimate younger individuals and underestimate older individuals.

Further calibration and validation are therefore necessary before interpreting RetiPhenoAge as a reliable individual biological age measurement.

Biological aging does not occur identically across all tissues and organs.

The retina is particularly interesting because its blood vessels and neural structures can be observed directly through non-invasive imaging.

Artificial intelligence may identify subtle patterns in retinal photographs that reflect aspects of cardiovascular, metabolic and systemic health.

This creates the possibility of using existing imaging technology to obtain additional information relevant to healthy aging.

However, an important distinction must be maintained.

Retinal biological age is not the same as chronological age, and a single retinal age estimate does not measure how quickly someone is aging.

A higher RetiPhenoAge score does not mean that future disease is inevitable.

Similarly, a lower score cannot exclude cardiovascular disease, cancer or other health risks.

For longevity medicine, the most important question is not whether artificial intelligence can generate an age-related number from a retinal photograph.

It is whether that number provides reliable, actionable information beyond established medical assessment.

This study represents meaningful progress in retinal aging research. Its findings justify further prospective validation, but they do not yet support replacing established preventive investigations with retinal age testing.

This study provides substantial observational evidence that retinal photographs contain information associated with systemic health and future mortality.

Its principal strength is the combination of a large UK Biobank dataset with external evaluation in two geographically distinct cohorts.

The findings are particularly interesting for cardiovascular prevention because associations with cardiovascular mortality and events persisted after adjustment for multiple established risk factors.

Nevertheless, several limitations are important.

First, association is not equivalent to clinical predictive utility.

The study reported hazard ratios and comparisons with selected aging-related measures. It did not establish whether RetiPhenoAge meaningfully improves discrimination, calibration or clinical decision-making beyond validated cardiovascular risk models.

Second, RetiPhenoAge is an image-derived estimate of PhenoAge, rather than an independently validated measure of biological aging.

The model was trained to reproduce a composite measure based on chronological age and blood biomarkers.

Its results may reflect retinal patterns associated with metabolic, vascular and inflammatory health rather than an independently measurable rate of biological aging.

Third, image quality and ocular conditions may influence results.

Cataracts reduced the magnitude of several associations. Severe lens opacity can also compromise retinal photography.

Fourth, external validation was encouraging but incomplete.

The independent cohorts demonstrated replication of selected associations, not complete validation of individual clinical risk prediction.

The results varied between populations and outcomes. Further studies are required to establish reproducibility across age groups, ethnicities, imaging systems and clinical settings.

Fifth, the study did not demonstrate better patient outcomes.

There was no evidence that using RetiPhenoAge to guide treatment or preventive interventions reduces cardiovascular events, cancer or mortality.

Finally, several authors disclosed commercial interests involving retinal artificial intelligence technologies. These disclosures do not invalidate the findings but reinforce the importance of independent replication.

RetiPhenoAge should therefore be regarded as a promising research biomarker, not an established clinical screening or risk-stratification test.