Fares Antaki.
Nima Ghadiri (NG): The ‘retinal age gap’ (RAG) is a relatively new concept. Based on your study [1], how confident are you in proposing this as a non-invasive, accessible biomarker for accelerated biological ageing in patients with neuropsychiatric disorders, and what are the next critical steps needed for its clinical validation?
Fares Antaki (FA): The RAG remains a research tool, though it may eventually help stratify risk in neuropsychiatric diseases such as schizophrenia. Its clinical use is still premature and requires further validation. To move toward routine application, key next steps include conducting longitudinal studies to assess its association with disease progression and severity, as well as its clinical utility. Validation should also extend to younger patients with schizophrenia or first-episode psychosis, since findings in older populations are influenced by metabolic comorbidities. Future work should explore the biological pathways and retinal features driving RAG predictions, potentially through explainable AI, to improve specificity and interpretation. Finally, standardised imaging protocols and hardware-agnostic algorithms are needed to ensure reproducibility across settings and devices.
NG: Your key finding is that the increased RAG in schizophrenia patients is largely attributable to the increased prevalence of hypertension and diabetes mellitus. Could you elaborate on why this distinction (that comorbidity, not just schizophrenia itself, is the primary driver) is so crucial for both the scientific understanding of the disease and for clinical practice?
FA: The finding that the increased RAG in schizophrenia is mainly driven by the higher prevalence of hypertension and diabetes mellitus is scientifically important because it refines our understanding of accelerated biological ageing in this population. While schizophrenia itself is linked to faster brain ageing, our study suggests that the retina (and the RAG) is capturing the effects of modifiable metabolic comorbidities and reflecting cumulative microvascular and neural damage. Our findings redirect research toward uncovering the shared biological mechanisms that connect metabolic dysfunction with accelerated senescence in neuropsychiatric diseases. Clinically, this highlights an actionable target for intervention: the clinical focus should shift toward more intensive management of diabetes and hypertension disorders to help slow biological ageing in this population.
NG: Your study uses a deep learning model trained on a healthy cohort to predict retinal age. Given that your study population (AlzEye) is an older, hospital-attending cohort with a higher burden of disease, how might the selection bias of the training data have influenced the observed RAGs, especially the high mean gap in the younger controls (7.93 years)?
FA: The deep learning model used to calculate the RAG was trained on over 130,000 fundus images from more than 29,000 healthy individuals aged 15–91, including participants from the UK Biobank and Chinese cohorts. Because this training population was relatively healthy, applying the model to the AlzEye cohort (an older, hospital-based group with more comorbidities) introduces bias that likely affects the observed RAG. In younger controls, for example, we found a mean gap of 7.93 years, even though retinal age should normally match chronological age. This can be explained by: (1) higher disease burden in the AlzEye population, even among controls; and (2) a known regression dilution effect, where the model overestimates age in younger subjects and underestimates it in older ones. These factors together highlight the need to control for chronological age in all subsequent analyses.
NG: Previous work by Krukow, et al. suggested that accelerated retinal ageing was associated with antipsychotic medication dosage [2]. While your study did not have data on smoking or medication, do you believe the pharmacological treatment of schizophrenia may be an unmeasured confounder contributing to the high prevalence of diabetes and hypertension, and thus, the larger RAG?
FA: Yes, it is likely that antipsychotic treatment acts as an unmeasured confounder contributing to the larger RAG observed in schizophrenia. Antipsychotic medications are known to induce metabolic side-effects such as weight gain, insulin resistance and dyslipidemia, leading to a higher prevalence of diabetes and hypertension in this population. These metabolic disturbances, in turn, drive the accelerated biological ageing signal captured by the RAG, as our study shows. The stronger effect of diabetes on the RAG in younger patients may further reflect a more aggressive metabolic profile, often accompanied by obesity and early microvascular complications. It is also possible that younger individuals with schizophrenia carry greater genetic vulnerability to neural and vascular atrophy, compounding these effects.
NG: If validated, could RAG screening be used clinically, for instance to identify schizophrenia patients at higher risk of systemic comorbidities or early mortality?
FA: If validated, RAG testing in clinic-attending individuals could help identify patients at higher risk of systemic disease. The RAG is a known marker of all-cause mortality, cardiovascular disease, stroke, metabolic disease and kidney failure. In schizophrenia, we found that the larger RAG mainly reflects modifiable comorbidities such as hypertension and diabetes. Clinically, the RAG could help flag patients who need tighter metabolic control. This approach targets preventable risk factors and may help slow biological ageing and reduce serious health complications in this vulnerable group.
NG: You identify that many psychiatric patients may not attend routine ophthalmic imaging. Might the true average RAG be higher for the entire schizophrenia population, and how could such oculomic biomarkers be integrated practically into mental health or general medical care?
FA: Indeed, the true average RAG in the general schizophrenia population may be higher than what we observed in the AlzEye cohort, as many individuals with schizophrenia do not regularly access eyecare (and are not in the AlzEye cohort). In practice, this non-invasive biomarker could be integrated into mental health or general medical settings to screen for patients at increased risk of accelerated biological ageing and metabolic dysfunction. Such use would support targeted interventions, including stricter management of metabolic disorders like diabetes and hypertension, to improve long-term systemic and ocular health in this vulnerable group.
NG: Do you foresee a future in which retinal imaging could help guide personalised management, for example flagging patients whose biological ageing is accelerated and who might benefit from metabolic interventions?
FA: Yes, the broader oculomics literature supports this vision [3]. Artificial intelligence-based retinal analytics like the RAG may enable early screening, monitoring and individualised care, helping clinicians tailor prevention or treatment strategies to each patient’s biological risk profile. Retinal imaging is fast and inexpensive, and it can be repeated to monitor biological responses to interventions over time. Conceptually, this could function like a Framingham-style risk score [4], where image-derived features are converted into a calibrated probability that guides the intensity of prevention and follow-up.
NG: Finally, what are the next steps for this field? Do you plan to explore longitudinal retinal ageing, or extend the approach to other psychiatric or neurodegenerative conditions?
FA: Future directions for oculomics, particularly for the RAG, involve additional validation, broader application and mechanistic insight. Large longitudinal studies are needed to confirm the RAG’s predictive value for disease onset and mortality, as suggested by associations with cardiovascular, metabolic and neurodegenerative diseases. Mechanistically, future work should investigate the biological pathways linking retinal and systemic ageing, apply explainable AI tools to reveal key predictive features, and use portable, low-cost imaging technologies such as smartphone fundus cameras to improve accessibility and real-world clinical integration.
References
1. Antaki F, Kerexeta-Sarriegi J, Reis APR, et al. The association of retinal age gap with schizophrenia: a cross-sectional analysis. Schizophr Res 2025:283:180–7.
2. Krukow P, Domagala A, Kiersztyn A, et al. The Retinal Age Gap as a Marker of Accelerated Aging in the Early Course of Schizophrenia. Schizophr Bull 2026;52(1):sbaf038.
3. Weinreb RN, Keane PA, Cooley A, et al. A Framework for Healthcare from the Eye: Oculomics as a Powerful Window to Systemic Health. Ophthalmology 2026:S0161-6420(26)00047-3.
4. D’Agostino Sr, RB, Pencina MJ, Massaro JM, Coady S. Cardiovascular Disease Risk Assessment: Insights from Framingham. Glob Heart 2013;8(1):11–23.
Declaration of competing interests: None declared.
