A deep-learning retinal aging biomarker for cognitive decline and incident dementia.
Level 3 - non-randomized controlled study
Prospective longitudinal cohort study with an external replication cohort
PubMed 40042460 · doi:10.1002/alz.14601
What was done
Researchers developed RetiPhenoAge, a deep-learning-derived retinal aging biomarker calculated from fundus photographs. They evaluated its association with 5-year incident cognitive decline and dementia using competing risk analyses in a Singapore memory-clinic cohort (n = 510). Findings were replicated for incident dementia in the UK Biobank (n = 33,495). The authors also tested associations between RetiPhenoAge, brain MRI markers (cerebral small vessel disease and neurodegeneration/atrophy), and plasma proteomic aging profiles.
What was found
In the memory-clinic cohort, RetiPhenoAge was associated with incident cognitive decline (subdistribution hazard ratio [SHR] 1.34, 95% CI 1.10–1.64, p = 0.004; 155 events) and incident dementia (SHR 1.43, 95% CI 1.02–2.01, p = 0.036). In the UK Biobank replication cohort, RetiPhenoAge similarly predicted incident dementia (SHR 1.25, 95% CI 1.09–1.41, p = 0.008). Significant associations were also observed between the biomarker, cerebral small vessel disease, brain atrophy, and aging-related plasma proteomic profiles.
Why it matters
Retinal imaging offers a fast, non-invasive approach to estimate biological aging and screen for dementia risk prior to symptom onset. Demonstrating predictive value across both a clinical memory cohort and a large population-based biobank reinforces the potential of retinal photography in neurological risk stratification.
Limits
The initial clinic cohort was relatively small (n = 510) and limited to a single geographic setting (Singapore), although supplemented by UK Biobank data. Specific effect sizes and numeric estimates for the MRI and proteomic correlations were not reported in the abstract. As an observational study, residual confounding cannot be excluded, and clinical utility for individual-level decision-making requires prospective clinical validation.
Cited by
- partial Artificial intelligence analysis of retinal photographs can predict the onset of Alzheimer's disease 5 to 7 years in advance.