**Background:** There is a need for novel, non-invasive, and inexpensive methods to study the early pathophysiology of dementia. The retina is considered a window to the brain, and retinal neurodegeneration, measured as thinning of inner retinal layers via optical coherence tomography (OCT), may precede clinical dementia. Prior prospective population-based data were limited, with only one study reporting on fewer than 90 dementia cases. This study aimed to investigate, in a large individual participant data meta-analysis, whether lower thickness of the macular retinal nerve fiber layer (mRNFL) and macular ganglion cell-inner plexiform layer (mGCIPL) is associated with incident all-cause dementia and Alzheimer's disease (AD), and whether these associations differ by age, sex, or APOE genotype.
**Methods:** Unpublished data from four prospective, population-based cohort studies were pooled: UK Biobank (n=82,860), Tromsø Study (n=10,180), Maastricht Study (n=7,689), and Alienor Study (n=613). The final analytic sample for mGCIPL included 69,955 participants (1,087 incident all-cause dementia cases; 520 incident AD cases) and for mRNFL included 69,566 participants (985 all-cause dementia; 450 AD). Median follow-up was 11.3 years (IQR 8.8–11.5). Retinal thickness was assessed at baseline using OCT; global mRNFL and mGCIPL thickness were calculated as the mean of both eyes. Dementia cases were identified via hospital records, death registries, or neuropsychological testing with clinical confirmation. Cox proportional hazards models were used, with results expressed as hazard ratios (HR) per standard deviation (SD) lower thickness. Models were adjusted for cohort, age, sex, educational level, spherical equivalent (Model 1), and additionally for BMI, diabetes, alcohol consumption, smoking, antihypertensive medication, and systolic blood pressure (Model 2). Multiple imputation was used for missing covariates. Secondary outcomes included vascular dementia and late-onset dementia (onset ≥65 years). Additional analyses examined change in retinal thickness over time (available in a subset of 3,309 Tromsø participants), exclusion of participants with eye diseases, and competing risk analyses.
**Key Results:** Baseline characteristics: mean age 58.1 years (SD 8.8), 47% men, 53% women. Participants with incident dementia were older (mean age 69.5 vs. 57.9 years) and had a more adverse risk profile. After full adjustment (Model 2), lower baseline mRNFL thickness was significantly associated with a higher incidence of all-cause dementia (HR 1.10 per SD lower; 95% CI 1.02–1.17) and AD (HR 1.11; 95% CI 1.001–1.23). Lower baseline mGCIPL thickness was not significantly associated with all-cause dementia (HR 1.04; 95% CI 0.99–1.10) or AD (HR 1.01; 95% CI 0.94–1.09). For secondary outcomes, lower mRNFL was significantly associated with late-onset dementia (HR 1.08; 95% CI 1.01–1.16) but not with vascular dementia (HR 1.08; 95% CI 0.85–1.36). Lower mGCIPL was not significantly associated with either secondary outcome. In the subset with repeated OCT measurements (n=3,309; 64 dementia cases), greater thinning over time in mRNFL (HR 1.19 per SD; 95% CI 1.05–1.35) and mGCIPL (HR 1.17; 95% CI 0.999–1.39) was associated with higher dementia risk. Associations did not differ significantly by age, sex, educational level, or APOE genotype. Results were robust in sensitivity analyses, including after excluding participants with eye diseases and in competing risk analyses.
**Clinical Implications:** This large-scale, prospective study provides strong evidence that retinal neurodegeneration, particularly thinning of the mRNFL, precedes the onset of clinical dementia by many years. The findings suggest that OCT-based retinal imaging could serve as a non-invasive, inexpensive, and widely accessible biomarker for early dementia risk assessment and for studying the early pathophysiology of Alzheimer's disease and other dementias. The association was modest (approximately 10% increased risk per SD lower mRNFL thickness), indicating that retinal thickness alone is not sufficient for clinical prediction but may be useful in combination with other risk factors. The lack of significant association for mGCIPL may reflect that synaptic loss (mRNFL) precedes neuronal soma loss (mGCIPL) in the neurodegenerative cascade. These results support the concept of the retina as a window to the brain and highlight the potential of retinal imaging in dementia research and prevention trials.