This study derives from the NORFACE project, established in 2014 at Ace Alzheimer Center Barcelona (Ace) with the goal to investigate retinal biomarkers in AD using OCT imaging [28,29,30]. The current work was designed as a cross-sectional observational study using data from a large cohort of consecutive patients evaluated due to cognitive decline at Ace between June 2016 and March 2019. Participants were recruited through various sources, including the 1) Ace´s memory unit [31], 2) the Open House Initiative [32], 3) the FACEHBI study [33], and 4) the BIOFACE study [34].

The current study inclusion criteria were: age 50–95 years, fluency in Spanish or Catalan, presence of a consensus-based clinical diagnosis about the participant’s cognitive status, ability to complete a full ophthalmological examination and OCT scan and signed informed consent. Exclusion criteria included advanced dementia, defined as a Global Deterioration Scale (GDS) [35] score > 6.

Study participants underwent neurological, neuropsychological, and social evaluations at Ace. A multidisciplinary team of neurologists, neuropsychologists, and social workers reached a consensus diagnosis on each participant’s cognitive status [31]. Cognitive evaluations included the Spanish version of the Mini-Mental State Examination (MMSE) [36, 37], the memory section of the 7-Minute Test [38], the Neuropsychiatric Inventory Questionnaire (NPI-Q) [39], the GDS [35], the Clinical Dementia Rating (CDR) [40], the Blessed Dementia Scale [41], and the neuropsychological battery of Fundació ACE (NBACE) [42, 43]. Demographic and medical history data, including age, sex, years of formal education, smoking, hypertension, diabetes mellitus, dyslipidemia, heart disease, stroke, and chronic obstructive pulmonary disease (COPD), were collected. Neuroimaging, either brain magnetic resonance imaging (MRI) or head computed tomography (CT scan), was performed on all patients in order to assess brain atrophy patterns, cerebrovascular disease (including brain infarctions and white matter vascular burden) or other brain lesions.

Alzheimer's disease dementia (ADD) was defined using NIA-AA criteria [4]. Vascular dementia (VaD) was diagnosed based on NINDS-AIREN International Workshop Criteria [6]. The two dementia groups included patients in the GDS stages 4–6. Mild cognitive impairment (MCI) was defined using Petersen’s criteria [44] and the Cardiovascular Health and Cognition Study [45]. In particular, the MCI-AD group was characterized by memory impairment and the absence of other comorbidities that could explain the cognitive decline (probable amnestic MCI) with suspected underlying AD [46]. The MCI-Va group was defined based on the suspected underlying etiology of CV pathology. The cognitively unimpaired (CU) group included cognitively healthy individuals and those with subjective cognitive decline (SCD), defined by self-reported cognitive problems without impairment on standardized cognitive tests [47, 48]. All CU participants had a CDR of 0, MMSE ≥ 27 and strictly normal performance on the NBACE.

Study participants underwent a comprehensive neuro-ophthalmological evaluation alongside their neurological assessment. Conducted by an optometrist, the evaluation lasted approximately 20 min and included: 1) a review of ophthalmological history, including previous treatments and surgeries; 2) monocular visual acuity assessment in the right eye using the participant’s usual optical correction and a pinhole occluder, evaluated with the Early Treatment Diabetic Retinopathy Study (ETDRS) chart [49, 50]; 3) intraocular pressure (IOP) measurement with a rebound tonometer (iCare model) [51]; and 4) retinal examination using SS-OCT.

Visual acuity was assessed uniformly, regardless of cognitive status. All evaluations were performed by a single optometrist, who received training from an ophthalmologist. The ophthalmologist reviewed the history, examination results, and OCT images if abnormalities were detected, and confirmed diagnoses as needed. Both the ophthalmologist and neurologist were blinded to cognitive diagnoses. Only OCT data from the right eye were analyzed.

Ophthalmological exclusion criteria were the following: 1) conditions affecting retinal and/or choroidal measurements, such as glaucoma, age-related macular degeneration (AMD), and amblyopia; 2) IOP (intraocular pressure) ≥ 24 mmHg; 3) history of retinal surgery; 4) presence of OCT image artifacts; 5) refractive errors: patients with high myopia (< -6D) or high hyperopia (> + 6D) (as these extreme refractive errors are associated with significant variations in axial length, which can influence choroidal thickness measurements); [52, 53] 6) other causes, including non-glaucomatous optic neuropathy, inability to complete the ophthalmological exam, or absence of right eye OCT data.

Retinal and choroidal images were captured using the DRI Triton—Swept Source (SS) OCT (Topcon Co., Tokyo, Japan), focusing on the right eye without using pupil dilation. The DRI Triton SS-OCT automatically measures the thickness of retinal and choroidal layers, producing a detailed map. Additionally, the integrated non-mydriatic color fundus camera allowed simultaneous acquisition of fundus photographs during the OCT scan. The CSI protocol, which measures CT from Bruch's membrane to the choroid-scleral interface (CSI) with a focus on this latter boundary, was used. The SMARTTRACK feature was enabled to minimize motion artifacts, and the follow-up and enhanced depth imaging (EDI) modes were deactivated. An optometrist trained in OCT image interpretation evaluated the quality of the images, and only those rated with good or very good quality (scores 3 or 4) were included in the analysis. The optometrist did not perform fundoscopy but thoroughly recorded the ophthalmic history of each participant. Any abnormal OCT images were reviewed by an ophthalmologist, and patients with suspected retinal pathologies were referred for further examination. Both the ophthalmologist and neurologist were blinded to diagnoses.

For data analysis, the TRITON DRI-OCT software (Capture Software v.1.1.4.45475, Analysis Software v.10.1.3.43469) was used. The software’s automatic segmentation method calculated CT in the 9 ETDRS quadrants, centered on the fovea (Fig. 1). The subfoveal choroidal thickness (SFCT), which refers to the thickness of the choroid directly beneath the fovea and is typically measured at the thinnest point of the retina using OCT, was used as the central reference point for aligning the grid. CT was measured within a 1 mm central radius (Center) and two concentric circles representing the Inner ring (3 mm) and Outer ring (6 mm), covering a 6 × 6 mm area. Measurements were classified into Nasal (N), Temporal (T), Inferior (I), and Superior (S), and further divided into Inner (In) and Outer (Out) regions, resulting in 12 total measurements, including the average thickness, total volume, and SFCT. Average thickness refers to the average CT measured across the entire region of interest, calculated from multiple points within the ETDRS grid, which is divided into nine subfields covering the inner and outer rings. The average thickness reflects the mean choroidal thickness across all these regions [54]. Total volume represents the cumulative choroidal volume across the ETDRS grid, derived from the average thickness and the area being analyzed. This volume measurement provides a global assessment of choroidal structure and is calculated for the full 6 × 6 mm area of the ETDRS grid [54]. Additionally, the numeric parameter "OCT image quality" was obtained from the software and used as a covariate in the analysis.

OCT imaging protocol. a Limits of automated CT measurements focused on SFCT, which refers to the thickness of the choroid directly beneath the fovea. The measurement of SFCT is obtained from Bruch's membrane (upper boundary) to the choroid-scleral interface (CSI) (lower boundary). b The 3 measurement radii are shown: a 1 mm radius corresponding to the central region, a 3 mm radius corresponding to the inner measurements, and a 6 mm radius corresponding to the outer measurements. c The ETDRS grid in the macular region of the right eye represents the 9 ETDRS quadrants with the respective CT measurements. The scan range is 6 × 6 mm. d along with the assigned names and their respective ETDRS quadrants. Abbreviations: CT = choroidal thickness; CSI = choroid-scleral interface; ETDRS = Early Treatment Diabetic Retinopathy Study quadrants; OCT = optical coherence tomography; SFCT = subfoveal choroidal thickness

From an initial sample of 1280 individuals, 140 manual measurements from the right eye were randomly selected using specialized software. The CT measurement technique used was based on Trebbastoni et al. [55], employing the "caliper tool" of the Triton-OCT software. Unlike Trebbastoni et al., we performed five measurements in the subfoveal area, 500 µm, and 1500 µm in the superior and inferior zones, omitting nasal and temporal measurements (Fig. 2). This modification aimed to better assess the correlation between manual and automated measurements by evaluating more individuals with fewer measurement points. The choroid was defined as the layer between the base of the retinal pigment epithelium and the hyperreflective boundary corresponding to the CSI. Measurements were taken at five specific points: the subfoveal choroid, and at 500 µm (± 10 µm) and 1500 µm (± 10 µm) along the vertical axis, in the superior and inferior zones. Manual measurements by a single blinded examiner were compared with the CSI protocol automatic tool of the OCT DRI-OCT software (Topcon Co., Tokyo, Japan).

Protocol for comparison of manual and automated CT measurements. a SFCT, which represents the thickness of the choroid directly beneath the fovea, was the first measurement performed. b Four additional CT measurement points were performed: two at 500 µm (Superior and Inferior) and two at 1500 µm (Superior and Inferior). These are measured manually using the “caliper tool,” visible at the top of the image. c Activation of the automated measurement software on the Triton DRI-OCT, using the CSI protocol, which measures CT from Bruch's membrane (upper boundary) to the choroid-scleral interface (CSI) (lower boundary) highlighting its defined boundaries. d Use of the “caliper tool” to perform automated measurements, facilitating comparison between manual and automated methods, detailed at the bottom of the image. Abbreviations: CT = choroidal thickness; CSI = choroid-scleral interface; DRI-OCT = Deep Range Image Optical Coherence Tomography; SFCT = subfoveal choroidal thickness

Data processing and analysis were performed using R software version 4.1.2 [56]. Normality, skewness, and range restriction were evaluated, confirming that all quantitative variables followed an approximately normal distribution.

Demographic (age, sex, education), clinical (hypertension, diabetes mellitus, dyslipidemia, heart disease, COPD, stroke, smoking), and OCT image quality variables were described using frequency analyses, measures of central tendency, and dispersion across the five diagnostic groups (CU, MCI-AD, MCI-Va, ADD, and VaD). Bivariate analyses (ANOVA) and Pearson's Chi-square tests were employed to characterize the distribution of these variables among the groups.

Three multinomial regression analyses were conducted to identify adjustment variables for the final multivariate model. The first examined demographic factors (age, sex, education); the second, clinical factors (hypertension, diabetes mellitus, dyslipidemia, heart disease, COPD, stroke, smoking); and the third, OCT image quality. The CU group served as the reference category in each of the analyses, and the significance level was set at 0.05.

The main analysis consisted of twelve multivariate regression analyses, one for each CT measure: subfoveal choroidal thickness, average thickness, total volume, and the nine ETDRS regions (center, inner temporal, inner superior, inner nasal, inner inferior, outer temporal, outer superior, outer nasal, outer inferior). The five diagnostic groups (CU, MCI-AD, MCI-Va, ADD, VaD) were used as discriminant factors, with adjustment for those factors that showed any significant effect in the former multinomial regression analysis. The CU group was considered the reference category. The following data were reported: regression coefficients, representing the average change in the outcome variable for each unit change in the predictor variable, holding other predictors in the model constant; beta coefficients, indicating the degree of change in the outcome variable for each unit change in the predictor variable; t-values, used to determine if the beta coefficient differs significantly from zero; and significance, expressed as the p-value, which indicates the probability of obtaining the observed results. A Bonferroni correction was applied to account for multiple comparisons performed in the twelve measurements, setting a corrected alpha level of p < 0.0042 to consider results as significant.

To avoid collider bias, all regression analyses were repeated without including the CVRF variables as adjustments, due to their stronger association with cognitive impairment in the MCI-Va and VaD groups.

A sensitivity analysis was conducted to assess the influence of extreme cases (defined as CT values ± 3 standard deviations from the mean) in the twelve multivariate regressions, using the CU group as the reference.

To examine the potential differential effects of sex in the relationship between CT and cognitive diagnosis, the twelve regressions were repeated considering the "diagnostic group x sex" interaction as the main factor.

Partial correlations between MMSE scores and each of the twelve CT measurements were also performed, adjusting for the same covariates, both in the total sample and by diagnostic group.

To assess the association between manual and automated CT measurements, the intraclass correlation coefficient (ICC) was calculated using a two-way mixed-effects model, where a single evaluator rated each target. This model estimates the ICC and its 95% confidence intervals using a single rating and consistency [57, 58]. The model was chosen because it is suited for studies where the selected raters are the only ones of interest. However, the results are specific to the raters involved and cannot be generalized to others, even with similar characteristics. ICC values below 0.5 indicate poor reliability, between 0.5 and 0.75 indicate moderate reliability, between 0.75 and 0.90 suggest good reliability, and values above 0.90 are considered excellent. These thresholds provide a framework for interpreting the level of agreement between manual and automated CT measurements based on the ICC value obtained.