This study is a cross-sectional observational study conducted at Northwestern Memorial Hospital, Chicago, USA. The study adhered to the tenets of the Declaration of Helsinki and was approved by the Northwestern University Institutional Review Board. Written informed consent was obtained from all participants prior to study enrollment, including consent for OCTA and fluorescein angiography. The study was carried out in compliance with the Health Insurance Portability and Accountability Act.

Recruitment occurred between October 2021 and September 2023. Eligible participants were adults with type 1 or type 2 diabetes, with or without DR spanning the full severity spectrum based on the International Clinical DR severity scales. Key exclusion criteria were: center-involving diabetic macular edema, defined as edema involving the central 1-mm circle (central subfield) around the fovea and operationalized as central subfield thickness ≥300 µm on Spectralis OCT (Heidelberg Engineering) ; prior intravitreal anti-VEGF or corticosteroid therapy within 6 months; ocular diseases that could confound DR evaluation or affect vision as determined by the investigator; OCT scans with signal strength index <6; major intraocular surgery within the preceding 3 months or planned within the following 6 months; concurrent enrollment in another investigational trial; or hemoglobin A1c (HbA1c) >10.0%. Eyes with edema within the 3 × 3 mm OCTA scan area that did not involve the foveal center (central 1-mm circle) were not excluded.

In addition, eyes with a history of retinal photocoagulation or an axial length outside the range of 22.00 to 26.00 mm were excluded to ensure reliable ischemia quantification.

At enrollment, all participants underwent a comprehensive ophthalmic evaluation, including lens status (history of cataract surgery), best-corrected visual acuity (BCVA, using the ETDRS protocol), and axial length measurements with the IOL Master 700 (Carl Zeiss Meditec). Additional assessments included ultrawidefield pseudocolor fundus photography and UWF-FA (Optos California system, Optomap Panoramic 200, Optos PLC), as well as spectral-domain OCT (Spectralis OCT) imaging. Multiple OCTA scans using the RTVue-XR Avanti system (Optovue Inc, version 2017.1.0.151) with 3 × 3 mm scan protocols were performed to assess macular perfusion. Demographic and systemic health information were collected, including age, sex, hypertension, ischemic heart disease, renal impairment, and cerebrovascular disease. Sex was defined as a biological variable as recorded in the medical records. Diabetes-specific data encompassed diabetes type, duration, and the most recent HbA1c level. These comprehensive assessments provided structural, functional, and systemic parameters for subsequent analyses.

UWF-FA images were acquired using the Optos California system (Optomap Panoramic 200, Optos PLC) following a standardized protocol. Prior to imaging, the study eye was pharmacologically dilated, and fluorescein dye was administered intravenously. A trained technician obtained high-quality angiographic images centered on the fovea.

Images were analyzed using the proprietary OptosAdvance software (version 4.4.33.107911, Optos PLC). For each eye, the earliest high-quality image demonstrating complete fluorescein filling of the peripheral vasculature was selected for evaluation. The total visible retina was manually outlined using the ROI-free annotation tool, which allows freehand delineation of irregular regions directly on the image rather than using predefined geometric regions of interest, with artifacts excluded to determine the total gradable area. A digital template of the ETDRS 7-fields mask was then superimposed, with its position automatically aligned after manually designating the locations of the fovea and optic disc, and its boundaries were delineated. The posterior retina was defined as the area within the ETDRS 7-fields, while the peripheral retina was defined as the area between the total visible retina and the ETDRS 7-fields.

All regions of retinal nonperfusion were manually traced with the ROI-free annotation tool. Nonperfusion was defined as areas lacking retinal arterioles or capillaries, often accompanied by pruning of adjacent arterioles and a darker underlying choroid. To account for the projection of the three-dimensional retina onto a two-dimensional image, the software automatically converted the outlined regions into a stereographic projection, corrected for magnification artifacts, and calculated absolute areas (mm²) based on a default axial length of 24 mm. For each eye, the percentage of nonperfusion was calculated as the ratio of nonperfused area to the total gradable retinal area. Nonperfusion within the ETDRS 7-fields was designated as posterior nonperfusion, while nonperfusion outside the ETDRS 7-fields was designated as peripheral nonperfusion. We have previously confirmed that the intergrader agreement for nonperfusion measurements was excellent. Figure 1 illustrates representative quantification, including the manual delineation of posterior and peripheral nonperfusion.

Quantification of retinal nonperfusion on ultrawidefield fluorescein angiography. This figure shows a representative ultrawidefield fluorescein angiography image analyzed using OptosAdvance software. The total visible retina was manually outlined (yellow solid contour) after excluding artifacts. A digital ETDRS 7-field template was automatically positioned after manually designating the fovea and optic disc, and its boundary was manually traced (yellow dotted line) to define the posterior retina. The peripheral retina was defined as the area outside the ETDRS 7-fields but within the total gradable retina. Areas of nonperfusion were manually traced (green outlines) and stereographically projected to correct for peripheral magnification, allowing calculation of absolute area and percentage of nonperfusion in posterior, peripheral, and total retina.

The severity of diabetic retinal changes was graded using pseudocolor fundus images acquired with the Optos California system in accordance with the International Clinical DR severity scales. ^, It categorizes disease into five levels: (1) no visible signs of retinopathy (no apparent DR), (2) mild nonproliferative DR (NPDR) defined by the presence of microaneurysms only, (3) moderate NPDR, which includes lesions beyond microaneurysms but not meeting the threshold for severe NPDR, (4) severe NPDR, characterized by any of the “4-2-1” features (≥20 intraretinal hemorrhages in each of four quadrants, venous beading in two or more quadrants, or prominent intraretinal microvascular abnormalities in at least one quadrant), and (5) proliferative DR, indicated by neovascularization or vitreous/preretinal hemorrhage.

We acquired multiple 3 × 3 mm (304 × 304 pixels) OCTA scans centered on the fovea using the RTVue-XR Avanti system with split-spectrum amplitude-decorrelation angiography for angiographic data. For each study eye, at least five repeat scans were acquired, and images with a signal quality index <6 or apparent artifacts were excluded. Default settings segmented the retinal microvasculature into DCP (10 µm above the inner plexiform layer to 10 µm below the outer plexiform layer), superficial capillary plexus (SCP; from internal limiting membrane to 10 µm above the inner plexiform layer), and full retina slabs (combining DCP and SCP).

En-face OCTA (functional) and corresponding en-face OCT (structural) scans were exported, registered, and averaged using Fiji software (ImageJ, National Institutes of Health) to enhance image quality and improve signal-to-noise ratio. ^, Structural OCT images were subsequently inverted to optimize capillary contrast. We used a semiautomated macro in Fiji to measure macular ischemia metrics, including GPDd, from averaged en-face OCTA images, as previously described in detail ( Figure 2 ). ^,

Averaged OCTA image and geometric perfusion deficit map of the deep capillary plexus. (A) Averaged en-face OCTA image of the deep capillary plexus (3 × 3 mm). Multiple frames were registered and averaged to enhance vessel visualization and reduce noise. (B) Corresponding GPD in the deep capillary plexus (GPDd), which was defined as regions >30 µm from the nearest capillary. These areas are highlighted in red, with the foveal avascular zone excluded. GPDd was expressed as the percentage of these areas relative to the total analyzed region, serving as a biomarker of ischemic burden in the DCP.

The structural FAZ was outlined on the averaged en-face OCT images, which depict the underlying foveal capillary architecture independent of flow detection, and therefore represent the baseline anatomic FAZ. The functional FAZ was delineated on the corresponding averaged en-face OCTA images, which visualize flow signal, and therefore represent the FAZ as the area defined by detectable blood flow at the time of imaging. Both measurements were performed manually using the polygon selection tool in Fiji by graders who were masked to clinical information. Excellent intergrader reliability was previously confirmed for both structural and functional FAZ measurements. FAZ enlargement was calculated as the difference between the functional and structural FAZ areas, thereby isolating the ischemia-related expansion of the FAZ while accounting for individual variability in baseline FAZ size ( Figure 3 ). This definition is consistent with prior work by Lynch et al, which described FAZ enlargement as the discrepancy between flow-based and structure-based FAZ measurements.

Averaged en-face OCT and OCTA images showing manual delineation of the structural and functional foveal avascular zone and quantification of foveal avascular zone enlargement. (A) Cropped averaged en-face structural OCT image of the macula, focusing on the central region. (B) Cropped averaged OCTA image from the same eye, demonstrating the perfused capillary network. (C) Manual delineation of the structural foveal avascular zone (yellow contour) on the averaged OCT image. (D) Manual delineation of the functional foveal avascular zone (green contour) on the averaged OCTA image, overlaid with the structural foveal avascular zone boundary (yellow contour). Foveal avascular zone enlargement was defined as the area difference between the functional and structural measurements, isolating the perfusion-related expansion.

All statistical analyses were performed using R (version 4.4.3; R Foundation for Statistical Computing) and RStudio (version 2024.04.2; Posit) with the appropriate packages. Linear mixed-effects models were constructed with either posterior, peripheral, or total retinal nonperfusion (expressed as a percentage of the gradable retinal area) as the dependent variables. Because some participants contributed both eyes, all eligible eyes were included in the analysis, and subject was included as a random intercept to account for within-subject intereye correlation. This approach allowed simultaneous evaluation of eye-level predictors and shared patient-level systemic covariates without requiring selection of a single eye per patient. ^, The base fixed-effects set comprised diabetes duration, HbA1c, and hypertension, together with age and sex, which we considered clinically relevant confounders. ^,, We evaluated the incremental contribution of the two biomarkers of interest, GPDd and FAZ enlargement, by fitting four nested models per region: Model 0 (m0): base covariates only; Model 1 (m1): m0 + GPDd; Model 2 (m2): m0 + FAZ enlargement; Model 3 (m3): m0 + GPDd + FAZ enlargement.

Model comparisons were conducted using maximum likelihood estimation, whereas the reported coefficients and 95% CIs were derived from restricted maximum likelihood estimates of the same models. To formally test whether adding GPDd and/or FAZ enlargement improved model fit, we performed incremental model comparisons using likelihood ratio tests (LRTs) and, to obtain small-sample–robust inference, parametric bootstrap LRTs with 2000 simulations and Kenward–Roger degrees-of-freedom correction. To complement these hypothesis tests, we also performed information-theoretic model selection, comparing small-sample corrected Akaike information criterion (AICc) and Bayesian information criterion (BIC) across all candidate models (m0-m3). We report ΔAICc/ΔBIC values (relative to the best-fitting model) and model weights, which represent the relative support for each model being optimal among the candidates.

To facilitate interpretation, we report standardized fixed-effect coefficients ( β ) with 95% CIs from the fully adjusted models (m3). Here, β reflects the expected change in the outcome, expressed in SD units, associated with one-SD increase in the predictor (where SD refers to the sample SD of each variable). To facilitate interpretation in the original measurement units, we report unstandardized fixed-effect coefficients with 95% CIs from the fully adjusted models. Multicollinearity was assessed using generalized variance inflation factors. Model assumptions were evaluated by visually inspecting residual-vs-fitted plots and quantile–quantile plots, which indicated approximate normality and homoscedasticity. Two-sided P values <.05 were considered statistically significant.