Macular Microaneurysms Are Associated With Chronic Kidney Disease and Deep Capillary Plexus Ischemia in Referable Diabetic Retinopathy

Abstract

Purpose

To identify retinal and systemic factors linked to macular microaneurysms (MAs) in referable diabetic retinopathy (refDR).

Methods

Eighty-three eyes from 65 patients with refDR without diabetic macular edema underwent multiple 3 × 3-mm optical coherence tomography angiography (OCTA) scans. The scans were averaged, and OCTA metrics were calculated, including vessel length density in the deep capillary plexus (VLD-DCP). Ultra-widefield fluorescein angiography was used to extract macular MA count within a 3-mm circle centered on the fovea and the nonperfusion index (NPI) across the entire visible retina. Chronic kidney disease (CKD) was defined as an estimated glomerular filtration rate <60 mL/min/1.73 m² for at least 3 months. Linear mixed models were used to assess the relationships with MA count, and receiver operating characteristic curve analysis evaluated MA diagnostic performance for detecting CKD.

Results

The median MA count was 6 (interquartile range, 2–12). In univariate analysis, macular MA count was significantly associated with CKD, NPI, and several OCTA parameters. In multivariable models adjusting for age, sex, CKD, NPI, and one OCTA parameter per model, CKD (P = 0.007–0.013) and VLD-DCP (P = 0.003) remained independently associated with MA count. A threshold of 14 macular MAs yielded the highest Youden index for CKD, with a sensitivity of 58.8%, a specificity of 86.4%, and an area under the curve of 0.755.

Conclusions

CKD and reduced VLD-DCP were independently associated with higher macular MA count in refDR, suggesting that MA count may serve as a marker for systemic microvascular disease.

Diabetic retinopathy (DR) remains the leading cause of preventable vision loss among working-age adults worldwide, affecting roughly one-third of people with diabetes.¹ Microaneurysms (MAs), saccular dilatations that form in areas of weakened vascular wall, as well as pericyte dropout and endothelial injury, are the earliest clinically visible signs of DR.² Their number and spatial distribution correlate closely with DR severity and progression, making MA burden an attractive biomarker for disease staging and longitudinal monitoring.^3–5

MAs are also important in the setting of DR complications, where higher macular MA burden increases the risk for diabetic macular edema (DME),^6–8 and a subset of perfused lesions often persists despite anti-VEGF therapy.^9,10 These residual MAs correlate with suboptimal anatomic responses and higher injection rates, suggesting a role in treatment resistance.¹¹ Conversely, focal or grid laser photocoagulation directed at macular MAs can treat macular edema and restore visual acuity, underscoring their pathophysiologic relevance.^12–14

Systemic comorbidities such as hypertension and dyslipidemia accelerate retinal microvascular injury through combined metabolic and hemodynamic stress.^15–18 Chronic kidney disease (CKD) is a common diabetic microangiopathy that mirrors retinal capillary changes, such as basement membrane thickening and endothelial dysfunction.¹⁹ Many studies have supported the relationship between CKD and DR or DME,^20–23 but the association between CKD and macular MA burden in DR has not been fully delineated.²⁴

Optical coherence tomography angiography (OCTA) offers depth-resolved, noninvasive visualization of the retinal microvasculature, compared to fluorescein angiography (FA). High-resolution 3 × 3-mm scans provide quantitative perfusion metrics, particularly within the deep capillary plexus (DCP), whose compromise is tied to visual decline and DR complications.^25,26 However, to date, no study has explored OCTA-derived macular perfusion metrics with MA counts, ultra-widefield FA (UWF-FA) ischemia, and systemic disease markers such as CKD.

To address these gaps, we asked whether the number of macular MAs in eyes with referable DR mirrors (1) systemic microvascular disease, particularly CKD, and (2) localized ischemia captured by OCTA-derived macular perfusion metrics. We hypothesized that a heavier MA burden would signal DCP ischemia and track with extraocular vascular comorbidity in high-risk patients with referable DR. Demonstrating such relationships could elevate MA counting from a purely ophthalmic marker of DR activity to a readily accessible proxy for an individual's broader microvascular health, enabling earlier interventions and closer collaboration between ophthalmologists and primary care providers.

Methods

Study Design

We conducted a cross-sectional observational study at Northwestern Memorial Hospital (Chicago, IL, USA). The protocol received institutional review board approval from Northwestern University and adhered to the Declaration of Helsinki and to all Health Insurance Portability and Accountability Act requirements. Written informed consent covering OCTA and FA was obtained from every participant before enrollment.

Recruitment spanned October 2021 to September 2023. Participants needed a diagnosis of type 1 or type 2 diabetes and could fall anywhere on the DR severity spectrum. Eyes were excluded if they (1) exhibited fovea-involving DME, defined as a central subfield thickness ≥300 µm on Heidelberg Spectralis OCT (Heidelberg Engineering, Heidelberg, Germany),²⁷ because center-involving DME can substantially degrade OCTA image quality; (2) had received intravitreal anti-VEGF agents or corticosteroids within the past 6 months; or (3) showed any ocular condition judged by the investigators as likely to influence DR status or visual acuity. Further exclusion criteria for study eye recruitment included an OCT signal-strength index <6, major ocular surgery within the prior 3 months or scheduled for the next 6 months, concurrent participation in another investigational study, and glycated hemoglobin (HbA1c) >10.0%.

In this study, we also excluded eyes with a history of retinal photocoagulation, no apparent DR, mild nonproliferative DR (NPDR), an axial length less than 22 mm or greater than 26 mm, or those with missing chart documentation of estimated glomerular filtration rate (eGFR).

OCTA Imaging and Analysis

Fovea-centered 3 × 3-mm OCTA scans (304 × 304 pixels) were acquired with the RTVue XR Avanti (software v. 2017.1.0.151; Optovue, Fremont, CA, USA) using split-spectrum amplitude-decorrelation angiography.²⁸ For each eye, we attempted at least five OCTA scans. We used the built-in algorithm default settings to segment the retinal microvasculature into the deep capillary plexus (DCP; 10 µm above the inner plexiform layer to 10 µm below the outer plexiform layer), superficial capillary plexus (SCP; internal limiting membrane to 10 µm above the inner plexiform layer), and full retina slabs (combining DCP and SCP).

Scans with a Q score <6 or with obvious motion artifacts were excluded. All remaining OCTA scans (median number of scans [interquartile range]: 5 [5–5]) were exported, registered, and averaged using Fiji software (National Institutes of Health, Bethesda, MD, USA) to enhance image quality and improve signal-to-noise ratio. Using a previously published semi-automated macro,²⁹ we then calculated vessel density (VD), vessel length density (VLD), and geometric perfusion deficits (GPDs) on the averaged images for the SCP and DCP slabs.^30,31 VD was computed as the proportion of vessel area relative to the total image area after binarization using the Huang thresholding method. VLD was calculated by measuring the total length of perfused capillaries on skeletonized images. GPD was determined by identifying regions located ≥30 µm from the nearest capillary on each retinal slab, with large superficial vessels subtracted from DCP images to avoid projection artifacts. The foveal avascular zone, which was manually delineated on the full retina slab, was excluded from all calculations. Representative examples of each image for calculating these OCTA metrics are shown in Figure 1.

Figure 1.

Quantification of optical coherence tomography angiography metrics in the superficial and deep capillary plexuses. (A, B) Single-frame OCTA images of the SCP and DCP, respectively. (C, D) Averaged OCTA images of the SCP and DCP after image registration to enhance signal-to-noise ratio. (E, F) Binarized images generated using the Huang thresholding method to compute VD as the proportion of vessel area relative to total image area. (G, H) Skeletonized images used to calculate VLD by measuring the total length of perfused capillaries. (I, J) Perfusion maps identifying regions ≥30 µm from the nearest vessel (highlighted in red), used to compute GPDs. In the DCP, large superficial vessels were subtracted before GPD calculation to avoid projection artifacts. The foveal avascular zone (FAZ) was excluded from all calculations.

Quantification of Macular Microaneurysms and Retinal Nonperfusion on Ultra-Widefield Fluorescein Angiography

UWF-FA images were acquired using the Optos California system (Optomap Panoramic 200; Optos PLC, Dunfermline, Scotland) following a standardized protocol. The earliest, high-quality angiography frame where fluorescein dye had reached the peripheral retina (typically 30–60 seconds postinjection) was selected for analysis of nonperfusion and MA. Stereographic projection was applied using OptosAdvance software (version 4.4.33.107911; OptosAdvance, Marlborough, MA, USA) to correct peripheral magnification distortion inherent to ultra-widefield retinal imaging, enabling accurate quantification of retinal areas.

A 3-mm diameter circle centered on the fovea (7.1 mm²) was overlaid using OptosAdvance software, chosen to approximate the 3 × 3-mm OCTA scan area, enabling direct comparison with OCTA-derived metrics. The image was magnified to fully include the 3-mm circle, then exported into Fiji software, where MAs within the 3-mm circle were manually counted, using the multipoint tool.

Separately, two graders independently outlined the total visible retinal area and areas of nonperfusion, as previously described.³² The nonperfusion index (NPI) was then calculated as the percentage of nonperfused area relative to the total visible retinal area: NPI (%) = (nonperfused area/total visible retinal area) × 100.

Both MA counting and NPI measurements were performed independently by different graders, each masked to clinical data and to each other's assessments during the analyses. Representative examples of each image for MA counting and NPI measurement are shown in Figure 2.

Figure 2.

Quantification of macular microaneurysm count and nonperfusion index on ultra-widefield fluorescein angiography. (A) Early-phase ultra-widefield fluorescein angiography image centered on the fovea, magnified so that the 3-mm-diameter analysis circle is fully visible. (B) Using the same image shown in (A), macular microaneurysms within the 3-mm circle were manually counted with the multipoint tool in Fiji. (C) Total visible retinal area and nonperfused regions were manually delineated; the NPI was calculated as the ratio of nonperfused area to total visible retinal area, with surface areas provided by OptosAdvance software.

Data Collection

Demographic and systemic variables, including age, sex, race (American Indian or Alaska Native, Asian, African American, Native Hawaiian or other Pacific Islander, Caucasian, or Prefer not to answer), ethnicity (Hispanic or Latino, Not Hispanic or Latino, or Prefer not to answer), hypertension, ischemic heart disease, and cerebrovascular disease, as well as diabetes-specific information (type, duration, and most recent glycated hemoglobin [HbA1c]), were recorded at the time of study enrollment. Smoking history was classified as never, past, or current smoker based on patient-reported history at the time of enrollment. CKD status was determined from existing medical records in accordance with the 2024 Kidney Disease: Improving Global Outcomes guideline.³³ CKD was defined as an eGFR persistently below 60 mL/min/1.73 m² for at least 3 months, either at the time of study enrollment or as documented in the medical records prior to enrollment.³³ Medication use (angiotensin-converting enzyme inhibitors, angiotensin receptor blockers, and lipid-lowering agents) was retrospectively reviewed from electronic medical records. Data on body mass index (BMI) were retrospectively retrieved from electronic medical records. For each eye, we documented lens status, best-corrected visual acuity, low-luminance visual acuity, and axial length measured with the IOL Master (Carl Zeiss Meditec, Jena, Germany). DR severity was evaluated using UWF-FA and UWF–color photography, following the International Clinical Diabetic Retinopathy severity scale, as previously described.^32,34 Although referable DR is commonly defined as the presence of DME and/or moderate nonproliferative DR or worse, this study excluded eyes with DME. Therefore, referable DR was defined solely based on DR severity. All data were securely entered into REDCap, a web-based platform for electronic research data capture.

Statistical Analyses

We conducted all analyses with R (version 4.4.3; R Foundation for Statistical Computing, Vienna, Austria) and RStudio (version 2024.04.2; Posit, Boston, MA, USA). Univariate linear mixed models (LMMs) were fitted for macular MA count with the following fixed effects: DR severity (moderate NPDR, severe NPDR, or proliferative DR [PDR]), axial length, lens status (phakia or pseudophakia), diabetes type (type 1 or 2), duration of diabetes (years), HbA1c, hypertension (yes or no), ischemic heart disease (yes or no), CKD (yes or no), cerebrovascular disease (yes or no), dyslipidemia (yes or no), VD in the SCP, VD in the DCP, VLD in the SCP, VLD in the DCP (VLD-DCP), GPD in the SCP (GPD-SCP), GPD in the DCP (GPD-DCP), and NPI. Patient was included as a random effect to account for intrasubject correlation.

Subsequently, multivariable linear mixed models were constructed, including all covariates significant on univariate analysis, together with age and sex, which are clinically relevant confounders based on previous literature.^35,36 To prevent multicollinearity, only one OCTA-derived metric was included per model. Furthermore, multicollinearity among the explanatory variables was assessed by calculating variance inflation factors. Model assumptions were verified by quantile–quantile plots and residual-versus-fitted plots, confirming that the residuals satisfied the assumptions of normality and homoscedasticity.

To explore the utility of MA count as a screening tool for systemic vascular comorbidities, we fitted generalized estimating equation models using continuous MA counts as predictors, with patient specified as the clustering variable. Each model assumed an exchangeable correlation structure and a binomial distribution. Predicted probabilities were used to generate receiver operating characteristic (ROC) curves and corresponding areas under the curve. Sensitivity and specificity were calculated across the range of MA values, and the Youden index (sensitivity + specificity – 1) was computed to assess diagnostic performance. The MA value with the highest Youden index was selected as the optimal threshold.

A P value of <0.05 was considered statistically significant.

Results

A total of 65 patients (83 eyes) met the inclusion and exclusion criteria. The cohort comprised 32 men (49%) with a mean age of 60.9 ± 12.4 years. The mean duration of diabetes was 23.5 ± 13.3 years; 16 patients (25%) had type 1 diabetes, and 49 patients (75%) had type 2 diabetes. CKD was present in 14 patients (22%, corresponding to 17 eyes). Patient characteristics are summarized in Table 1.

Table 1.

Patient Demographics and Systemic Characteristics (n = 65)

Characteristic	Value
Age, y	60.9 ± 12.4
Sex (male/female)	32/33
Race
Asian	4
African American	19
Caucasian	34
Prefer not to answer	8
Ethnicity
Hispanic or Latino	20
Not Hispanic or Latino	42
Prefer not to answer	3
Smoking history
Never smoker	39
Past smoker	22
Current smoker	4
DM duration, y	23.5 ± 13.3
HbA1c, %	7.4 ± 1.0
DM type (type 1/type 2)	16/49
Body mass index, kg/m²	31.9 ± 7.9
Hypertension	45
Ischemic heart disease	10
CKD	14
G1	25
G2	26
G3	8
G4	5
G5	1
Cerebrovascular disease	4
Dyslipidemia	39
Medications
ACE inhibitors	26
ARB	19
Lipid-lowering medications	53

ACE, angiotensin-converting enzyme; ARB, angiotensin receptor blockers; DM, diabetes mellitus. Data are expressed as number or mean ± standard deviation. CKD was graded based on eGFR as follows: G1, ≥90; G2, 60–89; G3, 30–59; G4, 15–29; G5, <15 (mL/min/1.73 m²). CKD was defined as G3, G4, or G5. Body mass index values were available for 60 participants.

Of the 83 eyes analyzed, 20 had moderate NPDR, 41 had severe NPDR, and 21 had PDR without DME or prior retinal photocoagulation. The median macular MA count within the 3-mm circle centered on the fovea was 6 (interquartile range, 2–12). Other ocular characteristics are summarized in Table 2.

Table 2.

Ocular Characteristics of Study Eyes (n = 83)

Characteristic	Value
BCVA	83.4 ± 6.1
LLVA	74.0 ± 8.0
DR severity
Moderate NPDR	20
Severe NPDR	41
PDR	22
Lens (pseudophakia)	25
Axial length, mm	23.73 ± 0.94
VD-SCP, %	42.72 ± 5.64
VD-DCP, %	46.41 ± 4.34
VLD-SCP, mm−1	14.72 ± 3.49
VLD-DCP, mm−1	18.24 ± 1.70
GPD-SCP, %	11.67 ± 4.94
GPD-DCP, %	5.03 ± 2.93
NPI, %	2.93 ± 5.19

BCVA, best-corrected visual acuity; LLVA, low-luminance best-corrected visual acuity. Data are expressed as number or mean ± standard deviation.

In univariate analyses, with MA count as the dependent variable and each parameter entered as a single fixed effect, higher MA count was significantly associated with CKD, NPI, VLD-DCP, GPD-SCP, and GPD-DCP (Table 3).

Table 3.

Summary of Univariate Analyses for Fixed Effects Related to Macular Microaneurysm Count

Explanatory Variables	Estimate	95% CI	P Value
DR severity = severe NPDR	4.48	−0.78 to 9.74	0.103
DR severity = PDR	5.74	−1.15 to 12.63	0.107
Axial length, mm	−0.78	−3.82 to 2.27	0.619
Lens status	3.09	−3.61 to 9.79	0.369
DM type	1.29	−5.92 to 8.49	0.727
DM duration, y	0.09	−0.15 to 0.32	0.463
HbA1c, %	0.81	−2.39 to 4.02	0.621
Age, y	−0.10	−0.35 to 0.15	0.447
Hypertension	0.96	−5.76 to 7.69	0.780
Sex	−3.46	−9.62 to 2.71	0.276
Body mass index, kg/m²	−0.20	−0.62 to 0.22	0.360
Ischemic heart disease	−2.47	−11.04 to 6.1	0.574
CKD	11.38	4.34 to 18.42	0.002**
Cerebrovascular disease	0.22	−12.53 to 12.96	0.974
Dyslipidemia	−0.42	−6.76 to 5.91	0.896
VD-SCP, %	−0.14	−0.56 to 0.27	0.496
VD-DCP, %	−0.50	−1.08 to 0.07	0.092
VLD-SCP, mm−1	−0.41	−1.21 to 0.38	0.312
VLD-DCP, mm−1	−3.06	−4.50 to −1.63	<0.001***
GPD-SCP, %	0.60	0.06 to 1.15	0.033*
GPD-DCP, %	1.37	0.48 to 2.26	0.003**
NPI, %	0.55	0.05 to 1.05	0.033*

CI, Confidence Interval; DR, Diabetic Retinopathy; NPDR, Non-Proliferative DR; PDR, Proliferative DR; DM, Diabetes Mellitus; CKD, Chronic Kidney Disease; VD-SCP, Vessel Density in the Superficial Capillary Plexus; VD-DCP, Vessel Density in the Deep Capillary Plexus; VLD-SCP, Vessel Length Density in the Superficial Capillary Plexus; VLD-DCP, Vessel Length Density in the Deep Capillary Plexus; GPD-SCP, Geometric Perfusion Deficits in the Superficial Capillary Plexus; GPD-DCP, Geometric Perfusion Deficits in the Deep Capillary Plexus; NPI, Nonperfusion Index.

Linear mixed models were fitted with patient as a random effect and one explanatory variable as a fixed effect. Analyses included 83 eyes from 65 patients, except for body mass index (78 eyes from 60 patients). CI, confidence interval.

*P < 0.05. **P < 0.01. ***P < 0.001.

To assess whether these associations were independent of potential confounders, we fitted three separate multivariable LMMs, each including one OCTA parameter (VLD-DCP, GPD-SCP, or GPD-DCP) together with all significant covariates from univariate analyses and the a priori confounders age and sex. CKD remained a significant independent predictor of higher macular MA count across all models (β ≈ 9.6–10.6; P ≤ 0.013). Among the OCTA metrics, only VLD-DCP remained significant (β = −2.46; P = 0.003), whereas GPD-SCP (P = 0.201) and GPD-DCP (P = 0.089) were not. Age showed a modest inverse association with MA count in the VLD-DCP model (P = 0.049) and a borderline trend in the other two models. Sex and NPI were not significantly associated with MA count in any model (Table 4). All variance inflation factors were below 1.3, suggesting that multicollinearity was not a concern. In a sensitivity analysis including BMI as an additional covariate (Supplementary Table S1), CKD and VLD-DCP remained independently associated with MA count, whereas BMI was not significant.

Table 4.

Multivariate Linear Mixed Models for Fixed Effects Associated With Macular Microaneurysm Count

Explanatory Variables	Estimate	95% CI	P Value
Model 1: Multivariable LMM including VLD-DCP as the OCTA parameter
(Intercept)	66.76	33.99 to 99.53	<0.001***
VLD-DCP	−2.46	−4.00 to −0.92	0.003**
CKD	9.60	2.84 to 16.36	0.007**
NPI	0.18	−0.30 to 0.66	0.462
Age	−0.23	−0.44 to −0.01	0.049*
Sex	−2.06	−7.32 to 3.20	0.447
Model 2: Multivariable LMM including GPD-SCP as the OCTA parameter
(Intercept)	16.70	1.25 to 32.14	0.039*
GPD-SCP	0.37	−0.19 to 0.93	0.201
CKD	10.57	3.19 to 17.94	0.007**
NPI	0.36	−0.13 to 0.85	0.155
Age	−0.21	−0.45 to 0.03	0.091
Sex	−3.43	−9.12 to 2.25	0.242
Model 3: Multivariable LMM including GPD-DCP as the OCTA parameter
(Intercept)	16.90	1.98 to 31.82	0.031*
GPD-DCP	0.85	−0.12 to 1.82	0.089
CKD	9.75	2.34 to 17.16	0.013*
NPI	0.32	−0.17 to 0.81	0.203
Age	−0.21	−0.45 to 0.02	0.078
Sex	−2.56	−8.17 to 3.04	0.374

CI, Confidence Interval; DR, Diabetic Retinopathy; NPDR, Non-Proliferative DR; PDR, Proliferative DR; DM, Diabetes Mellitus; CKD, Chronic Kidney Disease; VD-SCP, Vessel Density in the Superficial Capillary Plexus; VD-DCP, Vessel Density in the Deep Capillary Plexus; VLD-SCP, Vessel Length Density in the Superficial Capillary Plexus; VLD-DCP, Vessel Length Density in the Deep Capillary Plexus; GPD-SCP, Geometric Perfusion Deficits in the Superficial Capillary Plexus; GPD-DCP, Geometric Perfusion Deficits in the Deep Capillary Plexus; NPI, Nonperfusion Index.

Linear mixed models were fitted with patient as a random effect. Multivariable LMMs were fitted in 83 eyes from 65 patients.

*P < 0.05. **P < 0.01. ***P < 0.001.

We next assessed the diagnostic performance of MA count for detecting CKD, using MA count as the predictor and CKD status as the binary outcome. ROC analysis included 83 eyes (17 with CKD and 66 without CKD) and demonstrated that macular MA burden had a fair discriminative ability for identifying CKD, with an area under the curve of 0.755. A threshold of 14 macular MAs maximized the Youden index and yielded a sensitivity of 58.8% and a specificity of 86.4% for detecting CKD.

Discussion

In this cross-sectional study of patients with referable DR (excluding DME), we report three principal findings. First, the presence of CKD was independently associated with a higher macular MA count after controlling for age, sex, and ocular covariates, underscoring a systemic link between diabetic renal dysfunction and MAs. Second, among the OCTA-derived perfusion metrics, only VLD-DCP retained significance in multivariable models, with lower VLD-DCP predicting greater MA counts, implying that DCP ischemia is a strong predictor of MA. Third, ROC analysis identified an optimal macular MA count threshold to detect CKD, yielding a clinically useful specificity and sensitivity. Collectively, these results position macular MA counting as a potential surrogate marker that reflects local capillary damage and CKD in patients with referable DR.

Previous work on the retina–kidney axis in diabetes has focused on DR severity or DME, consistently showing that these retinal endpoints track with lower eGFR or higher albuminuria.^20–23 By contrast, few studies have examined whether retinal lesions, particularly macular MAs, mirror systemic microvascular disease like CKD. A notable exception is the community-based report by Wong et al.,²⁴ who followed more than 10,000 middle-aged adults in the US Atherosclerosis Risk in Communities cohort, where 13 % of the participants had diabetes at baseline. Interestingly, the presence of any retinal MA on 30° color fundus photographs independently doubled the 6-year risk of incident renal dysfunction, defined as either a ≥0.4-mg/dL increase in serum creatinine or hospitalization for CKD. Crucially, this association persisted after adjustment for diabetes, hypertension, and other vascular risk factors and was evident in both diabetic and nondiabetic subgroups, underscoring MAs as a pragmatic marker of systemic microangiopathy. Building on that population-based observation, our study adds several layers of novelty. First, we focused on the eyes of patients with referable DR, who already warrant close ophthalmic surveillance. In this high-risk population, we demonstrated that a higher macular MA burden independently signals concomitant CKD, underscoring the value of systemic risk stratification within a referral cohort. Second, after adjusting for age, sex, ultra-widefield FA–derived nonperfusion, and layer-specific OCTA metrics, the MA–CKD association remained significant, suggesting that macular MA burden captures microvascular injury beyond what is explained by DR severity alone. Together, these findings highlight macular MA quantification, when combined with multimodal retinal imaging, as a cross-disciplinary indicator of systemic microvascular health.

A biologically plausible framework underpins the clinical link we observed between CKD and macular MA burden. Diabetic nephropathy and diabetic retinopathy are prototypical microangiopathies that share histopathologic hallmarks—diffuse basement membrane thickening, loss of mural support cells (pericytes in the retina, podocytes in the kidney), and endothelial dysfunction—all of which narrow the capillary lumen and impair perfusion.³⁷ In the kidney, these changes precipitate glomerular hyperfiltration followed by eGFR decline³⁷; in the retina, they promote focal out-pouching of weakened capillary walls, manifesting as MAs.^2,3 Beyond these structural parallels, several systemic drivers amplify damage in both organs: chronic inflammation, oxidative stress, and activation of the renin–angiotensin–aldosterone system accelerate endothelial injury, hastening capillary rarefaction.¹⁹ Together, the association of reduced VLD-DCP and CKD with higher MA counts reinforces the notion that the retina and kidney undergo parallel microvascular injury in diabetes.^19,38

Only VLD-DCP remained significant among the OCTA metrics in multivariable analyses, likely reflecting its greater sensitivity to early and diffuse capillary rarefaction rather than advanced confluent nonperfusion. VLD-DCP, which quantifies the total length of perfused capillaries, can capture subtle microvascular remodeling preceding complete capillary dropout, whereas GPD-SCP and GPD-DCP primarily identify advanced, confluent ischemic zones. Given that MAs can arise early in DR,³⁴ VLD-DCP may better reflect the microvascular milieu underlying their formation. Most MAs are known to reside within the DCP, and the DCP is anatomically more vulnerable to ischemia owing to its distal location from arterioles and higher metabolic demand. This heightened susceptibility likely explains the statistical association observed between VLD-DCP and macular MA count.^10,39–41 Of note, our relatively small sample size may also have contributed to limited statistical power to detect independent associations for GPD-SCP and GPD-DCP, and future studies with larger cohorts are warranted to confirm these findings.

Our study has several strengths. We focused on eyes with referable DR but without center-involving DME, minimizing clinical heterogeneity and potential imaging artifacts. Using averaged images and high-resolution OCTA and UWF-FA allowed us to capture both local and panretinal ischemia, as well as explore their relationships with CKD.

However, there are also several limitations. Our cohort included patients with both type 1 and type 2 diabetes, with type 2 diabetes predominating (75%). Although diabetes type was not significantly associated with macular MA count in univariate analyses, retinal and renal complication patterns may differ between diabetes types. We quantified macular MAs on FA (rather than OCTA or OCT) due to its superior sensitivity to detect MA.^42–44 Moreover, maximizing the Youden index produced a cutoff with high specificity (86.4 %) but only moderate sensitivity (58.8 %). This outcome reflects the prevalence of macular MA in CKD-positive as well as CKD-negative eyes. Incorporating additional factors into the model may help improve diagnostic accuracy. Given the small cohort size in this study, the findings obtained require validation in larger cohorts. Furthermore, considering the eGFR threshold of 60 mL/min/1.73 m², further studies are required to validate the utility of macular MA count as a biomarker for more severe CKD stages. Albuminuria was not assessed in most patients, resulting in an incomplete evaluation of CKD. In addition, glomerular hyperfiltration was not considered in this study, which may have affected the assessment of early diabetic kidney disease. Because eyes with fovea-involving DME were excluded in this study, the clinical relevance of our findings needs to be further explored in referable eyes with DME.

Taken together, our findings highlight macular MA burden as a cross-disciplinary biomarker that encapsulates both localized deep capillary plexus ischemia and systemic microvascular injury reflected by CKD. By combining high-resolution OCTA with UWF-FA, we demonstrate that macular MAs can stratify systemic risk within a relatively high-risk cohort. Although our cross-sectional design precludes causal inference and albuminuria data were unavailable, the association between MA count, reduced VLD-DCP, and CKD supports a shared microvascular pathology and underscores the value of integrated ocular–renal screening pathways. As OCT and OCTA algorithms and software continue to advance, noninvasive and convenient quantification of macular MAs may facilitate the identification of patients with DR at high risk for CKD in clinical practice. Prospective longitudinal studies are needed to verify whether macular MA burden can predict renal function decline or guide timing of nephrology referral. Incorporating macular MA quantification into routine retinal imaging could reflect broader vascular health, fostering earlier multidisciplinary intervention and ultimately improving outcomes for patients with diabetes.