Peripheral arterial disease is associated with retinal fluid and microvascular alterations in type 2 diabetes

Abstract

BACKGROUND

Peripheral arterial disease (PAD) and diabetic retinal microvascular complications share systemic vascular pathways, but whether PAD is associated with retinal vessel density reduction and later retinal fluid development in type 2 diabetes mellitus (T2DM) remains unclear.

AIM

To determine whether PAD is associated with retinal microvascular impairment and retinal fluid risk in T2DM.

METHODS

This retrospective cohort included 212 eyes from 203 patients with T2DM without baseline macular edema who underwent complication screening at a tertiary center from December 2016 to February 2021. PAD was defined as ankle-brachial index ≤ 0.9. Optical coherence tomography angiography vessel density was compared by PAD status. Bayesian Cox regression estimated intraretinal fluid and diabetic macular edema (DME) risk, with inverse probability weighting as triangulation.

RESULTS

Among 212 eyes from 203 patients, 25 eyes (11.8%) had PAD. Patients with PAD were older and had lower parafoveal vessel density in the superficial capillary plexus (43.9% vs 46.4%) and deep capillary plexus (46.6% vs 49.7%). PAD was associated with higher intra-retinal fluid (IRF) risk (hazard ratio 5.44; 95% credible interval: 2.11-15.18) and DME risk (hazard ratio 3.54; 95% credible interval: 1.07-11.44), although the wide intervals indicate substantial uncertainty in effect magnitude. Exploratory mediation analysis estimated an indirect proportion of 34% through lower deep capillary plexus vessel density.

CONCLUSION

PAD in T2DM was associated with reduced retinal capillary vessel density and increased incidence of IRF and DME; these observational findings require prospective validation before informing retinal surveillance.

Key Words: Peripheral arterial disease; Type 2 diabetes mellitus; Retinal microvasculature; Intraretinal fluid; Optical coherence tomography angiography

Core Tip: Peripheral arterial disease in type 2 diabetes was associated with lower superficial and deep capillary plexus vessel density on optical coherence tomography angiography and with higher incidence of intraretinal fluid and diabetic macular edema. The association remained positive in the primary Bayesian model and sensitivity analyses, but the small peripheral arterial disease sample and sparse diabetic macular edema events produced substantial uncertainty in effect magnitude. The mediation analysis was exploratory. These findings identify a hypothesis for prospective validation rather than establish a retinal risk-stratification strategy.

INTRODUCTION

Diabetes mellitus is a major global health concern, affecting millions worldwide and leading to both microvascular and macrovascular complications[1,2]. Among the microvascular complications, diabetic retinopathy (DR) is particularly common and can progress to diabetic macular edema (DME), a leading cause of vision loss in patients with type 2 diabetes mellitus (T2DM)[3-5]. In parallel, macrovascular complications such as peripheral arterial disease (PAD) are frequently observed in patients with T2DM and are associated with an increased risk of cardiovascular events and poor clinical outcomes[2,6,7].

Recent advances in optical coherence tomography angiography (OCTA) have enabled detailed assessment of retinal microvascular alterations in patients with diabetes, offering a non-invasive method to evaluate capillary perfusion and vascular density (VD)[8,9]. Emerging evidence suggests a potential link between systemic macrovascular dysfunction, such as PAD, and retinal microvascular alterations, raising concerns about a potential link between systemic vascular disease and retinal health[10-13]. However, the association between PAD and retinal microvascular impairment and its role as a biomarker for the development of DME remains unclear.

This study aimed to characterize retinal microvascular alterations in patients with T2DM based on PAD status using OCTA. By evaluating retinal capillary plexus VD and analyzing its association with the development of intra-retinal fluid (IRF) and DME, we sought to determine the potential role of PAD as a predictor of vision-threatening retinal complications.

MATERIALS AND METHODS

Participants

The study was designed, analyzed, and reported in accordance with the STROBE statement for cohort studies. This retrospective cohort study included individuals who underwent screening for diabetic complications at Asan Medical Center’s Diabetes Center between December 2016 and February 2021. The study window reflected the institutional review board-approved retrospective dataset and the availability of baseline OCTA, ankle-brachial index (ABI), and follow-up data. The study was conducted in accordance with the principles of the Declaration of Helsinki and approved by the Institutional Review Board and Ethics Committee of Asan Medical Center, No. 2025-0825. Informed consent was waived due to the retrospective nature of the study. This study is reported in accordance with the STROBE guidelines.

Patients were excluded if they: (1) Were aged < 18 years; (2) Had high myopia (spherical equivalent ≥ -6 D); (3) Had a history of trauma, pan-retinal photocoagulation, major vitreoretinal surgery, retinal break or detachment, ocular inflammation, and other ocular pathologies that could significantly affect the retinal vasculature other than diabetes; (4) Had a history of recent (within 1 year) active treatment for retinal edema or vascular leakage; and (5) Had IRF at the baseline OCTA findings. Patients younger than 18 years were excluded because this study was designed as an adult T2DM cohort. When both eyes met eligibility criteria, both eyes were retained and patient-level dependence was handled analytically. In total, 212 eyes from 203 patients were included in the final analysis (Supplementary Figure 1).

Basic clinical parameters

Patient demographic data were obtained from a database managed by the Health Screening and Promotion Center at Asan Medical Center. Medical history related to diabetes, hypertension, medication use, and smoking status was gathered using a structured questionnaire completed prior to routine health examination. Height, body weight, body mass index (BMI), and blood pressure were measured during the examination. Additionally, biological parameters, including fasting plasma glucose, glycated hemoglobin (HbA1c), creatinine, total cholesterol, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, and triglycerides, were collected following a fasting period of at least 12 hours before sampling. PAD was defined as ABI ≤ 0.9, consistent with contemporary lower-extremity PAD guideline criteria[6]. ABI screening was performed in patients with T2DM who are considered high-risk as follows: Aged 50 years or older; cardiovascular risk factors such as smoking, hypertension, dyslipidemia, or a diabetes duration of ≥ 10 years; and a history of other atherosclerotic diseases, including coronary artery disease or cerebrovascular disease[14]. Because ABI screening in this program was risk-based rather than age-only, adults younger than 50 years were included when they met other high-risk criteria and had available ABI screening data.

Ophthalmic assessments

OCTA was performed using the AngioVue system (Optovue, Fremont, CA, United States) to acquire split-spectrum amplitude-decorrelation angiography, as previously described[9,15]. The scanning area was captured in 3 × 3 mm sections, centered on the fovea. The software automatically segmented the superficial capillary plexus (SCP) and deep capillary plexus (DCP). OCTA images with a signal strength index of < 50, significant artifacts caused by poor fixation, or errors in automatic layer segmentation were excluded from the analysis. Macular microvascular integrity was assessed using the foveal avascular zone (FAZ) area and parafoveal VD of the SCP and DCP. The FAZ area and VD were measured using the non-flow and density functions of Optovue software, respectively. VD, representing the percentage of the parafoveal area occupied by vessels, was calculated separately for the temporal, superior, nasal, and inferior parafoveal regions based on the Early Treatment DR Study grid sectors, with the average of the four measurements used for analysis.

Spectral-domain OCT scans of the macula were acquired using the Spectralis OCT system (Heidelberg Engineering, Heidelberg, Germany), as previously described and according to contemporary DR evaluation guidance[16]. Two independent retinal fellowship-trained specialists (Yang JM and Kim YJ), blinded to the patient’s clinical data, reviewed and analyzed the spectral-domain-OCT images.

Outcomes

The primary outcome was the new development of IRF. IRF was defined as the presence of hypo-reflective fluid in the inner retina, evidenced by OCT or OCTA examination[17]. Only IRF lesions with a diameter ≥ 50 μm were included to avoid image resolution biases[18]. The secondary outcome was new incidents of DME. DME was defined as increased central subfield retinal thickness (≥ 300 μm) accompanied by signs of vascular hyperpermeability and leakage (e.g., IRF or hard exudates)[19]. Early DME was defined as occurring within 6 months of the initial OCTA screening.

Statistical analysis

The statistical methods of this study were reviewed by Yon DK. The exposure variable was the presence of PAD in patients with T2DM who underwent OCTA examinations. OCTA parameters (SCP and DCP parafoveal VD) were categorized into quartiles from the lowest (quartile 1) to the highest (quartile 4) VD. Bayesian Cox regression (brms v2.23.0) served as primary analysis. We specified a skeptical prior Normal(0, 1) for the PAD log-hazard ratio (HR), reflecting Zhang et al[20] showing only a trend (non-significant) for PAD-DR association in T2DM patients. This null-centered, weakly regularizing prior allowed clinically important effects while limiting implausibly extreme estimates. Other covariates received weakly regularizing priors Normal(0, 1.5). MCMC sampling used four chains with 5000 iterations each, including 2000 warmup iterations; R-hat and effective sample size were monitored. The analyses were adjusted for known confounders that may have affected the outcomes. Posterior median HRs with 95% credible intervals (CrIs) and posterior probabilities P (HR > 2) were reported. The IRF model adjusted for age, sex, diabetes duration, HbA1c, DR severity, and estimated glomerular filtration rate; because only 15 DME events occurred, the primary DME model used a parsimonious covariate set of age and DR severity. Inverse probability of treatment weighting (IPTW) triangulated findings using propensity scores estimated from baseline covariates [age, sex, diabetes duration, HbA1c, DR stage, estimated glomerular filtration rate, log[albumin-to-creatinine ratio (ACR)], systolic blood pressure]. Average treatment effect weights were truncated at 1^st/99^th percentiles; covariate balance was confirmed by standardized mean differences < 0.1 (Supplementary Figure 2). Matching adequacy was validated by comparing the distributions from IPTW weighting and assessing the standardized mean difference, providing a more robust assessment of imbalance between groups compared to the traditional P values. All analyses were performed using R software (v 4.5.2, R Foundation for Statistical Computing, Vienna, Austria) and SAS software (v 9.4, SAS Institute Inc., Cary, NC, United States). Continuous variables are reported as mean ± SD, and categorical variables are reported as n (%). Normality was assessed using the Shapiro-Wilk test. For multiple comparisons, the Benjamini-Hochberg method was used to adjust P value. The significance threshold was set at a P value < 0.05. Both Bayesian Cox models included a patient-level random intercept to account for correlation between fellow eyes from the same patient. As a sensitivity analysis, the associations were also examined after retaining one eye per patient.

RESULTS

Baseline clinical characteristics

Table 1 presents the clinical, biological, and imaging parameters of patients with T2DM, comparing those without PAD (n = 187) and with PAD (n = 25). Patients with PAD were significantly older (61.6 years vs 56.9 years, P = 0.023) and had a lower estimated glomerular filtration rate (71.7 mL/minute/1.73 m² vs 82.5 mL/minute/1.73 m², P = 0.018). Systolic and diastolic blood pressure were also significantly higher in the PAD group [systolic blood pressure (144.0 mmHg vs 132.0 mmHg, P = 0.002); diastolic blood pressure (81.0 mmHg vs 74.7 mmHg, P = 0.006)]. Furthermore, OCTA parameters showed lower SCP (P = 0.003) and DCP (P < 0.001) parafoveal VD in the PAD group. No significant differences were observed between the groups in HbA1c levels, lipid profiles, BMI, DR stage, or OCT thickness. Additional descriptive summaries by age group, sex, and baseline DR stage are provided in Supplementary Table 1.

Table 1 Baseline demographic and clinical characteristics of all participants with type 2 diabetes mellitus based on peripheral arterial disease status, n (%)/mean ± SD.

	Patients with type 2 diabetes mellitus		P value
	PAD- (n = 187)	PAD+ (n = 25)
Age, years	56.9 ± 9.7	61.6 ± 8.9	0.023
Sex, male	120 (64.2)	19 (76.0)	0.345
Smoking status			0.267
Never	89 (51.4)	10 (40.0)
Current	42 (24.3)	9 (36.0)
Ex-smoker	42 (24.3)	6 (24.0)
Smoking, pack-years	9.5 ± 14.6	14.5 ± 17.0	0.459
Duration of diabetes, years	17.3 ± 7.7	19.6 ± 10.6	0.310
Medication			0.671
Insulin only	1 (0.5)	0 (0.0)
OHA	121 (64.7)	19 (76.0)
Insulin + OHA	65 (34.8)	6 (24.0)
HTN	79 (42.2)	15 (60.0)	0.202
DR stage			0.793
No DR	23 (12.3)	3 (11.5)
Mild NPDR	81 (43.3)	13 (50.0)
Moderate NPDR	41 (21.9)	6 (23.1)
Severe NPDR	32 (17.1)	4 (15.4)
PDR	10 (5.3)	0 (0.0)
Biological parameters
HbA1c	7.8 ± 1.3	7.5 ± 0.8	0.174
Glucose	148.0 ± 49.4	136.4 ± 30.7	0.111
Total cholesterol	144.1 ± 35.0	146.5 ± 33.4	0.746
Triglyceride	132.8 ± 75.9	125.4 ± 55.7	0.640
HDLc	45.1 ± 11.3	42.4 ± 8.7	0.257
LDLc	89.6 ± 28.2	96.1 ± 30.6	0.284
Creatinine	1.0 ± 0.8	1.1 ± 0.3	0.553
eGFR	82.5 ± 21.7	71.7 ± 18.8	0.018
Clinical parameters
BMI, kg/m2	25.6 ± 3.7	26.3 ± 4.2	0.327
SBP, mmHg	132.0 ± 17.8	144.0 ± 18.3	0.002
DBP, mmHg	74.7 ± 10.5	81.0 ± 12.9	0.006
Visual acuity, logMAR	0.07 ± 0.09	0.04 ± 0.07	0.158
OCTA parameters
SCP fovea VD, %	15.1 ± 4.7	15.6 ± 5.6	0.372
SCP parafoveal VD, %	46.4 ± 3.9	43.9 ± 3.7	0.003
DCP fovea VD, %	28.2 ± 6.0	29.5 ± 8.0	0.409
DCP parafoveal VD, %	49.7 ± 3.8	46.6 ± 3.6	< 0.001
Signal strength	65.8 ± 6.8	66.0 ± 7.4	0.090
OCT parameters
CFT, μm	251.0 ± 23.1	250.2 ± 25.3	0.872
PFT, μm	315.7 ± 16.8	313.7 ± 20.8	0.582
Inner retinal CFT, μm	70.0 ± 12.7	71.0 ± 14.1	0.702
Inner retinal PFT, μm	128.0 ± 9.8	126.6 ± 11.8	0.521

PAD: Peripheral arterial disease; OHA: Oral hypoglycemic agents; HTN: Hypertension; DR: Diabetic retinopathy; NPDR: Non-proliferative diabetic retinopathy; PDR: Proliferative diabetic retinopathy; HbA1c: Glycated hemoglobin; HDLc: High-density lipoprotein cholesterol; LDLc: Low-density lipoprotein cholesterol; eGFR: Estimated glomerular filtration rate; BMI: Body mass index; SBP: Systolic blood pressure; DBP: Diastolic blood pressure; logMAR: Logarithm of the minimum angle of resolution; OCTA: Optical coherence tomography angiography; SCP: Superficial capillary plexus; VD: Vascular density; DCP: Deep capillary plexus; OCT: Optical coherence tomography; CFT: Central foveal thickness; PFT: Parafoveal thickness.

Parafoveal retinal VD and PAD

As SCP and DCP parafoveal VD were significantly reduced in patients with PAD, we investigated whether lower values of these parameters were associated with PAD, as well as the development of vision-threatening DR. Participants were categorized into quartiles based on the SCP parafoveal VD (Tables 2 and 3). Lower SCP parafoveal VD was significantly associated with higher rates of proliferative DR (PDR) (P = 0.003). Lower SCP parafoveal VD was also correlated with a higher prevalence of PAD (P = 0.019) and IRF (P < 0.001). When participants were grouped according to DCP parafoveal VD, lower DCP parafoveal VD was similarly linked to a higher prevalence of PAD (P < 0.001), as well as increased risks of IRF (P < 0.001) and DME (P = 0.003).

Table 2 Clinical characteristics of participants by capillary plexus parafoveal vascular density-superficial capillary plexus parafoveal vessel density quartiles, n (%)/mean ± SD.

	SCP, quartile 1 (n = 53)	SCP, quartile 2 (n = 53)	SCP, quartile 3 (n = 53)	SCP, quartile 4 (n = 53)	P value
Parafoveal VD, %	41.0 ± 2.9	45.0 ± 0.6	47.7 ± 0.6	50.7 ± 1.6
Age, years	57.3 ± 11.7	60.1 ± 8.4	58.4 ± 8.8	54.0 ± 8.7	0.051
Sex, male	38 (71.7)	33 (62.3)	36 (67.9)	32 (60.4)	0.593
Smoking status					0.216
Never	25 (47.2)	30 (56.6)	28 (52.8)	30 (56.6)
Current	20 (37.7)	11 (20.8)	10 (18.9)	10 (18.9)
Ex-smoker	8 (15.1)	12 (22.6)	15 (28.3)	13 (24.5)
Smoking, pack-years	10.7 ± 14.2	12.6 ± 17.2	8.4 ± 14.3	8.1 ± 13.6	0.231
Duration of diabetes, years	17.0 ± 9.4	19.3 ± 7.9	18.1 ± 7.0	16.0 ± 7.7	0.385
Medication					0.527
Insulin only	0 (0.0)	1 (1.9)	0 (0.0)	0 (0.0)
OHA	31 (58.5)	37 (69.8)	36 (67.9)	36 (67.9)
Insulin + OHA	22 (41.5)	15 (28.3)	17 (32.1)	17 (32.1)
HTN	25 (47.2)	26 (49.1)	29 (54.7)	14 (26.4)	0.020
DR stage					0.003
No DR	3 (5.7)	5 (9.4)	7 (13.2)	10 (18.9)
Mild NPDR	15 (28.3)	22 (41.5)	27 (50.9)	30 (56.6)
Moderate NPDR	15 (28.3)	15 (28.3)	11 (20.8)	6 (11.3)
Severe NPDR	13 (24.5)	10 (18.9)	8 (15.1)	5 (9.4)
PDR	7 (13.2)	1 (1.9)	0 (0.0)	2 (3.8)
Biological parameters
HbA1c	7.9 ± 1.3	7.7 ± 1.1	7.8 ± 1.3	7.6 ± 1.2	0.181
Fasting glucose	144.8 ± 44.8	145.7 ± 46.6	147.7 ± 54.6	148.3 ± 45.4	0.669
Total cholesterol	145.5 ± 37.7	142.5 ± 32.2	150.7 ± 38.2	138.8 ± 29.8	0.583
Triglyceride	129.5 ± 76.4	127.7 ± 68.6	155.6 ± 92.8	114.9 ± 45.0	0.723
HDLc	44.1 ± 11.5	46.9 ± 12.6	42.5 ± 9.3	45.4 ± 10.3	0.942
LDLc	91.1 ± 32.4	87.4 ± 26.2	96.7 ± 30.8	86.3 ± 23.3	0.765
Creatinine	1.0 ± 0.7	1.1 ± 0.9	1.1 ± 1.0	0.9 ± 0.2	0.269
eGFR	81.4 ± 22.5	76.2 ± 24.2	78.6 ± 19.8	88.7 ± 18.0	0.067
Clinical parameters
BMI, kg/m2	25.3 ± 3.7	24.8 ± 3.7	26.4 ± 3.5	26.1 ± 3.9	0.081
SBP, mmHg	132.0 ± 19.8	132.9 ± 19.1	134.7 ± 15.7	133.9 ± 18.5	0.508
DBP, mmHg	74.4 ± 11.8	75.9 ± 11.9	75.5 ± 9.5	75.8 ± 10.6	0.562
Visual acuity, logMAR	0.09 ± 0.10	0.06 ± 0.09	0.08 ± 0.09	0.05 ± 0.07	0.067
OCTA parameters
FAZ size, mm2	0.34 ± 0.10	0.33 ± 0.01	0.30 ± 0.11	0.34 ± 0.10	0.516
FAZ perimeter, mm	2.4 ± 0.4	2.3 ± 0.3	2.2 ± 0.4	2.4 ± 0.4	0.433
SCP fovea VD, %	13.3 ± 3.9	14.6 ± 4.9	16.6 ± 4.8	16.0 ± 4.9	< 0.001
DCP fovea VD, %	25.8 ± 6.6	28.3 ± 5.9	30.3 ± 5.9	28.8 ± 5.7	0.004
DCP parafoveal VD, %	46.8 ± 4.4	49.2 ± 3.4	50.5 ± 2.8	51.0 ± 3.6	< 0.001
Signal strength	66.7 ± 6.9	66.1 ± 6.1	69.5 ± 6.9	70.3 ± 6.9	0.001
OCT parameters
CFT, μm	251.5 ± 25.1	250.0 ± 23.2	256.0 ± 24.5	246.3 ± 19.7	0.508
PFT, μm	314.7 ± 20.3	313.0 ± 19.0	316.3 ± 13.8	317.9 ± 15.3	0.218
Inner retinal CFT, μm	69.9 ± 13.0	69.8 ± 12.1	73.1 ± 14.6	67.7 ± 11.2	0.679
Inner retinal PFT, μm	125.5 ± 11.3	124.9 ± 10.6	128.9 ± 8.4	132.0 ± 8.1	< 0.001
Presence of PAD	11 (20.8)	8 (15.1)	5 (9.4)	1 (1.9)	0.019
Development of IRF	25 (47.2)	12 (22.6)	9 (17.0)	7 (13.2)	< 0.001
Development of DME	6 (11.3)	4 (7.5)	2 (3.8)	3 (5.7)	0.473

SCP: Superficial capillary plexus; VD: Vascular density; OHA: Oral hypoglycemic agents; HTN: Hypertension; DR: Diabetic retinopathy; NPDR: Non-proliferative diabetic retinopathy; PDR: Proliferative diabetic retinopathy; HbA1c: Glycated hemoglobin; HDLc: High-density lipoprotein cholesterol; LDLc: Low-density lipoprotein cholesterol; eGFR: Estimated glomerular filtration rate; BMI: Body mass index; SBP: Systolic blood pressure; DBP: Diastolic blood pressure; logMAR: Logarithm of the minimum angle of resolution; OCTA: Optical coherence tomography angiography; FAZ: Foveal avascular zone; DCP: Deep capillary plexus; OCT: Optical coherence tomography; CFT: Central foveal thickness; PFT: Parafoveal thickness; PAD: Peripheral arterial disease; IRF: Intra-retinal fluid; DME: Diabetic macular edema.

Table 3 Clinical characteristics of participants by capillary plexus parafoveal vascular density-deep capillary plexus parafoveal vessel density quartiles, n (%)/mean ± SD.

	DCP, quartile 1 (n = 53)	DCP, quartile 2 (n = 53)	DCP, quartile 3 (n = 53)	DCP, quartile 4 (n = 53)	P value
Parafoveal VD, %	44.0 ± 2.3	48.5 ± 1.0	51.2 ± 0.7	53.9 ± 1.1	< 0.001
Age, years	58.0 ± 10.5	58.1 ± 9.5	57.7 ± 8.9	55.9 ± 9.9	0.273
Sex, male	38 (70.4)	28 (52.8)	41 (77.4)	32 (60.4)	0.042
Smoking status					0.010
Never	22 (43.1)	35 (68.6)	17 (37.8)	25 (49.0)
Current	16 (31.4)	3 (5.9)	17 (37.8)	15 (29.4)
Ex-smoker	13 (25.5)	13 (25.5)	11 (24.4)	11 (21.6)
Smoking, pack-years	27.4 ± 8.1	33.8 ± 9.3	24.5 ± 11.6	28.4 ± 12.3	0.695
Duration of diabetes, years	17.1 ± 7.9	18.4 ± 8.5	18.9 ± 8.0	15.8 ± 8.1	0.521
Medication					0.563
Insulin only	0 (0.0)	0 (0.0)	1 (1.9)	0 (0.0)
OHA	36 (66.7)	33 (62.3)	38 (71.7)	33 (62.3)
Insulin + OHA	18 (33.3)	20 (37.7)	14 (26.4)	20 (37.7)
HTN	25 (46.3)	19 (35.8)	27 (50.9)	23 (43.4)	0.460
DR stage					< 0.001
No DR	4 (7.4)	3 (5.7)	6 (11.3)	13 (24.5)
Mild NPDR	15 (27.8)	23 (43.4)	28 (52.8)	28 (52.8)
Moderate NPDR	11 (20.4)	16 (30.2)	13 (24.5)	7 (13.2)
Severe NPDR	16 (29.6)	11 (20.8)	6 (11.3)	3 (5.7)
PDR	8 (14.8)	0 (0.0)	0 (0.0)	2 (3.8)
Biological parameters
HbA1c	7.9 ± 1.3	7.9 ± 1.2	7.6 ± 1.1	7.6 ± 1.3	0.037
Fasting glucose	150.5 ± 47.2	151.2 ± 51.6	143.9 ± 37.6	140.2 ± 53.0	0.191
Total cholesterol	146.4 ± 40.0	142.3 ± 29.2	143.9 ± 38.5	145.8 ± 31.2	0.982
Triglyceride	136.4 ± 70.3	129.4 ± 62.5	119.5 ± 69.7	144.8 ± 90.8	0.739
HDLc	43.9 ± 10.8	45.7 ± 11.2	45.5 ± 12.7	44.0 ± 9.3	0.998
LDLc	93.2 ± 33.7	87.9 ± 25.9	89.2 ± 29.9	91.9 ± 24.3	0.887
Creatinine	1.0 ± 0.7	1.0 ± 0.3	1.2 ± 1.3	0.9 ± 0.3	0.852
eGFR	80.6 ± 23.2	79.3 ± 20.2	81.9 ± 24.1	82.8 ± 19.0	0.485
Clinical parameters
BMI, kg/m2	25.5 ± 3.9	25.7 ± 3.8	25.9 ± 4.1	25.7 ± 3.1	0.694
SBP, mmHg	138.8 ± 19.6	131.5 ± 19.4	133.7 ± 18.6	129.2 ± 13.7	0.016
DBP, mmHg	78.4 ± 11.4	74.2 ± 11.2	74.4 ± 10.8	74.7 ± 10.0	0.107
Visual acuity, logMAR	0.08 ± 0.12	0.07 ± 0.09	0.06 ± 0.07	0.06 ± 0.07	0.083
OCTA parameters
FAZ size, mm2	0.34 ± 0.11	0.33 ± 0.10	0.32 ± 0.09	0.32 ± 0.10	0.119
FAZ perimeter, mm	2.4 ± 0.4	2.4 ± 0.4	2.3 ± 0.4	2.3 ± 0.4	0.158
SCP fovea VD, %	14.5 ± 5.0	15.6 ± 5.2	14.8 ± 4.1	15.8 ± 4.9	0.314
DCP fovea VD, %	43.6 ± 4.6	46.2 ± 3.3	46.6 ± 2.9	48.1 ± 3.5	< 0.001
DCP parafoveal VD, %	26.2 ± 7.2	29.1 ± 5.6	28.3 ± 5.1	29.8 ± 6.4	0.009
Signal strength	63.6 ± 5.5	66.8 ± 7.4	70.6 ± 5.2	71.9 ± 6.2	< 0.001
OCT parameters
CFT, μm	250.1 ± 25.2	252.4 ± 22.3	251.3 ± 23.3	250.0 ± 22.9	0.925
PFT, μm	319.3 ± 20.2	318.0 ± 15.5	313.9 ± 16.9	310.6 ± 15.0	0.004
Inner retinal CFT, μm	69.8 ± 14.4	70.1 ± 10.9	70.4 ± 13.8	70.1 ± 12.4	0.893
Inner retinal PFT, μm	128.5 ± 11.2	128.0 ± 9.7	128.0 ± 10.3	126.7 ± 9.1	0.368
Presence of PAD	14 (25.9)	9 (17.0)	0 (0.0)	3 (5.7)	< 0.001
Development of IRF	24 (45.3)	12 (24.0)	4 (8.2)	4 (8.2)	< 0.001
Development of DME	10 (18.9)	3 (5.7)	1 (1.9)	1 (1.9)	0.003

DCP: Deep capillary plexus; VD: Vascular density; OHA: Oral hypoglycemic agents; HTN: Hypertension; DR: Diabetic retinopathy; NPDR: Non-proliferative diabetic retinopathy; PDR: Proliferative diabetic retinopathy; HbA1c: Glycated hemoglobin; HDLc: High-density lipoprotein cholesterol; LDLc: Low-density lipoprotein cholesterol; eGFR: Estimated glomerular filtration rate; BMI: Body mass index; SBP: Systolic blood pressure; DBP: Diastolic blood pressure; logMAR: Logarithm of the minimum angle of resolution; OCTA: Optical coherence tomography angiography; FAZ: Foveal avascular zone; SCP: Superficial capillary plexus; OCT: Optical coherence tomography; CFT: Central foveal thickness; PFT: Parafoveal thickness; PAD: Peripheral arterial disease; IRF: Intra-retinal fluid; DME: Diabetic macular edema.

Clinical characteristics associated with the development of early DME

Figure 1 shows HRs for the association between PAD and development of IRF and DME. PAD was associated with an increased incidence of IRF [adjusted HR: 5.44; 95%CrI: 2.11-15.18; P(HR > 2) = 98%] and DME [adjusted HR: 3.54; 95%CrI: 1.07-11.44; P(HR > 2) = 83%]; the wide CrIs, particularly for DME, indicate limited precision in effect magnitude. Figure 2 shows the survival probability and cumulative hazard for the development of IRF and DME. Patients with PAD demonstrated both earlier onset and a higher risk of developing both IRF and DME (P < 0.001 and P = 0.003, respectively). Patients with early onset DME (within 6 months of the initial visit) had a significantly higher percentage of PAD (37.5% vs 9.7%) and lower SCP parafoveal VD (43.8 ± 4.6 vs 46.3 ± 3.8; P = 0.015) than those without early onset DME (Supplementary Table 2).

Figure 1 Association between peripheral arterial disease and retinal microvascular complications. Forest plot showing adjusted hazard ratios for intra-retinal fluid and diabetic macular edema development. Bayesian Cox estimates are shown with 95% credible intervals, whereas inverse probability of treatment weighting estimates are shown with 95% confidence intervals. Three methods: Bayesian Cox with skeptical prior N(0,1) (primary), Bayesian Cox with Zhang-based prior (sensitivity), and inverse probability of treatment weighting (triangulation). Dashed line = hazard ratio 1; dotted line = hazard ratio 2. PAD: Peripheral arterial disease; IRF: Intra-retinal fluid; DME: Diabetic macular edema; IPTW: Inverse probability of treatment weighting; HR: Hazard ratio; CI: Confidence interval.

Figure 2 Kaplan-Meier survival curves by peripheral arterial disease status. A: Intra-retinal fluid event-free survival; B: Intra-retinal fluid cumulative hazard; C: Diabetic macular edema event-free survival; D: Diabetic macular edema cumulative hazard. Black = peripheral arterial disease- (n = 187); blue = peripheral arterial disease + (n = 25). IRF: Intra-retinal fluid; DME: Diabetic macular edema; HR: Hazard ratio; PAD: Peripheral arterial disease.

Sensitivity analyses

To assess robustness, we evaluated results across five prior specifications and after adjustments (Supplementary Figure 3). Under all priors, IRF maintained P(HR > 2) exceeding 96%: Skeptical N(0,1) primary (HR 5.44, 98%), weakly informative (HR 7.11, 99%), non-informative (HR 8.72, 99.6%), moderate (HR 4.01, 97%), and Zhang-based (HR 4.16, 97%). For DME, P(HR > 2) ranged from 79% to 95% across priors, with lower certainty reflecting the sparse events (n = 15). Additional covariate adjustment did not alter the direction of association, but these sparse-event sensitivity analyses do not exclude residual or unmeasured confounding.

Subgroup analyses

Exploratory PAD estimates were positive across DR severity strata (Supplementary Figure 4). In mild-to-moderate non-PDR (NPDR; n = 166), PAD was associated with higher IRF incidence [HR 4.17; 95%CrI: 1.94-8.50; P(HR > 2) = 96%]. In severe NPDR/PDR (n = 46), the PAD estimate was positive [HR 2.87; 95%CrI: 0.91-8.08; P(HR > 2) = 77%], though with greater uncertainty because only four PAD-positive observations were available. The interaction P value was 0.652, but this underpowered analysis does not establish absence of heterogeneity. Exploratory stratification by DCP vessel density tertiles showed varying point estimates: PAD had the largest estimate in eyes with low baseline DCP (T1: HR 3.13; 95%CrI: 1.44-6.44), a smaller estimate in medium DCP (T2: HR 1.63), and a near-null estimate in high DCP (T3: HR 0.93). Therefore, the severe NPDR/PDR subgroup estimate should be interpreted as exploratory rather than definitive.

Methodological triangulation

IPTW produced directionally similar estimates, although the wide intervals also indicate limited precision. Covariate balance was achieved after IPTW (Supplementary Figure 2). IPTW yielded IRF HR 3.39 [95% confidence interval (CI): 1.44-7.99] and DME HR 5.37 (95%CI: 1.72-16.77), both exceeding the HR = 1 threshold (Figure 1). The prior-to-posterior shift shows updating under the fitted model but does not overcome sparse-event uncertainty (Supplementary Figure 5). E-values were retained only as descriptive sensitivity metrics: The values were 10.3 for IRF (3.7 for the lower CrI bound) and 6.5 for DME (1.34 for the lower CrI bound), but they do not eliminate bias from unmeasured or inadequately measured confounding.

Mediation analysis

We explored whether DCP vessel density statistically accounted for part of the PAD-IRF association. The mediation analysis estimated an indirect proportion of 34% (95%CI: 15%-73%) through lower DCP vessel density (Supplementary Figure 6). Because the retrospective analysis may be affected by unmeasured exposure-mediator and mediator-outcome confounding, this estimate is a statistical decomposition rather than proof of a causal direct or indirect pathway.

DISCUSSION

Key findings

This study showed that patients with T2DM and PAD had lower parafoveal VD in both the SCP and DCP, as measured by OCTA. Reduced parafoveal VD was associated with a higher incidence of IRF and DME, and PAD was also associated with these outcomes. These findings identify parafoveal VD as a candidate observational marker rather than a validated biomarker for screening or prognosis.

Comparisons with previous studies

Several recent studies have explored the relationship between PAD and retinal microvascular impairment using OCTA[10]. Prior research by Wintergerst et al[11] demonstrated that reduced SCP and DCP vessel density were correlated with PAD severity (as classified by the Fontaine stage) and systemic vascular dysfunction, showing a moderate positive correlation with ABI. Our study aligns with these findings, confirming that PAD, a representative marker of systemic atherosclerosis, is associated with lower retinal VD in patients with diabetes.

In T2DM, OCTA-derived parameters have been shown to correlate with specific systemic factors, including platelet count, renal function, and lipid profiles[21]. Other systemic factors, such as chronic kidney disease and cardiovascular diseases, have also been associated with significant reductions in retinal and choroidal vessel density, as well as enlargement of the FAZ[22,23]. We have previously demonstrated that subclinical carotid atherosclerosis is associated with retinal parafoveal VD decline[9]. Similarly, findings from the Fremantle Diabetes Study reported that markers of carotid arterial disease in individuals with T2DM are associated with reduced retinal microvascular parameters, as measured by OCTA[24]. Collectively, these studies suggest that systemic vascular conditions uniquely affect retinal microvasculature, with distinct OCTA signatures reflecting their mechanisms. This highlights OCTA as a valuable non-invasive tool for monitoring systemic vascular diseases via ocular manifestations. A recent United Kingdom Biobank study by Zhang et al[20] found that PAD was associated with composite diabetic microvascular complications, though the specific association with retinopathy alone showed only a trend in T2DM patients, suggesting potential effect modification by context.

Plausible underlying mechanisms

The observed relationship between PAD and DR can be explained through several interconnected mechanisms. As PAD is a systemic manifestation of generalized atherosclerosis and endothelial dysfunction, patients with PAD often have generalized microcirculatory damage, which often coexists with diabetes-induced chronic inflammation, oxidative stress, and endothelial impairment. The retinal microvasculature, owing to its end-arterial nature and high oxygen demand, is highly sensitive to systemic vascular insults, such as those prevalent in PAD, rendering it susceptible to ischemia[25]. Chronic retinal ischemia may lead to endothelial damage, impaired autoregulation, and capillary dropout, all contributing to reduced retinal VD.

Furthermore, this ischemic state due to the decline of retinal capillary increases the production of angiogenic and permeability factors, particularly vascular endothelial growth factor (VEGF)[26,27]. Elevated levels of VEGF and other factors that cause hyperpermeability weaken endothelial tight junctions and increase endothelial transcytosis, leading to a higher risk of IRF/DME development in patients with PAD, as observed in our study[28]. Thus, retinal microvascular alterations may represent both direct ischemic damage and indirect effects mediated by systemic inflammatory and angiogenic responses associated with PAD.

Additionally, diabetes-related microangiopathy and PAD-related macroangiopathy may share overlapping pathophysiological pathways, including enhanced oxidative stress, chronic inflammation, dyslipidemia, and hyperglycemia-induced endothelial dysfunction. These shared pathways further amplify the susceptibility of patients with diabetes and coexisting PAD to severe retinal microvascular damage and macular edema. Adjustment for albuminuria increased rather than attenuated the PAD-IRF estimate (HR: 5.44-6.68). Because this exploratory change may reflect model instability or covariance among measured factors, it should not be interpreted as proving a suppressor effect or an independent causal pathway.

Clinical and policy implications

These observational findings suggest that PAD may be a marker of systemic vascular burden in patients with T2DM, but they do not establish that PAD-specific OCTA screening or intensified retinal surveillance improves outcomes. Clinical monitoring should therefore continue to follow established diabetic retinopathy care standards. Prospective multicenter studies are needed before PAD status is used to alter retinal surveillance.

Strengths and limitations

A major strength of our study is its novel focus on both VD and DME characteristics in patients with diabetes and coexisting PAD, addressing a critical gap in prior research that largely excluded eyes with edema. Our study also employed a Bayesian-primary analytical framework with five-prior sensitivity analysis (Supplementary Figures 3 and 5) and IPTW triangulation. The Bayesian framework enabled posterior probability statements and regularization, but it cannot replace information absent from sparse DME events. Across five priors, the estimated direction was unchanged, but the wide posterior intervals indicate that effect magnitude remains uncertain. The mediation analysis provides an exploratory mechanistic hypothesis; it does not establish that one-third of a causal PAD effect operates through DCP capillary dropout, and unmeasured common causes of the mediator and outcome remain possible. However, our study has some limitations. These include a small PAD group (25 eyes from 16 patients), only 15 incident DME events, a single-center setting, and a retrospective design, resulting in wide intervals, uncertain effect magnitude, and limited causal interpretation. Nine patients contributed both eyes; the primary models included a patient-level random intercept, and a one-eye-per-patient sensitivity analysis preserved the direction of association. Diabetes duration was included in the IRF model, whereas smoking status and insulin therapy were available descriptively but were not included in the primary outcome models. Named antihypertensive agents, statin use, antiplatelet therapy, a validated PAD-severity measure, concomitant carotid disease, and overall cardiovascular disease burden were not available in analysis-ready form. Residual and unmeasured confounding therefore remain possible, and E-values cannot resolve this limitation. OCTA artifacts and media opacity in patients with DME may also affect vessel density measurements. Future multicenter prospective longitudinal studies are warranted to validate these findings and further explore the pathophysiological mechanisms linking PAD with diabetic retinal complications. ABI was obtained as part of real-world risk-based diabetic complication screening rather than universal screening of all patients with T2DM; therefore, the cohort may be enriched for vascular risk, which can limit generalizability to unselected T2DM populations and should be considered when interpreting absolute event rates and transportability of the effect estimates. Only four PAD-positive observations were available in the severe NPDR/PDR subgroup, so that subgroup estimate should be regarded as exploratory rather than definitive.

CONCLUSION

In this retrospective cohort, PAD was associated with lower retinal vessel density and higher incidence of IRF and DME in patients with T2DM. Because the PAD sample and DME event count were small and residual confounding remains possible, the magnitude and clinical utility of these associations are uncertain. Prospective multicenter studies are required before PAD status is used to modify retinal surveillance.