Optical Coherence Tomography Angiography Biomarkers in Thai Patients With Diabetic Nephropathy: A Diabetic Eye and Kidney Diseases (DEK-D) Study

Abstract

Purpose: To identify optical coherence tomography angiography (OCTA) biomarkers to predict the diabetic nephropathy (DN) and their associations with 24-hour urine albumin levels in diabetic patients.

Methods: This cross-sectional, observational study examined 186 eyes from 93 individuals subdivided into three groups according to 24-hour urine albumin levels: no DN, early DN, and late DN. Vessel density (VD), fractal dimension, foveal avascular zone area, intercapillary area, central retinal thickness, and subfoveal choroidal thickness were measured from OCTA images to determine their association with the DN stages.

Results: VD values of the superficial capillary plexus, deep capillary plexus, and whole retina were significantly lower in the early DN group compared to the no DN group (adjusted P = 0.042, 0.016, and 0.008, respectively). VD values for the deep capillary plexus and whole retina were significantly decreased in the late DN group compared to the no DN group (adjusted P = 0.025 and 0.021, respectively). Mean fractal dimension, intercapillary area, foveal avascular zone area, central retinal thickness, and subfoveal choroidal thickness were not statistically different among the three groups.

Conclusions: VD may be a useful parameter for the early non-invasive screening of DN. Further studies in larger populations are needed to establish a cutoff value for detection.

Translational Relevance: This study investigated the association of each retinal vasculature measurement by OCTA and diabetic nephropathy status which could serve as an alternative way to screen for albuminuria.

Introduction

It is estimated that more than 560 million adults suffer from diabetes mellitus (DM), and this number is expected to reach 783 million by 2045.¹ Approximately one-third of diabetic patients develop diabetic retinopathy (DR), and 40% of diabetic patients develop chronic kidney disease (CKD).¹ Diabetic nephropathy (DN) is the leading cause of end-stage renal disease, requiring renal replacement therapy in most of the patients with CKD.² The estimated overall cost for treating patients with CKD in Thailand was about US$285 million in 2020 and increased to US$294 million in 2021.³ 

Both DR and DN result from similar diabetic microvasculopathy.⁴ Hence, retinal microvascular abnormalities may be associated with renal vascular pathology and the development of CKD. Furthermore, measurement of anatomical changes, commonly detected earlier than functional changes, might offer the opportunity to prevent the development of CKD, which has been proven to be associated with excess mortality in diabetic patients.^5,6 

With the recent development of optical coherence tomography angiography (OCTA), retinal capillaries can be examined thoroughly without the need for invasive injection of fluorescein dye. OCTA has proven to be useful in detecting microaneurysm, enlarged foveal avascular zone (FAZ), and neovascularization in DR.⁷ Retinal and choroidal vascular disorders evidenced by OCTA might also be an early sign of anatomical changes in DN. These changes may precede functional changes detected by blood or urine chemistry, such as blood urea nitrogen (BUN), serum creatinine (SCr), and urine albumin-to-creatinine ratio.⁵ Early detection of DN can lead to earlier treatment and prevention of the progression of the disease, which would save lives, lessen the burden of disease for both patients and relatives, and eventually reduce costs within the healthcare system. However, data on OCTA parameters in the early stages of DN remain limited to date, and only a few studies have explored the relationship between retinal microcirculation and DN.⁵ The present study aimed to analyze OCTA parameters in association with albuminuria status. 

Methods

This cross-sectional, observational study of patients with type 2 diabetes was performed at an outpatient clinic at King Chulalongkorn Memorial Hospital, Bangkok, Thailand. This study was registered on the Thai Clinical Trial Registry (TCTR20210308001) and was approved by the institutional review board of the Faculty of Medicine, Chulalongkorn University, Bangkok, Thailand (COA No. 549/2021). The protocol of this study followed the tenets of the Declaration of Helsinki. All information regarding the study, including risks and benefits, was discussed with participants before informed consent was obtained. Participants who were 18 years of age or older and who had recent 24-hour urine albumin testing not older than 3 months were included and categorized into three groups according to their micro- and macroalbuminuria status: (1) no DN group (24-hour urine albumin level below 30 mg), (2) early DN group (24-hour urine albumin level from 30–300 mg), and (3) late DN group (24-hour urine albumin level higher than 300 mg).⁸ Participants with a history of ocular trauma, surgery other than uncomplicated cataract surgery, retinal anomaly other than DR, corneal disorders, strabismus, nystagmus, uveitis, glaucoma, optic nerve abnormalities, neurological disorders affecting the visual pathways, diabetic macular edema, tractional retinal detachment, retinal laser or intravitreal anti-vascular endothelial growth factor treatment, or media opacities were excluded from the study. Other exclusion criteria included active urinary tract infection, any non-diabetic renal diseases, end-stage renal disease requiring renal replacement therapy, and inadequate cooperation for the OCTA examination. Both eyes were included in the study if they were eligible. 

Data Collection

The 24-hour urine albumin levels were obtained from medical records. Blood samples were collected and analyzed for BUN, SCr, and serum glycated hemoglobin (HbA1c). Blood pressure was measured using an automated sphygmomanometer while the subject was in a sitting position. Distance visual acuity was tested using an Early Treatment of Diabetic Retinopathy Study chart (Precision Vision, Woodstock, IL). Slit-lamp and dilated fundus examinations were done by a single investigator (Nu.S.). Axial length was then measured using the IOLMaster 700 optical biometer (Carl Zeiss Meditec, Jena, Germany). Widefield fundus photographs were taken by a masked technician using a ZEISS CLARUS 700 ultra-widefield retinal camera (Carl Zeiss Meditec). The images were randomly graded by two independent masked assessors (P.P., W.K.) as no diabetic retinopathy, mild non-proliferative diabetic retinopathy (NPDR), moderate NPDR, severe NPDR, or proliferative diabetic retinopathy according to the International Clinical Diabetic Retinopathy and Diabetic Macular Edema Disease Severity Scales.⁹ In case of any disagreement between two graders, a decision was made by a third grader (Nu.S.). 

OCT images were taken using a SPECTRALIS OCT system (Heidelberg Engineering, Heidelberg, Germany) in enhanced depth imaging mode, with 100 images averaged by the Automatic Real Time averaging function for one horizontal line scans at the foveal center. Only images with signal strength of 20 or above were used for any analysis. Central retinal thickness (CRT) was measured using automatic segmentation from the vitreoretinal interface to the outer boundary of retinal pigment epithelium–Bruch membrane complex at the foveal center, and the location of the foveal center was confirmed by a masked investigator (Nu.S.). Subfoveal choroidal thickness (CT) was manually measured by a single masked assessor (Nu.S.) using the Heidelberg SPECTRALIS software measurement function. Subfoveal CT is defined as the perpendicular distance between the outer border of the retinal pigment epithelium–Bruch membrane complex and the chorioscleral border under the fovea.¹⁰ 

OCTA images were obtained by a masked technician using a ZEISS PLEX Elite 9000 Swept-Source OCT system with 6 × 6-mm field of view. The en face OCTA image was segmented with an inner boundary at 3 µm beneath the internal limiting membrane, and the outer boundary was set at 15 µm beneath the inner plexiform layer (IPL) to obtain images of the superficial capillary plexus (SCP). The en face image was segmented with an inner boundary of 15 µm beneath the IPL, and the outer boundary was set at 70 µm beneath the IPL to obtain images of the deep capillary plexus (DCP). Per the manufacturer's suggestion, only images with signal strength of 6 and above were used,⁷ and the investigator also examined the images again for any artifacts or missing portions. Participants were excluded from the study if their image quality failed to meet the standard after the second image acquisition. 

OCTA Image Analysis

To calculate vessel density (VD), OCTA images of the SCP, DCP, and whole retina were used. Images were converted into binary images using global Otsu's auto-threshold algorithm¹¹ for ImageJ (National Institutes of Health, Bethesda, MD) without any contrast enhancement (Fig. 1). Global threshold binarization was reported to be more efficacious for OCTA metrics in DR eyes.¹² VD is defined as the area occupied by all the blood vessels divided by the total image area. The fractal dimension (FD) was automatically calculated from skeletonized images using the FracLac plug-in for ImageJ (Figs. 1, 2). Intercapillary area is defined as a proportion of area without vessels over the total area in a binarized en face OCTA image. FAZ area was calculated using an automated macro function described by Ishii et al.¹³ The images were then inspected by a masked investigator (Nu.S.). If the outline of the FAZ area was not accurate, then the investigator manually drew a new FAZ outline using ImageJ (Fig. 3). 

Figure 1.

 

OCTA image (6 × 6 mm) of SCP (left column), binarized image (middle column), and skeletonized image by ImageJ (right column) from individuals with no DR in the no DN group (A), early DN group (B), and late DN group (C).

Figure 2.

 

OCTA images (6 × 6 mm) of SCP (left column), DCP (middle column), and whole retina (right column) from individuals with no DR in the no DN group (A), early DN group (B), and late DN group (C).

Figure 3.

 

OCTA images (6 × 6 mm) before (left) and after (right) the FAZ area was extracted.

Data Analysis

A sample size of 186 eyes with a power of 80% was calculated for a standard deviation (SD) of 3.4 in VD of SCP difference between normoalbuminuria and microalbuminuria from a previous study,⁵ and a minimal clinical difference of 2 was considered significant. Statistical analysis was performed using Stata 14.0 (StataCorp, College Station, TX). Data are presented as mean ± SD, frequency, or percentage where appropriate. The Shapiro–Wilk W-test was used to test for normality. Baseline characteristic data were compared using the χ² test for categorical variables. Continuous parametric variables were compared among the three groups using one-way analysis of variance (ANOVA), and non-parametric variables were tested using the Kruskal–Wallis test. Due to the limited number of participants with DR, participants with any stage of DR were categorized into the same DR group. Multilevel linear mixed-model analysis with post hoc Bonferroni pairwise comparison was used to compare VD, intercapillary area, FAZ area, FD, and subfoveal CT among the three groups with adjustment for DR status, age, sex, HbA1c level, and axial length. Two-tailed P values were calculated using the mixed-model approach, considering the eye as a unit of analysis and participant identification as the random effect. Global P < 0.05 was considered statistically significant. For correlation analysis between the OCTA parameters and urine albumin, data from both eyes were used to evaluate the association, and Spearman's rank correlation coefficient (r_s) was used to analyze the data. 

Results

The study enrolled 94 volunteers (188 eyes), but one participant (two eyes) from the early DN group was excluded due to media opacities. All baseline characteristics are shown in Table 1. More male participants were commonly identified in the no DN and the late DN groups. There were no statistically significant differences in demographics for age, sex, or ethnicity, as well as blood pressure (systolic blood pressure [SBP] and diastolic blood pressure [DBP]), among the three groups. The highest SCr levels were measured in the late DN group (2.0 mg/dL), followed by the early DN group (1.41 mg/dL) and the no DN group (1.03 mg/dL). SCr concentration was found to be significantly correlated with DN severity (r_s = 0.62, P < 0.001). Moreover, serum HbA1c levels differed significantly among the three groups. 

Table 1.

 

Baseline Characteristics of Participants

 

All of the participants categorized as no DN had similar 24-hour urine albumin levels of 6.8 mg, corresponding to the lower limit of detection of the assay. In the early and the late DN groups, the mean 24-hour urine albumin levels were 191.8 mg (range, 53.68–285.90) and 3236.84 mg (range, 368.60–10506.50), respectively. Numbers of participants requiring insulin injection were identical within the three groups. There were no statistically significant differences in visual acuity among the three groups, and the majority of participants in all the three groups were phakic. Significant differences were observed among the three groups for the levels of DR severity (Fig. 4). The agreement between the two fundus graders was excellent (intraclass correlation coefficient = 0.87; 95% confidence interval [CI], 0.82–0.9). 

Figure 4.

 

Stacked bar chart showing proportions of DR severity among participants in each group.

Vessel Densities and Diabetic Nephropathy

The mean VD values for the three groups are shown in Table 2. Mean VD values for the SCP were significantly different based on DR severity (P = 0.038), with lower VD values being measured in participants with moderate or severe DR compared to no DR (P = 0.045). A trend of reduction in mean VD values of all layers was associated with increasing 24-hour urine albumin excretion (Fig. 5). There were significant differences in the mean VD values for the SCP, DCP, and whole retina between the no DN group and the late DN group (SCP P = 0.02; DCP P = 0.008; whole retina P = 0.023). However, when the analysis was adjusted for age, sex, HbA1c, DR status, axial length, and lens status, significant differences were only found for the mean VD values for the DCP and whole retina (DCP P = 0.025; whole retina P = 0.021). Moreover, mean VD of SCP and DCP values were similar in the no DN group and the early DN group (SCP P = 0.061; DCP P = 0.051). But, following adjustment for the aforementioned factors, there was a significant reduction in values for mean VD of the SCP and DCP (SCP P = 0.011; DCP P = 0.006). Interestingly, a significant decrease was also found in the mean VD values for the whole retina between the no DN group and the early DN group (P = 0.032), and the significant difference persisted after adjusting for the confounding factors (P = 0.003). Additionally, similar VD values were measured in all eyes with DN severity. 

Table 2.

 

Mean VD Values

 

Figure 5.

 

Scatterplots showing correlations between vessel densities of retinal layers and 24-hour urine albumin levels.

Logistic regression analysis between VD values and risk of DN progression showed that the risk of developing DN increased with decreasing VD values of the SCP (odds ratio [OR] = 1.13; 95% CI, 1.04–1.24; P = 0.005), DCP (OR = 1.16; 95% CI, 1.06–1.28; P = 0.002), and whole retina (OR = 1.13; 95% CI, 1.04–1.23; P = 0.004). Using Spearman's correlation analysis, we found a weak but significant negative correlation between 24-hour urine albumin level and the average VD value of the SCP, DCP, and whole retina (SCP r_s = −0.26, P = 0.013; DCP r_s = −0.30, P = 0.004; whole retina r_s = −0.28, P = 0.007). 

Other OCTA Parameters and Diabetic Nephropathy

Mean FD, intercapillary area, FAZ area, CRT, and subfoveal CT values are summarized in Table 3. FD tended to decrease with increasing 24-hour urine albumin; however, the significant difference was only found between the no DN group and the late DN group (P = 0.022), and significance was not reached after adjusting for confounding factors (P = 0.056). Logistic regression analysis using mean FD in the first quartile as a cut-off indicated that participants with FD < 1.8401 Db had a significantly increased risk of developing macroalbuminuria (OR = 2.25; 95% CI, 1.12–4.5; P = 0.022). A similar trend in increasing intercapillary area with increasing 24-hour urine albumin was measured; however, a significant increase was only found between the no DN group and the late DN group (P = 0.02), but after adjusting for confounding factors there were no significant differences (P = 0.139). No significant differences were found in FAZ area, CRT, and subfoveal CT between any groups. In addition, no significant correlations were found between 24-urine albumin level and intercapillary area, FAZ area, FD, CRT, or subfoveal CT (P = 0.125, 0.150, 0.065, 0.946, and 0.724, respectively). 

Table 3.

 

Mean OCTA Parameters Values

 

Discussion

DN is one of the major diabetic microvascular complications, and almost all organs can be affected due to the microvascular, systemic complications of DM. Although causal or chronological relationships have not been fully established, a collection of studies have reported clinical correlations between the incidence of DN and the presence of DR.^6,14–20 Histopathological data suggest that thickening of the capillary basement membrane, loss of pericytes in DR, and loss of podocytes are associated with DN progression.^21,22 A strong association has been found between retinal lesions detected by fundus examination and fluorescein angiography and declined renal function.^4,6,23–28 

The development of a non-invasive OCTA imaging system greatly facilitates the diagnosis of pre-clinical DR.⁷ Retinal VD was found to be associated with both micro- and macrovascular diseases including coronary artery disease.²⁹ Our present data showed that the mean decreases in VD of the SCP, DCP, and whole retina were significantly different between the early DN and the no DN groups, after adjusting for confounding factors. Additionally, mean VD values of the DCP and whole retina were significantly decreased in the late DN group compared to the no DN group. Of note, a similar relationship was previously reported by Cankurtaran et al.⁵ but only in the SCP layer, and significant differences in mean VD values between the early DN group and healthy individuals were also described.⁵ We also found that a reduction in VD values significantly increased the risk of developing any DN stage. We propose that the mean VD might represent a useful marker to monitor microalbuminuria, as decreased VD may reflect systemic capillary loss from hyperglycemia. 

Although an association between intercapillary area and estimated glomerular filtration rate (eGFR) was initially reported by Cheung et al.,³⁰ we could not find any significant difference among the three groups after adjusting for confounding factors. A previous study by Sng et al.²⁸ demonstrated a U-shaped relationship between the prevalence of CKD and quintiles of retinal vascular FD calculated from fundus photography. In our study, only the SCP FD values obtained from OCTA were significantly different between the no DN group and the late DN group. Moreover, we found that any FD value lower than 1.8401 significantly increased the risk of developing DN. Although the use of different imaging protocols and algorithms can affect the FD threshold values, identifying other factors affecting vascular FD remains important. 

As previously observed in one study,⁵ we could not find any associations between mean FAZ area and DN. In contrast, a prospective case-control study by Ahmadzadeh Amiri et al.³¹ described significantly larger FAZ areas for the SCP, DCP, and whole retina in participants developing late DN compared to the no DN group. 

Microalbuminuria is known to be significantly associated with major cardiovascular diseases and progressive kidney dysfunction and could be an indicator of systemic endothelial dysfunction.³² Apart from DM, prior studies have explored the relationship between retinal microcirculation and hypertension. Although retinal capillary rarefaction was found to be correlated with reduced eGFR, the association of VD values and microalbuminuria in hypertensive patients were still inconsistent.^33,34 Our results indicate that decreased VD values determined by OCTA could represent an early marker of microalbuminuria that requires further investigation and treatment. With an appropriate collection of data, OCTA screening might be included in routine DR screening to detect potential kidney complications and other systemic diabetic microvascular complications. 

To our knowledge, this study is one of the first to investigate the relationship between VD values on OCTA and 24-hour urine albumin, especially among patients with early DN. The strength of this study is notably the inclusion of a sample size calculation to ensure enough power for result differentiation, together with the use of 24-hour urine albumin, which reflects kidney functional evaluation and is not affected by age, gender, extreme muscle mass, or collection time.³⁵ Additionally, most of the measurements from the present study were done by automated software, thus reducing bias. 

Some limitations have to be pointed out, however. First, we designed a cross-sectional study that can only establish associations, not causal or chronological relationships. Moreover, no control group with healthy individuals was included. Second, VDs were measured using the whole image area and not separately measured in parafoveal or perifoveal areas. Third, although the sample size was calculated for comparing VD values, larger sample sizes could evidence stronger associations or correlations with other parameters. Another weakness of our study is that DR severity was graded using fundus photographs for which the field of view was limited. Missing and inaccurate data did not allow us to include the history of DM duration in the analysis; however, DR severity might be partially related to the DM duration. 

Finally, the discrepancy between our results and earlier findings^28,30,31 might result from the imaging protocols of each OCTA system. With our OCTA instrument, we had to use other software to analyze the images. The absence of consensus about a standard binarization algorithm could induce variation in VD values among studies. Manufacturing calibration could ensure greater accuracy and consistency of the measuring tools, particularly as vascular layers could be defined differently among the various instruments. Standardized protocols and segmentation are critical to establishing an effective screening tool. Artifacts, such as motion or segmentation, can also affect the results of a study. The integration of automated image processing and analysis software into OCTA instruments could be useful for future investigations. 

Conclusions

OCTA is increasingly being adopted in daily clinical practice. Our study found that decreased VD values can be detected in the early stages of DR and can have an impact on clinical outcomes. Our research suggests that retinal microcirculation changes occur concomitantly with analogous pathological changes in kidney. More detailed studies are needed to explore these relationships in larger, different populations to determine a standardized protocol and cutoff values. 

Acknowledgments

The authors warmly thank the Excellence Center for Critical Care Nephrology (ECCCN) and Ophthalmology Clinic at King Chulalongkorn Memorial Hospital for their support. The authors also thank the English editing service, Research Affairs, Faculty of Medicine, Chulalongkorn University, for their useful advice. Nuntachai Surawatsatien thanks the late Waraluck Supawatjariyakul, MD, for her constant assistance while she was working on this study. 

Supported by the 90th Anniversary of Chulalongkorn University Fund (Ratchadapiseksomphot Endowment Fund). 

Disclosure: N. Surawatsatien, None; P.F. Pongsachareonnont, None; K. Kulvichit, None; A. Varadisai, None; T. Somkijrungroj, None; A. Mavichak, None; W. Kongwattananon, None; D. Suwajanakorn, None; N. Phasukkijwatana, None; N. Srisawat, None