Synchronous evaluation of early neurovascular changes in macula and optic nerve head in early diabetic retinopathy using combined optical coherence tomography angiography parameters

Background This study aim to evaluate neurovascular changes in both macula and optic nerve head (ONH) in early stages of diabetic retinopathy (DR) using combined quantitative optical coherence tomography angiography (OCTA) parameters. Methods We studied 194 right eyes from 64 type 2 diabetic patients without DR (DM), 64 diabetic patients with nonproliferative diabetic retinopathy (NPDR) and 64 age-matched healthy controls. OCTA parameters were analyzed using binary logistic regression models and receiver operating characteristic (ROC) curves. Results There was signicant reduction of vessel density (VD) in both macula and ONH comparing DM patients with and without NPDR to controls. The area under the curve (AUC) of the ROC curve for NPDR versus control was 0.963 ( p <0.001), with sensitivity of 93.8% and specicity of 90.6%. The ROC curves for NPDR patients versus NoDR patients (including DM and control groups) exhibited an AUC of 0.923 ( p <0.001), with sensitivity of 90.6% and specicity of 82.8%. The thickness of the retinal nerve ber layer (RNFL) in both macula and ONH was not signicantly different among these three groups. Conclusions VD in both macula and ONH was simultaneously decreased prior to RNFL thinning in DM patients through the course proceeding from preclinical DR to NPDR. Combined analysis of macula and ONH parameters was an comprehensive and accurate OCTA metric to distinguish NPDR patients from healthy controls and DM patients without DR. All patients were examined with best-corrected visual acuity (BCVA) (LogMAR visual acuity), intraocular pressure (IOP), slit-lamp biomicroscopy, and, after pharmacologically dilating pupils, indirect ophthalmoscopy, color fundus photographs, OCTA and FFA (only in patients with NPDR). The results were reviewed by two ophthalmologists (XW and ZJL) independently and the diagnoses were made according to severity scale[14].


Background
Diabetic retinopathy (DR) is a leading cause of acquired blindness in working-age adults worldwide [1,2]. DR is classically thought to be a microvascular disorder that is likely to occur in the posterior pole. Several lines of evidence show that neurodegeneration, including decreases in peripapillary retinal nerve ber layer (RNFL) thickness is present before clinical DR in patients of diabetes mellitus (DM) [3]. A recent study showed positive correlations between peripapillary blood ow and RNFL thickness around the optic nerve head(ONH) [4]. However, these studies analyze the vessel density (VD) and RNFL of either the macula or ONH separately and therefore cannot conclude whether the DR changes of macula and ONH are synchronized pathological processes or are separate pathological changes. They cannot determine whether retinal microvascular disorders occur prior to retinal neural complications or vice versa from a comprehensive view. results are controversial across various reports. Several studies suggest that OCTA parameters vary with the stage of DR [9][10][11]. Some suggest that vessel parameters in the SCP are signi cantly different among the various stages of DR [11] while others nd that the vessel parameters in the DCP, not in the SCP, are signi cantly different between eyes of nonproliferative DR (NPDR) and proliferative DR (PDR) [12]. There are also reports indicating that the VD in the ONH [13] as well as the thickness of RNFL [4] are signi cantly different between the DM and normal subjects.
In these studies, the OCTA parameters of macula or ONH were studied individually. Few studies produced integrated analyses of the dominating parameters to eliminate the interactions among parameters.
Therefore, the aim of this study was to synchronously study the parameters of macula and ONH and to identify the neuropathy and vasculopathy in early DR from a comprehensive perspective. Hopefully, such a study would be helpful to abtain a better understanding of pathological process in DR and the clinical application of identifying early signs of DR.

Study design and subjects
This cross-sectional case-control study recruited 64 type 2 DM (T2DM) patients without any DR (DM group), 64 T2DM patients with NPDR (NPDR group), and 64 healthy control subjects (control group) who presented between October 2017 and February 2019. Patients in these three groups were age-matched. The study was approved by the research ethics committee of Sun Yat-sen Memorial Hospital, Sun Yat-sen University, and adhered to the tenets of the Declaration of Helsinki. Written informed consent was obtained from all participants. The data from the right eyes of all patients were collected and analyzed. Patients with diagnosis of T2DM, disease duration 1 year were included. Exclusion criteria included: ocular disease with severe opacity of refractive media that would in uence fundus examination and any other ocular disease that would affect ocular circulation, including glaucoma, myopia with refractive error greater than 3 diopters, age-related macular degeneration, macular edema and retinal artery/vein occlusion. All patients were examined with best-corrected visual acuity (BCVA) (LogMAR visual acuity), intraocular pressure (IOP), slit-lamp biomicroscopy, and, after pharmacologically dilating pupils, indirect ophthalmoscopy, color fundus photographs, OCTA and FFA (only in patients with NPDR). The results were reviewed by two experienced ophthalmologists (XW and ZJL) independently and the diagnoses were made according to the proposed international diabetic retinopathy severity scale [14].

OCTA image acquisition
The OCTA images were acquired using an RTVue XR Avanti spectral domain OCT device (Optovue, Inc., Fremont, CA, USA) with a speed of 70,000 A-scans/second and a wavelength of 840 nm. A 3 mm × 3 mm area of macula and a 4.5 mm × 4.5 mm area of ONH were scanned. A split-spectrum amplitudedecorrelation angiography (SSADA) algorithm was used to detect erythrocyte movement in OCTA. Retinal vascular networks and the nerve bers of macula and ONH were analyzed. OCTA images without signi cant motion or shadow artifacts and with scan quality ≥ 6 were included. OCTA images with poor quality or large motion artifacts or segmentation errors were excluded.
The parameters of the macula and ONH were automatically measured and analyzed using prototype AngioVue software 2.0 without any attempts at manual alteration. The parameters of choroidal capillary were not included for the requirement of manual measurement. Macula-associated parameters included whole VD of SCP and DCP, FAZ area, retinal thickness of the super cial slab started from internal limiting membrane (ILM) to retinal pigment epithelium (RPE) and the deep slab started from RPE to Bruch's membrane (BRM). VD was de ned as the percentage of the area that was occupied by blood vessels. The area of vessels was calculated using the software by extracting a binary image from the grayscale OCTA image, and then calculating the percentage of pixels occupied by blood vessels in the de ned region. The SCP slab was segmented from the ILM to the inner plexiform layer (IPL). The DCP slab was segmented from 15μm posterior to the IPL to the outer plexiform layer (OPL). FAZ perimeter (PERIM), acircularity index (AI), foveal vessel density in a 300-µm region around FAZ (FD), foveal VD, parafoveal VD and perifoveal VD in the SCP and DCP were not included because the high correlation between any two of the above parameters would in uence the results of the statistical model. ONH-associated parameters included VD of the whole 4.5 mm × 4.5 mm area, inside the ONH, peripapillary area as well as the average RNFL thickness of the entire peripapillary area. The peripapillary area was de ned as a 1.0 mm-wide round annulus extending from the 2.0-mm circle of ONH.

Statistical analysis
The statistical analyses were performed using SPSS 22.0 (SPSS Inc. Chicago, IL, USA). One-way ANOVA with further Bonferroni analysis were performed to analyze the OCTA variables of patients in the control, DM and NPDR groups. For variables that were statistically signi cant, a multivariable binary logistic regression model using backward elimination for the nonsigni cant variables was further conducted, including models with comparison among eyes of controls, DM and NPDR groups as well as a model comparing eyes of NoDR (including eyes of both control and DM groups) with those of NPDR. The receiver operating characteristic (ROC) curves were generated based on the binary logistic regression models. Area under the curve (AUC), sensitivity, and speci city were calculated in the ROC curve. A value of P < 0.05 was considered statistically signi cant.

OCTA ndings
Univariate analysis and multivariable logistic regression of the OCTA parameters showed that whole VD in the DCP and ONH were signi cantly different between control group and DM group. ROC curves for these variables using the combined parameter model were generated. The AUC was 0.798 (p < 0.001) with sensitivity of 82.8% and speci city of 67.2% ( Fig 1A). The statistical model also showed that parameters such as whole VD of SCP and inside disc VD were signi cantly different between the DM and NPDR groups.
The AUC was 0.860 (p < 0.001) with sensitivity of 85.9% and speci city of 70.3% ( Fig 1B). Several parameters associated with the macula and ONH were signi cantly different when comparing the control group to the NPDR group, including whole VD in SCP and DCP, whole VD in ONH and Peripapillary VD. The same approach of ROC was carried out. We found an AUC of 0.963 (p < 0.001) with sensitivity of 93.8% and speci city of 90.6% (Fig 1C). When comparing the NoDR group with NPDR group, the differences were signi cant in whole VD of SCP and DCP, whole VD of ONH, inside disc VD and peripapillary VD. The ROC curve showed an AUC of 0.923 (p < 0.001) with sensitivity and speci city of 90.6% and 82.8%, respectively ( Fig 1D). There was no signi cant difference in average RNFL thickness of both ONH-associated and macula-associated parameters among these groups ( Table 2, Table 3  Recently, the relationship between microvascular changes and neural changes was also analyzed. A study reported that ONH perfusion in diabetic patients without DR was signi cantly lower than that of controls [4,18], and concluded that reduced ONH perfusion preceded peripapillary RNFL thinning. However, they did not include NPDR objects; therefore, they could not answer the question regarding whether microvascular changes or neural impairments occur rst in the early onset of DR. Our study comprehensively analyzed the retinal vessels of macula and ONH together with RNFL changes and demonstrated that vascular changes, speci cally the VD, occurred prior to detectable RNFL changes in both the macula and ONH. Another interesting issue that has been little studied is the relationship between OCTA parameters of macula and ONH. To the best of our knowledge, this was the rst study to analyze OCTA parameters in both the macula and ONH in preclinical DR and NPDR. We found that both VD in macula and ONH were simultaneously reduced prior to retinal neural changes and clinically-visible retinopathy. This could be explained anatomically as follows: the super cial layers of ONH deriving from the central retinal artery (CRA) circulation [19,20] and the radial peripapillary capillaries (RPCs), the density of which we detected using OCTA, might derive from the arterioles that were parallel to retinal ganglion cell axons [21] and were clustered radially to supply super cial RNFL around the ONH [22,23]. Previous histological reports also showed that RPCs were derived from the super cial vascular plexuses that supplied the macular area and ascended to the nerve ber layer [24,25]. Therefore, several mechanisms of DM may cause dysregulation of cellular physiology and may damage vascular cells, possibly leading to capillary occlusion and decreased VD in both ONH and macula.
VD could be in uenced by age and gender [26,27]. In this study, the factor of age was eliminated and the factor of gender was not signi cantly different. The correlation between VD and glycosylated hemoglobin (HbA1c) is controversial [11,26]. In this study, HbA1c was not signi cantly different between eyes with and without DR. VD examination in OCTA by automated method showed high reproducibility and reliability [7]. It is reasonable that this set of VD parameters may be able not only to detect diabetic eyes at a higher risk of NPDR but also to screen for NPDR even before systemic diagnosis is made [28]. Some studies found that VD in DCP on OCTA, not in SCP or choriocapillaris, was lower in diabetic patients without DR compared to nondiabetic individuals [29][30][31]. Other studies showed that not only VD in the DCP but also VD in the SCP were different between diabetic patients and healthy controls[10, 32, 33]. In this study, which took the parameters of both macula and ONH into consideration, both SCP and DCP were signi cantly different among DM patients with or without DR and healthy controls. Consistent with the anatomical fact that DR could affect retinal vessels in both SCP and DCP, such result indicated that this statistical model could reduce the chance of drawing one-side conclusions and could analyze the VD changes from a more comprehensive view. Consistent with results of other studies[7, 12], which showed a negative correlation between the logMAR BCVA and the VD in the SCP and DCP, our result also showed higher logMAR BCVA in NPDR group in which the VD of SCP and DCP were lower.
In this study, we identi ed a small set of selected OCTA parameters that accurately distinguished eyes of normal control and DM patients from those with NPDR. Compared to individual parameters with ROC ranging from 0.472 to 0.893 [26], the combined model had the advantage of higher sensitivity/speci city as well as accuracy, making it a useful tool for analyzing the neurovascular changes of DR [34]. The current study demonstrated that combining a small set of selected macula-associated OCTA parameters, the accuracy of which is similar to our results, improved the overall diagnostic e cacy for discriminating eyes with NPDR from those with PDR[34]. To the best of our knowledge, this is the rst study to analyze in an integrated fashion the widely-studied OCTA parameters of both the macula and the ONH in patients with preclinical DR. Furthermore, our results demonstrated a minimal set of VD parameters in the macula and ONH with signi cant difference and favorable sensitivity and speci city to detect the very early changes in preclinical DR. Such set of OCTA parameters was with high sensitivity and speci city to differentiate diabetic patients especially those with NPDR from healthy controls and DM patients without any DR. Interestingly, many variables that were signi cant in the univariate analysis of recent study were not signi cant in the nal model [11,35]. This suggested that individual analyses could neglect the interaction among the parameters. It also suggested that the combined models could highlight the dominating parameters affecting the pathogenetic process of DR in early stages and set up a reliable evaluation system of early vasculopathy in DM.
Based on our ndings, we suggested a new perspective for the pathophysiological investigation of DR.
Because microvascular alterations might occur earlier than neurodegenerative changes in both macula and ONH in the early onset of DR, therapeutic strategies should target the signaling pathways that could remedy microvascular dysfunction for the purpose of preventing the development of DR, and the target area should include both the macula and the ONH.
One limitation of this study was that it was an observational cross-sectional study, possibly missing a de nite conclusion. Another limitation was that automatic segmentation was applied to analyze the parameters; therefore, the possibility of segmentation errors could not be ruled out. Furthermore, we did not include all the possible parameters of macula and ONH as well as parameters of choroidal capillary, possibly leading to de ciencies of parameters with signi cant differences. Further longitudinal study with expanded parameters are needed to resolve these issues.

Conclusions
Our results indicated that reduction of VD simultaneously occurred in both the macula and ONH prior to RNFL thinning in the early stage of DM. We provided a statistical model combining the parameters of both macula and ONH to identify early vasculopathy of DM and to differentiate NPDR patients from healthy controls and DM patients without any DR.

Availability of data and materials
The datasets used and/or analysed during the current study were available from the corresponding author on reasonable request.

Ethics approval and consent to participate
This study had been approved by the research ethics committee of Sun Yat-sen Memorial Hospital, Sun Yatsen University and was in compliance with the Declaration of Helsinki. Written informed consent was obtained from all participants.

Consent for publication
Written informed consent for publication had been obtained from participants whose fundus images were present in Fig. 2. 3.Chen X, Nie C, Gong Y, Zhang Y, Jin X, Wei S, Zhang M: Peripapillary retinal nerve ber layer changes in preclinical diabetic retinopathy: a meta-analysis. PloS one 2015, 10(5):e0125919.  Tables   Table 1 Demographic Figure 1 Receiver operating characteristic (ROC) curves for the binary regression models of combined parameters.

Figures
Green lines represented reference lines. year-old male patient with NPDR. Gradually reduction of VD was observed in both macula and ONH when proceeding from normal fundus to NPDR. RNFL thickness was also decreased in these three eyes.
However, it was not statistically signi cant among the healthy controls, NoDR and NPDR groups.