Comparison of retinal layer thickness and microvasculature changes in patients with diabetic retinopathy treated with intravitreous bevacizumab vs panretinal photocoagulation

To compare changes in retinal layers and microvasculature in diabetic retinopathy (DR) patients after bevacizumab therapy and panretinal photocoagulation (PRP). This prospective study divided patients into two groups: patients treated with bevacizumab and those treated with PRP. Patients visited our retinal clinic at 1, 3, and 6 months after treatment. Retinal layer thickness and vessel density (VD) using optical coherence tomography angiography were analyzed. 37 eyes in the bevacizumab group and 36 eyes in the PRP group were enrolled. In the bevacizumab group, the parafoveal RNFL, GCL, and IPL thicknesses significantly decreased (P < 0.001, P = 0.013, and P = 0.017, respectively), whereas the thicknesses in the PRP group showed an increasing tendency over time (P = 0.087, P = 0.005, and P = 0.003, respectively). The VD of the superficial capillary plexus (SCP) and deep capillary plexus (DCP) in the bevacizumab group did not show significant changes, whereas the VD in the PRP group significantly increased over time (both P < 0.001). Additionally, RNFL (P = 0.001) and GCL thicknesses (P = 0.035) were significant factors affecting changes in BCVA, whereas the VDs of SCP and DCP did not. Patients who received bevacizumab therapy did not show a significant change in macular VD, whereas the VD of patients after PRP significantly increased after treatment. The increased macular VD in patients after PRP would be associated with the increased inner retinal layer thickness after treatment, which was significantly related to the impairment in visual acuity.

OCT and OCTA measurements. We performed spectral domain OCT (SD-OCT; Spectralis; Heidelberg Engineering, Heidelberg, Germany) using retinal thickness map analyses to display numeric averages of the measurements for each of the nine Early Treatment Diabetic Retinopathy Study (ETDRS) subfields to measure the thickness of the retinal layer. The subfoveal (inner ring of ETDRS subfields) and parafoveal (intermediate ring of ETDRS subfields) areas were analyzed. Automated retinal layer segmentation was conducted by the built-in software, Heidelberg Eye Explorer ver. 6.9a (Heidelberg Engineering, Heidelberg, Germany), and central macular thickness, and the thickness of the inner retinal layer including retinal nerve fiber layer (RNFL), ganglion cell layer (GCL), and inner plexiform layer (IPL) were measured. We excluded patients showing definite cystic changes in the parafoveal area, which may cause segmentation errors and inaccurate measurement of VD.
OCTA was performed using a Spectralis OCT2 device (Heidelberg Engineering). En face OCTA images were recorded with a 20 × 15 degrees angle and a lateral resolution of 5.7 μm/pixel, resulting in a retinal section of 6.0 mm × 4.5 mm. The images of the superficial capillary plexus (SCP), defined as the layer originating from the internal limiting membrane to the IPL, and the deep capillary plexus (DCP), defined as the layer starting from the outer border of the IPL to the outer plexiform layer, were visualized automatically by segmenting two separate slabs defined by arbitrary segmentation lines created by the device software. VD was calculated using ImageJ software ver. 1.52a (National Institutes of Health, Bethesda, MD, USA). The threshold adjustment tool was applied with the default settings, and the dark background option was selected. This tool automatically sets the lower and upper threshold values and segmented grayscale images into features of interest and the background. Using this binarized image, the VD was calculated by dividing the area of white pixels by the total number of pixels ( Fig. 1). Images with loss of fixation, segmentation errors, motion artifacts, and OCTA quality < 25 were excluded. Statistical analysis. Demographic characteristics and ocular parameters were compared via an independent t-test and chi-square test. Analysis of covariance was used to compare the OCT and OCTA parameters between groups after adjusting for age. Repeated-measures analysis of variance was used to analyze longitudinal changes in the thickness of the retinal layer and VD of SCP and DCP in each group. Linear mixed models were fitted to identify factors associated with changes in BCVA. The Pearson correlation test was performed to iden-

Results
Demographics. The study initially enrolled 86 eyes in which treatment was performed for DR. Of these, 13 eyes were excluded from the study: 5 eyes due to the occurrence of DME during the treatment or follow-up period in the PRP group, 4 eyes due to loss of follow-up (3 eyes in the bevacizumab group and 1 eye in the PRP group), 2 eyes for the occurrence of vitreous hemorrhage during the follow-up period (1 eye in the bevacizumab group and 1 eye in the PRP group), and 2 eyes due to low OCTA quality (1eye in the bevacizumab group and 1 eye in the PRP group). As a result, a total of 73 eyes were finally enrolled: 37 eyes for the bevacizumab group, and 36 eyes for the PRP group. The mean ages of the bevacizumab and PRP groups were 51.8 ± 8.9 and 57.1 ± 11.3 years, respectively (P = 0.029) ( Table 1). The baseline BCVA was 0.06 ± 0.09 and 0.09 ± 0.11 in each group, which was not significantly different (P = 0.111). Sex, laterality, spherical equivalent, and baseline IOP were not also significantly different between groups. The mean duration of T2DM was 8.9 ± 6.1 and 11.2 ± 6.6 years (P = 0.125), and the HbA1c level was 8.9 ± 2.3 and 8.1 ± 1.9% (P = 0.098) for the bevacizumab and PRP groups, respectively. The change in BCVA did not show a significant result in the bevacizumab group (P = 0.252) (Fig. 2), whereas, the change in BCVA in the PRP group was statistically significant (P < 0.001).

Thickness of central macula and parafoveal inner retinal layer in each group. The baseline CMTs
were 269.7 ± 22.6 and 268.8 ± 31.2 μm in the bevacizumab and PRP groups, respectively, which did not differ significantly (P = 0.881) ( Table 2). The baseline thicknesses of individual inner retinal layers in the parafoveal area, including the RNFL, GCL, and IPL, did not differ significantly between groups (P = 0.335, P = 0.330, and P = 0.665, respectively). In the bevacizumab group, the CMT showed a decreasing trend over time, but it was not statistically significant (P = 0.073). The parafoveal thicknesses of the RNFL, GCL, and IPL decreased over time, and they were statistically significant (P < 0.001, P = 0.013, and P = 0.017, respectively). In the PRP group, the CMT increased significantly over time (P = 0.035). Additionally, the parafoveal thicknesses of the RNFL, GCL, and IPL showed a similar trend to changes in the CMT (P = 0.087, P = 0.005, and P = 0.003, respectively) (Fig. 3).  www.nature.com/scientificreports/  www.nature.com/scientificreports/ VD of SCP and DCP in each group. The baseline VDs of the SCP for the bevacizumab group and the PRP group were 25.5 ± 5.8 and 22.9 ± 5.7%, respectively (P = 0.056) ( Table 3). The baseline VDs of the DCP 19.9 ± 4.8 and 18.6 ± 4.5, respectively, which was not also significantly different (P = 0.219). In the bevacizumab group, the VDs of the SCP and DCP did not show significant changes over time (P = 0.350 and P = 0.130, respectively). However, the VDs of the SCP and DCP showed a continuous increase over time in the PRP group and the change was statistically significant (both P < 0.001) (Fig. 4).

Discussion
We investigated the changes in retinal microvasculature of patients with DR after anti-VEGF therapy or PRP and compared the two groups. In the bevacizumab group, the inner retina showed a significant reduction after anti-VEGF therapy, whereas the VDs of the SCP and DCP did not significantly change over time. In the PRP group, the CMT and the inner retinal layer thickness increased significantly after PRP treatment, and the VDs of the SCP and DCP also showed a consistent increase over time. Additionally, the inner retinal layer thickness was significantly associated with BCVA change and final BCVA, whereas the VDs of the SCP and DCP were not. Sorour et al. 10 reported that the macular VDs of the SCP, DCP, and total retinal capillary plexus remained statistically unchanged following up to three intravitreal injections of anti-VEGF therapy in PDR patients. Another study also found that the macular VD and flow area of the SCP, DCP, and choriocapillaris did not change after monthly or quarterly intravitreal injections of aflibercept during 12 months 8 . Our study showed that the VDs of the SCP and DCP did not change significantly after three intravitreal injections of bevacizumab until the 6-month visit, which is consistent with previous studies. The release of VEGF from the ischemic retina could cause progressive vascular nonperfusion resulting in a consistent decrease in retinal VD. The anti-VEGF therapy may block this mechanism, which leads to a cessation of decreasing VD. Additionally, Nesper et al. 11 reported that retinal capillary nonperfusion in OCTA was correlated significantly and linearly with disease severity in DR patients. Therefore, anti-VEGF therapy would have a preventive effect for DR progression by cessation of an increasing nonperfusion area.
By contrast, the VDs of the SCP and DCP significantly increased over time with a significant increase in the CMT and inner retinal layer thickness in the PRP group. Recently, Kim et al. 12 also reported a significant increase in perfusion density and vessel length density using OCTA after PRP in DR patients. They explained that the improved flow in the remaining macular capillaries could potentially a re-establish macular microvasculature Table 5. Correlation between optical coherence tomography and optical coherence tomography angiography parameters and final best-corrected visual acuity. Values in boldface (P < 0.05) are statistically significant. OCT optical coherence tomography, CMT central macular thickness, RNFL retinal nerve fiber layer, GCL ganglion cell layer, IPL inner plexiform layer, OCTA optical coherence tomography angiography, VD vessel density, SCP superficial capillary plexus, DCP deep capillary plexus. www.nature.com/scientificreports/ from regression of peripheral neovascularization or intraretinal microvascular abnormalities. Additionally, the increased thickness of the inner retinal layer would affect the increased macular VD, which was well-known to have significant relationships between each other [13][14][15][16][17] . Although the PRP group showed a significant increase in macular VD, the BCVA became worse over time significantly. In the bevacizumab group, the final BCVA was better than the baseline BCVA, although it was not statistically significant. Notably, the BCVA changes and final BCVA were significantly associated with the inner retinal layer thickness, and not with the VD of the SCP or DCP. The PRP group showing a significant increase in inner retinal layer thickness exhibited a decrease in BCVA over time, whereas the bevacizumab group with a significant reduction of the inner retinal layer thickness showed an increasing tendency with respect to BCVA. In treatment with PRP, the laser energy is absorbed by the retinal pigment epithelium and generates thermal energy to the outer retina. These thermal damages cause the upregulation of cytokines, such as VEGF and interleukin-6. VEGF can cause the breakdown of the blood-retinal barrier, induce vessel dilation, and increase the ocular blood flow, which leads to an increase in vascular permeability and fluid leakage, eventually resulting in increased thickness of the inner retina [18][19][20] . Although the impairment of visual acuity due to subclinical DME in the early stage would be minimal, accumulation of inflammatory factors in the retina such as VEGF, interleukin-6, and tumor necrosis factor-α may eventually result in decreased visual acuity over time 21,22 . On the other hand, anti-VEGF therapy could cause the subclinical macular edema to subside, which would result in the reduction of the inner retina and maintenance of better visual acuity.
Previous studies have reported a significant correlation between macular VD and visual acuity in T2DM patients without clinical DR 14,17,23 . Samara et al. 24 also found a positive correlation between macular VD and the visual acuity of DR patients. However, we did not find any significant association between the VD of the SVP or DVP and BCVA; additionally, the PRP group that exhibited a significant increase in VD showed impaired visual acuity changes over time. This discrepancy may be due to a difference in the DR severity of enrolled patients. Our study, unlike previous studies, only included patients with severe DR requiring treatment. The difference in the scan area of OCTA images, which was larger in our study (6.0 mm × 4.5 mm scan area), may have also played a role. Damage to endothelial cells for various reasons may increase vascular permeability and vessel dilation. Erisgin 25 reported that 75 mg/kg melamine exposure results in an increase in the dilatation of brain blood vessels and endothelial degeneration via damage to the blood-brain barrier. Similarly, various inflammatory factors could impair the endothelial cells in the retina and cause damage to the blood-retina barrier, which may result in vascular permeability and vessel dilatation followed by an increase in retinal VD. Therefore, the increase in VD of patients with PRP would be affected not only by the increase in retinal perfusion but also by the dilatation of retinal vasculature by inflammatory factors. Further histophathological studies are needed to confirm this hypothesis.
After the publication of Protocol S, anti-VEGF therapy increased, whereas the PRP rate decreased for the treatment of PDR patients 26 . The CLARITY study found that aflibercept was superior to PRP in BCVA changes at 52 weeks after treatment 27 . Although the PRP group showed a significant increase in macular VD unlike the bevacizumab group, increased inner retinal layer thickness after PRP, which is related to an impairment in visual acuity, is also associated with an increase in macular VD [13][14][15]17 . Therefore, the idea that an increase in VD after PRP treatment implies only positive effects should be reconsidered. As such, although anti-VEGF therapy has various advantages over PRP for the treatment of PDR, the cost-effectiveness cannot be ignored in practice. The effects of anti-VEGF injection is not permanent; thus, continuous treatment is required, which could impose a greater economic burden on the patient 28 . Additionally, PDR could progress and cause permanent visual www.nature.com/scientificreports/ impairment if the patient becomes a follow-up loss. Therefore, physicians should understand the characteristics of the patient and choose the appropriate treatment. This study had several limitations. First, the treatment choice of patients was not randomized, which could cause some bias. Second, the number of cases was relatively small due to the strict inclusion criteria. Third, although we performed statistical analyses with adjustment for age, there may be some bias in analyses of VD changes due to the age difference between the two groups. The strength of this study was that we enrolled OCTA images with OCTA quality ≥ 25, allowing accurate analyses. Additionally, this is the first study to compare the changes in the macular VD using OCTA of patients after PRP and patients after anti-VEGF therapy.
In conclusion, patients after bevacizumab therapy did not show a significant change in macular VD, whereas the VD of patients after PRP significantly increased after treatment. However, increased macular VD in patients after PRP would be associated with the increased inner retinal layer thickness after treatment, which was significantly related to the impairment in visual acuity. Therefore, the increased VD after PRP may not include only positive effects. As both treatments have their pros and cons, physicians should choose the appropriate treatment for each patient.