Longitudinal changes in superficial microvasculature in glaucomatous retinal nerve fiber layer defects after disc hemorrhage

Glaucoma is a multifactorial optic neuropathy, possibly involving vascular dysfunction, leading to the death of retinal ganglion cells and their axons. Disc hemorrhage (DH) is known to be closely associated with the widening of retinal nerve fiber layer defect (NFLD); however, it has not been well elucidated how DH affects retinal microvasculature. We aimed to investigate the association between DH history and longitudinal changes in superficial retinal microvasculature in NFLD. We enrolled 15 glaucoma patients with DH history (32 glaucomatous NFLD locations, with or without DH history). NFLD-angle, superficial retinal vessel density (VD), and decreased superficial retinal microvasculature (deMv)-angle were assessed using optical coherence tomography angiography for at least three times over time. The mean follow-up period and OCT/OCTA scan interval were 21.3 ± 5.4 months (range, 12–28) and 6.8 ± 0.4 months (range, 2–18), respectively. Linear mixed-effects models showed that the presence of DH history was significantly associated with an additional NFLD-angle widening of 2.19 degree/year (P = 0.030), VD decrease of 1.88%/year (P = 0.015), and deMv-angle widening of 3.78 degree/year (P < 0.001). These changes were significantly correlated with each other (P < 0.001). Thus, the widening of NFLD was closely associated with deMv, and DH was associated with a subsequent decrease in superficial retinal microvasculature in NFLD.


Discussion
The current study showed that a history of DH was a significant factor for site-specific progression of blood flow impairment in NFLD. The decrease in VD and widening of deMv-angle were significantly correlated with the widening of NFLD, suggesting that assessments of microvasculature using OCTA could be helpful to detect glaucoma progression.
Previous studies have shown the association between glaucoma progressions according to function 9,11,12,17 and structure 13,14,18 . The location of DH also is known to relate spatially to the progressive localized thinning of the RNFL 13,14 , widening of the NFLD 19,20 , and focal visual field progression 9,11 . Recent advancements in OCTA Table 2. Comparison of clinical characteristics between NFLD with DH and NFLD without DH (N = 32 NFLD locations). Data are presented as mean ± standard deviation (SD) with the minimum and maximum values in parentheses. CMvD peripapillary choroidal microvasculature dropout, deMv decreased superficial retinal microvasculature, DH disc hemorrhage, IOP intraocular pressure, NFLD nerve fiber layer defect, OCTA optical coherence tomography angiography, VD vessel density. *Analyzed using linear mixed-effect modeling between NFLD without DH and NFLD with DH.

P value* Mean ± SD (Range) Mean ± SD (Range)
Age-baseline (years) 55.   22 showed that a decrease in peripapillary VD was significantly associated with DH occurrence in patients with normal tension glaucoma, which is consistent with our results. However, the association between microvasculature reduction and RNFL thinning has not been fully clarified. Longitudinal assessments may be necessary to elucidate this issue. We used the deMv-angle to investigate the longitudinal decrease in superficial retinal microvasculature. This parameter was first described by Lee et al. 8 using the term "vascular impairment. " They showed that "vascular impairment" (deMv) was almost identical to NFLD in primary open angle glaucoma (POAG) eyes having a localized RNFL defect. In contrast, we found that the deMv was 14.5 degrees larger on average than the NFLD measurements in the current study (deMv-angle, 35.74 ± 13.73; NFLD-angle, 21.23 ± 13.69; P < 0.001). The reason for this discrepancy is not clear. One possible reason is the difference in the method used for the measurement of NFLD. NFLD was evaluated using red-free fundus photographs in the previous report, whereas OCT en face images were used in the current study. Another reason might be the difference in the threshold level between these methods. In any case, the change rate of the widening of deMv-angle was significantly associated with that of the widening of NFLD, which indicated that superficial microvasculature reduction was highly correlated with structural damage in NFLD.
In the current study, the change rates of NFLD-angle, VD, and deMv-angle had significant correlations with each other, which suggested that both OCTA and OCT parameters could be useful to detect glaucoma progression. Both OCTA and OCT could be valid and feasible assessments for glaucoma progression, although it was not clarified in the current study. Shoji et al. 21 reported that a significant decrease was more detectable in macula VD than in ganglion cell complex thickness in some glaucomatous eyes. Moghimi et al. 23 showed the possibility that in OCTA, VD was less affected by floor effects, and no further structural change could be detected in OCTbased thickness measurements. This evidence indicates the possibility of the different utility of these methods. Further studies should be conducted to clarify this issue.
This study has several limitations. First, the number of eyes analyzable was small. Nonetheless, change rates of all examined parameters (NFLD-angle, VD, and deMv-angle) were significantly different between the two groups, and their close relationships were detected with this small sample size. Second, because the line between the normal retina and the area of deMv to detect deMv-angle were manually evaluated, the values for deMv-angle could be different between graders. However, because the ICCs for these parameters were excellent, we believe that the results can be acceptable.
In conclusion, significant changes in NFLD-angle, VD, and deMv-angle were detected in NFLD with DH history more than in NFLD without DH history, and change rates of these parameters were significantly correlated with each other. This suggests that OCTA and OCT measurements can be used to detect glaucoma progression. Further studies are needed to determine the relationship between vascular and structural measurements and the usefulness of OCTA measurements in clinical practice.

Methods
This longitudinal observational study was conducted at the Glaucoma Clinic of the Kyoto University Hospital. The study adhered to the tenets of the Declaration of Helsinki and was approved by the Institutional Review Board and Ethics Committee of the Kyoto University Graduate School of Medicine. Written informed consent was obtained from all the patients. NFLD without DH in subjects met the above inclusion criteria were included as a control group. Exclusion criteria of this study were: (1) eyes with coexisting uveitis, retinal disease, or non-glaucomatous optic neuropathy. (2) eyes with any history of intraocular surgery including cataract surgery and glaucoma surgery. The patients had undergone a comprehensive ophthalmic examination including measurement of best-corrected visual acuity (using a 5-m Landolt chart), slit-lamp examination, measurement of axial length (IOLMaster 500, Carl Zeiss Meditec, Dublin, CA), central corneal thickness (SP-3000, Tomay, Tokyo, Japan), Goldmann applanation tonometry, gonioscopy, indirect ophthalmoscopy, dilated slit-lamp examination of the optic nerve head, fundus photography, stereo disc photography (using a 3-Dx simultaneous stereo disc camera, Nidek, Optic coherence tomography angiography. The optic nerve and peripapillary area were imaged using a commercially available OCTA device (AngioVue; OptoVue, Fremont, CA, USA). Each image covered an area of 4.5 × 4.5 mm and 3.0 × 3.0 mm centered on the optic disc. Each B-Scan contained 216 A-scan. To produce images of perfused vessels, the Split Spectrum Amplitude Decorrelation Angiography software algorithm was employed 7,24 . The OCTA images were coregistered with OCT B-scans that were obtained concurrently to enable visualization of both the vasculature and structure in tandem. The area of deMv was assessed by determining the presence of a region of decreased vasculature in the inner retina using the en face angiogram. Using the internal limiting membrane as a plane of reference, a slab with a uniform thickness that included the RNFL, ganglion cell layer, and inner plexiform layer was manually determined from the entire OCTA data sets using the coregistered OCT B-scans in each eye 8 .
The peripapillary region was defined as a 500-μm-wide elliptical annulus extending from the optic disc boundary, and segmentation of the peripapillary area was performed using the intrinsic software provided by Opto-Vue. Vessel density was defined by the percentage area occupied by vessels, measured using the intensity-based thresholding feature of the software, which adopted the same method of calculation as that previously reported 24 . Measurement methods of NFLD-angle, VD, and deMV-angle. In reference to previous reports 8,19 , NFLD-angle was measured by identifying the two points at which the borders of an NFLD area met the clinical optic disc margin (Fig. 1a), using 3.0 × 3.0 enface image of OCT. Lines were then drawn that connected the disc center and the two points, and the angular distance between these two lines was defined as NFLD-angle (Fig. 1b).
In the same way, deMv-angle was measured by identifying the two points at which the borders of the deMv area met the clinical optic disc margin and the disc center, using 3.0 × 3.0 enface image of OCTA (Fig. 1c).
As for VD, the optic disc was divided into six areas (blue dotted lines) using the software. In this study, we measured the VD of the superotemporal or inferotemporal areas (white dotted lines) in the location of NFLD (Fig. 1d).

Statistical analysis.
Values were presented as mean ± SD for continuous variables. Variables were compared using linear-mixed effects modeling, where eyes were nested within subjects to properly adjust for eyes from the same individual exhibiting similar measurements. The significance of differences between the groups was determined after Bonferroni correction.
Linear mixed-effects modeling was used to evaluate the rates of changes in NFLD-angle (degree/year), VD (%/ year), and deMv-angle (degree/year). Details of the use of these models for assessment of longitudinal changes in glaucoma have been previously described 13,14,[25][26][27] . In brief, models were first fit with objective parameter measurements (NFLD-angle, VD, or deMv-angle) as a response variable, whereas time, group (the presence of DH history), and a time-group interaction were defined as fixed effects. The variable, time, was measured as time in years from the first OCT and OCTA examination. Random intercepts and slopes were used to account for repeated measurements over time, where eyes (right or left) and locations (superotemporal or inferotemporal) were nested within subjects to properly adjust for NFLDs from the same individual exhibiting similar measurements. Then, multivariable models were evaluated with mean IOP throughout the follow-up period and SSI (in case of VD), which potentially affect rates of change in objective parameters, to evaluate relationships between objective parameter measurements over time and the presence of DH history. Two-way interactions between time and mean IOP and SSI (for VD) were used to determine whether these were significantly associated with change www.nature.com/scientificreports/ in objective parameter measurements over time. IOP-adjusted rates of changes in NFLD-angle and deMv-angle, and IOP-SSI-adjusted rates of VD change, were used for analyses. The statistical analyses were performed using the R package 'lme4′ with R version 3.3.1 (http://www.r-proje ct.org) and SPSS Version 24 software (IBM Corp., Armonk, New York, USA. P values less than 0.05 were considered statistically significant. Interobserver reproducibility of NFLD-angle and deMv-angle. To evaluate the interobserver reproducibility of NFLD-angle and deMv-angle, all NFLDs included in the current study were evaluated independently by two examiners (YO and EN) blinded to any information other than OCT enface images or OCTA, and intraclass correlation coefficients [ICCs (2,1)] were calculated.