Nasalization of Central Retinal Vessel Trunk Predicts Rapid Progression of Central Visual Field in Open-Angle Glaucoma

Central visual field (CVF) loss is important in maintaining vision-related quality of life in eyes with open-angle glaucoma (OAG). The present study investigated whether nasalized location of central retinal vessel trunk (CRVT) at baseline is associated with rapid rate of CVF loss in early-stage OAG eyes. This study included 76 OAG eyes with high nasalization CRVT [HNL] group and 75 OAG eyes with low nasalization CRVT [LNL] group matched for glaucoma severity at baseline that showed progressive visual field (VF) loss. The rates of mean threshold changes at various regions were compared in the two groups using a linear mixed model. Clinical variables associated with rapid rate of CVF progression were also identified using a linear mixed model. The rate of CVF loss in the central 10° was significantly higher in the HNL group than that in the LNL group (−0.452 dB/year vs. −0.291 dB/year, P < 0.001). The average and inferior hemi-macular ganglion cell inner plexiform layer (GCIPL) progression rates were significantly faster in the HNL group than in the LNL group (P < 0.05). Nasalized location of CRVT was an independent predictor of a more rapid VF loss in the central 10° region (P < 0.05).

In the GHT map, the rates of VF loss were higher in the HNL group compared with those in the LNL group in the superior central region (GHT-S1; −0.601 dB/year vs. −0.345 dB/year; P < 0.001) and superior paracentral region (GHT-S2; −0.707 dB/year vs. −0.451 dB/year; P < 0.001). In contrast, LNL eyes had significantly more rapid rates of VF loss in the superior peripheral regions (GHT-S4; −0.524 dB/year vs. −0.226 dB/year; P < 0.001, LNL (n = 75) HNL (n = 76) P value www.nature.com/scientificreports www.nature.com/scientificreports/ and GHT-S5; −0.522 dB/year vs. −0.076 dB/year; P < 0.001) and inferior peripheral regions (GHT-I4; −0.335 dB/ year vs. −0.113 dB/year; P < 0.001, and GHT-I5; −0.206 dB/year vs. −0.001 dB/year; P < 0.001). There were no significant differences in the rates of VF loss between the LNL and HNL groups in the nasal regions, both superior (GHT-S3; −0.453 dB/year vs. −0.527 dB/year; P = 0.148) and inferior (GHT-I3; −0.402 dB/year vs. −0.397 dB/ year; P = 0.933) (Fig. 1). Representative cases of the two groups are shown in Fig. 2. An LNL eye shows progressive VF loss in the PVF area, whereas the initial CVF scotoma rapidly enlarges and extends into the PVF area in the HNL eye, despite having similar severity of VF loss at baseline. Table 3 shows the results of structural progression rates based on the guided progression analysis (GPA; Carl Zeiss Meditec, Dublin, CA) software provided by the Cirrus spectral-domain optical coherence tomography (SD-OCT, Carl Zeiss Meditec). There were no significant differences in the average and superior and inferior quadrant retinal nerve fiber layer (RNFL) progression rates between the two groups. However, the average and inferior hemi-macular ganglion cell inner plexiform layer (GCIPL) progression rates were significantly faster in the HNL group than in the LNL group.
Our linear mixed model that controlled for all covariates showed that higher NI as well as being in HNL group were significant independent predictors of more rapid VF loss in the central 10° region (P = 0.048, P < 0.001, respectively) and central GHT map region (GHT-S1 and GHT-I1, P = 0.048, P < 0.001, respectively). Other significant variables associated with faster VF loss in the central 10° and central GHT map regions included older age at baseline (P = 0.008 and 0.009, respectively), myopic refraction (P = 0.052 and 0.018, respectively), larger adjusted β-PPA area (P = 0.002 and 0.027, respectively), lower baseline MD (P = 0.029 and 0.019, respectively), and higher baseline PSD (P < 0.001, respectively) ( Table 4).

Discussion
Previous cross-sectional studies have reported that nasalized CRVT is consistently associated with CVF defects in glaucomatous eyes regardless of disease severity 13,14 . The proposed explanations for this finding include both mechanical and vascular theories. CRVT may act as a stabilization support preventing glaucomatous deformation in the LC. Therefore, a nasalized CRVT may result in less mechanical support for the LC in the temporal region, which corresponds to CVF area. A nasalized CRVT may also compromise the adequacy of vascular supply in the temporal region, leading to thinning of the RNFL in the macular region and CVF loss 14 . Furthermore, the correlation between CRVT location and CVF loss was significantly stronger in moderate to severe glaucoma than that in mild glaucoma, suggesting that CRVT nasalization is not the result of glaucoma progression, but rather a stable anatomic parameter that may be a risk factor for development of CVF loss in patients with glaucoma 14 . Our findings are consistent with the speculation that there is no significant difference in the amount of CRVT NI between baseline and last follow-up measurements in both LNL and HNL groups (Table 1).
In the current study, patients in the HNL group were younger than those in the LNL group at baseline. This finding is consistent with that of a recent study that reported that normal-tension glaucoma (NTG) patients with more nasalized CRVT were younger than those with less nasalized CRVT 15 . Moreover, eyes in the HNL group were more myopic and had higher prevalence of β-PPA than those in the LNL group. The current literature regarding the association between the location of CRVT and myopia is relatively scarce, but few studies have reported findings consistent with our results. In a prospective study of myopic children, CRVT location changed with myopic elongation and the major direction of dragging or displacement was nasal 16 . Our finding was further confirmed by a cross-sectional study with myopic NTG eyes, which demonstrated that eyes with nasalized CRVT were more myopic than those with less nasalized CRVT 15 .
In our study, both LNL and HNL groups had a similar degree of CVF MT superiorly and inferiorly at baseline, whether mapped on the 10-24° map or GHT map. Furthermore, there was no significant difference in the frequency of eyes with CVF defects between the two groups at baseline. However, the proportion of CVF defects was significantly higher in HNL eyes than that in LNL eyes at last follow-up (P = 0.01, Table 2). This indicates that CVF progression rates were significantly higher in HNL eyes than those in LNL eyes during the course of disease. These findings may be in agreement with previous reports, which showed that the magnitude of association between CRVT nasalization and CVF depression at presentation was significant but small in mild glaucoma (MD ≥ −6 dB). However, the association was more than three-fold in moderate glaucoma (−12 dB ≤ MD < −6 dB) and almost six-fold in severe glaucoma (MD < −12 dB) compared with that in mild glaucoma 14 .   In our study, CVF progression rates were significantly more rapid in the HNL eyes than those in the LNL eyes based on both 10-24° and GHT maps (P < 0.001, Table 1). In the GHT hemifield analysis, although the rates of CVF progression in the HNL group were significantly more rapid in the superior central (GHT-S1) and paracentral VF (GHT-S2) regions (P < 0.001, respectively) than those in the LNL group, they were not significantly different in the inferior central (GHT-I1) and paracentral (GHT-I2) regions (P = 0.288 and P = 0.395, respectively). A similar trend was also seen in the 10-24° map (Fig. 2). One of the explanations for our findings is that VF defects    www.nature.com/scientificreports www.nature.com/scientificreports/ progress more rapidly in the superior than those in the inferior hemifield of OAG eyes in the CVF area 17,18 . Cho et al. reported that the VF progression rate of the superior central 10° (−0.911 dB/year) was significantly more rapid than that of the inferior central 10° (−0.16 dB/year) in NTG eyes 17 . Despite differences in the study subjects and designs, our results are in agreement with these earlier findings. Another explanation is that the inferior central VF is typically affected at a later advanced stage of glaucoma, and eyes in the early stage of glaucoma were included at the time of enrollment in the current study.
In contrast, PVF progression rates were significantly more rapid in the LNL eyes than those in the HNL eyes based on both of the 10-24° map and the GHT map (−0.636 dB/year vs. −0.234 dB/year, 10-24° map; P < 0.001, Table 1). In the hemifield analysis, the rates of PVF progression in the LNL group were significantly more rapid in both the superior and inferior PVF regions (P < 0.001, respectively) based on both of the 10-24° map and the GHT map compared with those in the HNL group. One explanation for our findings is that more temporal location of the CRVT found in the LNL eyes may act as a stabilization support for papillomacular nerve fibers, preventing or minimizing CVF loss during the course of the disease. However, these eyes may be at a greater susceptibility of glaucomatous VF progression in the superior and inferior PVF areas due to lack of sufficient connective tissue support within the superior and inferior regions of the LC 19,20 . In our study, the rates of VF progression were similar in the two groups (GHT-S3 and GHT-I3; P = 0.14 and P = 0.93, respectively) in the nasal regions (Fig. 1). The explanation for this finding is that the nasal VF area is known typically to be affected first in glaucoma, which was the case in our patients with early-stage glaucoma.
In addition to VF progression, trend-based analysis was performed to compare the structural progression rates of RNFL and GCIPL in the two group. The HNL group showed significantly faster average GCIPL progression rates (−0.75 µm/year vs. −0.38 µm/year, P = 0.008) compared to LNL group. The rate of inferior hemi-macular GCIPL thickness loss was significantly faster in the HNL group compared to LNL group (−0.92 µm/year vs. −0.35 µm/year, P = 0.014), resulting in faster superior CVF progression rate in the current study. Faster rate of GCIPL thickness loss might have caused greater speed of CVF loss as seen in our HNL group as CVF is closely associated with structural integrity of GCIPL.
Another important finding in the present study was that a more nasalized CRVT as determined by NI as well as belonging to HNL group at baseline were independently associated with a greater velocity of CVF sensitivity loss, which has an important clinical implication as rapid CVF loss compromises vision-related quality of life (QOL) as measured by the National Eye Institute Visual Function Questionnaire 21 . Clinically, our findings may suggest that eyes with highly nasalized CRVT should be considered a candidate for more aggressive treatment to prevent early loss of CVF. Likewise, a poorer baseline VF as determined by MD and PSD was also a significant predictor of fast CVF progression in our OAG patients. In other words, CVF progression is more common in advanced disease than in early disease, which is in agreement with other findings that advanced stage of glaucoma is an important risk factor for VF progression [22][23][24][25] . Since advanced glaucoma often affects CVF, the rate of CVF progression correlates well with advanced glaucoma severity.
In our study population, more myopic refraction was also associated with faster CVF loss whether mapped on the 10-24° map or GHT map (P = 0.052 and P = 0.018, respectively, Table 4). During the myopic process, when the optic disk tilts temporally and becomes depressed, it may give rise to papillomacular RNFL defects due to the Effect Central 10°Central (GHT-S1 and I1)  www.nature.com/scientificreports www.nature.com/scientificreports/ shearing forces across the temporal sides of the LC [26][27][28][29] , which may further increase the vulnerability to glaucomatous VF progression in the CVF area. Furthermore, the extent of myopia has been associated with faster progression in the CVF region in NTG eyes 30 . Therefore, our results suggest that myopia may also have a significant effect on the faster progression of CVF, which is independent of the effect of CRVT nasalization. Larger adjusted β-PPA area at baseline was also associated with a faster rate of CVF progression (P = 0.002 for the 10-24° map, and P = 0.027 for the GHT map). β-PPA area is closely related to the degree of myopia, where the optic disk tilts temporally during posterior globe elongation and may weaken the structural integrity of the parapapillary sclera and LC and increase the risk of glaucoma progression in the macula and CVF area 15,18 . Of interest, in myopic children with ongoing axial elongation, enlargement of β-PPA was associated with the extent and direction of vascular trunk dragging 29 , and location of β-PPA has been shown to be associated with CRVT location in adult glaucoma patients 11 .
We must acknowledge several limitations in the current study. First, our study was retrospective in design. Consequently, there was a large number of OAG patients excluded from the initial patient list during initial screening due to failure of meeting our inclusion criteria. This could have introduced selection bias. Since our patients represent two groups of Korean OAG patients with different degrees of CRVT nasalization referred to a tertiary clinic, study results from a tertiary clinic using single ethnic group may not be applicable to other races or the general population. We classified our OAG eyes into two groups (LNL vs. HNL) and measured NI based on the location of CRVT on NA. Although there is currently no universally accepted method to classify CRVT location or quantify CRVT nasalization, the method used in the current study has been validated previously [13][14][15] and may minimize the subjectivity associated with manual localization or measurement of the CRVT location in the ONH. Our subjects were highly selected groups (HNL vs. LNL) of patients based on the location of the CRVT in the ONH according to the method described by Lee et al. 15 . Eyes with highly nasalized location of CRVT may be closely related to myopic disc as shown in our Table 1. Therefore, we have constructed a linear mixed model to assess whether the location of CRVT is significantly associated with rapid central VF progression independent of the effects of myopia, including SE, torsion degree, disk tilt ratio, and β-PPA area. Our study consistently showed that higher NI as well as being in HNL group were significant independent predictors of faster VF loss in the central 10° and central GHT map regions as noted in the Table 4. For the evaluation of CVF progression, Humphrey 24-2 VF testing was used. However, 24-2 VF testing may not detect subtle CVF progression due to distribution of large space between neighboring test spots in the central 10° area 31 . Ideally, Humphrey VF10-2 may better detect CVF progression rates as well as defects than Humphrey 24-2 VF. In the current study, we have included the subjects with visible CRVT origin in our enrollment. This could have limited our inclusion criteria to the larger discs with central cupping in which CRVT origin is not obscured by overlying neural tissue. Therefore, eyes with small discs might have been excluded from our study since they do not usually have visible CRVT origin. Finally, since our study was exploratory in the study nature, we could have used more stringent threshold for statistical significance in our multiple comparisons.
In conclusion, there are significant regional differences in VF progression rates among early-stage OAG eyes with different CRVT location. CVF loss in the 12 central-most points on 24-2 VF tests was significantly more rapid in eyes with nasalized CRVT. Our study indicates that CRVT nasalization may be an independent structural biomarker to predict rapid CVF deterioration in OAG.

Study participants. Our study was approved by the Institutional Review Board (IRB) of Asan Medical
Center while conforming to the principles of the Declaration of Helsinki. Written informed consent was waived by our IRB as our study was retrospective study design.
The medical records of 330 consecutive patients with OAG as seen by a glaucoma specialist (M.S.K.) between March 2008 and December 2012 at the glaucoma service of Asan Medical Center were retrospectively evaluated. Initially, all patients received comprehensive ophthalmologic examination including a review of medical history, followed by measurement of best-corrected visual acuity (BCVA), slit-lamp biomicroscopy, Goldmann applanation tonometry, gonioscopy, central corneal thickness (CCT) measurement, dilated fundus examination, digital color fundus photography, red-free RNFL photography, stereoscopic optic disk photography, VF examination using the Swedish Interactive Threshold Algorithm standard 24-2 program of the Humphrey Field Analyzer (HFA, Carl Zeiss Meditec), and imaging with Cirrus HD SD-OCT.
Diagnosis of OAG was made as follows 26,28 : BCVA ≥ 20/30; normal open anterior chamber on gonioscopic examination; glaucomatous ONH with diffuse or focal neural rim thinning; a difference in the vertical cup-to-disk ratio >0.2 between eyes, not explained by differences in disk size; and disk hemorrhage or RNFL defects along with compatible glaucomatous VF loss irrespective of intraocular pressure (IOP) level. Glaucomatous VF defects met the Anderson criteria 30 . A reliable VF had to meet the following criteria: a false-positive error <15%, a false-negative error <15%, and a fixation loss <20% 30 . When patients showed glaucomatous VF defects initially, a repeat VF was performed within 2 to 4 weeks to minimize the learning effect.
The following criteria were required to be included in the current study: newly diagnosed OAG without prior treatment; age at initial presentation >18 years; VF MD ≥ −6 dB at initial presentation (for the purpose of assessing VF progression rates in early-stage OAG eyes 32 ); follow-up at our clinic of at least 6 years with regular visits at 6 to 12 month intervals; availability of at least six reliable VF datasets after exclusion of the first perimetry dataset during follow-up; and documented VF progression as determined by event-based analysis. Eyes with visible CRVT origin were included while small discs without clearly visible CRVT origin and optic discs with anomalous vascular patterns with dual trunks were excluded from the study.
Other exclusion criteria included the following: a BCVA < 20/30, pathologic myopic macula, large β-PPA affecting BCVA and VF testing, lens opacities more than C2, N2, or P2 based on the lens opacities classification system III criteria during follow-up 33  www.nature.com/scientificreports www.nature.com/scientificreports/ surgery/laser treatment (including cataract surgery) were excluded from the study. In case of unilateral disease, the affected eye was selected while the first eye with progressive VF loss was included in patients with bilateral disease that met the inclusion criteria.

Measurement of nasalization index of cRVt location, optic disk tilt, and torsion.
Localization of CRVT has been described previously by Wang et al. and Lee et al. (Fig. 3) 14,15 . Briefly, the optimally fitted ellipse around the ONH border was drawn on Cirrus HD SD-OCT volume scan based on the Bruch membrane opening (BMO) margin by two independent raters (K.S. and Y.H.J) who were blinded to each other's results. The center of ellipse is automatically provided by Cirrus HD SD-OCT software. This ellipse was overlaid onto the color fundus photography to show the disk margin. The center of ellipse was to represent the disk center. The location of the CRVT was demarcated on the fundus photography without knowing the patients' clinical information including VF results. The nasalization axis (NA) was then drawn on the fundus photography to connect the disk center and the CRVT location from the temporal to nasal disk border at the initial and last visit by the same two examiners (K.S. and Y.H.J.) using ImageJ software (version 1.52; Wayne Rasband, National Institutes of Health, Bethesda, MD). NI of CRVT location was defined as the distance of the CRVT to the temporal disk border divided by the whole disk diameter on the NA 12,13 . Average values of the NI measurements from the two examiners (K.S. and Y.H.J) were used in analyses.
Optic disk tilt ratio was defined as the ratio between major and minor axis diameter, and torsion degree was defined as the angle between the major axis and the vertical line 90° from a horizontal line connecting the disk center and fovea as described in the previous study 34 . The β-PPA area was estimated to be the total number of pixels by using the ImageJ software in a circumferential pattern 29 . The β-PPA area-to-optic disk area ratio was calculated as an adjusted β-PPA area to minimize the effect of photographic magnification error and to represent the size of β-PPA 35,36 . www.nature.com/scientificreports www.nature.com/scientificreports/ patient grouping according to cRVt location. The study eyes were categorized into two groups on the basis of location of the CRVT in the ONH according to the method described by Lee et al. 15 . From the disk center, the distances were measured to the CRVT (a) and to the nasal disk border (b) on the NA. If the shift index (a/b) is ≥0.5, the patient/eye was categorized to HNL group 15 . Eyes with the shift index <0.5 were categorized to LNL group 15 . In cases where the CRVT is located temporal to the disk center on the NA, negative value for the distance (a) was applied. Both groups were well matched in terms of glaucoma severity (MD ≥ −6 dB) according to inclusion criteria. outpatient follow-up. Treatment was initiated based on the following clinical factors at the time of OAG diagnosis: age, glaucoma severity, baseline and target IOP, presence of disc hemorrhage, and other risk factors. Anti-glaucoma medication was started or increased in patients already receiving treatment in the case of VF progression during follow-up in order to obtain target IOP. At last follow-up, anti-glaucoma medications included prostaglandin analogs (68.9% of patients), dorzolamide/timolol fixed combination (52.3%), topical dorzolamide (19.8%), and alpha-adrenergic agonists (29.8%). Baseline IOP was measured before treatment. The mean, peak, and fluctuation of IOP during the follow-up period were calculated.
Location of visual field defects. We counted the VF defect clusters at baseline and the final visit, which were categorized according to their location in the central 10° or 10-24° regions based on a 10-24° map (Fig. 4) 22 . A central 10° VF defect was defined as clusters of three significant points in the central 10° with a probability <5% on the PD map or two significant points in the central 10° with a probability <1%, regardless of extension to the 10-24° VF area 22 . A peripheral 10-24° VF defect was defined as clusters in the 10-24° region without any extension into the central 10°2 2 .
Definition of visual field progression and visual field progression rates. VF progression was determined with HFA GPA (Carl Zeiss Meditec) using event-based analysis 37 . In the current study, the classification of "likely progression" was considered VF progression 37 . From each follow-up 24-2 VF test of eligible eyes, the MTs of the global, superior central 10°, inferior central 10°, peripheral superior 10-24°, and peripheral inferior 10-24° regions of the VF were collected (Fig. 1) 22 . Similarly, the MTs of the global and 10 regional clusters of the VF were also collected from the glaucoma hemifield test (GHT) map (Fig. 3) 30 . Average MTs were calculated by converting decibel values to apostilbs, averaging them, and then converting back to decibel units. VF progression rates were calculated as the changes in the average VF MT from baseline of each area of the same eye during follow-up 22,30,34,38,39 . RNFL and GCIPL progression analysis. Structural progression was determined with trend-based analyses for the parapapillary RNFL and macular GCIPL thickness parameters (at global, superior, and inferior regions) based on Cirrus HD SD-OCT measurement. Linear regression analysis (expressed in µm/year) was performed on the parapapillary RNFL and macular GCIPL thickness parameters using the GPA software. SD-OCT Images with poor centration, segmentation error, artifact, or a signal strength <7 were not included in the linear regression analysis.

Statistical analyses.
Inter-examiner agreements regarding the location of CRVT (HNL vs. LNL), disk tilt ratio, torsion degree, NI, and adjusted β-PPA area were assessed using Kappa statistics and intraclass correlation coefficients 26 . The two groups were compared with the t-test for continuous variables and Pearson's chi-squared test for categorical variables. To estimate global and regional VF progression rates, a linear mixed model was used to account for confounding effects of covariates 22,34,39 . Models were fitted for fixed effects of follow-up time (years), patient age (years), gender, laterality, spherical equivalent (SE), CCT, follow-up mean IOP, follow-up peak IOP, follow-up IOP fluctuation, and baseline MD and PSD, with a random intercept for each subject 22,34,39 . For the 10 clusters of the GHT map, statistical significance was set at P < 0.005 for the 10 clusters of the GHT map, P < 0.0125 for the superior and inferior central 10° and peripheral 10-24° zones, and P < 0.05 for the global www.nature.com/scientificreports www.nature.com/scientificreports/ 24-2 area 30,34 . A linear mixed model, after adjusting for covariates, were used to compare the rates of progression for the superior and inferior central 10° and peripheral 10-24°, and the 10 GHT clusters between the LNL and HNL groups 30,34 . For structural parameters, the independent t-test was used to compare the rates of parapapillary RNFL and macular GCIPL thinning based on the GPA software between the 2 groups.
Finally, linear mixed models were constructed in all eyes to predict independent variables influencing the rate of VF progression in the central 10° region and central GHT region (i.e., GHT-S1 and GHT-I1 sector) based on age, gender, NI, SE, adjusted β-PPA area, torsion degree, disk tilt ratio, baseline MD and PSD, follow-up mean IOP, peak IOP, and IOP fluctuation 22,34 . Commercially available SAS software version 9.1.3 (SAS, Inc., Cary, NC, USA) and SPSS software version 17.0 (SPSS, Inc., Chicago, IL, USA) were used to perform all statistical analyses.