Association of Nailfold Capillary Abnormalities With Primary Open-angle Glaucoma and Glaucomatous Visual Field Loss

Supplemental Digital Content is available in the text. Precis: Nailfold capillary abnormalities are associated with primary open-angle glaucoma (POAG) and increased severity of global and central glaucomatous visual field (VF) loss. Purpose: The purpose of this study was to investigate whether nailfold capillary abnormalities are associated with POAG and the severity of glaucomatous VF loss. Materials and Methods: A cross-sectional study of 83 POAG cases and 40 controls was conducted. Nailfold capillaroscopy images were assessed by masked graders for dilated capillaries >50 μm, crossed capillaries, tortuous capillaries, hemorrhages, avascular zones >100 μm, capillary density, and capillary distribution. VF loss in glaucoma cases was quantified using mean deviation and mean central pattern standard deviation (PSD) from the worst-affected eye. Results: Logistic regression analyses of cases and controls showed that avascular zones [odds ratio (OR)=1.24; 95% confidence interval (CI): 1.06, 1.47; P=0.005], capillary density (OR=0.63; 95% CI: 0.46, 0.83; P<0.001), and capillary distribution (OR=7.88; 95% 95% CI: 2.53, 28.40; P=0.001) were associated with POAG. Simple linear regression analysis of cases only showed that nailfold hemorrhages were associated with mean deviation (β=−5.10; 95% CI: −9.20, −1.01; P=0.015) and mean central PSD (β=−4.37; 95% CI: −8.18, −0.57; P=0.025), and this remained significant in the multiple linear regressions. After controlling for demographic and clinical factors, avascular zones were associated with both mean deviation (β=−0.76; 95% CI: −1.48, −0.04; P=0.040) and mean central PSD (β=−0.78; 95% CI: −1.45, −0.10; P=0.024), whereas capillary distribution was only associated with mean deviation (β=−4.67; 95% CI: −7.92, −1.43; P=0.017). Conclusion: Nailfold capillary abnormalities are associated with POAG as well as increased global and central vision loss.

However, the relationship between capillary morphology and POAG remains poorly understood. For example, previous studies that examined the association between dilated capillaries and POAG have shown mixed results. [40][41][42]44 Capillary density and capillary distribution are other NFC variables that are commonly assessed in rheumatology, [47][48][49][50][51] but their potential association with POAG has yet to be fully characterized. 42 Previous studies have also not found a significant association between NFC findings and the severity of glaucomatous visual field (VF) loss. [39][40][41][42] The aims of this study were to examine an extensive range of NFC findings to identify which NFC abnormalities occur in POAG and to determine whether NFC abnormalities were associated with POAG severity and central VF loss. Identifying patients with potential underlying vascular dysfunction may improve risk stratification for POAG and allow earlier initiation of treatments that preserve vision and quality of life. 3,52 Subjects were recruited from an academic glaucoma clinic in Auckland and enrolled after providing informed consent.
Inclusion criteria for patients with POAG were the presence of glaucomatous optic nerve head changes as assessed by a glaucoma specialist ophthalmologist (Helen Danesh-Meyer), glaucomatous VF loss on at least 2 consecutive tests consistent with the pattern of retinal nerve fiber layer (RNFL) loss, open angles on gonioscopy, and normal slit-lamp examination (to exclude secondary glaucomas). Control subjects were normal healthy adults who were the spouses, family members, or caregivers of patients with POAG, or patients with cataract with otherwise normal slit-lamp examination and no evidence of glaucomatous optic neuropathy on clinical examination.
Those with any other ocular or neurological disease that could cause VF loss were excluded, as were individuals with known connective tissue disorders, recent nail trauma, or cosmetic nail treatments that may have affected their NFC findings. Subjects with primary Raynaud phenomenon but no features of an underlying rheumatological disease were included, as primary Raynaud's is not associated with abnormal NFC. 31

Assessment and Analysis of NFC Findings
Subjects were assessed while seated, at an ambient temperature of 22 to 24°C with their hands placed on the examination table at heart level. A drop of cedar immersion oil was placed on the nailfold to improve epidermal translucency and maximize image resolution. 48 All digits of the nondominant hand except the thumb were examined using the Dino-Lite CapillaryScope 500 (MEDL4N5) at a fixed magnification of ×500. Images were captured, coded, stored, and analyzed using the DinoCapture Imaging Software.
Four images per subject were evaluated by 2 independent masked graders. One image per digit was used unless the image quality from a particular digit was too poor, in which case a nonoverlapping image from one of the other 3 digits was randomly chosen. Each image captured ∼0.466 mm 2 of nailfold; hence, the total nailfold area assessed per subject was 1.86 mm 2 .
For each image, the following parameters were assessed: (i) number of dilated capillaries > 50 μm; (ii) number of crossed capillaries with 1 to 3 crossing and hairpin shape intact; (iii) number of tortuous capillaries with > 3 crossings and/or loss of hairpin shape, as shown in Figure 1; (iv) number of hemorrhages; (v) number of avascular zones > 100 μm; (vi) capillary density defined as the number of capillaries per millimeter in the most distal row; and (vii) capillary distribution in the most distal row. [39][40][41][42][43][47][48][49][50][51] Capillary distribution was graded from 1 to 5 based on a scoring system adapted from Cheng et al. 33 A mean distribution score (MDS) was then calculated for each subject based on the distribution scores of all 4 graded images.

Collection of Covariate Data
Information on the subjects' demographic features, medical conditions, family history, and glaucoma-related parameters was extracted from medical records and patient questionnaires. Subjects were also asked specifically about their smoking status 53 and current use of blood thinning medications (antiplatelets and/or anticoagulants) as these factors may affect their NFC results. The resting blood pressure of each subject was also measured as systemic hypertension has been associated with NFC abnormalities. 32

Measurement of POAG Severity and Central Vision Loss
POAG severity was quantified using mean deviation (MD) from a reliable Humphrey SITA-standard 24-2 VF test. The degree of central vision loss was measured by calculating the mean pattern standard deviation (PSD) of the 12 central points on the PSD plot (mean central PSD). Peripheral vision loss was measured using the mean PSD of the remaining 40 peripheral points (mean peripheral PSD). This is similar to the methods used by Park et al 39 except that subjects with overlapping VF defects in both the central and peripheral sectors were not excluded from the analysis.
The average RNFL thickness and average ganglion cell layer plus inner plexiform layer (GCL+IPL) thickness were obtained from optical coherence tomography (OCT) scans of the optic nerve and macula using the Zeiss Cirrus spectral-domain OCT.
All data was recorded from the more severely affected eye and ∼90% of the VF tests and OCT scans were obtained within 1 year of NFC. The remaining POAG subjects had VF tests and OCT scans within 2 years of NFC and had stable POAG, with no significant progression noted by the glaucoma specialist ophthalmologist at the time of the study.

Statistical Analysis
Statistical analysis was performed using R, version 3.6.1 54 with the following packages: lme4, 55 glmperm, 56 lmperm, 57 and multcomp. 58 All tests were 2 tailed, and statistical significance was set at P-value < 0.05.
Demographic and clinical features in POAG cases and controls were compared using t tests, Mann-Whitney-Wilcoxon tests, χ 2 tests, or Fisher exact tests as appropriate.
The relationship between NFC findings and POAG status was assessed using simple and multiple logistic regression analyses. Subjects with Raynaud's were excluded, as Raynaud's was significantly more common among POAG cases than controls. The association between NFC findings and POAG severity was investigated using simple and multiple linear regression analyses among POAG cases only (including those with Raynaud's). NFC findings were the independent variables, and POAG status was the binary outcome variable in the logistic regressions. MD, mean central PSD, mean peripheral PSD, average RNFL thickness, and average GCL+IPL thickness were outcome variables in the linear regressions. Odds ratios (ORs), regression coefficients (β), 95% confidence intervals (CIs), and P-values were calculated for each NFC variable.
For the multiple logistic and linear regressions, 2 models were tested. Model 1 adjusted for demographic variables (age, sex, and ethnicity), whereas model 2 adjusted for demographic variables as well as clinical variables that may affect NFC findings (family history of POAG, hypertension, diabetes, dyslipidaemia, and use of blood thinners). In addition, Raynaud's was added as a control variable in model 2 of the linear regressions but not the logistic regressions, as subjects with Raynaud's were excluded from logistic regression analyses.
Permutation testing was used to control for multiple testing and overcome any distributional assumptions around normally distributed residuals and constant error variance. [59][60][61] For all logistic and linear regressions, a P-value was calculated for each NFC variable, based on random resampling of the data. P-value <0.05 was considered statistically significant.

Power and Reliability
Post hoc power analyses using G*Power 62 indicated 80% power to detect a significant univariate difference of a medium-sized effect between POAG cases and controls, based on a Cohen d = 0.5, and 80% power to detect a medium-sized effect in the simple linear regressions based on a Cohen f 2 = 0.10. 63 Interrater reliability (IRR) for numeric NFC variables was assessed with intraclass correlation coefficients (2-way mixed, average measures, consistency). IRR for categorical NFC variables was measured using Cohen κ 64 and Cohen weighted κ (κ w ) 65 with quadratic weights. (Table 1) showed no significant difference between POAG cases and controls except primary Raynaud's (P < 0.001) as there were 6 POAG cases with Raynaud's and none in the control group.

Analysis of baseline demographic and systemic clinical features
IRR was good to excellent for the NFC findings analyzed as continuous numeric variables (intraclass correlation coefficient ≥ 0.681), and there was near perfect agreement between the 2 graders for hemorrhages (κ = 0.96). However, there was only moderate agreement for capillary distribution (κ w = 0.42).
On average, a 1-unit increase in avascular zones was associated with a 0.76-point reduction in MD and a 0.78-point reduction in mean central PSD.
Avascular zones were not associated with mean peripheral PSD in the simple linear regression analysis or model 1 of the multiple linear regressions (Supplemental Table 4, Supplemental Digital Content 4, http://links.lww. com/IJG/A463). However, avascular zones had a significant negative association with mean peripheral PSD on permutation testing in model 2 of the multiple linear regressions (P = 0.039). This result was more reliable than the results of the linear regression analysis without permutation testing (P = 0.052), as the assumptions of the linear model were violated (non-normally distributed residuals and heteroscedasticity), whereas permutation testing is nonparametric and does not rely on these distributional assumptions. 66 Capillary density was significantly associated with POAG status in univariate (P = 0.002) and simple logistic regression analyses (OR = 0.63; 95% CI: 0.46, 0.83; P < 0.001) ( Capillary distribution was significantly associated with POAG status in the univariate analysis (P = 0.002) and the simple logistic regression analysis for MDS " > 3" compared with the reference level of MDS " ≤ 2" (OR = 7.88; 95% CI: 2.53, 28.40; P = 0.001) ( Table 2, Supplemental  Table 1, Supplemental Digital Content 1, http://links.lww. com/IJG/A460). Capillary distribution remained significant in the multiple logistic regressions and on permutation testing (Supplemental Table 1, Supplemental Digital Content 1, http://links.lww.com/IJG/A460). In model 2 of the multiple logistic regressions, having an MDS " > 3" compared with the reference level of MDS " ≤ 2" increased the odds of being a POAG case by 601%.
Post hoc testing using Tukey test also showed a significant difference between MDS " > 2 and ≤ 3" and MDS " > 3" (P = 0.045). Changing the reference level to MDS " > 2 and ≤ 3" in the simple logistic regression showed that having an MDS " > 3" compared with the new reference level of MDS " > 2 and ≤ 3" increased the odds of being a POAG case by 280% (OR = 3.80; 95% CI: 1.34, 12.59; P = 0.001). However, no other comparisons between MDS levels were significant in the multiple logistic regressions (P > 0.05).
Capillary distribution was not associated with mean central PSD (Supplemental Table 3, Supplemental Digital Content 3, http://links.lww.com/IJG/A462), nor was it associated with MD or mean peripheral PSD in the simple linear regressions (Supplemental Table 2, Supplemental Digital Content 2, http://links.lww.com/IJG/A461, Supplemental Table 4, Supplemental Digital Content 4, http://links.lww. com/IJG/A463). However, capillary distribution was significantly associated with MD and mean peripheral PSD after the addition of control variables in models 1 and 2 of the multiple linear regressions. These results suggest negative confounding and were confirmed by permutation testing (P < 0.05). Yet, there was no significant difference between having an MDS " > 3" compared with the reference level of MDS " ≤ 2" in the multiple linear regressions as the CIs for MD and mean peripheral PSD crossed 0 in both models 1 and 2 (Supplemental Table 2 However, post hoc testing using Tukey test found significant differences between MDS " > 2 and ≤ 3" and MDS    Dilated capillaries > 50 μm were not significantly associated with POAG status (  Table 3, Supplemental Digital Content 3, http://links.lww.com/IJG/ A462). Simple linear regression analysis showed that the presence of dilated capillaries had a significant positive association with MD (β = 0.62; 95% CI: 0.08, 1.17; P = 0.026) and mean peripheral PSD (β = 0.57; 95% CI: 0.05, 1.10; P = 0.033). However, dilated capillaries progressively lost significance as more covariates were controlled for in the multiple linear regressions (Supplemental Table 2, Supplemental Digital Content 2, http://links.lww. com/IJG/A461, Supplemental Table 4, Supplemental Digital Content 4, http://links.lww.com/IJG/A463). This positive confounding suggests that dilated capillaries were not independently associated with MD or mean peripheral PSD.
Crossed capillaries and tortuous capillaries were not significantly associated with POAG status ( Table 2, Supplemental Table 1, Supplemental Digital Content 1, http://links. lww.com/IJG/A460), MD (Supplemental Table 2, Supplemental Digital Content 2, http://links.lww.com/IJG/A461), mean central PSD (Supplemental Table 3, Supplemental Digital Content 3, http://links.lww.com/IJG/A462), or mean peripheral PSD (Supplemental Table 4, Supplemental Digital Content 4, http://links.lww.com/IJG/A463). There was also no significant association between NFC findings and either average RNFL thickness or average GCL+IPL thickness (data not shown). A total of 7 patients with POAG did not have OCT scans available during the study period and were excluded from this portion of the study.

DISCUSSION
To the best knowledge of the authors, this is the first study to demonstrate that reduced capillary density and abnormal capillary distribution are associated with POAG. This study also corroborated the findings of Pasquale et al 41 who identified that presence of avascular zones > 100 μm was significantly associated with POAG. Kosior-Jarecka et al 42 previously found no association between capillary density and normal-tension glaucoma (NTG). Kosior-Jarecka et al 42 also found that "some microvascular architecture alterations" were observed more frequently in controls than patients with NTG. This is in contrast to the results of this study despite there being significant overlap in the criteria used by Kosior-Jarecka et al 42 for assessing microvascular architecture and our grading system for capillary distribution.
As far as the authors know, this is also the first study to demonstrate a significant relationship between NFC abnormalities and the severity of VF loss in POAG. The presence of hemorrhages and avascular zones were associated with MD, mean central PSD, and mean peripheral PSD. On the contrary, abnormal capillary distribution was associated with MD and mean peripheral PSD but not mean central PSD. Park et al 39 previously found that patients with NTG with abnormal NFC have significantly deeper central VF defects than those with normal NFC. However, earlier studies did not find an association between NFC findings and MD. [39][40][41][42] Methodological differences may account for the results of this study, as MD, mean central PSD and mean peripheral PSD were analyzed as continuous variables using linear regression analyses in relation to NFC findings. 39 In contrast, Pasquale et al 41 categorized VF loss as mild (MD > −6 dB), moderate (MD ≤ −6 and ≤ −12 dB), and severe (MD < −12 dB), and then used multiple logistic regression analysis to compare NFC findings in subjects with mild VF loss versus those with moderate or severe VF loss. Park et al 39 and Kosior-Jarecka et al 42 classified their subjects into those with "normal" and "abnormal" NFC but found no difference in MD between these 2 groups.
Despite finding that nailfold hemorrhages are associated with VF loss in patients with POAG, this study did not show a significant association between hemorrhages and POAG status. This is in contrast to previous studies which showed that the presence of hemorrhages is significantly greater in POAG cases than controls. [40][41][42][43] This study also found relatively few hemorrhages in 28% of our POAG cases and 10% of our controls, whereas Pasquale et al 41 42 this study did not find a significant association between hemorrhages and POAG. Further studies are therefore needed to clarify the relationship between nailfold hemorrhages and POAG status.
In addition, despite the association between NFC findings and VF loss, it was interesting to note that there was no significant association between NFC findings and RNFL thickness or average GCL+IPL thickness. Park et al 39 similarly found no difference in RNFL thickness by OCT between NTG cases with normal and abnormal NFC. However, as far as the authors are aware, this is the first study to examine the relationship between NFC findings and average GCL+IPL thickness.
In this study, there was no significant association between POAG and the presence of dilated, crossed, or tortuous capillaries. Kosior-Jarecka et al 42 performed an in-depth analysis of capillary morphology in relation to POAG and reported that "tortuous or coiled" capillaries were not associated with POAG. They also found that "branching" capillaries occurred more in patients with POAG, whereas "meandering" capillaries were more common in controls. However, the results of this study cannot be directly compared with those from Kosior-Jarecka et al 42 as they did not provide specific definitions of the different capillary morphologies they assessed. Furthermore, this study did not classify capillaries as "branching" or "meandering," but it is likely that these 2 categories will have a degree of overlap with the crossed and tortuous capillaries described in the present study. Other studies have shown similarly mixed results in terms of the relationship between dilated capillaries > 50 μm and POAG, [40][41][42] whereas Cousins et al 44 found no difference in mean capillary diameter between patients with POAG and controls. This suggests that capillary morphology is unlikely to be associated with POAG.
This study was limited by its relatively small sample size with predominantly white participants. The study subjects may therefore not be representative of a nonwhite or population-based sample. The cross-sectional nature of this study also makes it difficult to determine whether the NFC abnormalities detected occurred before or after the onset of POAG. POAG cases were also not further classified based on presenting IOP (normal vs. high tension) as they were already receiving IOP-lowering therapy at the time of this study. IOP at enrollment was considered, but many subjects had already started POAG treatment at other centers before enrolling at the study clinic, and IOP at diagnosis was not available for all study participants. In addition, the control group included a small number of subjects with elevated IOP (treated and untreated), but these patients had all been followed up for several years at the study clinic and showed no evidence of glaucomatous changes in optic nerve structure or function.
This study did not control for disease duration in the linear regression analyses as some subjects had already been diagnosed with POAG before enrolling at the study clinic, whereas others were diagnosed with moderate or severe POAG on enrollment, which suggests that the disease had gone undetected for some time before their presentation at the study clinic. Previous studies [40][41][42] did not control for disease duration likely owing to similar difficulties in accurately quantifying disease duration. This study also did not control for IOP in the linear regression analyses as majority of the patients with POAG were already established on IOPlowering therapy, and there was no significant difference in mean IOP between POAG subjects and controls (Table 1). Pasquale et al 41 and Kosior-Jarecka et al 42 similarly did not control for IOP in their analyses. However, Park and colleagues 39,40 found no relationship between mean IOP and NFC findings, which suggests that the systemic microvascular abnormalities detected on NFC are independent of IOP. Last, the use of glaucoma medications and/or prior glaucoma surgery was not controlled for in the logistic or linear regression analyses. Previous similar studies also did not control for glaucoma treatments except Pasquale et al. 41 Furthermore, there are many subtypes of glaucoma medications and most patients with POAG were on 1 or more drugs in combination with 1 or more previous glaucoma surgeries. This would have resulted in very heterogenous and small subgroups for our sample size, which would make it difficult to control for glaucoma treatments in a meaningful sense.
The strengths of this study include its sufficient power despite its small sample size and masked NFC graders to minimize bias. There was good agreement between graders on all NFC variables except for MDS, but this may be owing to the arbitrary cutoffs used for the different MDS levels in the analysis. This study also minimized confounding by collecting data on multiple demographic and clinical covariates as well as controlling for them in the multiple logistic and linear regression analyses. Furthermore, this study examined an extensive range of NFC findings and controlled for multiple testing using permutations in the statistical analysis. Finally, this study establishes the utility of NFC as a safe and noninvasive adjunct to current glaucoma assessment by correlating NFC abnormalities to clinical markers of POAG severity.
In conclusion, NFC abnormalities were significantly and independently associated with POAG case status, POAG severity, central vision loss, and peripheral vision loss. These results support the hypothesis that systemic vascular factors play a role in POAG pathogenesis and progression, but the exact mechanism remains unknown. Future directions include longitudinal studies involving glaucoma suspects (with normal and elevated IOP, or a family history of POAG) to establish the utility of NFC as an adjunct to glaucoma screening and determine the timeline of NFC changes relative to POAG onset and/or progression. Other potential future directions include longitudinal observation of NFC abnormalities in treatmentresistant POAG, as well as developing an automated software for grading NFC images that objectively measures capillary density and identifies hemorrhages and/or avascular zones.