Diagnostic Accuracy of Three-Dimensional Neuroretinal Rim Thickness for Differentiation of Myopic Glaucoma From Myopia

Abstract

Purpose: To compare the diagnostic accuracy of three-dimensional neuroretinal rim (3D-NRR) and peripapillary retinal nerve fiber layer (RNFL) thickness for differentiation of myopic glaucoma from myopia.

Methods: Healthy myopic individuals (n = 193 eyes) and age-matched myopic glaucoma patients (n = 61 eyes) were enrolled. A 200 × 200-optic disc cube scan was performed with Cirrus HD-OCT. The rates of false-positive errors were compared between RNFL and 3D-NRR thickness measurements. The diagnostic accuracies of RNFL and 3D-NRR thickness for myopic glaucoma were compared by calculating the areas under receiver operating characteristic (AUROC) curves and the partial area under the curve (pAUC) for sensitivity ≥90%.

Results: The overall false-positive rate was significantly greater for RNFL thickness (26.9%) than for 3D-NRR thickness (2.1%, P < 0.001). False-positive RNFL-thickness errors were prevalent in the nasal peripapillary region. The 3D-NRR thickness relative to RNFL revealed a greater AUROC from the 12 to 6 o'clock and 9 o'clock sectors. Again, comparing 3D-NRR with RNFL thickness, the pAUC for sensitivity ≥90% was greater in the nasal quadrant, 12, 3, 4, and 5 o'clock sectors. Also, the sensitivity and specificity, based on the internal normative database, were greater for 3D-NRR than for RNFL thickness.

Conclusions: 3D-NRR thickness measurement reduced the false-positive rate for glaucoma diagnosis and demonstrated better accuracy for glaucoma detection in myopic eyes. Measurement of 3D-NRR can be complementary to RNFL thickness measurement for differentiation of myopic glaucoma from myopia.

Currently, spectral-domain optical coherence tomography (SD-OCT) is widely used to detect structural change to the optic nerve head (ONH) in cases of glaucoma. SD-OCT–measured peripapillary retinal nerve fiber layer (RNFL) thickness provides excellent glaucoma-diagnostic performance and reproducibility.^1–3 In practice, many ophthalmologists depend on the color-coded RNFL thickness map provided by SD-OCT, as it can provide clues to probable “outside normal limits” or “borderline” RNFL defects according to its internal normative database. However, at times, ophthalmologists are confused when confronted with ‘red' signs on the RNFL thickness map and rather equivocal looking optic discs in myopic eyes.⁴ 

Myopia has been well known as a risk factor for glaucoma development.^5–7 Nonetheless, myopia is notorious for causing false-positive errors on the OCT-measured RNFL thickness map.^4,8–10 This phenomenon can be attributed to (1) vessel temporalization during myopic axial elongation, which can affect the peak distribution of RNFL thickness,^11–13 (2) inherent measurement errors in eyes with myopic refraction or longer axial lengths (AXL), due to the OCT magnification effect^14–16, and (3) OCT scan-circle misalignment on the optic disc.^17,18 Kim et al.¹⁹ warned that greater AXL and smaller disc area is significantly associated with false-positive red signs in cases of SD-OCT–measured RNFL thickness. Macular ganglion cell inner-plexiform layer (GCIPL) thickness, which has been reported to offer excellent diagnostic performance for glaucoma,^20,21 also can induce false-positive errors in myopic eyes.²² 

The high incidence of false-positive errors for glaucoma in myopic eyes can lead to erroneous glaucoma diagnoses, thus resulting in unnecessary glaucoma management in healthy individuals. The appropriate diagnosis of glaucoma from myopic population is crucial, as the prevalence of myopia and high myopia are dramatically increasing.²³ Recent meta-analysis have estimated that 50% and 10% of the world (nearly 5 billion and 1 billion people, respectively) will have myopia and high myopia by 2050.²⁴ Due to the significant lifestyle changes for children and young people over the past decades, myopia is getting more prevalent in young ages, especially in Asian countries. 

Glaucoma can induce global or focal change to the neuroretinal rim architecture.^25,26 The neuroretinal rim (NRR) parameters from SD-OCT scans, referencing the Bruch's membrane opening (BMO) as an anatomic landmark, exhibit a good structure-function relationship and excellent diagnostic performance for glaucoma.^27,28 Cirrus HD-OCT (Carl Zeiss Meditec, Inc., Dublin, CA, USA) detects the minimum area of a surface from the optic disc margin (defined as the BMO) to the vitreoretinal interface (VRI) based on three-dimensional (3D) volume scan data. 3D-NRR thickness, which is defined as the distance between the BMO and VRI (which is associated with the minimum cross-sectional rim area in the given direction), is measured and supplemented with a color-coded map based on the internal normative database. This parameter can be less affected by migration of major vessels or scan circle location in myopic eyes. In this regard, the authors hypothesized that 3D-NRR thickness measurement can reduce glaucoma false-positivity rates for myopic eyes. Indeed, SD-OCT–measured 3D-NRR thickness relative to RNFL thickness can be more specific for glaucoma detection in myopic eyes. The purpose of the present study, then, was to determine whether 3D-NRR thickness compared with RNFL thickness reduces the rate of false-positive red signs and improves glaucoma-diagnostic accuracy for myopic eyes. 

Methods

The present study was approved by the Armed Forces Capital Hospital (AFCH) institutional review board and followed the tenets of the Declaration of Helsinki. The healthy participants in this study comprised subjects from the Armed Forces Myopia Study (AFMS), which is an ongoing prospective study on healthy myopic subjects at the AFCH. They were consecutive subjects who met the eligibility criteria and provided written informed consent to participate. Additionally, 61 age-matched myopic open-angle glaucoma (OAG) patients (mean deviation [MD] of visual field [VF] range: −1.28 to −30.42 dB) from the AFCH Glaucoma Clinic were retrospectively enrolled. 

Study Subjects

Subjects who were enrolled in the AFMS as well as OAG patients from AFCH Glaucoma Clinic underwent a comprehensive ophthalmic examination, including visual acuity assessment, slit-lamp biomicroscopy, gonioscopy, Goldmann applanation tonometry, refraction, dilated fundus examination, disc stereo-photography, red-free fundus photography by digital fundus camera (VISUCAM, Carl Zeiss Meditec, Inc.) and standard automated perimetry (Humphrey C 24-2 SITA-Standard visual field; Carl Zeiss Meditec, Inc.). Also, the central corneal thickness (NT-530; NIDEK Co., LTD, Gamagori, Aichi, Japan) and AXL (IOLMaster, Carl Zeiss Meditec, Inc.) were measured. 

To be included in the present study, eyes had to have myopia greater than −0.5 diopters (D) or AXL greater than 24 mm. Mild-to-moderate myopia was defined as eyes having spherical equivalence (SE) greater than −5 D or AXL less than 26 mm. High myopia was defined as eyes having SE −5 D or less or AXL 26 mm or more. Healthy participants had to show no abnormalities on disc stereo-photography, red-free fundus photography, or standard automated perimetry. Myopic OAG was defined as myopic eyes manifesting glaucomatous optic disc changes, such as focal notching and thinning, RNFL defects on disc stereo-photography and red-free fundus photography, glaucomatous VF defect, and an open angle confirmed by gonioscopic examination. Glaucomatous VF defect was defined as (1) glaucoma hemifield test values outside the normal limits or (2) three or more abnormal points with a probability of being normal of P < 5%, of which at least one point had a pattern deviation of P < 1%, or (3) a pattern standard deviation of P < 5%. The VF defects were confirmed on two consecutive reliable tests (fixation loss rate ≤ 20%, false-positive and false-negative error rates ≤ 25%). 

The present study excluded subjects with a history of any ocular surgery except photorefractive keratectomy (PRK) or LASIK, or a history of any retinal disease (e.g., diabetic retinopathy, retinal vein occlusion, AMD). Eyes with motion or blink artifacts (n = 3), algorithm segmentation failure including inappropriate detection of Bruch's membrane openings (BMO; n = 22), a signal strength less than 6 on OCT scans (n = 10), or any abnormalities (e.g., large peripapillary atrophy [PPA] in the circumpapillary region that affected the scan circle where the OCT RNFL thickness measurement was obtained; n = 5) also were excluded. 

SD-OCT Measurements and Determination of False-Positives

If both eyes were eligible, the more myopic eye was selected as the study eye. In cases of myopic OAG eyes, the eye with the greater MD of VF was selected. After obtaining 200 × 200-optic disc cube scans, the following parameters were automatically measured by Cirrus HD-OCT with the built-in analysis algorithm (software version 6.0; Carl Zeiss Meditec, Inc.): peripapillary average, four quadrants (superior, nasal, inferior, and temporal); 12 o'clock-hour sectoral RNFL thicknesses; and optic nerve head (ONH) parameters including rim area, disc area, average cup-to-disc (C/D) ratio, vertical C/D ratio, and cup volume. 

Cirrus HD-OCT detects the innermost termination of the BMO as the optic disc margin. The vector from the optic disc margin to points on the VRI is generated from the 3D volume scan data set. The vector that produces the minimum cross-sectional area for each point on the optic disc margin is identified and transformed into an optic disc plane. The 3D-NRR thickness is defined as the distance between the BMO and VRI (which is associated with the minimum cross-sectional rim area in the given direction; Fig. 1). A total of 180 measurement values of 3D-NRR thickness in TSNIT order, spaced in 2° (from 2°–360° in the Cirrus HD-OCT viewer) were extracted from the device and converted to average, four quadrant (superior, nasal, inferior, and temporal), and 12 o'clock-hour sectoral values. For example, the 3D-NRR thickness in the 12 o'clock sector was the mean of the 3D-NRR thickness values from meridians 76 to 106. The right-eye orientation was used in documenting and analyzing the OCT data. 

Figure 1

 

Schematic diagram showing measurement of 3D-NRR thickness by Cirrus HD-OCT. Cirrus HD-OCT detects BMO as an optic disc margin. The nerve fiber cross-section vectors at points around the optic disc margin in two-dimensional lead to a 3D volume (volumes shown in orange color). This is broken down into a set of trapezoids spaced in 2° (n = 180) around the neuroretinal rim. The 3D-NRR thickness was defined as the distance between the BMO and VRI (which is associated with the minimum cross-sectional rim area in the given direction; green arrow).

The Cirrus HD-OCT software analyzes the measurement values and compares them with the device's internal normative database, and generates a color-coded map to match the RNFL and 3D-NRR thicknesses: green for values within the normal range (P = 5%–95%), yellow for borderline values (P = 1%–5%), and red for values outside the normal range (P < 1%). If an eye had yellow or red codes on the clock-hour sectoral RNFL thickness map or 3D-NRR thickness map without any glaucomatous optic disc change or definite RNFL defects on disc stereo-photography or red-free fundus photography and glaucomatous VF defect, the result was considered to be a false-positive sign. Eyes with any false-positive ‘red' signs on the clock-hour RNFL thickness map were defined as having ‘red disease'. The presence of false-positive signs on RNFL and 3D-NRR thickness maps was determined by two experienced glaucoma specialists (YWK and KHP). Disagreements were resolved by consensus. 

The study eyes were categorized as follows according to the presence of any false-positive red signs on the 12 clock-hour RNFL thickness map: (1) group A (healthy myopic eyes without red signs, n = 141), (2) group B (healthy myopic eyes with false-positive red signs, i.e., red disease, n = 52), and (3) group C (myopic OAG eyes, n = 61). 

Statistical Analysis

Continuous variables were compared among the three groups by one-way ANOVA with Scheffe's post hoc analysis. Categoric variables were compared using a χ² test. The frequency of false-positive signs (at the levels of P < 5% and P < 1%) and their distribution according to the clock-hours on the RNFL and 3D-NRR thickness maps were determined and compared. With regard to the overall false-positive rate for RNFL and 3D-NRR thicknesses, the number of patients with at least one abnormal color code on the clock-hour map or 3D-NRR thickness map, respectively, was determined. The diagnostic performances of RNFL and 3D-NRR thicknesses for detection of myopic glaucoma were determined by calculating the areas under receiver operating characteristic (AUROC) curves. The best cut-off value was selected according to the Youden index value (which maximizes the value of ‘sensitivity + specificity − 1’).²⁹ To compare the two tests' specificities in confirming true normal findings on myopia, the partial area under the curve (pAUC) for sensitivity ≥ 90% was computed and compared between RNFL and 3D-NRR thickness. Partial areas were standardized by the method described by McClish.³⁰ A receiver operating characteristic (ROC) analysis was performed using the “pROC” package³¹ in the open platform R software.³² Confidence intervals for AUC and pAUC were computed with a bootstrap method with 2000 iterations. The AUCs were compared using the function “ROCtest” in the “pROC” package, based on the method introduced by DeLong et al.³³ 

The sensitivities and specificities of the Cirrus OCT-measured RNFL and 3D-NRR thicknesses were tested by comparison with the internal normative database. McNemar's test was used to compare the sensitivity and specificity for detection of myopic OAG. 

Except where stated otherwise, the data are presented as mean ± SD, and the level of statistical significance was set at P < 0.05. 

Results

Baseline Characteristics

The present study included 193 eyes of 193 healthy myopic individuals and 61 eyes of 61 age-matched (31.7 ± 7.8 years; range, 19–45 years) myopic glaucoma patients who met the eligibility criteria. Among the 193 healthy myopic eyes, 141 did not reveal any red signs on the RNFL clock-hour thickness map (group A), and 52 showed false-positive red signs on the RNFL thickness map (red disease, group B). The myopic glaucoma patients included more males than did the healthy individuals. The average age at diagnosis was 28.4 ± 7.5 years (19- to 45-years old) and the average IOP at diagnosis was 17.9 ± 5.5 mm Hg (9–33 mm Hg). 

Table 1 provides the demographics and comparative statistics of the three groups. The ANOVA revealed significant differences among the groups with respect to the refractive error (SE), AXL, disc area, rim area, cup volume, average C/D ratio, and MD of VF. The Scheffe's post hoc analysis revealed that the subjects in groups B and C had significantly longer eyes (26.11 ± 0.89 and 26.15 ± 1.09 mm, respectively) and greater myopia (−3.44 ± 3.19 and −4.15 ± 3.20 D, respectively) than those in group A (24.73 ± 1.12 mm and −1.88 ± 2.10 D; P < 0.001 for all). The disc area was significantly smaller in group B (1.70 ± 0.40 mm²) than in group A (1.97 ± 0.43 mm²) or group C (2.00 ± 0.48 mm²; P < 0.001). The signal strength of the OCT scan tended to be weaker in group B (8.1 ± 0.9) than in groups A (8.5 ± 1.0) or C (8.4 ± 1.0), but the post hoc analysis revealed no statistical significance. The myopic OAG eyes (group C) had a significantly smaller rim area, a greater cup volume, a greater average C/D ratio, and a lesser MD of VF compared with the healthy eyes (groups A and B, all P < 0.001, Table 1). Representative cases of groups B and C are provided, respectively, in Figures 2 and 3. 

Table 1

 

Subject Demographics

 

Figure 2

 

Representative cases of false-positive red signs of RNFL in healthy myopic eyes. (A) Right eye of 21-year-old healthy male in group B (SE = −4.50 D, axial length = 26.33 mm, disc area = 2.06 mm²). Red-free fundus photography and disc stereo-photography showed no definite glaucomatous neuroretinal rim change or RNFL defect. HVF test showed no definite glaucomatous visual field (VF) defect. Optic disc cube scan presented red signs at superior and inferior regions on both deviation and clock-hour map of RNFL thickness (average RNFL thickness = 82 μm). The 3D-NRR thickness map showed borderline abnormality in the superonasal area, but not any red signs (average 3D-NRR thickness = 301 μm). (B) Left eye of 21-year-old healthy male in group B (SE = −2.25 D, axial length = 24.96 mm, disc area = 1.78 mm²). Red-free fundus photography and disc stereo-photography showed no definite glaucomatous neuroretinal rim change or RNFL defect. HVF test showed no definite glaucomatous VF defect. The RNFL thickness revealed red signs on inferior quadrant map and in 3 o'clock sector as well as borderline abnormality in 6 and 7 o'clock sectors (average RNFL thickness = 89 μm). The 3D-NRR thickness was within the normal range (average 3D-NRR thickness = 674 μm).

Figure 3

 

Representative case of myopic glaucoma. Right eye of 26-year-old male in group C (SE = −5.25 D, axial length = 26.27 mm, disc area = 1.59 mm²). Red-free fundus photography and disc stereo-photography showed inferotemporal and superotemporal RNFL defect (yellow arrowheads). HVF test presented superior arcuate defect (mean deviation, −2.96 dB). The RNFL thickness map and 3D-NRR thickness map revealed red signs in superior and inferior regions (average RNFL thickness = 68 μm, average 3D-NRR thickness = 232 μm).

False-Positive Rates and Clock-Hour Distributions of RNFL and NRR Thickness

Among the 193 healthy myopic eyes, 52 showed at least one false-positive red sign on the clock-hour RNFL thickness map (total: 71 clock-hour red signs). By contrast, only 4 eyes showed at least one false-positive red sign on the 3D-NRR thickness map (total: 5 clock-hour red signs). The overall false-positive rate at P < 1% (red color) was greater for RNFL thickness (26.9%) than for 3D-NRR thickness (2.1%, P < 0.001). Figure 4 provides the frequency of false-positive errors at P < 5% (yellow and red colors) and P < 1% (red color only) on the clock-hour RNFL thickness maps. The false-positive red signs (P < 1%) of RNFL thickness were most frequent at 6 o'clock, followed by 5, 1, and 12 o'clock. The false-positive signs of RNFL thickness tended to be frequent inferiorly rather than superiorly, and nasally rather than temporally. The false-positive red signs (P < 1%) of 3D-NRR thickness were found in the 1 (n = 1), 2 (n = 2), 4 (n = 1), and 11 (n = 1) o'clock sectors. 

Figure 4

 

Frequency of false-positive errors of RNFL thickness at levels of P < 5% and P < 1% according to clock hours in healthy myopic eyes. The false-positive signs at the level of P < 5% (yellow and red color code) was most frequent at 5 o'clock (n = 41), followed by 1 (n = 38), 6 (n = 37), and 12 o'clock (n = 27). The false-positive signs (P < 1%) of RNFL thickness were most frequent at 6 o'clock (n = 29), followed by 5 (n = 14), 1 (n = 13), and 12 o'clock (n = 5). The false-positive signs of RNFL thickness tended to be frequent inferiorly rather than superiorly, and nasally rather than temporally.

The overall false-positive rate at P < 1% for RNFL thickness was significantly greater in high myopia (62.8%, n = 27 out of 43 eyes) than mild-to-moderate myopia group (16.7%, n = 25 of 150 eyes, P < 0.001). However, there were no significant differences in the rate of false-positive (P < 1%) for 3D-NRR thickness between high myopia (2.3%, n = 1 of 43 eyes) and mild-to-moderate myopia group (2.0%, n = 3 of 150 eyes, P = 0.90). 

RNFL and 3D-NRR Thicknesses in Healthy Myopia, Red Disease, and Myopic Glaucoma

Tables 2 and 3 provide the mean values of the average, 4 quadrant and 12 clock-hour sectoral measurements along with comparative statistics on RNFL and 3D-NRR thicknesses among the groups. The average RNFL thickness was significantly greater in group A (98.1 ± 8.0 μm), followed by group B (88.9 ± 7.5 μm) and group C (73.2 ± 10.7 μm, P < 0.001). The superior, nasal, and inferior RNFL thicknesses were significantly thinner in group B than in group A, whereas the temporal RNFL thickness was significantly thicker in group B than in group A (all P < 0.001, Table 2). On clock-hour analysis, the RNFL thicknesses from the 12 to 6 o'clock sectors were revealed to be greater in group A than in group B, whereas there were no significant differences in the 7, 9, 10, or 11 o'clock sectors (Table 2; Fig. 5). 

Table 2

 

Comparison of Retinal Nerve Fiber Layer Thickness Among Clock Hours Between Groups

 

Table 3

 

Comparison of Three-Dimensional Neuroretinal Rim Thickness Among Clock Hours Between the Groups

 

Figure 5

 

Clock-hour distribution of RNFL and 3D-NRR thickness between groups. *Statistically significant difference between group A (normal) and group B (red disease).

In contrast to RNFL thickness, the average 3D-NRR thickness showed even greater values in group B (401.0 ± 93.6 μm) compared with groups A (351.5 ± 95.0 μm) and C (203.9 ± 77.5 μm, P < 0.001). All four quadrants showed greater 3D-NRR thickness in group B than did those in group A. On clock-hour analysis, the 3D-NRR thicknesses from the 12, 1, 2, and 5 o'clock sectors revealed no significant differences between groups A and B, whereas those from the 3, 4, and 6 to 11 o'clock sectors showed greater 3D-NRR thickness in group B than in group A (all P < 0.001, Table 3; Fig. 5). The group C eyes showed significantly thinner 3D-NRR thickness than did group A or B, in all of the clock-hour sectors. 

ROC Analysis

There was no significant difference in AUROC for detection of myopic OAG between average RNFL (0.950) and average 3D-NRR thickness (0.923, P = 0.31), both of which showed excellent diagnostic performance. However, the AUROC of the nasal quadrant was significantly greater for 3D-NRR thickness (0.911) than for RNFL thickness (0.617, P < 0.001). The other quadrants revealed no significant differences between the two parameters (Table 4). On clock-hour analysis, 3D-NRR thickness had greater AUROC values than did RNFL thickness from the 12 to 6 o'clock sectors as well as in the 9 o'clock sector. RNFL thickness tended to have a greater AUROC value than did 3D-NRR thickness in the 7 o'clock sector, but with only borderline significance (P = 0.045). There were no significant parameter differences in AUROC values in the 8, 10, and 11 o'clock sectors. The sensitivity and specificity values determined by the best cut-off values ranged from 41.0% to 96.7% and from 37.3% to 91.7%, respectively, for RNFL thickness, and from 47.5% to 88.5% and 78.2% to 96.9%, respectively, for 3D-NRR thickness (Table 4). Figure 6 compares RNFL- and 3D-NRR-thickness ROC curves for detection of glaucoma according to clock-hour sector. 

Table 4

 

Comparison of Area Under Receiver Operating Characteristic Curve Between Retinal Nerve Fiber Layer and Three-Dimensional Neuroretinal Rim Thickness

 

Figure 6

 

ROC) curve of RNFL (blue line) and 3D-NRR (green line) thickness according to clock hours for diagnosis of glaucoma in myopic eyes. The light-bluish shaded box shows the maximum pAUC for sensitivities 90%–100%. P values are shown for comparison of AUROC between RNFL and 3D-NRR thickness.

From subgroup analysis, 3D-NRR thickness revealed greater AUROC value than RNFL thickness only at nasal quadrant for mild-to-moderate myopia, but superior as well as nasal quadrants for high myopia. On clock-hour analysis, the high-myopia group showed greater AUROC values at 9, 11, and 12 o'clock sectors as well as 1 to 6 o'clock sectors for 3D-NRR thickness, where in the region that the mild-to-moderate myopia group showed greater AUROC values (Supplementary Tables S1 and S2). 

To investigate how specific the two parameters are for glaucoma detection in myopic eyes, the corrected pAUC for the 90% to 100% sensitivity range were compared (Table 5). The nasal quadrant showed significantly greater pAUC for 3D-NRR thickness (73.6%) than for RNFL thickness (55.1%, P = 0.002), whereas the average and other quadrants revealed no statistical significance. Clock-hour analysis demonstrated that the 12, 3, 4, and 5 o'clock sectors had greater pAUC for 3D-NRR thickness than for RNFL thickness (Table 5). The pAUC for 3D-NRR thickness tended to be greater in the 1 and 2 o'clock sectors than those for RNFL thickness, but with only borderline significance (all P ≤ 0.10). The pAUC for RNFL thickness in the 7 o'clock sector tended to be greater than those for 3D-NRR thickness, with marginal significance (P = 0.07). None of the other clock-hour sectors showed any significant differences between the two parameters (Table 5). 

Table 5

 

Comparison of Corrected pAUC Between Retinal Nerve Fiber Layer and Three-Dimensional Neuroretinal Rim Thickness for Sensitivity Range 90%–100%

 

Subgroup analysis revealed that the pAUC values for 3D-NRR thickness were greater than RNFL thickness in wider clock-hour sectors, including 12 and 2 o'clock sectors, in the high-myopia compared with mild-to-moderate myopia group. However, RNFL thickness revealed greater pAUC values at 8 o'clock for mild-to-moderate myopia (73.2% vs. 61.2%, P = 0.013) and at 7 o'clock for high myopia group (93.9% vs. 71.9%, P = 0.036), respectively (Supplementary Tables S1 and S2). 

Sensitivity and Specificity Based on Internal Normative Database

Table 6 provides the sensitivities and specificities for RNFL and 3D-NRR thicknesses according to the clock-hour sectors based on the internal normative database. When using the less than 5% criterion for abnormality, the myopic OAG detection sensitivity was significantly greater for 3D-NRR thickness (range, 59.0%–90.2%) than for RNFL thickness (range, 8.2%–90.2%) in all of the clock-hour sectors except 6 and 7 (P = 0.08 and 1.00, respectively). When using the less than 1% criterion, 3D-NRR thickness showed greater sensitivity for detection of myopic OAG (range, 39.7%–82.0%) than did RNFL thickness (range, 3.3%–75.4%) in all of the clock-hour sectors except 7 (82.0% vs. 75.4%, respectively, P = 0.38). 

Table 6

 

Sensitivity and Specificity of Retinal Nerve Fiber Layer and Three-Dimensional Neuroretinal Rim Thickness for Diagnosis of Myopic Glaucoma Based on Internal Normative Database (P < 5% and P < 1%)

 

In terms of specificity, 3D-NRR thickness in the 12, 1, 4, 5, and 6 o'clock sectors showed superior values (range, 89.1%–95.3%) to those for RNFL thickness (range, 78.8%–94.8%) when using the less than 5% criterion for abnormality. For the less than 1% criterion, 3D-NRR thickness in the 12, 1, 3, 5, and 6 o'clock sectors exhibited greater specificities (range, 99.5%–100.0%) than did RNFL thickness (range, 85.0%–97.9%). The other clock-hour sectors showed comparable specificities between 3D-NRR (range, 91.7%–100.0%) and RNFL thickness (range, 88.6%–100.0%) when using the P < 5% and P < 1% criteria for abnormality. 

Discussion

The present study demonstrated that the false-positive rate was significantly lower for 3D-NRR thickness than for RNFL thickness when detecting glaucoma in myopic eyes. Moreover, the diagnostic performance for myopic glaucoma, as represented by AUROC values and pAUC for 90% or more sensitivity, was greater for 3D-NRR thickness than for RNFL thickness, especially in the nasal peripapillary area. Indeed, the sensitivity and specificity for diagnosis of myopic glaucoma based on the internal normative database consistently indicated better diagnostic accuracy for 3D-NRR thickness. 

The normative database in the majority of OCT devices is representative of emmetropic eyes. Cirrus HD-OCT provides a built-in normative database comprised of 271 healthy individuals with a mean spherical equivalence of −0.82 ± 1.96 D.³⁴ The low proportion of myopic eyes in the internal normative database is likely to be related to the high rates of false-positive errors in myopic eyes.^10,35 In the present study, the rate of false-positive errors (P < 1%) for RNFL thickness was significantly higher in high myopia (62.8%) than mild-to-moderate myopia group (16.7%, P < 0.001). The high incidence of false-positive errors for myopic eyes can be further explained, at least speculatively, as follows. First, myopia is well known to influence RNFL thickness measurement, due specifically to the OCT magnification effect.^14–16 The radius of the optic-disc scan circle (default, 1.73 mm) can, based on the ocular magnification effect, vary by refraction and AXL. The magnified scan circle in eyes with greater AXL can incur underestimation of RNFL thickness, as RNFL thickness decreases with distance from the optic disc margin. In fact, OCT-measured RNFL thickness has shown a positive correlation with SE and a negative correlation with AXL.¹⁶ However, this relationship is weakened after adjusting for the ocular magnification effect. Second, myopic eyes are reported to have a more temporally located peak of RNFL thickness, which is dependent on the distribution of major vessels.^12,13,36 This has caused thinning of superior, inferior, and nasal peripapillary RNFL thickness as well as thickening of temporal peripapillary RNFL thickness. A temporally converging RNFL bundle with increasing myopia has been associated with increased area of abnormal RNFL measurement.¹¹ This is consistent with the present findings, group B (red disease) eyes showing thinner RNFL thickness in the superior, nasal, and inferior quadrants and thicker RNFL thickness in the temporal quadrant. Lastly, inappropriate location of the scan circle in myopic eyes also can affect RNFL thickness measurement. It has been reported that a nasally shifted scan circle can increase the RNFL peak distance, which might induce thinning of the superior and inferior RNFL.¹⁷ In this regard, Chung and Yoo¹⁸ demonstrated, based on the contours of the neural canal opening in myopic tilted discs, that the rate of false-positive errors was reduced when the location of the scan circle was corrected in the temporal direction. 

In this regard, it is important not to rely only on color code maps when discriminating glaucoma from healthy eyes (especially myopic eyes) by using OCT device. Instead, we should look at the TSNIT map, assess the location of RNFL peaks, and consider the location of major vessels and their effect to the RNFL thickness measurements. The actual generated B scan as well as thickness and deviation map (in case of Cirrus OCT) also should be checked. 

Efforts are increasing to detect neuroretinal rim changes in glaucoma using SD-OCT scans. Chauhan et al.²⁷ argued that the BMO minimum rim width (BMO-MRW), defined as the minimum distance from the BMO to the ILM, exhibits better diagnostic performance for glaucoma than typical rim assessment from Moorfields regression analysis with confocal scanning laser tomography (CSLT) or RNFL thickness with SD-OCT. This BMO-based approach for detection of ONH change has shown comparable glaucoma-diagnostic power with RNFL thickness in both micro-³⁷ and macrodiscs.³⁸ Malik et al.³⁹ recently reported that BMO-MRW was more sensitive than optic disc margin rim area and similar to RNFL thickness for glaucoma detection in myopic eyes. From this perspective, BMO-MRW has been reported to reduce false-positive rates in cases of healthy myopia with tilted discs.^40,41 The present study, which enrolled young Koreans, also demonstrated an overall, significantly lower false-positive rate for 3D-NRR thickness (2.1%) relative to RNFL thickness (26.9%). East Asian countries, including Korea, are facing tremendously increasing prevalence of myopia.^23,24 The present study, for the first time investigated from Asian population, strengthens the finding that assessment of neuroretinal rim based on BMO can reduce the false-positive errors for glaucoma diagnosis in myopic eyes. 

The rate and clock-hour distribution of false-positive errors for RNFL thickness were similar to data from previous reports based on Cirrus OCT.^19,22 Kim et al.¹⁹ reported that eyes with longer AXL and smaller disc area are likely to have false-positive errors as incurred from RNFL thickness measurement. This is consistent with our finding that group B (red disease) eyes had a longer AXL and a smaller disc area. 

The present study found that 3D-NRR thickness offer better glaucoma-diagnostic performance (greater AUROC values) especially in the region where false-positive signs of RNFL thickness were prevalent (12, 1, 5, and 6 o'clock area). Interestingly, nasal RNFL thickness was thinner in group B (red disease) than in group A (healthy control) eyes, whereas 3D-NRR thickness was even thicker in group B than in group A eyes. The 3D-NRR thickness in the 7, 8, 10, and 11 o'clock sectors, however, showed comparable diagnostic power with RNFL thickness. This is consistent with the findings of Hwang et al.,⁴² who investigated the diagnostic power of NRR thickness for 80 glaucoma eyes and 80 healthy eyes, both of which groups were emmetropic. They concluded that NRR assessment in the nasal and temporal areas by Cirrus OCT can enhance glaucoma detection ability. Our current study further confirmed that 3D-NRR thickness improves specificity for glaucoma diagnosis in myopic eyes in regions where false-positive errors of RNFL thickness have been prevalent in the literature. 

The present study has some limitations. First, it investigated actual, nonmagnification-effect–corrected measurements. Nonetheless, its conclusions were not affected; in fact, when the magnification effect was corrected by AXL, even better AUROC values for NRR thickness were shown. Moreover, in the busy clinical setting, actual measurement values can be more practical, because correcting every patient's data is prohibitively time consuming. Second, the present study population was comprised of only Koreans, though the ONH architecture can differ by race or ethnicity.^43,44 However, another study revealed that there were no significant racial differences in ONH parameters measured by Cirrus OCT when adjusted for disc area.³⁴ Nevertheless, recruitment of study populations from other ethnicities could strengthen the generalizability of the present data. Third, most of the participants of the present study were in young age. This is due to the uniqueness of the military hospital (AFCH), where most of the visitors are youths. Considering that the rate of age-related change differs between RNFL and ONH NRR parameters,⁴⁵ our results may not be generalized to older population. Lastly, the internal normative database of Cirrus OCT adjusts for subject's age for RNFL thickness while subject's age and disc area for 3D-NRR thickness. Because eyes with smaller disc area are associated with thinner RNFL,^34,46 it may have biased to have higher prevalence of false-positive RNFL signs in eyes with smaller disc area. In addition, Cirrus OCT does not provide the normative database for 3D-NRR thickness in eyes with extreme range of disc area (<1.3 or >2.5 mm²), which make difficult to discriminate glaucoma from these eyes. 

In conclusion, 3D-NRR thickness measurement referencing the BMO as an anatomic landmark for neuroretinal rim assessment reduced the false-positive error rates relative to RNFL thickness and improved overall glaucoma-diagnostic performance for myopic eyes. The diagnostic accuracy of 3D-NRR thickness for glaucoma outperformed RNFL thickness especially in the region where false-positive signs of RNFL thickness were prevalent. NRR thickness assessment can be an effective and valuable complementary tool to RNFL thickness measurement for ruling out abnormality in detecting glaucomatous structural change in myopic eyes. 

Acknowledgments

Disclosure: Y.W. Kim, None; K.H. Park, None