Temporal Raphe Sign for Discrimination of Glaucoma from Optic Neuropathy in Eyes with Macular Ganglion Cell–Inner Plexiform Layer Thinning

Temporal Raphe Sign for Discrimination of Glaucoma from Optic Neuropathy in Eyes with Macular Ganglion CelleInner Plexiform Layer Thinning Jinho Lee, MD,1,2 Young Kook Kim, MD,1,2 Ahnul Ha, MD,1,2 Yong Woo Kim, MD,1,2 Sung Uk Baek, MD,1,2 Jin-Soo Kim, MD,1,2 Haeng Jin Lee, MD,1,3 Dai Woo Kim, MD,4,5 Jin Wook Jeoung, MD, PhD,1,2 Seong-Joon Kim, MD, PhD,1,3 Ki Ho Park, MD, PhD1,2 Purpose: To evaluate the potential of the temporal raphe sign on the macular ganglion celleinner plexiform layer (mGCIPL) thickness map for discriminating glaucomatous from nonglaucomatous optic neuropathy (NGON) in eyes with mGCIPL thinning. Design: Cross-sectional study. Participants: A total of 175 eyes of 175 patients with mGCIPL thinning on Cirrus (Carl Zeiss Meditec, Dublin, CA) high-definition OCT were retrospectively included. Glaucoma specialists and neuro-ophthalmology specialists evaluated the patients’ medical records for diagnosis of glaucomatous optic neuropathy (GON) or NGON. Finally, by consensus, 67 eyes with GON and 73 eyes with NGON were enrolled. Methods: A positive temporal raphe sign was declared in patients in whom there was a straight line longer than one-half of the length between the inner and outer annulus in the temporal elliptical area of the mGCIPL thickness map. Decision tree analysis was performed to formulate a diagnostic model. Main Outcome Measures: Area under receiver operating characteristic curve (AUC) with sensitivity and specificity. Results: The temporal raphe sign was observed in 61 of 67 GON eyes (91.0%), but in only 21 of 73 NGON eyes (28.8%) (P < 0.001; chi-square test). On this basis, the diagnostic ability of the temporal raphe sign for discriminating GON from NGON was judged to be good (AUC, 0.811; 95% confidence interval, 0.749e0.874; sensitivity, 91.0%; specificity, 71.2%). The diagnostic performance of the decision treeebased model (AUC 0.879; 95% confidence interval, 0.824e0.933; sensitivity, 88.1%; specificity, 87.7%) was better than that of the temporal raphe sign or the relative afferent pupillary defect (RAPD) alone (P ¼ 0.005, P < 0.001, respectively; DeLong’s test). The decision tree model revealed the following: (1) If the temporal raphe sign is positive and the RAPD is absent, the case should be diagnosed as GON; (2) if the temporal raphe sign is absent regardless of the presence or absence of the RAPD, or both the temporal raphe sign and the RAPD are present, the case should be diagnosed as NGON. Conclusions: In clinical practice, determining whether the temporal raphe sign appears on OCT macular scans can be a useful tool for discrimination of glaucomatous from nonglaucomatous mGCIPL thinning. Ophthalmology 2019;-:1e9 ª 2018 by the American Academy of Ophthalmology Supplemental material available at www.aaojournal.org.

Glaucoma, affecting more than 70 million people, is the leading cause of visual impairment worldwide.1,2 It comprises a group of diseases of multifactorial etiology defined by a characteristic pattern of structural damage to the optic nerve head (ONH) with visual field defect.2,3 Accurate diagnosis and proper management are essential for preservation of patients’ vision.4,5 However, a number of optic neuropathies mimic glaucomatous optic neuropathy (GON), including ONH cupping or retinal nerve fiber layer (RNFL) defect.6e8 Careful history taking and clinical examination with ancillary tests are helpful for discrimination of GON ª 2018 by the American Academy of Ophthalmology Published by Elsevier Inc.

from such cases of nonglaucomatous optic neuropathy (NGON). Nevertheless, they are still difficult to discriminate, even for well-trained ophthalmologists.7 Progressive retinal ganglion cell (RGC) axonal thinning occurs in glaucoma.9e13 Moreover, mounting evidence in the investigation of glaucomatous damage implicates macular involvement, even in the early stage.14e19 Therefore, for successful monitoring of glaucoma, assessment of all RGC components is vital. However, RGC axonal thinning is detectable not only in GON but also in NGON.20e23 Thus, additional clinical findings on discrimination of https://doi.org/10.1016/j.ophtha.2018.12.031 ISSN 0161-6420/19

1

Ophthalmology Volume -, Number -, Month 2019 glaucomatous from nonglaucomatous macular ganglion celleinner plexiform layer (mGCIPL) thinning are warranted. Recently, our group explored the glaucoma diagnostic ability of a step-like configuration near the temporal raphe on the Cirrus (Carl Zeiss Meditec, Dublin, CA) highdefinition OCT (HD-OCT) mGCIPL thickness map (the so-called temporal raphe sign). We concluded that because of the well-known asymmetry of glaucomatous structural damage around the macula,24e26 the temporal raphe sign can be effective for detection of GON.27,28 The potential of the mGCIPL thickness map’s temporal raphe sign for distinguishing glaucoma from various types of optic neuropathy was further investigated in the present study. Glaucoma specialists and neuro-ophthalmologists closely collaborated to ensure precision diagnosis of GON and NGON.

on stereo disc photography or the presence of RNFL defect on redfree fundus imaging, regardless of the presence or absence of glaucomatous visual field defects. Glaucomatous visual field defects were defined as (1) a cluster of 3 points with probabilities <5% in at least 1 hemifield on the pattern deviation map, including at least 1 point with a probability <1%; (2) glaucomatous hemifield test results outside of the normal limits; or (3) pattern standard deviation (PSD) <5%, as confirmed by at least 2 reliable examinations (falsepositive/negatives <15%, fixation losses <15%). Cases of NGON manifested compressive optic neuropathy (CON),29 nonarteritic anterior ischemic optic neuropathy (NAION),30e32 optic neuritis33 with or without systemic inflammatory conditions, traumatic optic neuropathy, toxic optic neuropathy, and undefined optic neuropathy with optic disc pallor. Data obtained at least 6 months after intracranial surgery for CON or acute attack of optic neuritis or NAION were applied for further analysis.

Methods

Optic-disc (optic disc cube 200200 protocol) and macular scans (macular cube, 512128 protocol) using HD-OCT software (Cirrus, version 6.0; Carl Zeiss Meditec) for RNFL and mGCIPL thickness measurements, respectively, were carried out. Subjects with high-quality OCT images of signal strength 7 and without motion artifacts, involuntary saccade, overt poor centration, or algorithm segmentation failure were included in the study. The analyzed mGCIPL thickness parameters were the average, minimum, and 6 sectoral thicknesses (superotemporal, superior, superonasal, inferonasal, inferior, inferotemporal). The evaluated RNFL parameters were average thickness (360 measure), 4quadrant thickness (temporal, superior, nasal, inferior), and thickness at each of the 12 clock-hour positions (12 o’clock [superior], 3 o’clock [nasal], 6 o’clock [inferior], 9 o’clock [temporal]).

Preparatory to this study, the patients’ electronic medical records of the Glaucoma and Neuro-ophthalmology clinics of Seoul National University Hospital (compiled from January 2017 to May 2018) were reviewed retrospectively. The study, as approved by the Institutional Review Board of Seoul National University Hospital, fully adhered to the Declaration of Helsinki. Informed consent was waived because of the retrospective nature of the study.

Subjects All of the subjects underwent common ophthalmologic examinations, including best-corrected visual acuity measurement, intraocular pressure (IOP) measurement by Goldmann applanation tonometry, refractive error measurement with an autorefractor (KR-890; Topcon Corporation, Tokyo, Japan), corneal pachymetry (Pocket II Pachymeter Echo Graph; Quantel Medical, ClermontFerrand, France), slit-lamp biomicroscopy, gonioscopy, and dilated fundus examination. They also underwent assessment for relative afferent pupillary defect (RAPD) with swinging flashlight method with neutral density filters. Upon maximal pupil dilation, the subjects additionally underwent stereo optic disc photography, red-free RNFL photography (Vx-10; Kowa Optimed Inc, Tokyo, Japan), Cirrus HD-OCT (Carl Zeiss Meditec, Dublin, CA), and standard automated perimetry using the Swedish interactive threshold algorithm according to the 30-2 standard program (Humphrey Field Analyzer II; Humphrey Instruments, Inc, Dublin, CA). For inclusion, subjects were required to have spherical equivalent refraction >6 diopters (D) and <þ3 D, IOP 21 mmHg, a normal open anterior chamber angle, and reliable visual field test results. Also, 2 or more sectors showing red on the mGCIPL deviation map were deemed to represent the presence of mGCIPL thinning; cases meeting the inclusion criteria described were selectively enrolled. Individuals were excluded from further analysis on the basis of the following criteria: (1) the existence of any retinal diseases that could alter mGCIPL thickness and (2) a history of intraocular surgery (except cataract surgery) or retinal laser photocoagulation. If both eyes were eligible for the study, 1 eye was randomly selected. Subsequently, eligible patients’ medical records were evaluated by both glaucoma specialists (J.L. and Y.K.K.) and neuro-ophthalmology specialists (H.J.L. and S.J.K.) for diagnosis of GON or NGON. If the diagnosis was ambiguous, the decision was made by consensus of all of the specialists. GON was defined by the presence of a characteristic optic disc (localized or diffuse neuroretinal rim [NRR] thinning or notching)

2

Cirrus HD-OCT Measurement

Determination of Temporal Raphe Sign We previously evaluated mGCIPL hemifield test using a MATLAB-based (MathWorks Inc, Natick, MA) computer program and validated its diagnostic accuracy.27,28 In this study, the temporal raphe sign was deemed to be identified if there was a horizontal straight line longer than one-half of the inner-to-outerannulus length on the macular mGCIPL thickness map. Determination of temporal raphe sign positivity was made by 3 glaucoma specialists (J.L., Y.K.K., and K.H.P.) independently. If their opinions differed, the pertinent eye was excluded from subsequent analysis. Representative cases of GON and NGON in this study are shown in Figure 1.

Decision Tree Analysis A decision tree is a decision support tool that uses a tree-like model of partitioning rules for a given dataset.34 It also can be applied to classification or regression problems. On the basis of data features, decision tree models learn a series of questions to infer the correct class labels. They can easily be interpreted and understood by general physicians who are not accustomed to machine learning or related statistics. Each question is contained in a decision node, and every child node corresponding to the answer to the question (typically, a “yes” child or a “no” child) is connected to the internal node. As such, the series of questions constitute a hierarchy that can be expressed as a tree. The nodes include a “root node,” the topmost node holding all items, internal nodes that have their child nodes, and leaf nodes that cannot be further divided by any questions. Each item follows the flow of nodes according to the answer of each question and is finally assigned to the category of the leaf node it reaches.34,35

Lee et al



Glaucoma vs. Optic Neuropathy on Macular OCT

Figure 1. Representative cases of glaucomatous optic neuropathy (GON) (AeE) and nonglaucomatous optic neuropathy (NGON) (FeO). In a glaucomatous eye, a horizontal straight line at the temporal macula (red arrowheads) was shown on the (C) mGCIPL thickness map and was consistent with the superior defect on the (E) HVF pattern deviation map. Therefore, diagnosis of glaucoma was made. A patient with compressive optic neuropathy (CON) (FeJ) caused by pineal cyst had no straight line in the temporal macula. The patient showed structural and functional defect respecting vertical meridian on (H and I) the mGCIPL HD-OCT and (J) HVF pattern deviation maps. A patient presenting with a history of optic neuritis (K-O) showed a pale disc on (K) colored disc photography, (L-N) diffuse mGCIPL and RNFL atrophy, and corresponding visual field defect (O). HD-OCT ¼ high-definition OCT; HVF ¼ Humphrey Visual Field; mGCIPL ¼ macular ganglion celleinner plexiform player; RNFL ¼ retinal nerve fiber layer.

In this study, a decision tree analysis for discrimination of GON from NGON was carried out with recursive partitioning algorithms, using RAPD, the temporal raphe sign, the MD and PSD of the visual field test results, and OCT thickness parameters including the average, minimum, and 6 sectoral mGCIPL thicknesses, as well as the average, quadrantal, and clock-hour RNFL thicknesses.36 The decision tree analysis used the “ctree” function of the “party” R software package with only the default settings (no additional specific parameters).37 From all of the entered clinical parameters as decision node candidates, the computer program automatically selected, recursively (based on a cutoff value for numeric variables), the variable for the best split condition, until a leaf node, which is not dividable, was reached. The performance of the identified decision tree was then assessed.

Statistical Analysis All of the statistical analyses were performed using the R software package (version 3.5.1).38 Between the GON and NGON groups, differences in age, IOP, spherical equivalent of refractive error, thickness profiles of mGCIPL and RNFL OCT, mean deviation (MD), and PSD of visual field test results were analyzed by t test; meanwhile, sex, temporal raphe sign, and RAPD differences were computed by chi-square testing. A subgroup analysis subsequently was performed for the NGON group with KruskaleWallis testing followed by Tukey’s post hoc testing (normality test failed). The diagnostic abilities were compared on the basis of area under receiver operating characteristic curve (AUC).39 The sensitivities and specificities of the numeric values were calculated according

to the optimal cutoff point that maximized Youden index (obtained as J ¼ max [sensitivity þ specificity e 1]). P values < 0.05 were considered statistically significant. The data ranges were recorded as mean  standard deviations.

Results A total of 175 eyes of 175 patients with mGCIPL thinning were consecutively enrolled in the present study. From those, 140 patients were selected on the basis of the inclusion and exclusion criteria. These patients were reevaluated by both glaucoma specialists and neuro-ophthalmology specialists for diagnosis of GON or NGON. Finally, 67 patients with GON and 73 patients with NGON were enrolled. The study population’s clinical characteristics are described in Table 1. The NGON group consisted of 4 disease subtype groups: CON (n = 18); central nervous system inflammation (i.e., neuromyelitis optica) (n ¼ 11); NAION (n ¼ 9); optic neuritis without any associated diseases (n ¼ 16); and other conditions (n ¼ 19). The mean age significantly differed between the GON and NGON groups (GON: 65.812.2 years; NGON: 53.914.5 years; P < 0.001), but sex, IOP, spherical equivalent refractive error, MD, and PSD in visual field did not.

Cirrus High Definition-OCT Index Table 2 shows the thickness profiles for the mGCIPL (average, superotemporal, inferotemporal, and absolute difference of mGCIPL thickness between the superotemporal and inferotemporal sectors) and circumpapillary RNFL (average, 4

3

Ophthalmology Volume -, Number -, Month 2019 Table 1. Descriptive Characteristics of Study Population Parameters Age, yrs (mean  SD) Male (%) IOP, in mmHg (mean  SD) SE, in diopters (mean  SD) VF MD, in dB (mean  SD) VF PSD in dB (mean  SD) Presence of RAPD

Glaucoma (N[67) 65.812.2 21 (31%) 14.93.0 0.62.4 7.36.3 7.64.4 3 (4.5%)

Nonglaucoma (N[73) 53.914.5 27 (37%) 15.53.4 1.01.9 7.48.8 6.54.6 37 (50.7%)

Table 2. Comparison of Cirrus High-Definition OCT Parameters between Two Groups P

<0.001 0.668 0.363 0.419 0.907 0.045 <0.001

dB ¼ decibels; IOP ¼ intraocular pressure; MD ¼ mean deviation; PSD ¼ pattern standard deviation; RAPD ¼ relative afferent pupillary defect; SE ¼ spherical equivalent; SD ¼ standard deviation; VF ¼ visual field.

quadrants, and 12 clock-hours). The average mGCIPL and RNFL thicknesses did not significantly differ between the 2 groups. In the GON group relative to the NGON group, however, the mGCIPL thickness was significantly thinner in the inferior sector (59.19.2 mm vs. 63.211.2 mm; P ¼ 0.021) and inferotemporal sector (59.010.6 mm vs. 64.713.3 mm; P ¼ 0.006). In addition, the absolute difference of mGCIPL thickness between the superotemporal and inferotemporal sectors was far greater in the GON group (12.49.2 mm) than in the NGON group (5.35.9 mm; P < 0.001). According to the RNFL comparison, the 2 groups significantly differed in the inferior quadrant (73.019.3 mm in the GON group vs. 90.629.9 mm in the NGON group; P < 0.001).

Diagnostic Ability of Temporal Raphe Sign Among the GON eyes with mGCIPL thinning, temporal raphe sign positivity was shown in 91.0% (61/67) of cases; this rate was significantly higher than that among the NGON eyes (28.8%, 21/73; P < 0.001). In the subgroup analysis for the severe stage (MD <12.0 decibels), temporal raphe sign positivity was 73.3% in GON (11/15 eyes) and 0.0% in NGON (0/6 eyes); the difference was significant (P ¼ 0.0039, Fisher exact test). The AUC value of the temporal raphe sign for discriminating GON from NGON was good (0.811), showing a sensitivity of 91.0% and a specificity of 71.2%, and was greater than that for RAPD (AUC 0.731, sensitivity 91.0%, specificity 71.2%; P ¼ 0.084 with DeLong’s test, Fig S1, available at www.aaojournal.org). In the decision tree analysis, although the MD and PSD of the visual field test results and all of the OCT thickness parameters, as described earlier, had been entered as decision node candidates, only the temporal raphe sign and RAPD were identified as decision nodes (Fig 2). Accordingly, the following new decision model was derived: (1) If the temporal raphe sign is absent or both the temporal raphe sign and RAPD are present, the case is diagnosed as NGON; (2) if the temporal raphe sign is positive but RAPD is absent, the case is diagnosed as GON. The sensitivity and specificity of the decision treeebased model were 88.1% and 87.7%, respectively. The AUC of the decision treeebased model (Fig S1) was 0.879, which was significantly greater than the AUCs of the temporal raphe sign (AUC, 0.811; P ¼ 0.0005) and RAPD (AUC, 0.731; P < 0.001). None of the OCT thickness parameters showed an AUC result superior to that of the decision treeebased model (Table 3).

Subgroup Analysis of Nonglaucomatous Optic Neuropathy Group The NGON group was divided into 4 subgroups according to the underlying conditions. The detailed clinical characteristics are

4

Parameters mGCIPL thickness, in mm (mean  SD) Average Minimum Sector Superotemporal Superior Superonasal Inferonasal Inferior Inferotemporal ST-IT difference* cpRNFL thickness, in mm (mean  SD) Quadrant Superior Nasal Inferior Temporal Clock-hour 11 o’clock 12 o’clock 1 o’clock 2 o’clock 3 o’clock 4 o’clock 5 o’clock 6 o’clock 7 o’clock 8 o’clock 9 o’clock 10 o’clock Temporal raphe sign positivity (%)

Glaucoma (N[67)

Nonglaucoma (N[73)

65.46.6 52.010.6

62.710.8 53.312.2

0.075 0.507

67.07.9 69.410.3 72.210.3 65.810.9 59.19.2 59.010.6 12.49.2 67.810.3

61.911.8 62.011.9 61.412.7 60.912.1 63.211.2 64.713.3 5.35.9 71.121.8

0.003 <0.001 <0.001 0.013 0.021 0.006 <0.001 0.256

85.119.3 59.510.0 73.019.3 53.611.6

87.737.1 58.911.3 90.629.9 52.022.1

0.600 0.740 <0.001 0.592

82.128.3 88.524.4 84.518.3 68.415.1 53.99.3 56.410.9 71.717.9 75.125.3 72.429.6 52.713.5 47.212.2 60.818.0 91.0%

96.852.1 84.238.2 81.529.5 65.617.6 53.710.6 57.512.1 74.222.1 97.136.4 100.841.5 54.626.5 45.115.2 57.229.5 28.8%

0.038 0.418 0.470 0.324 0.908 0.558 0.470 <0.001 <0.001 0.586 0.358 0.376 <0.001

P

cpRNFL ¼ circumpapillary retinal nerve fiber layer; IT ¼ inferotemporal; mGCIPL ¼ macular ganglion celleinner plexiform layer; SD ¼ standard deviation; ST ¼ superotemporal. *Absolute difference of mGCIPL thickness between superotemporal and inferotemporal sectors.

summarized and compared in Table 4. Eight of 9 patients with NAION showed temporal raphe sign positivity (88.9%); thus, in those cases, the temporal raphe sign was not helpful for discrimination of GON from NAION. In the post hoc analysis, the average mGCIPL thickness of the optic neuritis group was significantly different from that of the NAION group (P < 0.001) and CON group (P ¼ 0.038). The superonasal mGCIPL thickness of the optic neuritis group differed from those of the CNS inflammation and NAION groups (P ¼ 0.024 and 0.032, respectively). The inferior and inferotemporal mGCIPL thicknesses differed significantly between the NAION and optic neuritis groups (P ¼ 0.005, 0.030, respectively). The absolute difference of mGCIPL thickness between the superotemporal and inferotemporal sectors significantly differed between the NAION and other conditions groups (P ¼ 0.022). In the other conditions NGON group, most of the cases were optic neuropathy or optic atrophy with unknown causes, except for 4 with traumatic optic neuropathy and 2 with ethambutol-induced toxic optic neuropathy. In the other conditions group, all (100%) of the eyes with toxic neuropathy showed temporal raphe sign

Lee et al



Glaucoma vs. Optic Neuropathy on Macular OCT NGON

Temporal raphe sign P < 0.001

Negative

GON

Positive

RAPD P < 0.001

Absent

n = 58

1

n = 68

Present

1

n = 14

1

0.8

0.8

0.8

0.6

0.6

0.6

0.4

0.4

0.4

0.2

0.2

0.2

0

0

0

Figure 2. Decision tree analysis. The decision tree was formulated by unbiased recursive partitioning based on a permutation test. The terminal nodes are drawn as stacked bar charts. Although various OCT parameters were entered as decision node candidates, only the temporal raphe sign and relative afferent pupillary defect (RAPD) were identified as decision nodes. Each node showed P < 0.001. GON ¼ glaucomatous optic neuropathy; NGON ¼ nonglaucomatous optic neuropathy.

positivity and were misdiagnosed as GON in the decision tree model. One of 4 eyes (25%) with traumatic optic neuropathy showed temporal raphe sign positivity; nevertheless, the decision tree model correctly classified it as NGON.

Discussion In this study, we determined the diagnostic validities of the temporal raphe sign and a proposed decision tree model for discrimination of glaucomatous structural loss from nonglaucomatous damage. We found that the diagnostic efficacy of the temporal raphe sign alone for discrimination of glaucomatous structural loss was good and that the efficacy of the decision tree model was even better. Some kinds of NGON show structural and functional loss that appears similar to glaucomatous change.6e8 Several differences of structural and functional loss for the purposes of differential diagnosis have been reported, including loss of visual acuity out of proportion to optic disc cupping40 and deficiencies in color vision testing.41,42 In optic disc assessment, pallor of the remaining NRR favors NGON,43 whereas focal loss of the NRR, normal color of the remaining NRR, vertical cup elongation, and splinter hemorrhage are relatively specific to GON.44,45 Characteristic visual field defects in glaucoma and their relationships to structural

change, also well described already, can be applied for differential diagnosis as well.9,46 Although case history and careful clinical examination are essential, formulation of additional discriminating criteria is warranted. The temporal RNFL, the origins of which strictly respect the horizontal raphe, enters into the superotemporal and inferotemporal aspects of the optic disc. Because of this superior/inferior segregation, glaucomatous structural damage and functional loss corresponding to the anatomic arrangement of the RNFL are often asymmetric across the horizontal meridian, especially in the early stages.24,26,47 Asymmetry of glaucomatous structural loss can be expressed in the form of the temporal raphe sign on the mGCIPL OCT thickness map. In this study, we judged the presence of the temporal raphe sign by a horizontal demarcated line longer than one-half of the length between the temporal inner elliptical annulus and outer elliptical annulus. This methodology is similar to the mGCIPL Hemifield Test previously introduced;27,28 however, it has been difficult to apply to clinical practice for compatibility with MATLAB-based computer programs. In this study, glaucoma specialists determined temporal raphe sign positivity on the mGCIPL thickness map. The temporal raphe sign showed, despite the absence of any computer program, great diagnostic ability for discrimination of GON. The AUC (0.811), sensitivity (91.0%), and specificity (71.2%) of the temporal raphe sign were found to be good

5

Ophthalmology Volume -, Number -, Month 2019 Table 3. Areas Under the Receiver Operating Characteristic Curve of Temporal Raphe Sign, Relative Afferent Pupillary Defect, OCT Parameters, and Decision TreeeBased Model Parameters

AUC (Range)

Sensitivity, % (Range)

Specificity, % (Range)

Temporal raphe sign RAPD Decision treeebased model mGCIPL thickness Average Minimum Superotemporal Superior Superonasal Inferonasal Inferior Inferotemporal Superotemporal-inferotemporal* cpRNFL thickness Average Superior Nasal Inferior Temporal

0.811 (0.749e0.874) 0.731 (0.668e0.794) 0.879 (0.824e0.933)

91.0 (84.2e97.9) 95.5 (90.6e100) 88.1 (80.3e95.8)

71.2 (60.8e81.6) 50.7 (39.2e62.1) 87.7 (80.1e95.2)

0.611 0.529 0.656 0.703 0.754 0.637 0.613 0.626 0.771

(0.516e0.707) (0.432e0.625) (0.563e0.748) (0.617e0.790) (0.672e0.835) (0.542e0.732) (0.520e0.706) (0.533e0.718) (0.694e0.848)

74.6 98.5 73.1 86.6 89.5 65.7 47.8 76.1 83.6

(64.2e85.1) (95.5e100) (62.7e83.6) (77.6e94.0) (82.1e95.5) (53.7e77.6) (37.3e59.7) (65.7e86.6) (74.6e92.5)

52.0 54.8 60.3 49.l3 57.5 67.1 71.2 42.5 60.3

(41.1e63.0) (1.4e11.0) (49.3e71.2) (38.4e61.6) (46.6e68.5) (56.1e78.1) (61.6e80.8) (31.5e53.4) (49.3e71.2)

0.515 0.470 0.536 0.676 0.619

(0.389e0.582) (0.434e0.627) (0.439e0.632) (0.587e0.764) (0.524e0.713)

4.48 4.48 67.2 73.1 64.2

(0e8.95) (0e10.4) (55.2e79.1) (62.7e83.6) (52.2e76.1)

82.2 87.7 43.8 54.8 61.6

(74.0e90.4) (79.4e94.5) (32.9e54.8) (42.5e67.1) (50.7e72.6)

AUC ¼ area under receiver operating characteristic curve; cpRNFL ¼ circumpapillary retinal nerve fiber layer; mGCIPL ¼ macular ganglion celleinner plexiform layer; RAPD ¼ relative afferent pupillary defect. *Absolute difference of mGCIPL thickness between superotemporal and inferotemporal sectors.

and potentially helpful in clinical practice. Although temporal raphe sign positivity in severe-stage glaucoma was lower than overall (73.3% vs. 91.0%), it was still significantly larger than that of severe-stage NGON and might be a useful clue for discrimination.

In previous reports, the presence of RAPD in GON ranged from 9% to 82%.48 Relative afferent pupillary defect is detectable once the RNFL has been thinned to 83% compared with the fellow eye.49 Relatively symmetric optic neuropathy including GON, relative to asymmetrical optic

Table 4. Descriptive Statistics of Subgroup Analysis of Nonglaucomatous Optic Neuropathies by Underlying Conditions

Age, yrs (mean  SD) Male (%) Cup-to-disc ratio (mean  SD) Presence of RAPD Temporal Raphe sign positivity mGCIPL thickness, in mm (mean  SD) Average Superotemporal Superior Superonasal Inferonasal Inferior Inferotemporal Superotemporal-inferotemporaly cpRNFL thickness, in mm (mean  SD) Average Superior Nasal Inferior Temporal

CON (N[18)

CNS Inflammation (N[11)

NAION (N[9)

Optic Neuritis (N[16)

Others (N[19)

P

60.310.0 9 (50%) 0.40.1 38.9% 11.1%

52.810.3 2 (18%) 0.40.1 63.6% 18.2%

56.29.9 5 (56%) 0.30.2 55.6% 88.9%

52.816.1 4 (25%) 0.40.3 62.5% 31.2%

48.318.6 7 (37%) 0.50.2 42.1% 21.1%

0.372* 0.258 0.167 0.512 0.001

66.813.8 67.315.9 66.416.6 64.417.7 63.616.7 66.613.6 69.617.1 5.35.5

63.79.9 62.010.5 64.810.4 63.611.1 62.510.6 64.89.8 64.410.7 4.75.5

67.36.8 61.46.8 63.87.7 64.410.3 64.912.2 70.210.0 72.012.0 12.88.0

54.66.4 55.18.5 54.17.2 53.15.4 53.25.9 55.48.2 56.110.4 4.44.8

62.99.7 62.610.4 61.99.6 62.711.3 61.99.9 62.39.0 64.29.9 2.83.3

0.007* 0.282* 0.052* 0.040* 0.078* 0.014* 0.026* 0.017*

82.231.8 103.658.5 62.615.1 101.131.3 60.736.7

67.614.3 79.719.9 55.69.3 85.829.2 50.58.2

67.111.3 69.815.2 56.410.3 92.031.6 50.311.1

64.923.8 87.133.3 56.87.4 85.337.4 49.121.1

69.511.0 86.124.0 60.211.3 87.120.0 47.911.0

0.127* 0.202* 0.405* 0.429* 0.246*

CON ¼ compressive optic neuropathy; CNS ¼ central nervous system; cpRNFL ¼ circumpapillary retinal nerve fiber layer; mGCIPL ¼ macular ganglion celleinner plexiform layer; NAION ¼ nonarteritic anterior ischemic optic neuropathy; RAPD ¼ relative afferent pupillary defect; SD ¼ standard deviation. *KruskaleWallis test. y Absolute difference of mGCIPL thickness between superotemporal and inferotemporal sectors.

6

Lee et al



Glaucoma vs. Optic Neuropathy on Macular OCT

neuropathy, rarely shows RAPD. Thus, we evaluated the sensitivity and specificity of RAPD and compared the results with those for the temporal raphe sign. Because RAPD generally favors NGON over GON, “absence of RAPD” was used for diagnosis of GON. Because most GON cases in this study did not show RAPD, the sensitivity was high (95.5%); however, the specificity was very low compared with that of the temporal raphe sign. Of note, the frequency of RAPD was only 50.7% in the NGON group, which would make the specificity and AUC of RAPD significantly lower. The presence of RAPD reflects the laterality of disease, and 47 of 73 (64.4%) showed bilaterality in our NGON group. The results of the study could be different if the composition of the unilateral versus bilateral cases were changed in both groups. Although the temporal raphe sign showed good diagnostic ability, there were some cases of misclassification. Notably, most cases (8/9, 88.9%) of NAION showed temporal raphe sign positivity misdiagnosed as GON. Taking into account the vascular anatomy, the arterial circle of Zinn-Haller, which is branched from the short posterior ciliary artery and supplies blood flow to the ONH, is separated into superior and inferior halves.50 Because these superior and inferior parts are not reconnected distally, NAION can mimic glaucomatous change in patterns of RNFL thinning, such as predilection for the superior and inferior quadrants8 and altitudinal asymmetry.51 In some previous reports of analyses of mGCIPL OCT parameters of NAION,20,52,53 mGCIPL thickness also is decreased and correlated with the visual field. However, because the frequency of RAPD, which was included as a decision node, significantly differed between GON and NAION, the decision treeebased model could detect most GON cases. Also, all of the cases of toxic neuropathy (2/2) showed temporal raphe sign positivity. However, considering the relatively small number of subjects, further research including a greater number of participants with toxic neuropathy is warranted. We subsequently performed a decision tree analysis to achieve even higher diagnostic accuracy. Decision trees are a means of representing rules recursively by partition of hierarchical data. A decision tree can be used not only for description or classification of existing data but also for generalization, which is useful for prediction of the values of dependent variables (in our study: GON diagnosis). Composed of a sequence of easy-to-understand tests, decision trees are known to yield robust results.35 Among the many methods of decision tree formulation, we selected unbiased recursive partitioning,36 which is known to have good performance comparable to other established exhaustive search procedures. In the end, only 2 explanatory variables (temporal raphe sign and RAPD) were adopted. Generally, decision tree analysis is performed in the following sequence: (1) Find an explanatory variable that best partitions the population according to the dependent variable; (2) make a partitioning rule and divide the population into 2 groups accordingly; and (3) repeat steps 1 and 2 to divide the space repeatedly and organize these rules into a tree form. According to the present study’s results, the best explanatory variable was the temporal raphe sign, followed by RAPD. In each node divided by the presence of the temporal raphe sign and RAPD, there were

no OCT thickness parameters that significantly differed between the GON and NGON groups. Thus, no OCT parameters were adopted as explanatory variables. Although the model was simple to apply directly for clinical practice, it showed robust diagnostic performance. The AUC, sensitivity, and specificity of the model were superior to those of the temporal raphe sign or RAPD alone. Several points need to be considered when interpreting the results of this study. First, although the judgment of the presence or absence of the temporal raphe sign was confirmed by consensus of 2 glaucoma specialists, the determination was, nonetheless, subjective. There is a potential risk of misclassification, especially in cases in which the horizontal line length is very close to one-half of that of the temporal inner-to-outer annulus. Second, in both groups, we included only patients who had an IOP 21 mmHg, spherical equivalent refraction >6 D and <þ3 D, and no retinal disease that could alter mGCIPL thickness. Considering that NGON typically has IOP within the normal range, we considered that normal-tension glaucoma might be more difficult to discriminate than high-tension glaucoma (HTG) in clinical practice. Also, in cases of high myopia accompanied by myopic maculopathy, mGCIPL thickness could be affected, thus making results less reliable. Thus, we excluded subjects who had HTG or high myopia as well as any retinal disease that could alter mGCIPL thickness. However, our results might not be generalizable to other population, such as HTG or high myopia. Third, additional logistic regression for exploration of clinical parameters possibly affecting misclassification by the decision tree model was not performed. Further study with a larger population for investigation of the clinical factors associated with misclassification and improved performance is warranted. Fourth, considering the lower prevalence of optic neuropathies compared with glaucoma, the negative predictive power of the temporal raphe sign might be worse in the general population. Fifth, in the severe stage, temporal raphe sign positivity declined relative to the early-tomoderate stage, which means that it would be less useful in patients with diffuse mGCIPL atrophy. In conclusion, the temporal raphe sign and the decision treeebased model both showed good diagnostic efficacy for application to the HD-OCT mGCIPL thickness map. In the case of mGCIPL thinning, the temporal raphe sign can be a simple but useful tool for discriminating between glaucomatous and nonglaucomatous structural change.

References 1. Dandona L, Dandona R. What is the global burden of visual impairment? BMC Med. 2006;4:6. 2. Foster PJ, Buhrmann R, Quigley HA, Johnson GJ. The definition and classification of glaucoma in prevalence surveys. Br J Ophthalmol. 2002;86:238e242. 3. Casson RJ, Chidlow G, Wood JP, et al. Definition of glaucoma: clinical and experimental concepts. Clin Exp Ophthalmol. 2012;40:341e349. 4. Leske MC, Heijl A, Hyman L, Bengtsson B. Early Manifest Glaucoma Trial: design and baseline data. Ophthalmology. 1999;106:2144e2153.

7

Ophthalmology Volume -, Number -, Month 2019 5. Anderson DR. Collaborative normal tension glaucoma study. Curr Opin Ophthalmol. 2003;14:86e90. 6. Kuo DS, Asrani S. Normal-tension glaucoma masqueraders: detection using optical coherence tomography. J Glaucoma. 2017;26:e153ee156. 7. O’Neill EC, Danesh-Meyer HV, Kong GX, et al. Optic disc evaluation in optic neuropathies: the optic disc assessment project. Ophthalmology. 2011;118:964e970. 8. Pasol J. Neuro-ophthalmic disease and optical coherence tomography: glaucoma look-alikes. Curr Opin Ophthalmol. 2011;22:124e132. 9. Hood DC, Kardon RH. A framework for comparing structural and functional measures of glaucomatous damage. Prog Retin Eye Res. 2007;26:688e710. 10. Balaratnasingam C, Morgan WH, Bass L, et al. Axonal transport and cytoskeletal changes in the laminar regions after elevated intraocular pressure. Invest Ophthalmol Vis Sci. 2007;48:3632e3644. 11. Calkins DJ, Horner PJ. The cell and molecular biology of glaucoma: axonopathy and the brain. Invest Ophthalmol Vis Sci. 2012;53:2482e2484. 12. Crish SD, Sappington RM, Inman DM, et al. Distal axonopathy with structural persistence in glaucomatous neurodegeneration. Proc Natl Acad Sci U S A. 2010;107:5196e5201. 13. Buckingham BP, Inman DM, Lambert W, et al. Progressive ganglion cell degeneration precedes neuronal loss in a mouse model of glaucoma. J Neurosci. 2008;28:2735e2744. 14. Mwanza J-C, Durbin MK, Budenz DL, et al. Glaucoma diagnostic accuracy of ganglion celleinner plexiform layer thickness: comparison with nerve fiber layer and optic nerve head. Ophthalmology. 2012;119:1151e1158. 15. Nouri-Mahdavi K, Nowroozizadeh S, Nassiri N, et al. Macular ganglion cell/inner plexiform layer measurements by spectral domain optical coherence tomography for detection of early glaucoma and comparison to retinal nerve fiber layer measurements. Am J Ophthalmol. 2013;156:1297e1307.e2. 16. Choi YJ, Jeoung JW, Park KH, Kim DM. Glaucoma detection ability of ganglion cell-inner plexiform layer thickness by spectral-domain optical coherence tomography in high myopia. Invest Ophthalmol Vis Sci. 2013;54:2296e2304. 17. Hwang YH, Jeong YC, Kim HK, Sohn YH. Macular ganglion cell analysis for early detection of glaucoma. Ophthalmology. 2014;121:1508e1515. 18. Jeoung JW, Choi YJ, Park KH, Kim DM. Macular ganglion cell imaging study: glaucoma diagnostic accuracy of spectraldomain optical coherence tomography. Invest Ophthalmol Vis Sci. 2013;54:4422e4429. 19. Baek SU, Kim KE, Kim YK, et al. Development of topographic scoring system for identifying glaucoma in myopic eyes: a spectraldomain OCT study. Ophthalmology. 2018;125:1710e1719. 20. Larrea BA, Iztueta MG, Indart LM, Alday NM. Early axonal damage detection by ganglion cell complex analysis with optical coherence tomography in nonarteritic anterior ischaemic optic neuropathy. Graefes Arch Clin Exp Ophthalmol. 2014;252:1839e1846. 21. Park SW, Ji YS, Heo H. Early macular ganglion celleinner plexiform layer analysis in non-arteritic anterior ischemic optic neuropathy. Graefes Arch Clin Exp Ophthalmol. 2016;254:983e989. 22. Rebolleda G, de Dompablo E, Muñoz-Negrete FJ. Ganglion cell layer analysis unmasks axonal loss in anterior optic neuritis. J Neuroophthalmol. 2015;35:165e167. 23. Shin H-Y, Park H-YL, Choi J-A, Park CK. Macular ganglion celleinner plexiform layer thinning in patients with visual field defect that respects the vertical meridian. Graefes Arch Clin Exp Ophthalmol. 2014;252:1501e1507.

8

24. Um TW, Sung KR, Wollstein G, et al. Asymmetry in hemifield macular thickness as an early indicator of glaucomatous change. Invest Ophthalmol Vis Sci. 2012;53: 1139e1144. 25. Asrani S, Rosdahl JA, Allingham RR. Novel software strategy for glaucoma diagnosis: asymmetry analysis of retinal thickness. Arch Ophthalmol. 2011;129:1205e1211. 26. Yamada H, Hangai M, Nakano N, et al. Asymmetry analysis of macular inner retinal layers for glaucoma diagnosis. Am J Ophthalmol. 2014;158:1318e1329.e3. 27. Kim YK, Yoo BW, Kim HC, Park KH. Automated detection of hemifield difference across horizontal raphe on ganglion cell–inner plexiform layer thickness map. Ophthalmology. 2015;122:2252e2260. 28. Kim YK, Yoo BW, Jeoung JW, et al. Glaucoma-diagnostic ability of ganglion cell-inner plexiform layer thickness difference across temporal raphe in highly myopic eyes. Invest Ophthalmol Vis Sci. 2016;57:5856e5863. 29. Nakano E, Hata M, Oishi A, et al. Quantitative comparison of disc rim color in optic nerve atrophy of compressive optic neuropathy and glaucomatous optic neuropathy. Graefes Arch Clin Exp Ophthalmol. 2016;254:1609e1616. 30. Saito H, Tomidokoro A, Sugimoto E, et al. Optic disc topography and peripapillary retinal nerve fiber layer thickness in nonarteritic ischemic optic neuropathy and open-angle glaucoma. Ophthalmology. 2006;113:1340e1344. 31. Chan CK, Cheng AC, Leung CK, et al. Quantitative assessment of optic nerve head morphology and retinal nerve fibre layer in non-arteritic anterior ischaemic optic neuropathy with optical coherence tomography and confocal scanning laser ophthalmoloscopy. Br J Ophthalmol. 2009;93: 731e735. 32. Bellusci C, Savini G, Carbonelli M, et al. Retinal nerve fiber layer thickness in nonarteritic anterior ischemic optic neuropathy: OCT characterization of the acute and resolving phases. Graefes Arch Clin Exp Ophthalmol. 2008;246:641e647. 33. Costello F, Coupland S, Hodge W, et al. Quantifying axonal loss after optic neuritis with optical coherence tomography. Ann Neurol. 2006;59:963e969. 34. Kingsford C, Salzberg SL. What are decision trees? Nat Biotechnol. 2008;26:1011e1013. 35. Murthy SK. Automatic construction of decision trees from data: a multi-disciplinary survey. Data Min Knowl Discov. 1998;2:345e389. 36. Hothorn T, Hornik K, Zeileis A. Unbiased recursive partitioning: a conditional inference framework. J Comput Graph Stat. 2006;15:651e674. 37. Hothorn T, Hornik K, Strobl C, Zeileis A. Party: a laboratory for recursive partytioning. 2010. Computer program and documentation available at https://cran.r-project. Available at: org/web/ packages/party/vignettes/party.pdf. Accessed September, 12. 38. R Core Team (2013). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. Available at: http://www.R-project.org/. 39. Robin X, Turck N, Hainard A, et al. pROC: an open-source package for R and Sþ to analyze and compare ROC curves. BMC Bioinformatics. 2011;12:77. 40. Kupersmith M, Krohn D. Cupping of the optic disc with compressive lesions of the anterior visual pathway. Ann Ophthalmol. 1984;16:948e953. 41. Drance SM, Lakowski R, Schulzer M, Douglas GR. Acquired color vision changes in glaucoma: use of 100-hue test and Pickford anomaloscope as predictors of glaucomatous field change. Arch Ophthalmol. 1981;99:829e831.

Lee et al



Glaucoma vs. Optic Neuropathy on Macular OCT

42. Pacheco-Cutillas M, Edgar D, Sahraie A. Acquired colour vision defects in glaucomadtheir detection and clinical significance. Br J Ophthalmol. 1999;83:1396e1402. 43. Trobe JD, Glaser JS, Cassady J, et al. Nonglaucomatous excavation of the optic disc. Arch Ophthalmol. 1980;98: 1046e1050. 44. Healey PR, Mitchell P, Smith W, Wang JJ. Optic disc hemorrhages in a population with and without signs of glaucoma. Ophthalmology. 1998;105:216e223. 45. Healey PR, Mitchell P. Presence of an optic disc notch and glaucoma. J Glaucoma. 2015;24:262e266. 46. Kitazawa Y, Yamamoto T. Glaucomatous visual field defects: their characteristics and how to detect them. Clin Neurosci. 1997;4:279e283. 47. Walsh T. Visual Fields: Examination and Interpretation. Vol. 3. Oxford: Oxford University Press; 2010. 48. Chang DS, Xu L, Boland MV, Friedman DS. Accuracy of pupil assessment for the detection of glaucoma: a systematic review and meta-analysis. Ophthalmology. 2013;120:2217e2225.

49. Chew SS, Cunnningham WJ, Gamble GD, Danesh-Meyer HV. Retinal nerve fiber layer loss in glaucoma patients with a relative afferent pupillary defect. Invest Ophthalmol Vis Sci. 2010;51:5049e5053. 50. Olver J, Spalton D, McCartney A. Microvascular study of the retrolaminar optic nerve in man: the possible significance in anterior ischaemic optic neuropathy. Eye. 1990;4:7. 51. Hayreh SS, Zimmerman B. Visual field abnormalities in nonarteritic anterior ischemic optic neuropathy: their pattern and prevalence at initial examination. Arch Ophthalmol. 2005;123:1554e1562. 52. Fard MA, Afzali M, Abdi P, et al. Comparison of the pattern of macular ganglion cell-inner plexiform layer defect between ischemic optic neuropathy and open-angle glaucoma. Invest Ophthalmol Vis Sci. 2016;57:1011e1016. 53. Lee YH, Kim KN, Heo DW, et al. Difference in patterns of retinal ganglion cell damage between primary open-angle glaucoma and non-arteritic anterior ischaemic optic neuropathy. PLoS One. 2017;12:e0187093.

Footnotes and Financial Disclosures Originally received: August 23, 2018. Final revision: November 20, 2018. Accepted: December 12, 2018. Available online: ---.

No animal subjects were used in this study. Author Contributions: Manuscript no. 2018-1932.

1

Department of Ophthalmology, Seoul National University College of Medicine, Seoul, Korea.

2

Division of Glaucoma, Department of Ophthalmology, Seoul National University Hospital, Seoul, Korea.

3

Division of Neuro-Ophthalmology, Department of Ophthalmology, Seoul National University Hospital, Seoul, Korea.

4

Department of Ophthalmology, Kyungpook National University School of Medicine, Daegu, Korea. 5 Division of Glaucoma, Department of Ophthalmology, Kyungpook National University Hospital, Daegu, Korea. Presented at: the American Academy of Ophthalmology Annual Meeting, Chicago, Illinois, October 27e30, 2018. Financial Disclosure(s): The author(s) have no proprietary or commercial interest in any materials discussed in this article. Supported by the National Research Foundation of South Korea grant funded by the Korea government (MSIP; Ministry of Science, ICT & Future Planning) (No. 880-20180281). HUMAN SUBJECTS: Human subjects were included in this study. The human ethics committees at the Seoul National University Hospital approved the study. All research adhered to the tenets of the Declaration of Helsinki. All participants provided informed consent.

Conception and design: Y.K. Kim, S-J. Kim, Park Analysis and interpretation: J. Lee, Y.K. Kim, Ha, Y.W. Kim, Baek, H.J. Lee, D.W. Kim, Jeoung, S-J. Kim Data collection: J. Lee, Y.K. Kim, Baek, D.W. Kim, S-J. Kim, Park Obtained funding: Y. K. Kim, K.H. Park. Overall responsibility: J. Lee, Y.K. Kim, H.J. Lee, S-J. Kim Abbreviations and Acronyms: AUC ¼ area under receiver operating characteristic curve; CON ¼ compressive optic neuropathy; D ¼ diopters; GON ¼ glaucomatous optic neuropathy; HD-OCT ¼ high-definition OCT; HTG ¼ high-tension glaucoma; IOP ¼ intraocular pressure; MD ¼ mean deviation; mGCIPL ¼ macular ganglion celleinner plexiform layer; NAION ¼ nonarteritic anterior ischemic optic neuropathy; NGON ¼ nonglaucomatous optic neuropathy; NRR ¼ neuroretinal rim; ONH ¼ optic nerve head; PSD ¼ pattern standard deviation; RAPD ¼ relative afferent pupillary defect; RGC ¼ retinal ganglion cell; RNFL ¼ retinal nerve fiber layer. Correspondence: Young Kook Kim, MD, Department of Ophthalmology, Seoul National University Hospital, Seoul National University College of Medicine, 101 Daehak-ro, Jongno-gu, Seoul 03080, Korea. E-mail: [email protected].

9