Macular Ganglion Cell Imaging Study: Glaucoma Diagnostic Accuracy of Spectral-Domain Optical Coherence Tomography

Abstract

Purpose.: We evaluated the diagnostic accuracy of macular ganglion cell-inner plexiform layer (GCIPL) measurements using a high-definition optical coherence tomography (Cirrus HD-OCT) ganglion cell analysis algorithm for detecting early and moderate-to-severe glaucoma.

Methods.: Totals of 119 normal subjects and 306 glaucoma patients (164 patients with early glaucoma and 142 with moderate-to-severe glaucoma) were enrolled from the Macular Ganglion Cell Imaging Study. Macular GCIPL, peripapillary retinal nerve fiber layer (RNFL) thickness, and optic nerve head (ONH) parameters were measured in each subject. Areas under the receiver operating characteristic curves (AUROCs) were calculated and compared. Based on the internal normative database, the sensitivity and specificity for detecting early and moderate-to-severe glaucoma were calculated.

Results.: There was no statistically significant difference between the AUROCs for the best OCT parameters. For detecting early glaucoma, the sensitivity of the Cirrus GCIPL parameters ranged from 26.8% to 73.2% and that of the Cirrus RNFL parameters ranged from 6.1% to 61.6%. For the early glaucoma group, the best parameter from the GCIPL generally had a higher sensitivity than those of the RNFL and ONH parameters with comparable specificity (P < 0.05, McNemar's test).

Conclusions.: There were no significant differences between the AUROCs for Cirrus GCIPL, RNFL, and ONH parameters, indicating that these maps have similar diagnostic potentials for glaucoma. The minimum GCIPL showed better glaucoma diagnostic performance than the other parameters at comparable specificities. However, other GCIPL parameters showed performances comparable to those of the RNFL parameters.

Introduction

The essential pathologic process of glaucoma is the loss of retinal ganglion cells (RGCs) and their axons. Previous studies have demonstrated that considerable loss of RGCs can occur before visual field (VF) deficits are detected clinically. ^1–3 Therefore, direct assessment of the RGC loss is of vital importance for the diagnosis of glaucoma. 

During the past years, the value of assessing the macular inner retinal structure for diagnosing glaucoma has been the focus of research in many studies. Advances in ocular imaging technology, particularly the advent of optical coherence tomography (OCT), have enabled assessment of RGC axons by quantifying retinal nerve fiber layer (RNFL) damage. The recent introduction of spectral domain OCT has improved image resolution, imaging speed, and sensitivity. ^4–6 Moreover, recent advances in segmentation algorithms have enabled the quantitative assessment of individual retinal layers in the macular region. ^7–11 Since a significant proportion of the RGC population resides in the macula, measuring the macular ganglion cell complex (GCC), which includes the RNFL, ganglion cell layer (GCL), and inner plexiform layer (IPL), may be a promising target for noninvasive imaging to assess RGC loss clinically. To date, several studies have demonstrated the high diagnostic power of macular GCC and shown that it is comparable to that of RNFL analysis. ^12–15  

Cirrus HD-OCT (Carl Zeiss Meditec, Dublin, CA) is a commercially available OCT device that can measure macular ganglion cell-inner plexiform layer (GCIPL) thickness using a ganglion cell analysis (GCA) algorithm. Unlike GCC measurements, the GCA algorithm segments the GCIPL without including the RNFL. Therefore, the GCIPL is thought to be less influenced by RNFL thickness variation. Previous studies have demonstrated that the GCA algorithm by Cirrus HD-OCT can measure the macular GCIPL thickness with excellent intervisit reproducibility. ^16,17 Furthermore, a study that examined the performance of macular GCIPL parameters in early glaucoma showed promising results. ¹⁸ This study was designed to evaluate the diagnostic accuracy of macular GCIPL measurements by the Cirrus OCT GCA algorithm for detecting early and moderate-to-severe glaucoma. 

The Macular Ganglion Cell Imaging Study is an ongoing prospective study designed in 2011. Its primary objective was to evaluate quantitatively macular GCIPL and assess its diagnostic performance in various stages of glaucoma. Its secondary objectives included establishing a relationship between structural and functional parameters, elucidating the clinical role of macular imaging devices, and providing practical guidelines for the follow-up of glaucoma patients. 

Methods

This investigation is based on the Macular Ganglion Cell Imaging Study, an ongoing prospective study of glaucoma patients and healthy individuals at the Glaucoma Clinic of Seoul National University Hospital. The study followed the tenets of the Declaration of Helsinki, and was approved by the Institutional Review Board of the Seoul National University Hospital. In cases in which both eyes of a subject were eligible for the study, only one eye was chosen randomly for inclusion. Informed consent was obtained from all subjects. 

Study Subjects

Glaucoma patients and healthy control subjects were enrolled from the Glaucoma Clinic of Seoul National University Hospital. This study began enrollment in December 2011. All subjects underwent a complete ophthalmologic examination, including visual acuity, manifest refraction, axial length (Axis II PR; Quantel Medical, Inc., Bozeman, MT), IOP measurements using Goldmann applanation tonometry, slit-lamp examination, gonioscopy, dilated fundus examination with a 78 diopter (D) lens, color disc photography, red-free RNFL photography (TRC-50IX; Topcon Corporation, Tokyo, Japan), Swedish interactive thresholding algorithm (SITA) 30-2 perimetry (Humphrey field analyzer II; Carl Zeiss Meditec), and Cirrus HD-OCT. All examinations were conducted within a 1-month period. 

Inclusion criteria were ages between 20 and 79, best-corrected visual acuity of ≥20/40 in the study eye, refractive error within ±6.00 D equivalent sphere and ±3.00 D astigmatism, and no history or evidence of retinal (diabetic retinopathy, macular degeneration, retinal detachment, epiretinal membrane) or nonglaucomatous optic nerve diseases, treatment that might be toxic to the retina or optic nerve (e.g., chloroquine, ethambutol), laser therapy, or ocular surgery except noncomplicated cataract surgery. Eyes with consistently unreliable VFs (defined as false negative >33%, false positive >33%, and fixation losses >20%) were excluded from the study. 

Glaucomatous eyes were defined as those with a glaucomatous VF defect confirmed by 2 reliable VF examinations and by the presence of glaucomatous optic disc cupping irrespective of the level of IOP. Glaucomatous optic disc cupping was defined as neuroretinal rim thinning, notching, excavation, or RNFL defect with corresponding VF deficit. Color disc and red-free RNFL images were evaluated independently by 2 observers in a random order and masked fashion, without knowledge of the clinical information. The presence of glaucomatous optic disc cupping was determined by a consensus agreement between the 2 observers. 

The severity of glaucomatous damage was classified into early and moderate-to-severe according to the Hodapp-Parrish-Anderson criterion. ¹⁹ Healthy control eyes had an IOP of ≤21 mm Hg with no history of increased IOP, absence of glaucomatous disc appearance, no visible RNFL defect according to red-free RNFL photography, and a normal VF using standard automated perimetry. 

OCT Measurements

Imaging was obtained using the Cirrus HD-OCT. Two scans, including one macular scan centered on the fovea (macular cube 200 × 200 protocol) and one peripapillary RNFL scan centered on the optic disc (optic disc cube 200 × 200 protocol), were acquired through dilated pupils by the same operator on the same day. The macular GCIPL and peripapillary RNFL thicknesses, and optic nerve head (ONH) parameters were measured automatically using the internal ganglion cell, RNFL, and ONH analysis algorithms, respectively. 

The GCA algorithm was included in the 6.0 software version of Cirrus HD-OCT. It detects and measures the thickness of the macular GCIPL within a 6 × 6 × 2 mm elliptical annulus area centered on the fovea. The algorithm has been described previously in detail. ^16,17 In brief, the annulus has an inner vertical diameter of 1 mm, which was chosen to exclude the portions of the fovea where the layers are very thin and difficult to detect accurately, and an outer vertical diameter of 4 mm, which was chosen according to where the GCL again becomes thin and difficult to detect. The GCA algorithm identifies the outer boundaries of the RNFL and IPL. The difference between the RNFL and the IPL outer boundary segmentations yields the combined thickness of the RGC layer and IPL. 

The following GCIPL thickness measurements were analyzed: average, minimum, and sectoral (superonasal, superior, superotemporal, inferotemporal, inferior, and inferonasal). For peripapillary RNFL thickness measurements, average thickness; superior, inferior, temporal, and nasal quadrant thickness; and RNFL thickness in the 12 clock hour sectors were included in the analysis. For the ONH analysis, the following parameters were included: disc area, rim area, and cup-to-disc ratio. 

The software analyzes the values, compares them to the device's internal normative database, and generates a thickness map, deviation map, and significance map color-coded to match RNFL thickness, with values within the normal range in green (P = 5%–95%), borderline values in yellow (P = 1%–5%), and values outside the normal range in red (P < 1%). Subjects with scans showing algorithm segmentation failure, signal strength <6, or artifacts due to eye movements or blinking were excluded from the study. 

Quality assessments of the Cirrus GCIPL scans were performed by using the criteria of Ishikawa et al. ⁷ In detail, segmentation failures were defined as obvious disruption of the detected border, and/or border wandering (detected border jumping to and from different anatomic structures), within >5% consecutively (i.e., an uninterrupted error) or 20% cumulatively (i.e., adding up all errors amounted to 20% of the image width) of the entire image. 

Visual Field Testing

The SITA standard 30-2 of the Humphrey Field Analyzer II 750 was performed. Glaucomatous VF loss was defined as the consistent presence of a cluster of 3 or more nonedge points on the pattern deviation plot with a probability of occurring in <5% of the normal population, with 1 of these points having the probability of occurring in <1% of the normal population; a pattern standard deviation (PSD) with P < 5%; or a glaucoma hemifield test result outside normal limits. VF defects had to be repeatable on at least 2 consecutive tests. 

Data Analysis

All analyses were performed using SPSS for Windows Version 19.0 (SPSS, Inc., Chicago, IL) and MedCalc Version 10.0 (MedCalc Software, Ostend, Belgium). P values <0.05 were considered statistically significant. 

Statistical Analysis Based on the Quantitative OCT Parameters.

The average values of GCIPL, RNFL, and ONL parameters were compared between normal and glaucomatous eyes using Student's t-test. We used receiver operating characteristic (ROC) curves to describe the diagnostic ability of OCT parameters to differentiate each stage of glaucoma from normal eyes. An area under the ROC curve (AUROC) of 1.0 represents perfect discrimination, whereas an AUROC of 0.5 represents chance discrimination. The method described by DeLong et al. ²⁰ was used to compare AUROCs. 

Statistical Analysis Based on the Internal Normative Database.

Sensitivity and specificity of the Cirrus GCIPL, RNFL, and ONH parameters were tested by comparison with the internal normative database. McNemar's test was used to compare the sensitivity and specificity for detecting glaucoma between the different OCT parameters. 

We performed an analysis of whether a selective combination of OCT parameters could enhance the glaucoma diagnostic accuracy. The RNFL and GCIPL diagnostic parameters were combined using or-logic or and-logic approaches. We grouped the parameter on two different levels: (1) overall (average RNFL, average GCIPL, and minimum GCIPL), and (2) RNFL quadrants and GCIPL sectors. The RNFL quadrants and GCIPL sectors were grouped according to their corresponding areas (e.g., superior RNFL and/or superior GCIPL, inferior RNFL and/or inferior GCIPL). Combinations within each group were evaluated for diagnostic accuracy. Glaucoma diagnosis based on the or-logic or and-logic approaches have been described previously in detail. ²¹ In brief, for the or-logic combination, glaucoma diagnosis is made if any of the component parameters were abnormal according to the internal normative database. In the and-logic combination, all of the component parameters must be judged abnormal for the combination to be classified as abnormal. 

Results

During the enrollment period, this study initially included 545 eyes of 545 subjects (408 eyes with glaucoma and 137 normal control eyes). Eyes were excluded because of diabetic retinopathy (n = 36), macular degeneration (n = 28), epiretinal membrane (n = 20), and ocular surgery history (n = 8). We excluded from the analysis 17 eyes (GCA, n = 9; RNFL, n = 8) in which an erroneous GCA or RNFL profile of 0.0 μm was computed by poor delineation. Of the remaining 436 eyes, 11 (2.5%) with unacceptable Cirrus OCT scans were excluded from further analysis. Therefore, the final study sample included 425 eyes of 425 subjects (306 eyes with glaucoma and 119 normal control eyes). According to the Hodapp-Parrish-Anderson classification, 306 eyes with glaucoma were classified as having early (n = 164) and moderate-to-severe (n = 142) glaucoma. 

Subject Characteristics

The baseline demographics are detailed in Table 1. There was no significant difference in the mean age between the normal control subjects (57.1 ± 12.3 years), and those with early (58.7 ± 10.2) and moderate-to-severe (59.2 ± 13.1) glaucoma. The intraocular pressure without medication, spherical equivalent, and axial length were similar between the groups. As expected, the mean deviation (MD) of VF and PSD were significantly different between control and glaucomatous eyes at various stages. 

Table 1

 

Demographic Characteristics of Study Subjects

	Normal Control, n = 119	Early Glaucoma, n = 164		Moderate-to-Severe Glaucoma, n = 142		P ‡
	Mean ± SD	Mean ± SD	P *	Mean ± SD	P †
Age, y	57.1 ± 12.3	58.7 ± 10.2	0.27	59.2 ± 13.1	0.20	0.71
Female, n (%)	66 (55.5)	86 (52.4)	0.62	60 (42.3)	0.03	0.08
Visual acuity, logMAR	0.06 ± 0.11	0.06 ± 0.12	0.73	0.09 ± 0.11	0.02	0.04
Baseline IOP, mm Hg	13.4 ± 2.6	13.0 ± 2.3	0.20	12.9 ± 2.6	0.16	0.77
SE, D	−0.73 ± 2.04	−0.78 ± 2.06	0.87	−0.92 ± 2.09	0.56	0.62
Axial length, mm	23.8 ± 1.4	24.2 ± 1.1	0.13	24.2 ± 1.3	0.09	0.73
Humphrey C30-2 visual field
MD, dB	−0.22 ± 1.73	−2.68 ± 1.76	<0.01	−12.41 ± 5.92	<0.01	<0.01
PSD, dB	1.97 ± 0.82	5.47 ± 2.80	<0.01	12.20 ± 3.16	<0.01	<0.01

 LogMAR, logarithm of the minimum angle of resolution; SE, spherical equivalent.

*  Values for comparing normal and early glaucoma group.

†  Values for comparing normal and moderate-to-severe glaucoma group.

‡  Values for comparing early and moderate-to-severe glaucoma group.

OCT Measurements

The RNFL, GCA, and ONH parameters computed in the control and glaucoma groups are presented in Table 2. As expected, the mean RNFL thickness was highest in the control group and decreased as glaucoma severity increased (normal, 93.5 μm; early, 76.7 μm; and moderate-to-severe, 67.7 μm). Average GCIPL thickness followed the same pattern (normal, 80.4 μm; early, 72.0 μm; and moderate-to-severe, 65.5 μm). Statistically significant differences were detected in the OCT RNFL parameters of the 1, 2, 5, 6, 7, 10, 11, and 12 o'clock sectors, all four quadrants, and average RNFL thickness. Likewise, for the OCT GCIPL parameters, statistically significant differences were detected in all GCIPL parameters (P < 0.01). For the OCT ONH parameters, there were statistically significant differences in rim area, cup area, and cup-to-disc ratio (P < 0.05). 

Table 2

 

Peripapillary RNFL Thickness, GCIPL Thickness, and ONH Parameters Obtained Using Cirrus OCT

	Normal Control, n = 119 Mean (SD)	Early Glaucoma, n = 164		Moderate-to-Severe Glaucoma, n = 142		P Value‡
		Mean (SD)	P Value*	Mean (SD)	P Value†
Main RNFL thickness parameters, μm
Average	93.5 (8.6)	76.7 (10.5)	<0.01	67.7 (11.0)	<0.01	<0.01
Superior	118.6 (15.3)	96.7 (19.1)	<0.01	82.6 (20.3)	<0.01	<0.01
Nasal	67.3 (8.3)	62.1 (9.0)	<0.01	60.0 (9.1)	<0.01	0.02
Inferior	119.5 (15.3)	86.7 (21.2)	<0.01	71.4 (19.8)	<0.01	<0.01
Temporal	68.5 (12.2)	60.8 (12.0)	<0.01	56.6 (14.0)	<0.01	<0.01
RNFL clock hours, μm
12 superior	118.9 (24.5)	99.4 (25.4)	<0.01	85.2 (24.5)	<0.01	<0.01
1	111.0 (22.4)	92.2 (21.8)	<0.01	81.5 (22.8)	<0.01	<0.01
2	79.7 (14.6)	71.7 (13.5)	<0.01	66.9 (14.4)	<0.01	<0.01
3 nasal	57.0 (8.2)	55.2 (9.2)	0.07	54.6 (11.3)	0.05	0.65
4	64.5 (11.8)	58.6 (10.5)	<0.01	56.1 (9.9)	<0.01	0.03
5	101.8 (23.1)	80.8 (17.3)	<0.01	70.8 (18.9)	<0.01	<0.01
6 inferior	130.1 (21.8)	94.4 (25.9)	<0.01	72.9 (24.6)	<0.01	<0.01
7	128.3 (25.0)	87.2 (32.4)	<0.01	70.0 (26.2)	<0.01	<0.01
8	70.1 (17.2)	61.1 (14.9)	<0.01	57.6 (15.7)	<0.01	0.05
9 temporal	55.3 (8.5)	52.2 (11.3)	<0.01	53.0 (13.8)	0.11	0.57
10	80.1 (14.6)	68.4 (18.7)	<0.01	61.9 (20.1)	<0.01	0.03
11	123.2 (28.5)	98.2 (28.4)	<0.01	81.4 (30.2)	<0.01	<0.01
GCIPL thickness parameters, μm
SN	82.1 (6.9)	78.0 (10.2)	<0.01	72.7 (13.2)	<0.01	<0.01
S	81.3 (7.2)	74.4 (10.4)	<0.01	68.4 (15.0)	<0.01	<0.01
ST	80.1 (7.3)	71.8 (9.3)	<0.01	64.0 (13.2)	<0.01	<0.01
IT	80.2 (8.0)	66.2 (11.0)	<0.01	58.8 (12.9)	<0.01	<0.01
I	78.5 (7.4)	67.5 (10.2)	<0.01	61.0 (11.3)	<0.01	<0.01
IN	80.2 (6.7)	74.0 (9.0)	<0.01	67.7 (11.0)	<0.01	<0.01
Average	80.4 (6.7)	72.0 (7.4)	<0.01	65.5 (10.0)	<0.01	<0.01
Min	77.2 (8.6)	61.2 (9.9)	<0.01	53.0 (10.6)	<0.01	<0.01
Optic nerve head parameters
Rim area	1.12 (0.22)	0.84 (0.19)	<0.01	0.67 (0.20)	<0.01	<0.01
Cup area	0.35 (0.22)	0.49 (0.28)	<0.01	0.58 (0.41)	<0.01	0.02
Cup/Disc ratio	0.67 (0.08)	0.74 (0.10)	<0.01	0.79 (0.83)	<0.01	<0.01

 I, inferior; IN, inferonasal; IT, inferotemporal; Min, minimum; S, superior; SN, superoasal; ST, superotemporal.

*  Value for comparing normal and early glaucoma group.

†  Value for comparing normal and moderate-to-severe glaucoma group.

‡  Value for comparing early and moderate-to-severe glaucoma group.

Relationship Between Perimetry Global Indices and RNFL/GCA Parameters

The correlation between the perimetry global indices (MD and PSD) and various RNFL/GCIPL parameters was evaluated using Pearson's correlation coefficient (Table 3). The MD on Humphrey field analysis was found to have a significant positive correlation with average and all 4 quadrants (P < 0.05). Likewise, PSD showed a significant negative correlation with average, superior, inferior, and temporal RNFL thickness (P < 0.001). All GCIPL parameters showed a significant positive correlation with MD and a significant negative correlation with PSD (P < 0.001 and P < 0.01, respectively). 

Table 3

 

Pearson's Correlation Coefficient for MD and PSD Assessed on a Humphrey Visual Field Analyzer with RNFL Parameters on OCT

	MD			PSD
	R	R 2	P	R	R 2	P
RNFL thickness
Average	0.505	0.255	<0.001	−0.348	0.121	<0.001
Superior	0.430	0.185	<0.001	−0.227	0.052	<0.001
Nasal	0.173	0.030	0.002	−0.100	0.010	0.081
Inferior	0.456	0.208	<0.001	−0.383	0.147	<0.001
Temporal	0.192	0.037	0.001	−0.136	0.018	0.017
GCIPL thickness
SN	0.334	0.112	<0.001	−0.170	0.029	0.003
S	0.337	0.114	<0.001	−0.155	0.024	0.007
ST	0.456	0.208	<0.001	−0.258	0.067	<0.001
IT	0.390	0.152	<0.001	−0.358	0.128	<0.001
I	0.350	0.123	<0.001	−0.326	0.106	<0.001
IN	0.403	0.162	<0.001	−0.278	0.077	<0.001
Average	0.473	0.224	<0.001	−0.325	0.106	<0.001
Min	0.433	0.187	<0.001	−0.405	0.164	<0.001

ROC Analysis

Table 4 shows the AUROCs for all OCT parameters. Of the OCT parameters, the average RNFL (AUROC = 0.897), minimum GCIPL (AUROC = 0.902), and rim area (AUROC = 0.855) were best able to discriminate between early glaucoma and normal eyes. Average and inferior RNFL (AUROC = 0.958), minimum GCIPL (AUROC = 0.960), and rim area (AUROC = 0.943) were the most accurate parameters for the diagnosis of moderate-to-severe glaucoma. 

Table 4

 

AUROC Values Between Normal and Glaucomatous Eyes

	Early Glaucoma, n = 164 Mean (SD)	Moderate-to-Severe Glaucoma, n = 142 Mean (SD)	P Value
Main RNFL thickness parameters
Average	0.897 (0.019)	0.958 (0.012)	<0.01
Superior	0.815 (0.025)	0.915 (0.017)	<0.01
Nasal	0.664 (0.032)	0.730 (0.031)	0.14
Inferior	0.890 (0.019)	0.958 (0.013)	<0.01
Temporal	0.673 (0.032)	0.771 (0.029)	0.02
RNFL clock hours
12 superior	0.710 (0.030)	0.830 (0.024)	<0.01
1	0.726 (0.030)	0.818 (0.026)	0.02
2	0.660 (0.032)	0.751 (0.030)	0.04
3 nasal	0.577 (0.035)	0.591 (0.035)	0.49
4	0.645 (0.032)	0.715 (0.031)	0.12
5	0.778 (0.027)	0.860 (0.022)	0.02
6 inferior	0.849 (0.023)	0.942 (0.015)	<0.01
7	0.830 (0.024)	0.933 (0.016)	<0.01
8	0.657 (0.032)	0.736 (0.031)	0.08
9 temporal	0.599 (0.033)	0.598 (0.035)	0.98
10	0.685 (0.031)	0.799 (0.028)	<0.01
11	0.743 (0.029)	0.847 (0.024)	<0.01
GCIPL thickness parameters
SN	0.619 (0.033)	0.741 (0.031)	<0.01
S	0.714 (0.030)	0.816 (0.027)	0.01
ST	0.780 (0.027)	0.889 (0.021)	<0.01
IT	0.857 (0.022)	0.938 (0.016)	<0.01
I	0.817 (0.027)	0.902 (0.020)	0.01
IN	0.705 (0.031)	0.837 (0.026)	<0.01
Average	0.817 (0.025)	0.906 (0.020)	<0.01
Min	0.902 (0.018)	0.960 (0.012)	<0.01
Optic nerve head parameters
Rim area	0.855 (0.023)	0.943 (0.016)	<0.01
Cup area	0.650 (0.033)	0.712 (0.031)	0.17
Cup/disc ratio	0.721 (0.032)	0.858 (0.035)	<0.01

The Figure plots the ROC curves of the best GCIPL, RNFL, and ONH parameters in the early and moderate-to-severe glaucoma groups. In the early glaucoma groups, the diagnostic value of minimum GCIPL thickness appeared to be greater than that of average RNFL thickness, but the difference was not significant (P = 0.80). However, the AUROC of minimum GCIPL thickness was significantly greater than that of OCT RNFL clock hours (P = 0.03) and that of rim area (P = 0.04). In the moderate-to-severe glaucoma group, no significant difference was found between the AUROCs for the best OCT parameters. 

Figure

 

ROC curves of the best GCIPL, RNFL, and ONH parameters in the early (A) and moderate-to-severe (B) glaucoma groups.

Sensitivity and Specificity Based on the Internal Normative Database

Table 5 presents the sensitivity and specificity for the overall OCT parameters. The severity of glaucoma had a significant influence on the sensitivity of Cirrus OCT parameters. In the early glaucoma group, using a criterion of abnormal at the <5% level, the sensitivity of the various RNFL parameters ranged from 6.1% to 61.6%, and that of the GCIPL parameters ranged from 26.8% to 73.2%. In the moderate-to-severe glaucoma group, the sensitivity of the RNFL parameters ranged from 5.6% to 86.6%, and that of the GCIPL parameters ranged from 53.5% to 90.8%. Of all the severity grades of glaucoma, the minimum GCIPL (sensitivity of 73.2% in the early and 90.8% in the moderate-to-severe group) and inferior RNFL (sensitivity of 61.6% in the early and 86.6% in the moderate-to-severe group) had the highest sensitivity among the GCIPL and RNFL parameters with comparable specificities (88.2% specificity for GCIPL and 94.6% specificity for RNFL). Of the ONH parameters, the highest sensitivity was yielded with the rim area (sensitivity of 61.0% in the early and 80.5% in the moderate-to-severe group). In the early glaucoma group, the best parameter from GCIPL had generally higher sensitivity than those from RNFL and ONH parameters with comparable specificity (P < 0.05, McNemar's test). 

Table 5

 

Sensitivity and Specificity of OCT RNFL Thickness, GCIPL Thickness, and ONH Parameters for Diagnosis of Glaucoma

	Normal versus Early		Normal versus Moderate-to-Severe
	Sensitivity	Specificity	Sensitivity	Specificity
Main RNFL thickness parameters
Average	50.0	96.6	83.1	96.6
Superior	43.3	97.5	68.3	97.5
Nasal	12.8	99.2	19.7	99.2
Inferior	61.6	94.6	86.6	94.6
Temporal	18.3	99.2	23.9	99.2
RNFL clock hours
12 superior	17.7	97.5	36.6	97.5
1	25.0	91.6	39.4	91.6
2	12.8	96.6	23.9	96.6
3 nasal	6.1	99.2	5.6	99.2
4	7.3	98.3	12.0	98.3
5	18.9	96.6	43.0	96.6
6 inferior	44.5	94.6	79.6	94.6
7	56.1	98.3	77.5	98.3
8	14.0	99.2	20.4	99.2
9 temporal	11.0	100.0	13.4	100.0
10	23.2	99.2	38.7	99.2
11	36.0	97.5	64.8	97.5
GCIPL thickness parameters
SN	26.8	94.1	58.5	94.1
S	37.8	92.4	53.5	92.4
ST	47.0	86.6	79.6	86.6
IT	64.6	91.6	81.7	91.6
I	57.9	90.8	81.7	90.8
IN	34.8	89.9	69.0	89.9
Average	50.6	89.9	77.5	89.9
Min	73.2	88.2	90.8	88.2
Optic nerve head parameters
Rim area	61.0	86.6	80.5	86.6
Cup area	48.2	84.0	61.0	84.0
Cup/disc ratio	57.9	84.9	78.0	84.9

Table 6 demonstrates the diagnostic performances of combined RNFL and GCIPL parameters for glaucoma detection. The or-logic combination of average RNFL and minimum GCIPL had the highest sensitivity (sensitivity of 81.1% for the early and 95.8% for the moderate-to-severe groups) among the combined parameters with comparable specificities (specificity of 97.5% for both groups). The or-logic combination of average RNFL and minimum GCIPL had higher sensitivity than the best single RNFL and GCIPL parameters (P < 0.05, McNemar's test). 

Table 6

 

Sensitivity and Specificity of Combined Parameters of OCT RNFL and GCIPL Thickness for Diagnosis of Glaucoma

	Normal versus Early*		Normal versus Moderate-to-Severe*
	Sensitivity	Specificity	Sensitivity	Specificity
Avg RNFL or avg GCIPL	64.0	97.5	90.1	97.5
Avg RNFL and avg GCIPL	37.8	89.1	70.4	89.1
Avg RNFL or minimum GCIPL	81.1	97.5	95.8	97.5
Avg RNFL and minimum GCIPL	42.1	88.2	78.2	88.2
Sup RNFL or sup GCIPL (SN, S, ST)	62.8	100.0	87.3	100.0
Sup RNFL and sup GCIPL (SN, S, ST)	15.9	84.9	38.7	84.9
Inf RNFL or inf GCIPL (IN, I, IT)	79.3	98.3	95.1	98.3
Inf RNFL and inf GCIPL (IN, I, IT)	24.4	83.2	59.2	83.2

 Avg, average; Inf, inferior; Sup, superior.

*  All combined parameters were evaluated at the 5% level.

Discussion

Our study was designed with the main objective of evaluating the diagnostic accuracy of macular GCIPL, peripapillary RNFL thickness, and ONH parameters by Cirrus HD-OCT for detecting early and moderate-to-severe glaucoma. Based on the ROC analysis, we confirmed that the macular GCIPL and RNFL thickness parameters had similar diagnostic values for glaucoma detection. Of the GCIPL parameters, minimum GCIPL was the most accurate for the diagnosis of early and moderate-to-severe glaucoma. 

Our results are consistent with previous studies on the performance of macular OCT GCIPL parameters for detecting glaucoma. Mwanza et al. demonstrated that the ability of macular GCIPL parameters to discriminate between normal eyes and eyes with early glaucoma is excellent. ¹⁸ They reported that the GCIPL parameters with the best AUROCs in early glaucoma were the minimum (AUROC = 0.959), inferotemporal (AUROC = 0.956), and average (AUROC = 0.935) GCIPL. Takayama et al. reported that minimum GCIPL thickness AUROC (0.896) was significantly higher than average GCIPL thickness (0.821) in early glaucoma. ²² In our study, the minimum GCIPL had the highest AUROCs for detecting early and moderate-to-severe glaucoma (AUROC = 0.902 and 0.960, respectively). 

The minimum GCIPL was designed to be sensitive to focal RGC loss, which was the best parameter for accurately detecting early glaucoma in our study. Our results are similar to those of recent studies. ^18,22 Also, the minimum GCIPL was the most sensitive parameter in moderate-to-severe glaucoma. To explain this, we proposed that the glaucoma progression occurred focally and there are regional differences in vulnerability to glaucomatous damages, and this had less effect on average and sector GCIPL. Averaging the thickness values obtained in the average or sector parameters may mask the regional variation of GCIPL. Since the minimum GCIPL does not depend on averaging the thickness of the area, this parameter can be completely free of the effects of averaging. ^22,23  

We tested the sensitivity and specificity of the various OCT parameters for the detection of glaucomatous change. For detecting early glaucoma, we observed that the macular GCIPL parameters had generally higher sensitivities (26.8%–73.2%) than the OCT RNFL parameters (6.1%–61.6%) at comparable specificities. Our results suggested that macular GCIPL thickness may be a better diagnostic marker for early glaucoma than OCT RNFL thickness, although the AUROC difference was not significant. We speculated that macular GCIPL parameters have a theoretic advantage over OCT RNFL parameters, because early RGC loss typically gives rise to isolated damage in the paracentral areas. Moreover, macular GCIPL topography is less variable among normal individuals than other diagnostically important structures, such as the optic disc and RNFL, ²⁴ which may result in a superior diagnostic accuracy of macular GCIPL parameters in the early stage of glaucoma. However, direct comparison between OCT GCIPL and RNFL parameters may be limited because the OCT RNFL map does not include any “minimum” parameter corresponding to the minimum GCIPL. 

In our study, we adapted combined parameters using or-logic and and-logic. The or-logic combined parameter of average RNFL and minimum GCIPL showed higher sensitivity than the single best parameters in both groups. Using combinations of RNFL parameters from Stratus OCT, Budenz et al. suggested that the best combinations of sensitivity and specificity seem to be any quadrants abnormal or clock hours abnormal at the 5% level. ²⁵ Lu et al. demonstrated that the highest AUROC was achieved by the or-logic combination of overall, inferior, and superior quadrant RNFL thicknesses. ²¹ However, previous studies focused mainly on the RNFL parameters and were conducted mostly with time-domain OCT. ^21,25 Our results suggested that the combined application of GCIPL and RNFL parameters can improve the diagnostic precision of the OCT examination. 

In our study, we should note that the inferotemporal sector had a high AUROC (AUROC = 0.938) for moderate-to-severe glaucoma. Our results corresponded well with those of earlier studies that reported that the inferotemporal area in the macula is the most susceptible to glaucomatous changes. ^12,26–28 The papillomacular bundle usually is spared until the advanced stage, indicating that a greater number of RGCs in the nasal macular area are less susceptible to glaucomatous damage. ²² The difference in susceptibility may result in different degrees of thinning in the GCIPL, which could lead to differences in diagnostic performance among the various GCIPL parameters. 

In conclusion, there were no significant differences between the AUROCs for Cirrus GCIPL, RNFL, and ONL parameters, indicating that these maps have similar diagnostic potentials for glaucoma. The minimum GCIPL showed better glaucoma diagnostic performance than the other parameters at comparable specificities. However, other GCIPL parameters showed performances comparable to those of the RNFL parameters. Further investigation may be necessary to develop new parameters with increased diagnostic performance for the detection of early glaucomatous damage and its progression over time. 

Acknowledgments

Supported by Grant No. A121615 from the Korea Health technology R&D Project, Ministry of Health & Welfare, Republic of Korea, and by Grant No. 2009-0091931 from the National Research Foundation of Korea (NRF) funded by the Korea government (MEST). The authors alone are responsible for the content and writing of the paper. 

Disclosure: J.W. Jeoung, None; Y.J. Choi, None; K.H. Park, None; D.M. Kim, None 

Footnotes

 JWJ and YJC contributed equally to the work presented here and should therefore be regarded as equivalent authors.