The present study included consecutive OAG patients who visited the Glaucoma Clinic of Seoul National University Hospital (Seoul, Korea) from October 2011 to April 2014. This study adhered to the tenets of the Declaration of Helsinki and the study was approved by the institutional review board of the Seoul National University Hospital. 

Patients underwent a comprehensive ophthalmologic examination, including measurement of best-corrected visual acuity, IOP measurement by Goldmann applanation tonometry, spherical equivalent, slit-lamp examination, and gonioscopic examination. They were also imaged by color stereo-disc photography, red-free RNFL photography (Vx-10; Kowa Optimed, Tokyo, Japan), SD-OCT (Cirrus-HD; Carl Zeiss Meditec, Inc., Dublin, CA, USA), and standard automated perimetry (SAP) with the 30-2 Swedish interactive threshold algorithm (Humphrey Field Analyzer II; Carl Zeiss Meditec, Inc.). 

Open-angle glaucoma patients who had (1) undergone three serial SD-OCT scans at 6-month intervals without structural or functional deterioration during the study period, and (2) a localized RNFL defect in the superior or inferior hemifield with concurrent GCIPL defect on respective RNFL and GCIPL deviation maps were included in the present study. The eligibility criteria included best-corrected visual acuity better than 20/40, spherical equivalent within ± 5.0 diopters (D), cylindrical correction of less than 3.0 D, open anterior chamber angle on initial gonioscopic and slit-lamp examinations, glaucomatous optic disc appearance (i.e., increased cup-to-disc ratio, neuroretinal rim thinning, or notching), and glaucomatous visual field defect. Glaucomatous visual field defect was defined as presence of a cluster of three or more nonedge points with P less than 5% probability of being normal, one with P less than 1% on pattern deviation plot; pattern standard deviation with P less than 5%; or glaucoma hemifield test outside the normal limits, as confirmed on two consecutive tests within a month. Patients with high-quality stereo disc and RNFL photographs with well-focused and evenly illuminated images as well as reliable visual field testing results (fixation loss < 20%, false-positive errors < 15%, and false-negative errors < 15%) were included. For cases in which both eyes were eligible, one eye was randomly chosen prior to the analysis. 

Two glaucoma specialists (KEK and JWJ) masked to all of the patients' clinical information evaluated the disc and RNFL photographs in an independent manner. Sixty degree–wide-angle views of the optic disc focused on the retina using a built-in split-line focusing device and centered between the fovea and the optic disc were reviewed on a liquid crystal display (LCD) monitor (FLATRON; LG Display Co., Ltd., Seoul, Korea).¹³ A localized RNFL defect was defined as a dark wedge-shaped defect, its tip touching the optic disc border and its width larger than that of the major retinal vessels at a 1–disc diameter distance from the edge of the optic disc in the brightly striated pattern of the surrounding RNFL.¹⁴ The two observers classified glaucomatous eyes into one of the following categories: localized defect, diffuse atrophy, or ambiguous. Only cases classified independently by both observers as having a localized defect were included. 

The exclusion criteria were as follows: any history of disease that might cause nonglaucomatous optic neuropathy or RNFL damage, any macular disease, diseases other than glaucoma that can affect the visual field, neurologic disease, and intraocular surgery other than simple cataract extraction (but not within the study period). 

For the purpose of the longitudinal reproducibility, clinically stable OAG patients meeting the eligibility criteria were included in this study. Clinical stability was determined on the basis of serial stereo optic disc and RNFL photographs and VF progression analysis assessments by the independent masked observers (KEK and JWJ). The presence of a glaucomatous disc and/or RNFL change or functional change was confirmed by consensus between the two observers; eyes thereupon were classified into one of the following categories: nonprogressor, progressor, or ambiguous. 

Glaucomatous structural progression included (1) progressive optic disc changes, such as focal or diffuse narrowing or notching of the neuroretinal rim, increased cup-to-disc ratio, adjacent vasculature position shift, or optic disc hemorrhage; and (2) progressive changes in an RNFL defect,¹⁵ including an appearance of a new defect or an increase in the width or depth of the RNFL defect. The method determining RNFL change has already been described in detail elsewhere.^15,16 Widening was defined as a recognizable increase in the distance between the border of the RNFL defect and the adjacent vessels. Retinal nerve fiber layer defect depth was determined after equalizing the RNFL brightness and contrast at the posterior pole between two images on an LCD monitor. Deepening was defined as an increase in the severity of the RNFL defect, determined using the method by Quigley et al.,¹⁶ or sharpness of the temporal and nasal borders of the RNFL defect. 

Functional progression was based on serial evaluation of visual fields using “event-based” and “trend-based” method. Event-based analysis determines whether significant visual field progression has occurred. This method is used by the commercially available Guided Progression Analysis (GPA) software form the Humphrey Field Analyzer (Carl Zeiss Meditec, Inc.). Event-based progression by GPA was defined as a significant decrease from baseline (two examinations) pattern deviation at three or more of the same test points on three consecutive tests, which is classified as “possible progression” or “likely progression.” Trend-based analysis provides estimates of rates of visual field progression based on linear regression analysis of visual field index. In this method, visual field progression was defined as a rate showing significantly negative slope (P < 0.05).¹⁷ Data from at least five visual fields performed during the OCT scan period plus 1 year before and after the scan period were analyzed. Patients were categorized as progressors if any evidence of structural or functional glaucomatous change was identified. Only patients who were confirmed by both observers as nonprogressors were included in the analysis. 

All of the SD-OCT images were obtained by a single, well-trained technician using the Cirrus-HD OCT device (software version 6.0), the details of which have been described previously.^8,11 Subjects with high-quality OCT images of signal strength greater than or equal to 8 and without motion artifacts, involuntary saccade, overt misalignment of decenteration, or algorithm segmentation failure were included in the study. 

One macular scan centered on the fovea (macular 200 × 200 cube protocol) and one peripapillary scan centered on the optic disc (optic disc 200 × 200 cube protocol) were acquired though dilated pupils. Detailed information on the GCA and RNFL algorithm is available elsewhere.^2,11 The peripapillary scan allowed measurement of peripapillary RNFL thickness including the average, four quadrant sectors, and 12 clock-hour sectors. The macular scan allowed measurement of macular GCIPL thickness including the average, minimum, and six (superotemporal, superior, superonasal, inferonasal, inferior, and inferotemporal) sectors. The right-eye orientation was used in documenting all of the OCT data; the left-eye images were all converted to the right-eye format for analysis. For both the right and left eyes, 12 o'clock corresponded to the superior region, 3 o'clock, the nasal region, 6 o'clock, the inferior region, and 9 o'clock, the temporal region. 

For the GCIPL deviation map, the software compares the measurements with an age-matched normative database and presents the probability of their being within the normal range using the three-level color-coding system: yellow and red to indicate areas (superpixels) in which GCIPL thickness is less than 5% and less than 1% of the normal-distribution percentiles, respectively, and no color to indicate GCIPL thickness within the normal range. 

The main GCIPL defect on the GCIPL deviation map was arbitrarily defined as at least 10 contiguous color-coded superpixels in an area and more than a boundary of one superpixel from the inner annulus.¹⁸ If the largest area was less than 10 superpixels, the image was excluded to discard the artifact. Figure 1 presents how the angular width and area of GCIPL defect on the GCIPL deviation map was assessed. The reference line was considered to be the horizontal line passing through the center of the inner and outer annulus on the GCIPL deviation map. The temporal point of the line was set as 0°, and the angle was assessed in a clockwise direction for the right-eye format: the superior point set as 90°, the nasal point as 180°, and the inferior point as 270°. Two yellow lines were drawn by a blind observer (BWY) from the center to contain the outermost limits of the yellow or red color-coded GCIPL defects. According to the angle between those two lines, the angular width of the GCIPL defect was assessed using the computer program written in MATLAB R2011a (The MathWorks, Inc., Natick, MA, USA). The area of the GCIPL defect was assessed by yellow or red color-coded pixel counting, also using the computer program written in MATLAB R2011a. 

The interobserver agreement in the discrimination of nonprogressors from progressors was assessed by Kappa (κ) statistics. The strength of agreement was categorized according to the method of Landis and Koch¹⁹: 0 = poor, 0 to 0.20 = slight, 0.21 to 0.40 = fair, 0.41 to 0.60 = moderate, 0.61 to 0.80 = substantial, and 0.81 to 1.00 = almost perfect. The intersession RNFL and GCIPL thickness measurements as well as the GCIPL defect's angular width and areal measurements were compared by one-way ANOVA. The long-term reproducibility of macular GCA was assessed by the standard error of measurement, which is a measure of the within-subject variation or SD of values that would have been obtained from a single individual if tested multiple times.²⁰ This, the within-subject standard deviation (S_w), is the same as the test–retest standard deviation (TRTSD), and is calculated as the square root of the within-subject mean square for error.²¹ Additionally, in the present evaluation, the intraclass correlation coefficient (ICC), coefficient of variation (CV), and tolerance limit were determined to assess the reproducibility. The ICC, the ratio of the subject variance to the total variance (within-subject variance + between-subject variance), summarizes the reproducibility of a measurement process for a given group of subjects.¹ The ICC ranges from theoretical 0 to 1: a higher ICC (close to 1) value indicates smaller fluctuations among repeated measurements for the same subjects. The ICC cutoff value of 0.70 is considered acceptable, but greater values mean better reproducibility. The CV (%) is calculated as 100 × S_w /overall mean, and a result less than 10% is considered to represent good reproducibility. The tolerance limit was defined as 1.96 × Display Formula × intervisit S_w.²¹ The Mann-Whitney U test was run to compare the CVs and TRTSDs between the subject groups. To assess any relation of the present reproducibility with the degree of glaucomatous damage, the associations between severity of glaucomatous damage (represented by visual field mean deviation [MD], mean average GCIPL thickness, and mean average RNFL thickness), and TRTSD of GCIPL and RNFL maps were examined by linear regression analysis. The correlation coefficient (r²) was used to demonstrate the strength of association between variables. In addition, a logistic regression analysis was performed for the same purpose after classifying patients into low and high thickness reproducibility groups according to the median TRTSD values of the average GCIPL and RNFL thicknesses. The statistical analyses were performed using SPSS version 21.0 for Windows (IBM Corp., Armonk, NY, USA). Statistical significance was defined as P less than 0.05.  

The present study initially involved 164 eyes of 164 patients, each with a localized RNFL defect and a concurrent GCIPL defect on three qualifying serial SD-OCT images. After masked evaluation by the two observers, 40 patients showing progression were excluded, as were 15 with interobserver disagreement, leaving a final sample of 109 eyes of 109 patients. The interobserver agreement on the presence of any structural or functional progression was excellent (κ = 0.812). 

The baseline characteristics of included patients are summarized in Table 1. The average OCT scan intervals were 6.2 ± 0.2 months, and the average follow-up period for confirmation of disease stability was 2.4 ± 0.3 years. The mean age was 56.2 ± 12.9 years, and the mean baseline MD was −4.90 ± 3.70 dB. The mean average RNFL thickness and mean average GCIPL thickness were 75.9 ± 10.5 and 71.1 ± 8.1 μm, respectively. 

No significant intersession difference was found in GCIPL thickness parameters as well as GCIPL defects on deviation map (all P > 0.05). The ICCs of the GCIPL thickness parameters ranged between 0.985 (superotemporal and inferior sector) and 0.992 (superior sector), showing an excellent long-term reproducibility (Table 2). The long-term reproducibility of the angular width and area of GCIPL defect on the deviation map also was excellent, showing ICC values of 0.993 and 0.930, respectively. The area of GCIPL defect had the highest CV (6.7%), and all of the CV values for the GCIPL thickness parameters and angular width of GCIPL defect were equal to or less than 5%. The TRTSD values for the angular width and area of GCIPL defect were 5.32° and 0.27 mm², respectively. The TRTSD value of the minimum GCIPL thickness (3.42 μm) was the highest, followed by the inferior (2.16 μm), inferotemporal (2.12 μm), superotemporal (2.01 μm), inferonasal (1.99 μm), superonasal (1.64 μm), superior (1.54 μm), and average (1.45 μm) GCIPL thickness parameters. 

In a comparative analysis of the superior and inferior GCIPL defects, the CV for the angular width of GCIPL defect was similar, but that for the area of inferior GCIPL defect was higher (7.1%) than that for the area of superior defect (4.3%). The TRTSD for the area of inferior GCIPL defect (0.30 mm²) also was higher than that for the area of superior GCIPL defect (0.16 mm²), but the difference was insignificant (P = 0.143). Also no significant difference in the TRTSD values of angular width was found between the two GCIPL defects (P = 0.747). 

The average and quadrant RNFL thickness parameters showed a high level of agreement: The ICCs ranged between 0.934 (nasal quadrant) and 0.978 (inferior quadrant), as indicated in Table 2. The CV values for the RNFL thickness parameters were all less than 10%, those for the nasal quadrant (6.5%) being the highest, and the average thickness (3.1%) the lowest. There was no significant difference in any of the intersession RNFL thickness measurements. The average RNFL showed the lowest measurement variability, with a TRTSD of 2.36 μm. With respect to the quadrants, the TRTSD values varied from 3.48 (temporal) to 5.19 μm (superior). 

To investigate the effect of glaucomatous severity on long-term reproducibility of GCIPL and RNFL maps, the visual field MD and global thickness parameters (average GCIPL and RNFL thicknesses) were used to represent the extent of glaucomatous damage. The TRTSD of average GCIPL thickness showed no significant association with MD (r² = 0.001, P = 0.968; Fig. 2A), mean average RNFL thickness (r² = 0.001, P = 0.952; Fig. 2B), and mean average GCIPL thickness (r² = 0.003, P = 0.604; Fig. 2C). Minimum GCIPL thickness (r² = 0.001, P = 0.934), angular width of GCIPL defect (r² = 0.026, P = 0.128), and area of GCIPL defect (r² = 0.003, P = 0.625) had no association with MD. Moreover, their associations with mean average GCIPL and RNFL thicknesses showed similar results (all P > 0.05). Sectoral GCIPL maps also showed no association with glaucomatous severity. Similar to GCIPL measurements, the degree of glaucomatous damage had no significant effect on the long-term variance of average and quadrant RNFL thickness measurements (Figs. 2D–F). 

An additional logistic regression analysis was performed, preparatory to which, patients were assigned to one of two groups according to the median TRTSD values, which were 1.0 and 2.0 for GCIPL and RNFL thickness, respectively. Accordingly, patients were divided into low (TRTSD < 1.0; n = 52) and high (TRTSD ≥ 1.0, n = 57) GCIPL thickness reproducibility groups; as well as low (TRTSD < 2.0; n = 53) and high (TRTSD ≥ 2.0; n = 56) RNFL thickness reproducibility groups. Similar to the results by linear regression analysis, the associations were not significant (Table 3). The average GCIPL and RNFL thickness reproducibility values both tended to show a weak inverse though insignificant relationship with disease severity. 

Advancements in SD-OCT have provided a new strategy for monitoring glaucoma progression, following the change in the macular GCIPL thickness measurements. Previous studies have reported good short-term inter- and intravisit reproducibility of macular GCIPL thickness in different study populations.^11,12,22 Nevertheless, it is essential to confirm longitudinal variance in order to determine the extent to which GCIPL thickness loss represents clinically significant progression in glaucoma. To the best of our knowledge, the present study is the first to report on the long-term reproducibility of GCA, including all of the GCIPL thickness parameters and, especially, the GCIPL deviation map. Our results are in line with previous studies,^11,12,22 showing good agreement as well as less fluctuation of GCIPL thickness parameters and of angular width and area of GCIPL defect on deviation map. More importantly, variability results were not influenced by the degree of glaucomatous damage, suggesting that GCA can be used as an effective tool for tracking macular change in glaucoma patients. 

Test–retest variability of GCIPL thickness has varied according to studies, depending on the study design, interval time between examinations, and study population.^11,12 Mwanza et al.¹¹ reported good short-term intervisit reproducibility of GCIPL thickness: all of the ICCs ranged between 0.94 and 0.98, and all of the CVs were less than 5%. In the present study, GCIPL maps also showed a high level of intersession agreement with ICCs ranging between 0.96 and 0.99 and CVs less than 8%. In consideration of CV dependent on the study population, TRTSD was evaluated to assess reproducibility. For short-term TRTSD of average GCIPL thickness, Mwanza et al.¹¹ reported it as 1.16 μm and Francoz et al.¹² as 1.03 μm. Based on our longitudinal data, average GCIPL thickness TRTSD was 1.45 μm. In addition, previous studies reported different parameters showing the largest thickness variability. Minimum GCIPL showed the highest TRTSD (2.28 μm) in the study by Mwanza et al.,¹¹ as opposed to the inferior sectoral GCIPL (2.22 μm) in the study by Francoz et al.¹² In the present study, the minimum GCIPL showed the highest TRTSD among the GCIPL thickness parameters, indicating the largest thickness variation cutoff. This probably is attributable to the minimum GCIPL thickness's greater likelihood of reflecting early local RGC loss in glaucoma.²³ Moreover, the superotemporal, inferior, and inferotemporal sectors showed higher TRTSD values than the other parameters. Those regions' higher susceptibility to glaucomatous damage^18,23,24 are more likely to be the cause of this divergence. 

Another important purpose of this longitudinal study is to determine the cutoff value for variation, which may help in early detection of progression. On the basis of our findings, average GCIPL thickness change exceeding approximately 4.0 μm is more likely to be considered as significant progression. Along with the GCIPL thickness parameters, angular width and area of GCIPL defect on the deviation map also showed a good longitudinal reproducibility. The results of the present study suggest 15° as an approximate cutoff limit for angular width of GCIPL defect. For the area of GCIPL defect, the tolerance limit was 0.76 mm². Considering the size of the inner annulus of the GCA elliptical scan area, approximately 0.94 mm² (vertical and horizontal diameters of 1 and 1.2 mm, respectively), the areal measurement variability was approximately 80% of the size of the inner annulus. Thus, change in GCIPL defect beyond these extents possibly can be regarded as progression. 

However, caution is necessary in application of our data in the clinical setting for the following reasons. First, our results on the measurement variance of GCIPL defect and thickness measurements could have been somewhat overestimated. Our structural progression evaluation was based on the disc and RNFL change. However, because the sequence of glaucomatous structural damage remains in question, change in GCIPL defect and thickness (at least in some patients) could have been a sign of real progression preceding RNFL change, rather than mere variation. Second, for GCIPL defect, an areal enlargement does not always entail an increase in angular width. The early GCIPL defect is predominantly located in the temporal macular region and expands toward the optic disc with a horizontal raphe as a base.^19,25 Due to such characteristics, progression should be determined in consideration of both angle widening and areal expansion. Third, the current GCIPL deviation map, due to the limitation of the color-coding system, is imperfect in its representation of GCIPL defect deepening. Therefore, even in stable cases within tolerance limit of areal or angular change of the GCIPL defect, the possibility of deepening of the defect should not be overlooked. 

In the present study, the reproducibility of the GCIPL parameters was not influenced by the level of glaucomatous damage. We used mean average GCIPL and RNFL thicknesses as well as visual field MD to represent the degree of glaucomatous damage. The strength of association between all GCIPL measurements and glaucomatous severity was weak and insignificant, which was in line with the study by Mwanza et al.¹¹ However, as only patients with a localized RNFL defect were included, the number of patients with far advanced glaucoma was limited. Further investigations with large number of patients with advanced glaucoma are warranted. 

Our previous study¹ on the long-term reproducibility of SD-OCT–derived RNFL deviation map and thickness parameters reported good agreement between intervisit RNFL measurements. The results of the present study correspond well with the previous investigations,^1–3 demonstrating excellent longitudinal reproducibility of average and quadrant RNFL thickness parameters. A previous study showed significantly higher measurement variability of inferior-quadrant RNFL thickness in mild glaucoma than in moderate-to-severe glaucoma.¹ However, none of the average and quadrant RNFL measurements in this study indicated any association with the disease severity. This may in some degree be related to differences in the characteristics of the study population: the proportion of patients with superior or inferior RNFL defect, the spectrum of glaucomatous damage, and the definition of glaucomatous severity. 

The present study has several limitations. First, we included “clinically stable” glaucoma patients based on conventional examinations. Nevertheless, and notwithstanding our two masked, independent observers applying strict progression-evaluation criteria, patients with undetected progressive changes might possibly have been included. Second, our results are based on patients meeting fine selection criteria (e.g., good-quality image with high signal strength, apparent GCIPL defect of more than 10 superpixels, concurrent GCIPL defect corresponding to the RNFL defect in all 3 serial scans). Thus, in clinical practice where adequate OCT images are not always achievable, larger variability is expected than in the current investigation. Third, three OCT scans at 6-month intervals were used to assess longitudinal reproducibility. However, the different number and interval of OCT scans could have resulted in different measurement variance outcomes. Lastly, as Cirrus OCT used in our study does not include an eye-tracking system, motion artifacts caused by head movement, blinking, or poor fixation could have influenced the variability. We tried to minimize these effects by excluding images altered by such variables and including only those of signal strength greater than or equal to 8. Moreover, the fast acquisition speed of the Cirrus OCT, could have compensated for the lack of eye tracking system. 

In conclusion, macular GCA algorithm of SD-OCT demonstrated excellent long-term reproducibility, with a high level of intervisit agreement and a low level of measurement variability. The macular GCA, with GCIPL thickness and deviation maps, can be used as an effective tool for monitoring progressive macular damage in glaucoma patients. Our findings can be of value in understanding the different measurement variance among GCIPL thickness parameters and angular width and area of GCIPL defect on the deviation map. 

The authors thank the Medical Research Collaborating Center at Seoul National University College of Medicine/Seoul National University Hospital for the statistical consultation. 

Supported by a grant from the Korea Health Technology R&D Project, Ministry of Health & Welfare, Sejong City, South Korea (Grant HI13C2061). 

Disclosure: K.E. Kim, None; B.W. Yoo, None; J.W. Jeoung, None; K.H. Park, None