This was an observational, cross-sectional study. The study has been approved by the Ethics Committee of the First Affiliated Hospital of Kunming Medical University and was implemented following the Declaration of Helsinki. All participants gave written informed consent before the study. All patients underwent a detailed and comprehensive medical history and ophthalmologic examinations, namely, ophthalmic routine slit-lamp examination, diopter examination, anterior chamber angle, and fundus photography, A-scan measures central corneal thickness and axial length, Goldmann intraocular pressure (IOP), and average RNFL thickness, and visual field examination. A total of 89 eyes of patients diagnosed with primary open-angle glaucoma (POAG) in the First Affiliated Hospital of Kunming Medical University from February 2015 to February 2017 were included, according to the Advanced Glaucoma Intervention Study (AGIS) visual field scoring system, 20 eyes with preperimetric glaucoma, 20 eyes with the mild stage of disease (1–5 points), 20 eyes with the moderate stage (6–12 points), and 29 eyes with the severe stage (13–20 points). The normal control group includes 30 eyes of healthy examined subjects at the same period. Inclusion criteria were: the best-corrected visual acuity was not less than 20/40, spherical refraction ranged from −5 to +5 D, cylinder correction ranged from −3 to +3 D, and subjects were excluded if they presented with nonglaucomatous visual field defective diseases, such as optic neuritis, history of ocular trauma, macular lesions, or intraocular surgery.

All subjects were examined with the Humphrey Visual Field Analyzer II (Carl Zeiss Meditec, Dublin, CA, USA) using Swedish Interactive Threshold Algorithm (SITA)-FAST 24-2 and 10-2 programs. Each visual field test was performed by the same physician. Visual fields with a solid fixation loss rate of <20%, a false-negative rate of <33%, and a false-positive rate of <15% were considered reliable results. The visual field that the false-negative rate was higher than 33% but indicative of severe disease progression (mean deviation (MD) value below −12 dB) was also considered for the calculation. Repeated visual field examinations were required if eyelid masking, fatigue, incorrect fixation, or learning effects were present.

Subjects underwent optic nerve head (ONH) scanning program for ONH examination and GCC program for macula using the ultrahigh speed 70 kHz Fourier domain OCT system (RTVue XR Avanti, Optovue, Inc., Fremont, CA, USA). Peripapillary RNFL and GCC thickness probability maps with eight partitions were obtained by measuring the RNFL thickness of the optic disc and the macular GCC thickness. High-quality images with signal intensity index >50 without motion and vitreous floaters were considered for analysis. OCT testing and visual field testing were performed on the same day.

Harwerth et al. (19) proposed an empirical formula for estimating the counts of corresponding RGCs based on experimental studies on rhesus monkeys, namely, SAPrgcs from the visual field (Humphrey 24-2) detection as function measurements and OCTrgc obtained OCT detection as peripapillary structure measurements. The experimental model was translated to clinical perimetry in humans and has been validated in glaucoma and normal IOP glaucoma (5, 14, 20). Considering the effects of aging on axonal degeneration and varying severity of the disease, factors such as age and MD have been adjusted in OCTrgc. The use of this model would translate region-specific visual field sensitivity into estimate RGC for the corresponding regions.

In the above formulas, ec is the retinal eccentricity, which refers to the distance from the central fixation to a certain point on the retina. Based on the results, the values of ec were taken separately as 4.2 (3 × 3 deg), 12.8 (9 × 9 deg), 21.2 (15 × 15 deg), and 24 (nasal side of the visual field 21 × 3 deg and 27 × 3 deg). m and b represent the slope and intercept of the linear relationship relating ganglion cell count and visual field sensitivity at a given ec value. s refers to visual acuity at a certain locus of the visual field in dB. d refers to the axonal density (axons/μm²). c is highly related to the severity of the disease and is used to correct the ratio of axonal to nonaxonal cell composition of RGC cells in the model.

Considering that the mild manifestation of glaucoma derived mainly from RNFL damage while the severe stage reveals a more variable MD damage in the visual field, Medeiros et al. (15–17) performed a weighted integration by which the model relates structure to function.

Hood et al. found that mild glaucoma damages were detected in the macula (21, 22). Based on the previous model for eRGC estimation, Medeiros et al. (18) have modified the visual function test. The scattered loci in the Humphrey 24-2 SAP were changed to focused 16 loci of the retinal sensitivity within 10° of the macular fixation areas.

In the above formulas, ec was denoted as 4.2 (3 × 3 deg), 9.5 (3 × 9 deg and 9 × 3 deg), and 12.8 (9 × 9 deg). TD is the mean value of retinal sensitivity at 16 loci corresponding to 10° of the center of total deviation using 24-2 SAP. Corresponding estimated RGC count, namely, the macular SAPrgc was obtained by applying the value of each point detected by SAP to the above formula. Similarly, structural parameters were changed to mean RNFL of the temporal, superior, and inferior temporal regions of the optic disc corresponding to the macula.

The aforementioned formulas gave the OCT measurements in the macula, namely, macular OCTrgc. The formulas were weighted and integrated according to the previous description by Medeiros et al. and the obtained RGC remained both structural and functional components.

Three innermost layers of the retina are preferentially involved when there is glaucomatous damage in the macular: the nerve fiber layer, the ganglion cell layer, and the inner plexiform layer. These three layers contain ganglion cell axons, ganglion cell body, and ganglion cell dendrites, which are collectively called the GCC. The latest frequency-domain OCT uses Fourier technology and broadband light source technology to acquire and process data, which has higher resolution, sensitivity, and faster scanning speed than time-domain OCT, and can acquire and analyze the structural images and data information of GCC in the macula in a short time. The estimation model built by Harwerth and Medeiros was adopted in the modified macular structure-function model. SAP test was changed to Humphrey 10-2 program which focuses on 16 loci within 10° of the macula (2° interval). ec was taken as 1.4 (1 × 1 deg), 4.2 (3 × 3 deg), 7.0 (5 × 5 deg), and 8.6 (7 × 5 deg). Macular SAPrgc, corresponding to the estimated RGC count was obtained by applying the value of each point detected by SAP to the formula.

Ganglion cell complex thickness of the macula was taken as the structural parameter. MD is the value used for the center of the visual field in Humphrey SAP 10-2 program (interval 2°). Macular OCTrgc was obtained by the OCT measurements of the macula for RGC estimates.

The formula for modified macular SAPrgc is the same used in the above model for RGCs estimate.

Continuous variables were checked to meet the normality conditions of the Shapiro-Wilk test. ANOVA test and the least significant difference post-hoc test with Bonferroni adjustment were used for intergroup comparisons for normally distributed variables. Dichotomic variables were analyzed using the χ² test or the Fisher's exact test. Receiver operating characteristic (ROC) curve analyses were used to assess the diagnostic accuracy of eRGC, mRGC1, and mRGC2. The area under the ROC curve (AUC), sensitivity, specificity were calculated to assess the performance of the prediction model. Yorden's index, defined as the sensitivity + specificity – 1, was used to determine the optimal cut-off values to maximize diagnostic efficiency. Statistical significance was set at P ≤ 0.05. By setting the alpha to 0.05 to declare the slope of RGC loss as statistically significant, we were able to maintain a specificity of 95%, as demonstrated in the previous study that estimated rates of RGC loss obtained from SAP and OCT (16). At 95% specificity, approximately six of the 119 eyes would be expected to show significant slopes just by chance. The SPSS program package version 21.0 (Statistic Package for the Social Science, Inc., Chicago, IL, USA) was used for statistical analysis.

A total of 89 eyes with POAG meeting the diagnostic criteria were enrolled. Healthy control included 30 eyes of healthy examined subjects. According to the AGIS scoring system, POAG eyes were divided using Humphrey SAP 24-2 program, namely, 20 eyes with preperimetric glaucoma, 20 eyes with the mild stage of glaucoma, 20 eyes with moderate stage, and 29 eyes with advanced stage. The demographic and clinical characteristics of subjects among the groups were shown in Table 1. The glaucomatous eyes have significantly worse visual field index (VFI), MD (24-4 and 10-2), GCC, RNFL, eRGC, mRGC2, and mRGC1 than healthy eyes (P < 0.001).

Compared with normal individuals (Figure 1), eRGC, mRGC1, and mRGC2 obtained from the three models decreased markedly in glaucoma, and as glaucoma progressed, RGC estimates reduced gradually from mild to severe stage.

Compared with the estimated RGCs in healthy eyes, preperimetric glaucomatous eyes had mRGC1 reduced to 13.2%, mRGC2 reduced to 14.5%, and eRGC reduced to 18%. Glaucomatous eyes with mild stages had mRGC1 reduced to 28%, mRGC2 reduced to 34%, and eRGC reduced to 38%. In the severe stage, the reduction proportion for mRGC1, mRGC2, and eRGC was 81, 85, and 92%, respectively, indicating that mRGC1 has a good staging performance and may serve as a candidate model for estimating RGC counts in the macula.

To analyze the capability of eRGC, mRGC1, and mRGC2 and also structural parameters (GCC and RNFL) and functional parameters (VFI and MD24-2) in differentiating between POAG and healthy eyes, ROC curves were conducted. The results showed that the AUCs of all parameters were above 0.9. As shown in Figure 2, AUC of eRGC, mRGC1, and mRGC2 was 0.992 (95% CI, 0.954~1), 0.982 (95% CI, 0.939~0.998), and 0.972 (95% CI, 0.924~0.994), respectively. The AUC of structural parameter GCC was 0.995 (95% CI, 0.960~1) and the AUC of peripapillary RNFL was 0.992 (95% CI, 0.954~1). The AUCs for the functional parameters VFI and MD (24-2) were 0.906 (95% CI, 0.839–0.952) and 0.927 (95% CI, 0.865–0.967), respectively. All these indicators offered relatively good predictive performances to distinguish glaucoma from a healthy population.

For a 95% specificity as shown in Table 2, eRGC had a sensitivity of 95.51%, with a relatively low sensitivity of 86.52 and 87.64% for mRGC1 and mRGC2, respectively. The sensitivity of structural parameters GCC and RNFL were 98.88 and 95.51%, respectively, while the sensitivity of the functional parameter MD24-2 is 80.90%, the VFI sensitivity is 73.03%, much lower than the sensitivity of mRGC1 and mRGC2.