A Combined Index of Steady-State Pattern Electroretinogram and Optical Coherence Tomography Improved the Detection of Early Glaucoma

Glaucoma is a progressive disease caused by irreversible loss of retinal ganglion cells (RGCs) [1]. The early detection of RGC damage is necessary for preventing deterioration of the visual field (VF) as significant RGC damage precedes VF loss [2]. Therefore, detecting early RGC damage is fundamental for preserving vision in patients with glaucoma.

Pattern electroretinogram (PERG) was first introduced in 1982 as a direct indicator of RGC function [3, 4]. Moreover, PERG is classified into two types: steady-state response (8≥ rev/s) and transient response (<4 rev/s), based on the frequency of checkboard reversal [4]. Several studies have demonstrated that steady-state PERG (ssPERG) at temporal frequencies between 10 and 20 (rev/s) is more efficacious for the detection of incipient glaucoma damage compared to the transient response PERG [4‒7]. ssPERG reduces some of the effects of poor fixation on the recording, increasing inter-test reproducibility, and it has been optimized for the early detection of glaucoma [8]. To reduce inter-individual variability, the Freiburg group suggested PERG_ratio by combining two check sizes, 0.8° and 16° [4]. They reported that early-stage glaucoma can be detected using the ratio of the amplitude responses to 0.8° checks and 16° checks of ssPERG [4, 9].

Optical coherence tomography (OCT) is widely used for the detection of glaucoma and its progression since the modality provides an objective and reproducible quantitative evaluation of the optic disc and inner retinal layers [10‒13]. Both the thickness of the circumpapillary retinal nerve fiber layer (cpRNFL) and the ganglion cell inner plexiform layer (GCIPL) have demonstrated excellent capabilities in the diagnosis of glaucoma [14, 15]. In particular, the diagnostic capabilities of the inferior sector of the cpRNFL and GCIPL thickness have demonstrated superior performance compared to other sectors [14‒16].

Several studies support the “dysfunction-preceding-death hypothesis,” which suggests that RGC dysfunction precedes RGC death [17‒20]. Nagaraju et al. [19] demonstrated that PERG losses precede loss of cpRNFL thickness in a mouse model. Further, Ventura et al. [20] confirmed that PERG amplitude only weakly correlated with cpRNFL thickness, as determined by OCT, in groups of glaucoma suspects and early glaucoma patients. In this context, PERG can serve as a useful surrogate measure of RGC function during the stage of RGC dysfunction before RGC death.

A previous study supporting the dysfunction-preceding-death hypothesis suggested that a reduced response from viable RGCs could be detected by combining PERG and OCT. This is because OCT can detect focal defects that may not be reflected in the summed activity of PERG generators in the central retina [20]. In our previous study, we demonstrated that the amplitude of transient PERG was correlated with macular GCIPL thickness in early glaucoma [21]. However, few studies exist on the relationship between ssPERG, OCT, and VF tests.

We hypothesized that combining two complementary functional and structural tests would improve the accuracy of early glaucoma detection. Especially, ssPERG, which is an objective functional test, was expected to better reflect the RGC function than the subjective VF test, thus improving the accuracy of early glaucoma detection. In this study, we investigated indicators combined with structural and functional measures that could more effectively detect early-stage primary open-angle glaucoma (POAG) and evaluated the structure-function relationship between ssPERG, OCT, and the VF test.

In this retrospective cohort study, patients with early POAG including preperimetric stage and normal subjects who visited the Glaucoma Clinic at Pusan National University Hospital between April 1, 2018, and May 31, 2020, were included. The data were accessed for research purposes from March 25, 2022, to December 31, 2023. The study was approved by the Institutional Review Board (IRB) of Pusan National University (Approval No. 2203-017-113) and was conducted in accordance with the tenets of the Declaration of Helsinki. The requirement for patient consent was waived by the IRB because of the retrospective nature of the study. The data were de-identified, and the authors did not have access to information that could identify individual participants.

All participants underwent thorough ophthalmologic examination, including best corrected visual acuity, slit-lamp examination, intraocular pressure (IOP) measurement using Goldmann applanation tonometry, gonioscopy, dilated fundus examination, biometry using the IOL Master (Carl Zeiss Meditec, Dublin, CA, USA), and keratometry using an auto-kerato-refractometer (ARK-510A; NIDEK, Hiroshi, Japan). Regarding OCT and VF tests, only the tests performed within 6 months of ssPERG were included in the analysis. The inclusion criteria were age >18 years, best corrected visual acuity ≥20/32, and clear ocular media. Exclusion criteria included a diagnosis of diabetes, uveitis, secondary glaucoma, corneal abnormalities, non-glaucomatous optic neuropathies, previous trauma, ocular surgery except uncomplicated cataract surgery, laser treatment, or any eye disease other than glaucoma. When both eyes satisfied the inclusion criteria, one eye per individual was randomly selected for this study.

The diagnosis of early POAG was based on the following eligibility criteria: (1) presence of glaucomatous optic nerve appearance, corresponding VF loss; (2) 24–2 VF mean deviation (MD) ≥ −6 dB; and (3) open angles on gonioscopy. Glaucomatous optic nerve damage was defined as the vertical cup-to-disc ratio ≥0.7 or asymmetry between the vertical cup-to-disc ratio of both eyes >0.2 or the presence of focal neural rim notching or generalized loss of the neural rim or RNFL defect. Preperimetric glaucoma was defined as having a normal VF with glaucomatous optic nerve damage described above. Glaucomatous VFs were those that met at least one of the following criteria: ≥3 contiguous points with a significance level of p < 0.05, including ≥1 point significant at the p < 0.01 level, on the same hemifield in the pattern deviation plot and/or glaucoma hemifield test (GHT) outside the normal limits and/or a pattern standard deviation (PSD) probability outside 95% of the normal population.

Healthy controls were defined as those with no history of ocular disease, IOP <21 mm Hg, an absence of glaucomatous optic disc appearance, and a normal VF. Normal VF was defined as a GHT within the normal limits, and a MD and PSD within 95% of the healthy population.

The ssPERG was performed using the RETI-port/scan21 (Roland Consult, Brandenburg, Germany), according to Bach and Hoffmann [4] and standards of the International Society of Clinical Electrophysiology and Vision [9]. The participant was seated 50 cm away from the test monitor, with appropriate corrections for refractive errors. A ground electrode was attached to the forehead, a reference electrode was attached to the lateral canthus, and an active electrode (H-K loop; Avanta, Ljubljana, Slovenia) was attached to the lower conjunctival sac of each eye. Pattern stimuli were presented as alternating black and white squares, with widths of 0.8° and 16°, and the mean luminance was 85 candela/m². The examination was measured at a reversal rate of 15 reversals per second, and the results of 200 tests were combined. Pattern electroretinogram ratio (PERG_ratio) was defined as the ratio of the amplitude response to 0.8° checks to that of 16° checks.

PERG_ratio = (PERG amplitude to 0.8° checks)/(PERG amplitude to 16° checks).

All the ssPERG tests were conducted by one experienced technician. The intraclass correlation coefficient for intraobserver variability of PERG_ratio was 0.915.

The thicknesses of the cpRNFL and macular GCIPL were measured using the cirrus HD-OCT Macular Cube 200 × 200 and Optic Disc Cube 200 × 200 protocol (Carl Zeiss Meditec, Dublin, CA, USA). The optic nerve head and RNFL analysis algorithm automatically delineate the optic disc margin. The average, 12 clock-h, and four-quadrant cpRNFL thicknesses along with a 3.46-mm-diameter circle from the optic disc center were analyzed. The ganglion cell analysis algorithm reports the GCIPL thickness in six wedge-shaped sectors within a 4.8 × 4.0 mm ellipse, excluding an inner elliptical annulus measuring 1.2 × 1.0 mm in diameter. The average, minimum, and six-sectoral (superotemporal, superior, superonasal, inferonasal, inferior, and inferotemporal) GCIPL thickness values were obtained. For quality control, we set the minimum signal strength of all included SD-OCT scans to 6.0.

Automated perimetry was performed using a Humphrey Visual Field Analyzer 750i instrument (Carl Zeiss Meditec, Dublin, CA, USA) utilizing the Swedish interactive thresholding algorithm standard 24-2. Only eyes with reliable VFs (fixation loss ≤20%, false-positive rate ≤33%, and false-negative rate ≤33%) were included.

Logistic regression models were used to predict the probability of early glaucoma. The model’s ability to discriminate between a healthy eye and an eye with early glaucoma was evaluated by the area under the receiver operating characteristics curve (AUC). The selection of variables was guided by AUC. The parameters of ssPERG, OCT with an AUC ≥0.8 were selected as candidate variables for developing the composite prediction models. Four variables were included in the final model: PERG_ratio, average cpRNFL thickness, 7 o’clock cpRNFL thickness, and inferior quadrant cpRNFL thickness. As the model combining PERG_ratio and 7 o’clock cpRNFL thickness had the highest AUC, the combined index of those two parameters was compared with other parameters. We also compared the AUC of a model combining PERG_ratio and 7 o’clock cpRNFL thickness to those of models combining 7 o’clock cpRNFL thickness and visual field index (VFI), MD, and PSD, respectively.

The normality of data distribution was evaluated using the Kolmogorov-Smirnov test. Differences between normal subjects and patients with glaucoma were analyzed using the Mann-Whitney U test. Moreover, the independent-sample t test was used for continuous variables and the chi-squared or Fisher’s exact test for categorical variables. AUCs were used to evaluate the diagnostic performance for early glaucoma. DeLong’s test was used to evaluate the difference between AUCs [22]. The diagnostic performance was tested with 10-fold cross validation to optimize generalizability. All statistical analyses were performed using R (version 4.0.5; R Project for Statistical Computing, Vienna, Austria). Furthermore, p values of <0.05 were considered statistically significant.

A total of 135 eyes from 135 patients (57 normal subjects and 78 patients with POAG) were included in the study. Among the 78 patients with POAG, 13 patients had preperimetric glaucoma. No significant differences were observed between the groups in terms of age, sex, axial length, spherical equivalent, central corneal thickness, or IOP (all p ≥ 0.096) (Table 1). However, significant differences were observed in ssPERG, VF, and OCT parameters between the groups. The PERG_ratio was greater in the normal group compared to the early POAG group (1.11 ± 0.25 vs. 0.69 ± 0.25, p < 0.001). Although no significant difference was observed in ssPERG amplitude to 16° checks (5.54 ± 1.89 μV vs. 5.62 ± 1.84 μV, p = 0.690), a significant difference in ssPERG amplitude to 8° checks (6.18 ± 2.58 μV vs. 3.88 ± 1.92 μV, p < 0.001) was identified. The mean MD, PSD, and VFI of the normal group were −1.51 ± 1.27 dB, 1.88 ± 0.73 dB, and 98.35 ± 1.40%, respectively. In contrast, in the early POAG group, they were −2.89 ± 1.77 dB, 3.60 ± 2.30 dB, and 94.44 ± 4.97%, respectively (all p ≤ 0.002).

In the normal group, the mean GCIPL, cpRNFL, inferior quadrant cpRNFL, and 7 o’clock cpRNFL thicknesses were 78.37 ± 7.28 μm, 91.75 ± 10.22 μm, 115.67 ± 17.30 μm, and 136.19 ± 20.96 μm, respectively, while in the early POAG group, they were 72.41 ± 7.44 μm, 78.18 ± 10.26 μm, 88.91 ± 20.82 μm, and 97.11 ± 31.14 μm, respectively. All the OCT parameters in the normal group were significantly greater than those in the early POAG group (all p < 0.001).

Table 2 summarizes the diagnostic accuracies of ssPERG, OCT, and VF parameters, as well as a combined index for the detection of early POAG. Four parameters with AUCs higher than 0.8 included: PERG_ratio (AUC = 0.890), average cpRNFL thickness (AUC = 0.827), 7 o’clock cpRNFL thickness (AUC = 0.844), and inferior quadrant cpRNFL thickness (AUC = 0.830). Although the AUC of PERG_ratio was the highest, no significant difference compared to the other three parameters was observed (p ≥ 0.132). However, the AUC of the index combining PERG_ratio and 7 o’clock cpRNFL thickness significantly improved diagnostic accuracy (AUC = 0.951) compared to those of the best four parameters (all p ≤ 0.004) and, as well as those of indices combining three VF parameters (MD, PSD, and VFI) with 7 o’clock cpRNFL thickness (AUC = 0.850, 0.862, and 0.863, respectively) (all p = 0.001) (Fig. 1 a). The index combining PERG_ratio and 7 o’clock cpRNFL thickness showed consistently high values for sensitivity (0.885), specificity (0.912), and likelihood ratios (10.085). The combined index of 7 o’clock cpRNFL thickness and VFI showed high specificity (0.982) and likelihood ratios (38.000), but its sensitivity was low (0.667).

Online supplementary Table 1 (for all online suppl. material, see https://doi.org/10.1159/000545094) shows the diagnostic accuracies with age, refractive error, and central corneal thickness adjusted for the same parameters. Adjusted PERG_ratio showed good sensitivity (0.811), specificity (0.926), likelihood ratio (10.946), and AUC (0.917). The OCT parameter with the highest AUC was 7 o’clock cpRNFL thickness (AUC = 0.879). It showed good specificity (0.865) and sensitivity (0.759), but the likelihood ratios were low (3.593). There was no significant difference in AUC between PERG_ratio and 7 o’clock cpRNFL thickness (p = 0.322). However, when PERG_ratio and 7 o’clock cpRNFL thickness combined, the AUC value was significantly better than that of 7 o’clock cpRNFL thickness alone (p = 0.006). In addition, the AUC of the index combining PERG_ratio and 7 o’clock cpRNFL thickness (AUC = 0.967) significantly improved diagnostic accuracy compared to those of indices combining three VF parameters (MD, PSD, and VFI) with 7 o’clock cpRNFL thickness (AUC = 0.882, 0.890, and 0.887, respectively) (all p ≤ 0.011).

Figure 1 b displays the predicted probability in the binary logistic regression model of the four most accurate parameters and combined indices: PERG_ratio, average cpRNFL thickness, 7 o’clock cpRNFL thickness, inferior quadrant cpRNFL thickness, and four combined indices (PERG_ratio + 7 o’clock cpRNFL thickness, 7 o’clock cpRNFL thickness + VFI, 7 o’clock cpRNFL thickness + MD, and 7 o’clock cpRNFL thickness + PSD). The greater the diagnostic ability, the greater the difference between the healthy and glaucoma groups. Among all parameters, the combined index had the largest difference.

Table 3 summarizes the correlations between various OCT and VF parameters and PERG_ratio. Across the entire study sample, all functional measures including PERG_ratio, MD, and VFI were positively correlated with average cpRNFL and macular GCIPL thickness (all p ≤ 0.002). The correlation coefficients between VF and OCT parameters were higher than those between PERG_ratio and OCT parameters. Furthermore, PERG_ratio was not correlated with any OCT and VF parameters, whereas VF parameters were positively correlated with OCT parameters in the early POAG group.

In the linear regression analysis model, we explored the relationship between functional parameters (PERG_ratio, MD, and VFI) and structural parameters (average cpRNFL thickness, 7 o’clock cpRNFL thickness, inferior quadrant cpRNFL thickness, and average GCIPL thickness) in the entire study sample (Table 4). The total variance explained by each model was estimated using R². The percentages of variance explained by average cpRNFL thickness were 14%, 24%, and 32% for PERG_ratio, MD, and VFI, respectively. The percentages of variance explained by 7 o’clock cpRNFL thickness were 11%, 25%, and 38% for PERG_ratio, MD, and VFI, respectively. The percentages of variance explained by inferior quadrant cpRNFL thickness were 11%, 22%, and 32% for PERG_ratio, MD, and VFI, respectively. The percentages of variance explained by average GCIPL thickness were 5%, 9%, and 16% for PERG_ratio, MD, and VFI, respectively (online suppl. Fig. 1).

Figure 2a shows the ssPERG, disc and red free photographs, and VF tests of normal subjects, while Figure 2b shows those of a patient with preperimetric POAG. The PERG_ratio of Figure 2a and Figure 2b were 0.991 and 0.615, respectively. Both Figure 2a and Figure 2b had no signs of glaucomatous VF damage. There were no disc cupping or cpRNFL defects on the fundus photo of Figure 2a. In Figure 2b, enlargement of the cup-to-disc ratio and inferotemporal RNFL defect were observed on the disc and red free photograph, respectively, although VF was normal.

In this study, we investigated the combined index of structural and functional measures which is the most capable of detecting early POAG, including preperimetic stage, and explored the structure-function relationship between structural (OCT parameters) and functional (ssPERG and VF parameters) measures. Among single ssPERG, VF parameters, and OCT parameters, PERG_ratio demonstrated the greatest accuracy in detecting early glaucoma. The diagnostic accuracy significantly improved when combining PERG_ratio and OCT parameters (7 o’clock cpRNFL thickness) and the combined index showed significantly better diagnostic performance compared to those combining OCT with VF parameters such as MD, PSD, and VFI. Furthermore, the index demonstrated consistently high values in sensitivity, specificity, and likelihood ratio, proving that it could be a valuable tool for diagnosis. In the additional analysis adjusting age, refractive error, and central corneal thickness, the PERG_ratio was proven to be an excellent parameter for the detection of early glaucoma. When combined with the best OCT parameter, the diagnostic performance of the index combined with PERG_ratio was persistently better than those with the VF indices even after adjusted. Although a significant correlation was observed between functional and structural measures in the entire group, the percentages of variance in PERG_ratio explained by OCT parameters were lower than those for VF parameters.

The OCT findings of this study are consistent with the results of previous studies, which demonstrated that the thickness of the inferior cpRNFL and GCIPL offer better diagnostic accuracy compared to other sectors [14‒16]. We identified that the cpRNFL thickness at the 7 o’clock position and in the inferior quadrant of cpRNFL thickness had the best diagnostic performance for distinguishing between normal subjects and those with early POAG. As the inferior quadrant of the optic disc is more vulnerable to early glaucomatous damage compared to other regions of the disc [23], the diagnostic performance of the aforementioned parameters to discriminate early glaucoma from a healthy eye would have been better compared to the other sectors.

Clinical and experimental studies have reported that RGC dysfunction precedes RGC death in glaucoma [17‒20, 24, 25]. Ventura et al. [20] reported that ssPERG amplitude was only weakly correlated with cpRNFL thickness in glaucoma suspects, and was non-significant in the early glaucoma group. Porciatti and Ventrua [24, 25] suggested that in the RGC dysfunction phase, the electrical activity of RGCs, measured by PERG, may be altered before measurable changes in the RNFL thickness. However, they stated that PERG is not necessarily a superior diagnostic tool for detecting glaucoma compared to OCT as cpRNFL thickness represents focal defects. This is unlike PERG, which is generated from summed activity in the central retina [20].

In our previous study, we evaluated the relationship between transient state PERG N95 amplitude and macular GCIPL thickness [21]. In patients with early-stage glaucoma, PERG N95 amplitude was positively correlated with macular GCIPL thickness. In this study, however, no significant correlation between PERG_ratio and macular GCIPL thickness in early glaucoma was identified. This difference may be partly explained by the different types of PERG used. In the previous study, we used transient PERG N95 amplitude as a functional measure, whereas in this study, we focused on the PERG_ratio of ssPERG, which potentially led to different results.

The PERG is a direct indicator of ganglion cell function and is therefore a promising approach to assist early detection of glaucoma [4]. Bach and Hoffmann [4] and Gallo et al. [7] reported that in patients with glaucoma, the amplitude reduction is more significant in ssPERG than in transient PERG, indicating its superiority in detecting glaucoma. However, the amplitude has its disadvantage that it shows high inter-individual variability. To overcome this problem, previous research has suggested PERG_ratio which is a better biomarker than raw ssPERG amplitude as it reduces inter-individual variability [4, 5, 26]. In this study, and as reported by Ventura et al. [20], no significant correlation was observed between ssPERG amplitude and cpRNFL or GCIPL thickness in early glaucoma. However, it is important to note that in our study, we used PERG_ratio instead of ssPERG amplitude, which was the focus of Ventura et al. [24]’s study.

Improving the early diagnosis of glaucoma is of the utmost importance since preservation and restoration of RGCs is possible only in the RGC dysfunction phase [19, 24]. We hypothesized that combining two complementary tests, functional (ssPERG and VF) and structural (OCT), would improve the accuracy of early glaucoma detection. Among the two functional tests, we consider that the objective ssPERG better reflects the function of RGCs than the subjective VF. As a result, we identified a new combined index of PERG_ratio and cpRNFL thickness at the 7 o’clock position, which demonstrated significantly better AUC than those of any other parameters of ssPERG, OCT, VF tests and combined indices with OCT and VF tests. The AUC of PERG_ratio and 7 o’clock cpRNFL thickness were 0.890 and 0.844, respectively. However, the combined index of PERG_ratio and OCT 7 o’clock cpRNFL thickness had excellent discriminatory ability with an AUC of 0.951. We believe that this result may be explained by the fact that ssPERG and OCT, with a weak correlation and linear structure-function relationship, provide distinct valuable information to detect early POAG in terms of RGC function and focal structural change.

A limitation of this study is the usability of the combined index. To calculate the combined index, computing predicted probability using logistic regression is necessary. However, doing so in a clinical setting poses challenges. In future studies, we plan to develop methods to facilitate the calculation process, thereby enhancing the generalizability of our method for detecting early POAG in clinical settings. Second limitation is that this study is restricted to early-stage glaucoma since the objective was to detect changes during the RGC dysfunction phase at an early stage. Further research is needed to investigate the structure-function relationship and diagnostic accuracy of ssPERG, OCT, and VF parameters in moderate and advanced stages of glaucoma. Third, we used only the 24-2 VF test in our analysis as there was a lack of sufficient 10-2 VF tests. ssPERG mainly covers the central macular region, thus it is expected to have a strong relationship with the 10-2 VF test. We aimed to include 10-2 VF test in the analysis in future research.

The combined index, incorporating PERG_ratio and 7 o’clock cpRNFL thickness together, demonstrated significantly enhanced diagnostic accuracy compared to using a single PERG_ratio or OCT parameter alone, or combining 7 o’clock cpRNFL thickness with VF parameters. These findings suggest that ssPERG, reflecting RGC dysfunction, can contribute to the early detection of glaucoma, along with OCT, which represents focal structural damage.

The study was approved by the Institutional Review Board (IRB) of Pusan National University (approval No. 2203-017-113) and was conducted in accordance with the tenets of the Declaration of Helsinki. The requirement for patient consent was waived by the IRB of Pusan National University because it was a minimal-risk retrospective study which had no patient interaction.

The authors have no conflicts of interest to declare.

This research was supported by a grant from the Patient-Centered Clinical Research Coordinating Center (PACEN) funded by the Ministry of Health and Welfare, Republic of Korea (Grant No. HC19C0276), and by the National Research Foundation of Korea (NRF) grant funded by the Korean government (MSIT) (Grant No. RS-2023-00247504). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Jiwoong Lee contributed to conceptualization and involved in writing – review and editing. Jiwoong Lee, EunAh Kim, and Hwayeong Kim contributed to methodology. Hwayeong Kim contributed to investigation and involved in writing – original draft. Jiwoong Lee, Hwayeong Kim, Sangwoo Moon, and Jinmi Kim contributed to data curation. All authors have read and approved the final manuscript and agreed to publish the manuscript.

The data that support the findings of this study are not publicly available due to their containing information that could compromise the privacy of research participants but are available from the corresponding author (J.L.) upon reasonable request.