This study is a secondary analysis of data from the “Swept-Source OCT (SS-OCT) Myopia and Glaucoma Study,” a multicenter, cross-sectional investigation. The parent study was designed to differentiate morphologic changes between glaucoma and myopia by enrolling 4 distinct groups: (1) primary open-angle glaucoma (POAG) with myopia, (2) POAG without myopia, (3) healthy subjects with myopia, and (4) healthy subjects without myopia. ^,,, Only the POAG patients from the original study were included in the current study. Consequently, 185 eyes from 147 patients with POAG were reviewed for eligibility. Inclusion criteria were age 30 years or older, abnormal VF defined by the Hodapp-Anderson-Parrish criteria (abnormal GHT, PSD ≤ 5%, or 3 continuous points with a P value ≤5% with 1 point ≤1% in the upper or lower hemifield in the pattern deviation plot), best corrected visual acuity (BCVA) of 0.7 or better, and intraocular pressure (IOP) of 21 mm Hg or lower in the study eye at the time of examination.

Exclusion criteria include (1) presence of any ocular pathology other than glaucoma, including suggestive pathologic myopia defined by the presence of optic disc tilt (ovality index > 1.30), AXL of 28 mm or longer, posterior staphyloma, ridge at temporal side of the disc, parapapillary intrachoroidal cavitation, localized and/or diffuse chorioretinal atrophy; (2) cylinder refractive error greater than 2 diopters; (3) narrow angle with Shaffer grade 2 or less; (4) previous ocular surgery or laser treatment; (5) inability to obtain images of acceptable quality because of subject noncooperation or media opacities; (6) unreliable VF test results, defined as fixation losses >20% or false positives >15%; (7) history of leukemia, dementia, or multiple sclerosis; (8) subjects with uncontrolled hypertension (systolic blood pressure >150 mm Hg and diastolic pressure >95 mm Hg); and (9) subjects with hypotension (systolic blood pressure <100 mm Hg for age <60 years and systolic blood pressure <110 mm Hg for age ≥60 years). Participants with extremely advanced VF defect (mean deviation of–12 dB or worse) were also excluded.

Participants were recruited from 8 institutes in Japan (Kanazawa University, Osaka University, Tajimi Iwase Eye Clinic, Toho University Ohashi Medical Center, Tohoku University, The University of Tokyo), South Korea (Seoul National University Bundang Hospital), Hong Kong (Hong Kong Eye Hospital), and the United States (University of California San Diego). Study protocols were approved by the institutional review board of Kanto Central Hospital (R1-06-005) and adhered to the tenets of the Declaration of Helsinki. All participants provided written informed consent before participation to the study.

This clinical study is registered with the University Hospital Medical Information Network (UMIN: https://www.umin.ac.jp ) as “Myopic and Glaucomatous Optic Neuropathy Changes Using Swept Source OCT” under the registration number UMIN000024047. Participants underwent comprehensive ophthalmic examinations that include BCVA, refractometry, and keratometry, slitlamp biomicroscopy, ophthalmoscopy, gonioscopy, Goldmann-applanation tonometry, AXL measurements, standard automated perimetry (Humphrey Field Analyzer 24-2 SITA Standard), and photographs of the optic disc and the macula.

For each participant, 6.0 × 6.0-mm raster scans centered on the optic disc, 24-line Bruch membrane opening (BMO)–centered radial scans (BMO was manually identified in 2 perpendicular ONH radial scans prior to acquisition to determine BMO center), and 12.0 × 9.0-mm-wide scans including both the ONH and macula were obtained. Measurements were repeated 3 times and data with the best quality factor were used. Ocular magnification effects were corrected based on refractive error, corneal curvature, and AXL with a software provided by the manufacturer described previously.

In brief, an approximate optical model of each eye based on its refractive error, corneal curvature, and AXL were constructed, then the refractive power of the lens in the model was iteratively adjusted until the focusing error on the fundus was minimized using a paraxial ray-tracing, and the resultant final model was used to calculate the magnification correction factor. Images influenced by involuntary blinking or saccade or those with a quality factor <60% were excluded. All scans of all study participants were sent to a reading center for evaluation and review.

Details of the generation of ROTA images and the algorithm of ROTA have been described previously. ^,,,, In brief, RNFL reflectance and RNFL thickness measurements of individual eyes were exported from SS-OCT (DRI OCT Triton, Topcon, Japan) via a proprietary research browser provided by Topcon to a computer for ROTA. ROTA integrates RNFL reflectance and RNFL thickness measurements with a series of nonlinear transformations to reveal the optical texture and trajectories of axonal fiber bundles. Axonal fiber bundle defects in ROTA represent altered intrinsic optical properties of axonal fiber bundles or loss of axonal fiber bundles.

RNFLDs in ROTA images were identified on subjective assessment of RNFL reflectivity. Angular width and the location of RNFLDs were measured on the ROTA images using ImageJ version 1.54g (National Institutes of Health, USA; Figure 1 ). BMO locations were identified manually based on a previous study, but simplified in that we only use 4 (TSNI) BMO margins compared with 12 BMO margins in the previous paper. Pixel coordinates of the 4 BMO margins and the fovea, defined as the point with the smallest retinal thickness, were identified on original OCT images using the OCT viewer software. These locations were subsequently superimposed and depicted onto corresponding ROTA maps using ImageJ, referencing the predefined OCT coordinates.

Identification and measurement of RNFLD in ROTA images. A. Bruch membrane opening (BMO) margin was delineated (white circle) by transferring pixel coordinate values identified in the OCT B-scan images using ImageJ software. B. The fovea and the BMO center were marked on the ROTA image by transferring pixel coordinates of them identified on the OCT B-scan images to define the fovea-disc line (white line). C. An RNFLD border (white line) on the BMO circle is marked, and its angular location is calculated using ImageJ. D. Another RNFLD border (white line) was identified and its angular location was similarly calculated. OCT = optical coherence tomography, RNFLD = retinal nerve fiber layer defect, ROTA = retinal optical texture analysis.

The BMO margin was delineated with ellipse fitting, and the BMO center was defined as the midpoint of the major and minor axes of the outlined ellipse. Then the fovea-BMO center axis was delineated on the ROTA map as a reference for measuring the angular locations of the borders of RNFLDs. The angular location of the RNFLDs were depicted in the ROTA maps, and the angular width of the RNFLDs were calculated as the difference of the angular locations of the borders. Two graders (A.M., M.N.) independently evaluated the ROTA images. Discrepancies between the 2 graders were resolved by adjudication of a third grader (M.A.). The agreement of the initial 2 graders were evaluated using the kappa statistics.

The definitions and the measurement methodologies of the optic nerve head (ONH) parameters have been reported previously. ^,, In brief, RNFL thickness was measured on a magnification-corrected 3.4-mm-diameter BMO-centered annulus. The retinal pigment epithelium (RPE) edge, BMO and anterior scleral canal opening (ASCO) were manually segmented on reconstructed radial scans. BMO-minimum rim width was measured as the average of the shortest distance between the BMO and peripapillary inner limiting membrane on 24 radial scans. Other parameters measured from the OCT scans were BMO area/ovality, ASCO area/ovality, ASCO/BMO offset magnitude (representing the magnitude of ASCO centroid misalignment in reference to the BMO centroid), ASCO/BMO offset direction (representing the direction of misalignment of ASCO centroid in reference to the BMO centroid), and the parapapillary beta zone area and the gamma zone area.

Commercial artificial intelligence software, Reflectivity (Abyss Processing, Singapore), was used for lamina cribrosa (LC) parameter measurements. ^, LC depth was defined as the distance between the BMO plane and anterior LC surface at the BMO center. Prelaminar thickness was defined as the minimum distance between the anterior surface of the prelaminar tissue to the anterior surface of the LC. Scleral angle is the angle between 2 extended anterior scleral boundaries defined on a single horizontal B-scan in the nasal-temporal direction between 1200 and 1800 μm from the BMO centroid. The fovea-disc distance (FDD, distance between the fovea and the centroid of the ellipse-fitted disc contour) was measured on a shadowgram OCT image of the best macular OCT scan using ImageJ.

Descriptive statistics were computed to describe the study population. The number and the prevalence of RNFLDs as assessed by ROTA were calculated. Detection rate of RNFLDs was also calculated within early and moderate VF stage subgroups defined by the Hodapp-Anderson-Parrish criteria, as well as within highly myopic and non–highly myopic subgroups to assess the influence of stage and high myopia on the detective ability of ROTA. Based on previous literature, we defined high myopia as an axial length of 26 mm or greater. ^, Baseline factors of the highly myopic vs non–highly myopic subgroups are summarized in Supplementary Table S1 . As severe VF damage was one of the exclusion criteria for the study, there was no eye with severe VF damage. Total width of the RNFLDs (RNFLD-width) as the sum of angular width of all the RNFLD within a single eye and total number of the RNFLDs (RNFLD-number) as the sum of RNFLD counted within a single eye were calculated in each eye.

Correlation matrix among all parameters was first created ( Supplementary Table S2 ). Then, the best-fit multivariable linear mixed effects model (lowest corrected Akaike information criterion, AICc) was created to determine explanatory variables independently associated with RNFLD width or RNFLD number taking inter-eye correlations into account was created as described below. Explanatory variables were selected based on their clinical relevance and statistical meanings.

We took multiple VF parameters (MD, average antilog sensitivity of all VF test points, and antilog of MD) and OCT parameters (cpRNFLT, GCCT, and BMO-minimum rim width) in the study. Given the strong intercorrelation and shared underlying meaning among the VF parameters and OCT parameters, one representative parameter from each for the analysis was adopted as a surrogate marker of functional and structural glaucomatous damage, that is, the mean value of the anti-logarithmically transformed thresholds (in decibels) of all VF test points (average VF sensitivity (1/Lambert) and cpRNFL thickness measured on a BMO-centered annulus.

Other considered explanatory variables were baseline clinical factors including age, central corneal thickness (CCT), AXL, IOP, and image quality of the OCT, and ONH structural parameters including disc area, disc ovality, disc torsion, BMO area, BMO ovality, ASCO area, ASCO ovality, ASCO/BMO offset magnitude, ASCO/BMO direction, beta zone area, gamma zone area, LC depth, prelaminar thickness, scleral angle, and FDD. Linear mixed effects models of all possible combinations of these 22 parameters ([2 ²² – 1] models) were generated and the model with the lowest AICc was selected.

Because the degrees of freedom value in a multivariable regression model decreases with many variables, the use of model selection methods is recommended to improve the model fit by removing redundant variables. The other advantage of this method of the best fit model selection is that intercorrelations among the included explanatory variables do not need to be taken into account, except for those among the explanatory variables included in the best fit model. All statistical analyses were performed using the statistical programming language R (The R Foundation for Statistical Computing) and contribution of explanatory variables, with P <.050 considered statistically significant.