This study recruited consecutive primary open-angle glaucoma (POAG) patients who had been followed by a glaucoma specialist (M.H.S.) between October 2019 and March 2025. This study was approved by the Haeundae Paik Hospital Institutional Review Board, and informed consent was waived due to the retrospective nature of the study. This study adhered to the Declaration of Helsinki.

Initial ophthalmic examinations included best-corrected visual acuity (BCVA), IOP measured by Goldmann applanation tonometry, refraction, slit-lamp biomicroscopy, gonioscopy, and central corneal thickness (CCT) measurement using the Pentacam Scheimpflug imaging system (Oculus Optikgeräte GmbH, Germany), axial length (AXL) was measured using IOL Master (Carl Zeiss Meditec, Dublin, CA, USA), dilated stereoscopic examination of the optic disc, simultaneous color and red-free fundus photography (TRC NW8; Topcon, Tokyo, Japan), standard automated perimetry (SAP) (Humphrey Field Analyzer; 24-2 Swedish Interactive Threshold Algorithm; Carl-Zeiss Meditec), spectral-domain optical coherence tomography (SD-OCT) (Spectralis; Heidelberg Engineering GmbH, Germany), SS-OCTA (Carl Zeiss Meditec, Germany).

To be included in the study, subjects were required to have POAG, to have best corrected visual acuity ≥ 20/30, to be aged > 18 years, and to have ≥ 5 reliable SD-OCT, SS-OCTA, and VF tests conducted ≥ 3 years at 4 to 6-month intervals. All VF tests and OCT/OCTA imaging sessions had to be performed within 4 months of each other. POAG was defined as an open angle by gonioscopy and glaucomatous optic disc changes (ie, neuroretinal rim thinning, notching, or an RNFL defect) with a compatible VF defect. Glaucomatous VF defects were defined as (1) outside the normal limits on a glaucoma hemifield test, (2) 3 abnormal points with a < 5% probability of being normal, with one point with 1% by pattern deviation, or (3) pattern SD (PSD) of P < 5% on 2 consecutive reliable tests (≤33% fixation losses and false negatives, ≤15% false positives). ^,,, If ocular surgery (eg, cataract or glaucoma surgery) was performed during study follow-up, only data before the surgery were included. In addition, participants with a history of intraocular or neurologic diseases (eg, myopic maculopathy, pituitary tumor) that could lead to VF loss were excluded. Subjects with diabetes mellitus and/or systemic hypertension without diabetic or hypertensive retinopathy were included.

Glaucomatous eyes were categorized into early (mean deviation (MD) ≥ −6 dB), moderate (−12 dB ≤ MD < −6 dB), and advanced (MD > −12 dB) based on Hodapp-Parish-Anderson grading scale. If both eyes of the subject met the inclusion and exclusion criteria, eyes categorized as advanced stage were preferentially selected, followed by those with moderate and early glaucoma.

IOP at baseline was measured before IOP-lowering treatment or as indicated in the referral notes, and the average and fluctuation of IOP were calculated as the average and SD of all IOPs obtained at 6-month intervals during the entire follow-up period. ^,,, DH, defined as an isolated splinter- or flame-shaped hemorrhage of the optic disc, was recorded by 2 masked observers (M.H.S. and Y.J.L.) during the follow-up period on both fundus photographs and clinical examinations performed at intervals of 3 to 6 months. ^,,,,,,,

Based on Spectralis OCT2 enhanced depth imaging 20° × 20°-sized scan pattern comprised of 48 radial B-scan, focal LC defect defined as laminar holes or disinsertion violating the normal U- or W-shaped contour of the anterior laminar surface and β-zone defined as an area devoid of retinal pigment epithelium with a temporal width ≥ 200 µm on at least 1 radial scan were determined by the 2 masked observers (M.H.S. and Y.J.L.). ^,,,,

Circumpapillary RNFL thickness was measured utilizing a 3.5 mm-sized circle of Spectralis OCT2 Glaucoma Module Premium Edition software, and RNFL segmentation accuracy was reviewed and errors were corrected manually by one masked reviewer (J.W.S.). ^, The Spectralis built-in automated realignment procedure of SD-OCT, described in the system documentation (referred to as the follow-up examination), ensures that RNFL scans are acquired at the same location for each follow-up scan. ^,,, Poor quality images due to low quality score (<15), poor clarity, focal weak signal, segmentation failure, or falsely thick RNFL (ie, retinoschisis) were excluded.

VF progression was determined using both event- and trend-based analysis. Event-based progression based on Glaucoma Progression Analysis criteria was defined as “likely progression,” having ≥ 3 locations flagged as showing significant deterioration across ≥ 3 consecutive tests, with these changes being observed at the latest follow-up visit. ^,,, The trend-based analysis was defined as having a significant negative slope of either VF index (VFI) or VF MD ( P <.05). ^,, To mitigate uncertainty in identifying VF progression, eyes with ‘‘possible progression’’ having ≥ 3 test points with significant deterioration over 2 consecutive tests but lacking a significant negative slope in VFI or VF MD, were excluded from the analysis.

The optic nerve head (ONH) and peripapillary areas were imaged using OCTA which is integrated into the PLEX Elite 9000 SS-OCT producing 6.0ⅹ6.0 mm ² cubes. ^,,,, Details of the SS-OCTA imaging have been described elsewhere. ^,,,, Briefly, image acquisition and processing were performed using the device’s proprietary Optical Microangiography (OMAG) algorithm (software version 2.0.1.47652). To mitigate projection artifacts and potential motion during imaging, built-in projection artifact removal techniques and FastTrac motion tracking were employed. ^,,,, Poor-quality SS-OCTA images (eg, signal strength < 7, segmentation error, poor clarity, motion artifact, poor visualization of deep-layer microvasculature [ie, shadowing of neuroretinal rim or superficial vessels, limited penetration of OCT beam]) were excluded. ^,,,, ODVD and PDVD were measured by the two masked reviewers (M.H.S. and H.J.K.) using whole-signal-mode images below the internal limiting membrane (ILM) ( Figure 1 , B). ^,, Given that parapapillary deep-layer vascular dropout is accompanied by full-thickness dropout, including the superficial retinal layer, PDVD can be measured based on all OCTA signals below the ILM. En-face SS-OCTA images were binarized using Otsu’s method with the aid of ImageJ software (National Institutes of Health, Bethesda, MD, USA; https://imagej.nih.gov/ij/ ) relying on the principle by finding an optimal threshold to separate pixels into two classes (foreground and background) by minimizing intra-class variance based on a bimodal distribution ( Figures 1 , C and 2 , A2, and Figure 3 , A2). ^,,,,,

Measurement of optic disc vessel density (ODVD) and parapapillary deep vessel density (PDVD) using swept-source optical coherence tomography angiography (SS-OCTA). Fundus photography ( A ), enface SS-OCTA image ( B ), and a binarized OCTA image using Otsu’s method ( C ). B2 shows the horizontal OCTA images overlaid on B-scan showing all OCTA signals below internal limiting membrane surface (yellow dotted lines) on the enface SS-OCTA image ( B1 , blue line). C2 is the same as C1 , marking the optic disc margin (green circle), temporal half of the optic disc cup for ODVD (red dotted semicircle), β-zone area (red solid semicircle), and the line connecting the fovea-Bruch’s membrane opening center for dividing the superior and inferior hemispheres (blue dotted line) ( C2 ).

Left eye of a 56-y-old male with advanced primary open-angle glaucoma (visual field (VF) mean deviation (MD) at baseline of −15.93 dB) that progressed on VF and optical coherence tomography (OCT) angiography during 4 y of follow-up. Serial swept-source OCT angiography images before ( A1 ) and after image binarization using Otsu’s method ( A2 ), and scatterplot showing the changes of optic disc vessel density (ODVD) ( B ), serial standard automated perimetry grayscale plot ( C ) and Guided Progression Analysis ( D ), and scatterplot showing the rate of visual field index (VFI) ( E ), OCT circumpapillary retinal nerve fiber layer (RNFL) scan ( F ), and scatterplot showing the changes of global RNFL thickness ( G ). Note that remarkable ODVD deterioration (red arrows, A and B ) and subsequent VF progression (blue arrows, C, D , and E ) were observed, whereas there was no notable change in RNFL thickness on OCT ( F and G ).

Left eye of a 60-y-old female with advanced primary open-angle glaucoma (visual field (VF) mean deviation (MD) at baseline of −12.36 dB) that progressed on VF and optical coherence tomography (OCT) angiography during 5.3 y of follow-up. Serial swept-source OCT angiography images before ( A1 ) and after image binarization using Otsu’s method ( A2 ), and scatterplot showing the changes of optic disc vessel density (ODVD) ( B ), serial standard automated perimetry grayscale plot ( C ) Guided Progression Analysis ( D ), and scatterplot showing the rate of visual field index (VFI) ( E ), OCT circumpapillary retinal nerve fiber layer (RNFL) scan ( F ), and scatterplot showing the changes of global RNFL thickness ( G ). ODVD deterioration (red arrows, A and B ) and subsequent VF progression (blue arrows, C, D, and E ) in the superior and inferior areas were observed, whereas there was no notable change in RNFL thickness on OCT ( F and G ). Note that VF showed high long-term fluctuation (orange arrows) despite a statistically significant negative slope in the trend-based analysis ( P =.045).

Both ODVD and PDVD quantification were computed as the ratio of black pixels representing vessels to the total pixels in the region of interest (ROI) located in the temporal ONH and parapapillary areas ( Figure 1 , C2). ^, The ROI was manually designated as the cup area after excluding the neuroretinal rim for ODVD, and as the β-zone or the parapapillary area within 200 µm (minimum width of the β-zone ) of the optic disc margin for cases not exhibiting a β-zone for PDVD ( Figure 1 , C2). ^, ODVD and PDVD measurements were categorized into global, superior, and inferior areas based on the line connecting fovea and BMO center ( Figure 1 , C2). ^, To assess the rate of change of ODVD and PDVD over time (expressed in % per year), linear regression analysis was performed against time. Yes-or-no binary variables of ODVD reduction were defined as a statistically significant negative slope ( P <.05) for any of the global, superior, or inferior areas. ^, Disagreements between the 2 masked reviewers regarding parameters such as DH, focal LC defect, and ODVD reduction were resolved by consensus. If consensus could not be reached, the subjects were excluded. Consensus could not be reached in 8 subjects for determining focal LC defect, 4 for DH, and 10 for ODVD reduction.

For comparing demographics and baseline characteristics among the three groups, a one-way analysis of variance (ANOVA) test was employed for continuous variables, and the Chi-square test for categorical variables. Logistic regression analysis was conducted to identify factors associated with VF progression. To assess the variables obtained longitudinally with variable follow-up duration and differing number of visits per eye, longitudinal changes were analyzed using a linear mixed-effects (LME) model with random intercepts and slopes to account for within-eye correlation and variable follow-up duration. ^, The interaction term (time × progression status) was used to quantify differences in longitudinal rates of change between progressing and non-progressing eyes (interaction β). Additionally, the Cox Proportional Hazards (PH) model was performed for the time-to-progression analysis. Variables were included in the multivariable model using a backward elimination approach when they exhibited a P value <.1 in the univariable regression. Simulation-based power analysis demonstrated adequate power (≥80%) for all analyses except the Cox PH model in the moderate and advanced groups, which showed limited power (<70%) and model instability for most covariates. Therefore, moderate and advanced eyes were pooled for the Cox PH model to improve statistical robustness. In addition, the Cox PH model with interaction term of the disease severity was evaluated across all 345 eyes.

Variables were included in the multivariable model using a backward elimination approach when they exhibited a P value <.1 in the univariable regression. Inter-observer agreement for DH, focal LC defect, and binary yes-or-no variable of ODVD reduction, and inter-observer reproducibility of the ODVD and PDVD values were measured using κ values and intraclass coefficient (ICC), respectively. Statistical analysis was conducted using MedCalc (MedCalc, Inc., Mariakerke, Belgium) and R software version 4.2.3 ( R Project for Statistical Computing, Vienna, Austria), with P values <.05 considered statistically significant.