Abstract

Purpose: Accurately identifying moderate to severe cases of glaucoma is crucial to prevent irreversible glaucoma-related blindness. Assessing glaucoma severity relies on multimodal data from multiple sources, including demographics, clinical assessment, optical coherence tomography (OCT), and visual fields (VF), which can be represented by standardized data concepts. In some eye care environments, not all modalities are available or practical; VFs are particularly time intensive. We test the relative importance of different diagnostic domains of standardized data concepts at differentiating moderate to severe glaucoma (VF mean deviation [MD]≦-6) from mild glaucoma (MD>-6) or glaucoma suspects.
Methods: 736 patients diagnosed with primary open angle or angle closure glaucoma or glaucoma suspects by fellowship-trained glaucoma specialists at the USC Roski Eye Institute were identified using ICD-10 codes. Patients were stratified by MD≦-6 and an ICD-10 glaucoma code (N= 145) vs MD>-6 or an ICD-10 glaucoma suspect code (N= 591). Logistic regression models were developed using demographic (age, gender, ethnicity), clinical (visual acuity [VA], intraocular pressure, cup-to-disc ratio [CDR]), OCT (retinal nerve fiber layer [RNFL] thickness), and visual field (mean deviation [MD], glaucoma hemifield test [GHT]) domain data. Performance was assessed in a test set (N= 146) using area under the curve (AUC).

Date

June 30, 2025

Authors

Kimberley Yu, Zhiwei Li, Sreenidhi Munimadugu, Carl Kesselman, Ryan Scott Shean, Alexander T Hong, Tanner Frediani, Alan Tang, Nova Dea, Jose-Luis Ambite, Michael Pazzani, Kyle Bolo, Lauren Daskivich, Sophia Y Wang, Michael V Boland, Swarup Sai Swaminathan, Brian Craig Stagg, Aiyin Chen, Sally Liu Baxter, Benjamin Xu

Journal

Investigative Ophthalmology & Visual Science

Volume

66

Issue

8

Pages

4585-4585

Publisher

The Association for Research in Vision and Ophthalmology