Pointwise Modeling for Predicting Visual Field Progression in Korean Glaucoma Patients
10.3341/jkos.2022.63.11.918
- Author:
Jae Hyun KIM
1
;
Sang Woo MOON
;
Choongrak KIM
;
Jiwoong LEE
Author Information
1. Department of Ophthalmology, Pusan National University School of Medicine, Busan, Korea
- Publication Type:Original Article
- From:Journal of the Korean Ophthalmological Society
2022;63(11):918-927
- CountryRepublic of Korea
- Language:Korean
-
Abstract:
Purpose:To evaluate the utility of pointwise modeling for predicting visual field (VF) progression in Korean glaucoma patients.
Methods:Open-angle glaucoma or glaucoma suspect patients with VFs ≥ 10 times, who were followed-up for ≥ 6 years, wereincluded. Linear, exponential, and polynomial regression of threshold values at each test point against time were performed. Model fit was evaluated based on root mean squared error (RMSE) for the entire longitudinal VF series. To evaluate prediction ability, VFs from the first 5 years were used to estimate model parameters, followed by calculation of threshold values for 1, 2, 3, 5 years to obtain RMSE. Prediction ability was compared regarding initial threshold value and also central and peripheral VF area.
Results:Four hundred thirty-nine eyes (280 patients) were included. The mean follow-up duration and number of VF tests were 9.64 years and 13.02, respectively. When fitting the entire VF series, polynomial model had the lowest RMSE (p < 0.001). For 1-year predictions, linear model had the lowest RMSE, while exponential model had the lowest RMSE for 3- and 5-year predictions (p < 0.001). For 1- and 2-year predictions, exponential and linear models had the lowest RMSEs, with initial sensitivities of 0-7 and 20-27 decibel (dB), respectively (p < 0.001). Compared to exponential model, linear model had lower RMSE for 1-year, but higher RMSE for 3- and 5-year at peripheral VF area (p < 0.001). For central VF area, exponential model had lower RMSEs for 2-, 3-, and 5-year predictions compared to linear model (p ≤ 0.015).
Conclusions:The linear model outperformed the exponential model for short-term predictions, while the exponential model was better for long-term predictions. The prediction performance of the exponential model was superior to that of the linear model for central VFs, and for test points with lower initial sensitivities.
