Abstract
Purpose :
To investigate whether specific binocular patterns of visual field loss, derived from unsupervised machine learning classification, are related to patient-reported disability in glaucoma.
Methods :
The study included 677 patients with glaucomatous visual field damage in at least one eye. Patient-reported vision-related quality of life was assessed by the National Eye Institute Questionnaire (NEI VFQ-25) questionnaire. A latent class analysis procedure was used to divide subjects into two classes (disabled versus non-disabled) based on NEI VFQ-25 results. Total deviation values for the right and left eyes were extracted from the 52 locations measured by standard automated perimetry (SAP), and the binocular summation was done following a previously described method. An unsupervised machine learning method was used to classify binocular visual field abnormalities into different patterns.
Results :
One hundred eleven of the 677 (16.4%) subjects were classified as having vision-related disability based on NEI VFQ-25 results. Among the 5 patterns of binocular visual field loss, 4 of them showed significant association with disability (P<0.001, P=0.008, P=0.004, and P=0.009, respectively).
Conclusions :
Machine learning-derived binocular patterns of visual field loss are related to patient-reported vision-related quality of life in glaucoma.
This abstract was presented at the 2019 ARVO Annual Meeting, held in Vancouver, Canada, April 28 - May 2, 2019.