Spatial Patterns of Preclinical Visual Fields in Glaucoma

Abstract

Purpose : To investigate spatial patterns of preclinical visual fields and their association with future development of glaucoma.

Methods : We selected reliable 24-2 preclinical VFs for each eye from a multi-center dataset. Preclinical VFs were defined as mean deviation (MD) ≥ -1 dB, glaucoma hemifield test (GHT) within normal limits and pattern standard deviation (PSD) probability > 5%. An unsupervised artificial intelligence method termed archetypal analysis was applied to determine the preclinical VF patterns from total deviation values. The VF patterns and global indices (age, MD, PSD) of the baseline VF were associated with the incidence of subsequent glaucoma diagnosis (defined by MD ≤ -3 dB and abnormal GHT and PSD probabilities on VFs measured at least 3 months from the baseline) using Cox survival regression with mixed effects addressing the issue of inter-eye correlations. Model selection was performed using Bayesian information criterion (BIC).

Results : We determined 10 preclinical VF archetypal patterns from 44,410 VFs including two variants of normal VF archetypes (ATs 1 and 5), superior and inferior loss (ATs 4 and 9), superior and inferior peripheral loss (ATs 10 and 8), nasal and temporal loss (ATs 3 and 7), nasal-temporal peripheral loss (AT 6) and central scotoma (AT 2). Among the 17,192 eyes with follow-up tests, 7.1% of the eyes were subsequently diagnosed with glaucoma at an average of 4.3 years with a standard deviation of 2.9 years. From univariable regressions, subsequent glaucoma diagnoses were positively associated (p < 0.001) with age, PSD, and ATs 3, 4, 6, 7, and negatively associated (p < 0.001) with MD and ATs 1, 2 and 5. From multivariable regressions with AT features, glaucoma diagnoses were positively associated (p < 0.001) with ATs 3, 4, 6, 7 and 9 after feature selection. When combining ATs with global indices, glaucoma diagnoses were positively correlated (p < 0.001) with age, PSD, and AT 3, and negatively correlated (p < 0.04) with ATs 1 and 8. AT 10 remained in the optimal model with an insignificant p value. The model combining the VF patterns with global indices strongly outperformed the model using global indices alone (BIC value lowering of 160; BIC lowering of > 6 indicates strong model improvement).

Conclusions : Preclinical VF patterns were assessed and quantified for the first time to our knowledge. These VF patterns can be used to improve prediction of future glaucoma diagnoses.

This is a 2021 ARVO Annual Meeting abstract.