Baseline Visual Field (VF) Patterns are Predictive of Global and Central Loss in 24-2 Visual Fields

Abstract

Purpose : To study whether and how visual field (VF) patterns are related to global and central loss in 24-2 VFs.

Methods : Eyes with both ≥ 5 reliable 24-2 VFs and 5-year follow-ups (intervals ≥ 6 months) were selected. One eye was included randomly if both eyes qualified. 2 types of features from 2 baseline VFs were extracted: (1) Pattern features, including 16 VF pattern coefficients quantified by 16 VF archetypes (ATs) derived by machine learning in our prior work (Fig. 1), and the mean squared error (MSE) of VF archetype reconstruction; (2) VF global indices, including baseline mean deviation (MD) and pattern standard deviation (PSD), MD and PSD differences between the 2 baseline VFs, glaucoma hemifield test (GHT), and the total deviation (TD) difference (MSE) between 2 baseline tests. Linear regression was applied to predict the MD slope and slope of average TD at central 4 locations adjusting for baseline age and follow-up time. Stepwise regression via Bayesian information criterion (BIC) was used to select the optimal features.

Results : For the 7,360 eyes from 7,360 patients, the baseline age and MD, follow-up time, and the number of VFs were 65.2±12.7 years, -4.4±5.2 dB, 7.5±1.9 years and 6.6±1.7, respectively. The correlation between MD slope and central TD slope was 0.79 (P<0.001). Eyes with higher coefficients of temporal wedge (AT 4), near total loss (AT 6), central scotoma (AT 7), concentric peripheral defect (AT 11) and TD difference were less likely to develop MD worsening (Fig. 2a). MD worsening was related to higher AT MSE, abnormal baseline GHTs, and worsened MD/PSD between 2 baselines. Central TD worsening was more associated with higher coefficients of superonasal step (AT 3), inferior altitudinal defect (AT 13), inferior paracentral defect (AT 16) and AT MSE, and was less likely for eyes with higher coefficients of near total loss (AT 6), central scotoma (AT 7) and TD difference (Fig. 2b). Abnormal baseline GHTs, and worsened MD between 2 baselines were associated with central TD worsening. Compared to VF global indices alone, adding VF pattern features improved the prediction of MD slope and mean central TD slope measured by a decrease in BIC with 18 and 78 (lowered BIC > 6: strong model improvement).

Conclusions : Adding VF pattern features improves the prediction of the global and central 24-2 VF worsening, which might result in better glaucoma progression detection.

This abstract was presented at the 2019 ARVO Annual Meeting, held in Vancouver, Canada, April 28 - May 2, 2019.

 

Fig. 1

 

Fig. 2