Dual-Level Pattern Tree for Visual Field Characterization in Glaucoma Improves Predicting Progression and Polygenic Risk Score

Abstract

Purpose : To develop a dual-level pattern tree that more adequately represents visual field loss subtypes in glaucoma, which may better forecast progression and correlate with genetic variations.

Methods : Three datasets were used in this study: (1) the Glaucoma Research Network (GRN) dataset excluding the Massachusetts Eye and Ear (MEE)partition for the dual-level pattern tree model training, (2) the MEE longitudinal dataset for progression forecasting, and (3) the Nurses' Health Studies and Health Professionals Follow-up Study datasets for polygenic risk score (PRS) analyses. We first applied archetypal analysisto cluster 24-2 VFs into parent patterns and their own child patterns. The Cox regression modelforecasted the VF progression with four widely used progression definitions: slope of MD, MD-Fast, VFI, and TD pointwise. Pearson's correlations analyzed relationships between VF patterns and PRS.

Results : 182,548 VFs from 103,856 glaucoma patients in GRN were used to train the dual-level archetypal model. It contains 17 parent patterns with 118,059 VFs as the largest group (Parent 1) and 1,161 VFs as the smallest group (Parent 11). The second level contains a total of 169 child patterns (child pattern numbers under each parent pattern: 9.9 ± 1.6). Figure 1 shows all parent patterns and their child patterns. 119,856 VFs from 8,442 MEE patients were used for Cox progression prediction. Figure 2(A) shows parent-child patterns are consistently if modestly superior to parent patterns only for progression forecasting measured by the average area under the curve (average AUC): MD (0.62 vs 0.60), MD-Fast (0.79 vs 0.75), VFI (0.67 vs 0.64) and TD Pointwise (0.71 vs 0.69) with all p < 0.001. PRS relationship analysis with 732 VFs shows the child patterns are generally more strongly associated with PRS than parent patterns, as shown in Figure 2(B), in which the P3-C7 pattern (Child Pattern 7 under Parent Pattern 3) (R = 0.35; p = 0.003) shows a higher correlation than P3 (R = 0.15, p = 0.213) and the P7-C3 pattern (R = 0.57; p = 0.03) shows a higher correlation than P7 (R = 0.33, p = 0.24).

Conclusions : The dual-level pattern characterization for VFs predicts VF progression better and shows a stronger association with PRS which may facilitate a better understanding of glaucoma subtypes related to different progression trajectories and genomic variants..

This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.

 

Figure 1

 

Figure 2