Comparison of the Global Visit Effect model and Pattern Deviation over a varying number of affected visual field points

Abstract

Purpose : One of the difficulties in modeling visual field (VF) data is the sometimes large and correlated measurement errors (‘visit effects’) in the point-wise sensitivity estimates. Previously, we proposed to model these errors from the same VF as Global Visit Effects (GVEs) (Bryan et al. IOVS 2015). For the pattern deviation, the VF is adjusted such that the 85^th percentile of the VF matches the 85^th percentile of a normal, reference VF. This is done to compensate a general reduction of retinal sensitivity throughout the VF (due to, for example, cataract), but may hide diffuse progression by doing so. Our aim was to evaluate and compare the performance for estimating progression for both approaches as a function of the number of affected VF locations.

Methods : For the analysis, we simulated 100 eyes, each with 7 VFs. Censoring (at 0 dB) and a sensitivity-dependent measurement error were taken into account. Censored Bayesian hierarchical linear regression models were used. An example of the percentile and GVE models for one simulated eye is shown in Figure 1. The mean absolute errors (MAE) in the slopes (simulated – estimated) for the two models were computed for a varying number of damaged locations, different disease severities and progression rates. Disease severity was defined as severe, moderate or mild for an intercept at presentation of 10, 20 and 30 dB, respectively. Progression rates of -1, -2 and -3 dB/year were used.

Results : The MAE of the estimated slopes were similar when only a limited number of points progressed. It was lower for the GVE model than for the percentile model after 20, 30 or 40 locations were progressing for cases with severe, moderate and mild glaucoma, respectively. This was irrespective of the progression rate.

Conclusions : The pattern deviation (adjustment at 85^th percentile) model overcorrects the entire VF height when a number of points are affected by loss (depending on the level of loss), thereby masking progression. In contrast to the pattern deviation, the GVE model allows us to compensate for visit effects without disrupting the estimation of the progression.

This is an abstract that was submitted for the 2017 ARVO Annual Meeting, held in Baltimore, MD, May 7-11, 2017.

 

Figure 1: Illustration of the two main components which make up the model fit and how these differ between the percentile and GVE models

 

Figure 2: Mean absolute errors in estimating the slopes for the percentile and GVE models over a varying number of damaged locations within each eye