An Exponential Regression Model to Measure Perimetric Rates of Decay in Glaucoma

Abstract

Purpose : To develop and validate a Progression Index (PI) for estimating the rates of visual field (VF) decay in glaucoma.

Methods : Eyes in the AGIS VF database with at least 6 VF exams and 3 years of follow-up were included. A pointwise exponential regression (PER) was used to model the rates of change at each VF location. The sums of those rates were normalized on a scale of -1 to 1, defining a progression index (PI): 0 is no change, negative values represent decay and positive values improvement. The extremes of -1 and 1 were defined with simulated VF series. To determine if different techniques correlate, the raw AGIS score, Mean Deviation (MD) rate of change, and PI change were linearly regressed against each other. Eyes with PI scores of < −0.03 were defined as progressing; this value was based on an ROC curve developed from a separate VF database of three expert evaluations. AGIS progression was defined as 3 consecutive VF scores of >4 from baseline, or a score of >19. Progression based on MD rate was set at a rate of >0.53 dB/year (Gustavo De Moraes et al 2016).

Results : Serial VFs of 509 eyes of 402 patients were included. Mean (±SD) follow-up was 7.3 ± 1.7 years with 15.4 ± 3.8 VFs. There was a statistically significant correlation between the AGIS score and the rate of change in the mean deviation (r² =0.4, P<0.001), PI and MD rate of change (r² = 0.6, P<0.001) and between the PI and the rate of change of the AGIS score (r² = 0.4, P<0.001) (Figure 1). The proportion of eyes progressing with each technique was: AGIS (152, 29.9%), PI (214, 42.0%), and MD rate (176, 34.6%) (Figure 2).

Conclusions : Visits that scored worse on PI correlated with worse AGIS scores and MD rates of change in a statistically significant manner. This was true across a wide range of disease severity with long follow-up. PI identified a higher proportion of progressive fields compared to the AGIS criterion. MD rate seemed less specific than PI or AGIS scores to identify true VF worsening. Future work will determine relative sensitivities and specificities of these and other approaches, and determine the best approach to identify early progression.

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

 

Plots correlationing AGIS score, Mean Deviation rate and Progression Index. Linear fit with R² values (P<0.001) are reported. Visits scoring worse on PI statistically correlated with AGIS scores and MD change rates

 

Venn diagram showing the proportion of eyes identified as progressing.