Purchase this article with an account.
Susan R Bryan, Paul Eilers, Emmanuel M.E.H. Lesaffre, Hans G Lemij, Koenraad Arndt Vermeer, Rotterdam Ophthalmic Institute; Sensitivity improvement in detecting visual field progression by modeling Global Visit Effects. Invest. Ophthalmol. Vis. Sci. 2015;56(7 ):3182.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Previously, we proposed to model the spatially correlated errors which affect all locations from the same VF as Global Visit Effects (GVE) (Bryan et al., ARVO Abstract 3007, 2014). The model including the GVE provides a more robust estimate of the rate of progression than the conventional model, which does not contain the GVE. To evaluate the GVE for detecting VF progression, we determined the increase in sensitivity for the GVE model compared to the conventional model.
Series of 6 VFs (24-2 Full Threshold) of 30 eyes of 15 normal individuals from an ongoing study at the Rotterdam Eye Hospital were included in the healthy group. The VFs of each healthy individual were randomly resampled with replacement to obtain a total of 100 eyes from 50 individuals. Series of 15 VFs of 250 eyes of 125 individuals with primary glaucoma were included in the glaucoma group (data available at http://rod-rep.com).<br /> <br /> Limits of stability for a certain specificity were determined from the estimated sensitivity loss per year for each VF location in the healthy group. In the glaucoma group, sensitivity loss was computed from 6 VFs for each VF location. Locations were then classified as progressing based on the limits obtained from the healthy group (see Figure 1). Because no ground truth was available, the true sensitivity loss was estimated from the full series of 15 VFs for each location. The criterion for progression was then defined by using a varying aging effect factor multiplied by the average sensitivity loss for the glaucoma group.
For a specificity of 95% and an aging effect factor of 3, the sensitivity to detect progression in a series of 6 VFs increased from 47% to 62% by including the GVE. The sensitivity of the GVE model was approximately 15% higher than the conventional model for the tested specificities and aging effect factors (see Figure 2).
By including the GVE, we were able to improve the sensitivity to detect VF progression. The GVE takes into account a large part of the variability, produces more robust slope estimations and hence is able to correctly detect more locations as progressing.
This PDF is available to Subscribers Only