Abstract
Purpose :
The Global Visit Effects (GVE) model takes into account a large part of the variability in visual field (VF) data, thereby producing more robust and accurate progression rate (slope) estimations (Bryan et al. IOVS 2015). To evaluate the GVE model for detecting VF progression, we determined the sensitivity for this model compared to a reference model without the GVE.
Methods :
Two data sets (stable and progressing) were simulated, each consisting of 250 eyes with 52 VF locations, a varying number of VFs and intercepts based on data an early study (mean MD = -7.8 dB, IQR = -11.7 to -2.3 dB). All locations in the stable group had a slope of 0 dB/year. In the progressing group all locations had a slope of -0.5 dB/year. An ageing effect of -0.1 dB/year was assumed for all cases. Censoring (at 0 dB) and a sensitivity-dependent measurement error were taken into account.
For the analysis, censored Bayesian hierarchical linear regression models were used. Point-wise sensitivity losses (dB/year) were estimated for both data sets with both models using 7 VFs. Limits of stability for a 95% specificity were determined from the stable group. In the progressing group, locations were then classified as progressing if the loss exceeded this limit. To determine the effect of the number of VFs, this was repeated for a varying number (3-7) of VFs and different progression rates.
Results :
At a specificity of 95%, the sensitivity to detect progression in a series of 7 VFs increased from 20% to 80% by including the GVE (Figure 1). The sensitivity to detect progression improved for both models by including more VFs (Figure 2). However, the GVE model performed better than the reference model over all scenarios.
Conclusions :
By including GVEs in the model, we were able to greatly improve the sensitivity to detect VF progression in simulated data. The more robust slope estimations from the GVE model enable us to correctly detect more locations as progressing. These results suggest that it would be useful to apply to the GVE model to real data for detecting disease progression.
This is an abstract that was submitted for the 2017 ARVO Annual Meeting, held in Baltimore, MD, May 7-11, 2017.