Abstract
Purpose :
To compare the variability of multiple visual fields (VFs) results obtained using Compass fundus perimeter (CMP, Centervue, Italy) with and without the application of a dedicated post-processing filter based on the spatial interpoint correlations of thresholds values
Methods :
45 patients with glaucoma (Mean Sensitivity Range [5.81; 28.63] dB; Mean Deviation Range [-22.03; 0.98] dB) underwent six VFs with CMP over a period of about two months on one eye (randomly chosen). 24-2 VFs with additional macular points (24-2+ grid) were collected using Zippy Estimation by Sequential Testing (ZEST) algorithm. Unreliable visual field tests (FP > 18%, FN > 30%) were excluded and VFs assessed at the first visit were not included to avoid learning effect. A total of 225 VFs (5 VFs for each patient) were analyzed before and after the post-processing elaboration for both the 24-2+ Grid and the standard 24-2 grid (subset of 24-2+). Test-retest variability of 180 couples of VFs was evaluated both globally with Mean Sensitivity (MS) and pointwise (thresholds) with Bland-Altman technique. The results were compared in order to evaluate the impact of the filter procedure on test-retest variability
Results :
Variability of unfiltered CMP data changed as a function of the thresholds values. Pointwise test-retest limits of agreement for the 24-2+ and the 24-2 grids were reduced by 22.5% and 25.9% respectively (Figure 1). Global test-retest variability of MS was increased by 3.25% and 3.53% respectively (Figure 2). The Bland-Altman dynamic range of pointwise thresholds pre and post filter was the same: [-1; 37] dB for 24-2+ grid and [-1; 35] dB for 24-2 grid. For the MS the dynamic range pre and post filter was [6.5; 28.3] dB and [6.2; 28.2] dB respectively for 24-2+ grid and [4.9; 27.4] dB and [4.5; 27.2] dB respectively for 24-2 grid
Conclusions :
The filter procedure improves pointwise test-retest variability, preserving the dynamic range. As expected, the effect on MS was minimal. Further validation on longitudinal series is needed to assess its effect on progression detection
This is a 2020 ARVO Annual Meeting abstract.