Identifying Visual Field Progression with High Dimensional Analysis of Variance

Abstract

Abstract: : Purpose: To identify visual field progression with a mathematical technique that takes into account the spatial correlations between test locations. Methods: Fifty eyes from 50 patients from the Advanced Glaucoma Intervention Study (AGIS) with ≥ 14 Humphrey 24–2 visual field (VF) exams over ≥ 6 years of follow up were evaluated. None of the eyes had a history of cataract extraction. All exams had an initial AGIS score of ≤ 17 and a reliability rating of 2 or better. The AGIS scoring system and Pointwise Linear Regression (PLR) were independent criteria used to identify progression. Our PLR criteria classified VF series as progressing if ≥ 2 points in a Glaucoma Hemifield cluster had a slope ≥ 1 dB/yr and p ≤ 0.01. The blind spot and the two most nasal locations were not used. A Fourier transform was applied to each series to reduce the dimensionality of the data. To decrease high frequency noise in a series, only the least noisy 1/64th of the transformed data was analyzed by applying Fan’s Adaptive Neyman test and the p value was calculated. Results: The patients had a mean follow up of 8.6 years (± 1.0) with an mean of 18.4 (± 2.1) VF exams. AGIS identified 25 eyes (50%) and PLR identified 17 eyes (34%) as progressing while the Neyman test identified 20 eyes (40%) as progressing (p < 0.5). Thirteen of the 25 eyes (52%) identified as progressing and 18 of the 25 eyes (72%) identified as stable by the AGIS criteria were also identified by the Neyman test. The kappa between the AGIS and the Neyman test was 0.24 (62% agreement). The kappa between PLR and the Neyman test was 0.58 (79% agreement). Conclusions: Visual field progression may be more accurately detected by taking into account spatial correlations between VF test locations. The Adaptive Neyman test takes into consideration the spatial correlations and provides new information not otherwise available. This new method shows fair agreement with the AGIS standard and moderate agreement with PLR to detect visual field progression.  

Keywords: visual fields