Purchase this article with an account.
David F Garway-Heath, Haogang Zhu, David Paul Crabb, ; Imaging improves the accuracy of visual field progression analysis in glaucoma: structure-guided ANSWERS. Invest. Ophthalmol. Vis. Sci. 2015;56(7 ):2058.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
A new, more accurate method for visual field (VF) progression analysis, called ‘ANSWERS’ (Analysis with Non-Stationary Weibull Error Regression and Spatial Enhancement; Zhu et al, PLoS ONE 2013), is a form of ‘robust regression’ which takes into account increasing measurement variability as glaucoma progresses. The rate of structure progression can be incorporated, as a prior probability distribution, with VF measurements in a Bayesian framework: structure-guided ANSWERS (sANSWERS). The rate of progression may be used to predict future VF sensitivity. The difference between the prediction and true VF is a measure of the accuracy of the progression rate estimate. The aim was to compare the accuracy of sANSWERS, ANSWERS and ordinary linear regression (OLR) in predicting future VFs .
VF series of up to 2 years from participants in the UK Glaucoma Treatment Study (Garway-Heath et al, Ophthalmology 2013) were selected. Only VFs with false positive responses <15% and observation period >6 months were used, resulting in 9104 VFs from 659 glaucomatous eyes of 437 participants. SITA standard 24-2 VFs were acquired with the Humphrey Field Analyzer (Carl Zeiss Meditec, CA) and optic disc rim areas (RA) were acquired with the Heidelberg Retina Tomograph (Heidelberg Engineering, Germany). sANSWERS (with the RA rate prior), ANSWERS and OLR were used to predict the VF at the next visit using subseries that are within 7, 13, 18 or 22 months from the baseline visit. The averaged prediction error across the VF locations was compared.
Across all subseries, ANSWERS produced significantly (p<0.01%, Wilcoxon signed rank test) more accurate predictions than OLR. The prediction accuracy of sANSWERS was better than ANSWERS at all series lengths and the improvement was particularly marked in shorter series (Figure). 75% and 86% of VFs were better predicted by ANSWERS and sANSWERS, respectively, compared with OLR. The average prediction error of ANSWERS was 18% lower, and sANSWER 33% lower, than that of OLR.
ANSWERS predicts future VF loss better than OLR across a wide range of observation periods. sANSWERS further improves the prediction by incorporating estimates of structure progression. The gain in accuracy is especially large over short observation periods.
This PDF is available to Subscribers Only