Abstract
Purpose: :
To develop and evaluate a methodology to reduce variability in glaucoma follow-up by linking retinal structure and visual function measurements.
Methods: :
A model to predict the visual field (VF) from a retinal nerve fibre layer thickness image was applied to a test-retest dataset. The model was developed using a Bayesian Radial Basis Function from measurements of 535 subjects from 3 centres (Zhu et al; IPS2006). The test-retest dataset comprised 48 glaucomatous eyes with 5 repeat GDxVCC scans and Humphrey SITA VF tests. A combined VF (CVF) was calculated as the pointwise weighted mean of the GDx-predicted and the measured VF, where the weighting was derived from a structure-function concordance index (Zhu et al; IPS2008). Reproducibility of CVF was compared against that of single-, and the mean of 2-, VFs, repeated. We also examined the false positive (FP) rate, when detecting progression, in simulated no-change time series where the 5 repeat tests were multiply re-ordered and assumed to be taken over 2 years. Progression was defined as ≥2 contiguous points decreasing by ≥1dB/year at p<0.05 in linear regression trend analysis.
Results: :
CVF showed much better reproducibility than single VF and was better than the mean of 2 VFs repeated (p<0.01 in Wilcoxon test on standard deviation (SD) of mean sensitivities; Figure). The SD of individual thresholds in repeat CVFs was smaller than that of single VFs (p<0.0001). CVF had lower FP rate (3.3%) compared to the single VF series (5.5%; Figure).
Conclusions: :
Sources of variability in structural and functional measurements are different, so combining them should reduce overall variability. Combining a VF predicted from structure with a real VF considerably reduced variability compared with single, and the mean of 2, VFs. This suggests that taking an image with a VF is better than repeating the VF on the same day. The lower FP rate indicates the CVF may improve progression detection.
Keywords: visual fields • nerve fiber layer • imaging/image analysis: clinical