ABSTRACT Purpose To compare concordance between ganglion cell‐inner plexiform layer metrics acquired using widefield optical coherence tomography (OCT) and visual function assessed using Goldmann (G) II, III, and V stimulus sizes, in turn evaluating the role of spatial summation properties in binary classification of visual field (VF) results. Methods Eighty three glaucoma and 34 healthy participants underwent widefield OCT scans, segmented to generate ganglion cell‐inner plexiform layer measurements, and 24‐2 VF assessment using GII, GIII, and GV in full threshold mode. Accuracy was assessed using mean weighted absolute error and 95% prediction interval width from mixed effects models between ganglion cells per stimulus area estimated from ganglion cell‐inner plexiform layer thicknesses and VF thresholds compared using mixed effects models and post hoc analyses of estimated marginal means. Results Across healthy and glaucoma eyes, mean weighted absolute error and 95% prediction interval widths were smallest with GV ( p < 0.0001), suggesting the least model variability with GV. With VF locations in glaucoma cohort subclassified into VF nondefective and VF defective, although significant differences in mean weighted absolute error and 95% prediction interval widths were noted within stimulus sizes ( p < 0.0001), larger values indicating poorer model accuracy were noted in glaucoma VF defective locations relative to both healthy and VF nondefective models. Larger mean weighted absolute errors and 95% prediction interval widths were observed with increasing disease stage in VF defective locations across all stimulus sizes ( p < 0.0001). Conclusions Overall, structure–function models in healthy eyes and VF nondefective locations were similar across all stimulus sizes, but larger deviations were observed in VF defective locations and with worsening glaucoma stage. Our findings suggest that GIII sufficiently balances measurement variability and VF defect detection, but that disease stage‐specific variations in the structure‐function relationship exist and the ability to monitor VF defect progression over time is poor regardless of stimulus size.
Tong et al. (Sat,) studied this question.