Prediction of repeatable glaucomatous visual field defects based on cluster characteristics.

Abstract

This study evaluates if characteristics (eg, location, size, volume) of clusters of defects on an initial visual field (VF) test were predictive of a repeatable defect in the subsequent two tests. Retrospective cohort study of 197 eyes of 103 patients with healthy, suspect or early glaucoma. Using the initial VF pattern deviation probability grid, we defined the number of clusters (≥1 location of p<5%) and associated size (number of adjoining defect locations) and volume (sum of corresponding total deviation values) for each cluster stratified by the four probability levels (ie, p<5%; p<2%; p<1% and p<0.5%). Of 4424 locations with a defect of p<5%, only 1189 (26.9%) were repeatable. The size [area under the receiver operating characteristic curve (AUC) 0.80, CI 0.76 to 0.85)] and volume (AUC 0.80, CI 0.76 to 0.85) of clusters were predictive of a repeatable defect within the cluster. The optimal thresholds for predicting a repeatable location within each cluster at 95% specificity based on initial cluster size were >6 locations at p<5%, >4 locations at p<2%, >3 locations at p<1% and >2 locations at p<0.5%. Defining cluster defects by involvement of central or peripheral rim locations improved the predictive value compared with the entire 24-2 grid. The location, size and volume of clusters of defects on an initial VF test may be predictive of subsequent repeatability. This may help distinguish eyes with a higher risk of repeatable defects.