An evaluation of algorithms designed to classify the results from frequency doubling perimetry.

All previously published algorithms for the interpretation of frequency doubling perimetry test results were compared in full-threshold mode in a large group of glaucoma patients (n = 452) and normal subjects (n = 237). Areas under the receiver-operating characteristic (ROC) curve ranged from 0.86 to 0.92. None of the algorithms performed substantially better than simply counting the number of defects p < 1% in the total deviation plot. For this algorithm, we found a sensitivity of 0.90 and a specificity of 0.81 at a cut-off point of >1 defect, and an area under the ROC curve of 0.92.