Macular sector-wise decision tree model for the prediction of parafoveal scotoma not detected by 24-2 visual field...

Development of a macular sector-wise decision tree model (DTM) for the prediction of parafoveal scotoma.

This prospective study enrolled 126 patients with early-stage open-angle glaucoma (mean deviation ≥-6 decibels) without the signs of parafoveal scotoma on the 24-2 visual field (VF) test (i.e., any abnormalities at the four innermost points). Based on the central 36 points of the 10-2 pattern deviation plot, patients were classified as being with or without 10-2 parafoveal scotoma. For the discrimination of patients from those without 10-2 parafoveal scotoma, a macular ganglion cell-inner plexiform layer (mGCIPL) sector-wise DTM analysis was performed.

Among 126 eyes without 24-2 parafoveal scotoma, 10-2 parafoveal scotoma was detected in 77 (61.1%) eyes. The balanced accuracy of DTM was best in the inferotemporal sector (0.9286; 95% CI, 0.7458-0.9697) and worst in the inferior sector (0.8373; 0.6484-0.9204). DTM revealed that even in the absence of VF abnormalities at the innermost 4 points on the 24-2 test, (1) 10-2 parafoveal scotoma should be strongly suspected when the adjacent 24-2 perifoveal point in the correlated sector is abnormal; (2) if the 24-2 perifoveal point is normal, and if the probability colour codes of the correlated mGCIPL sector are green, the probability of 10-2 parafoveal scotoma is very low.

In clinical practice, the evaluation of the 24-2 perifoveal test points along with the probability colour codes of mGCIPL can be a useful decision-support tool in determining whether 10-2 tests are needed for a given patient.