American journal of ophthalmology vol:116 issue:6 pages:684-91
We used a statistical cluster analysis to analyze patterns of loss in 76 visual fields with typical glaucomatous defects to identify natural groupings of test locations in the visual field. Eleven clusters in the Octopus Program G1 (Interzeag, Inc., Northboro, Massachusetts) visual field were thus defined. In a separate population of 70 early glaucomatous and 70 age-matched normal visual fields, the local mean defects within these clusters and the global mean defect were calculated to assess their relative abilities to discriminate between the two groups. The 11 clustered mean defects collectively had a sensitivity of 90% and a specificity of 93%; the global mean defect had a sensitivity of 81% and a specificity of 91%. Additionally, we examined the long-term fluctuation clusters of test locations compared to the long-term fluctuation of individual test locations. Four hundred ten visual fields of 93 clinically stable eyes of 67 patients with glaucoma, as well as 210 visual fields of 105 eyes of 105 normal subjects were studied. In the stable glaucoma group, mean fluctuation of clustered test locations was 3.5 dB2, and mean fluctuation of individual test locations was 7.0 dB2. In the normal group, the respective values were 0.6 dB2 and 1.8 dB2. Cluster analysis was effective in detecting localized loss and in dampening long-term fluctuation. We studied the use of clusters in distinguishing normal from glaucomatous as well as stable from deteriorating visual fields.