Purchase this article with an account.
BA Ford, PH Artes, TA McCormick, MT Nicolela, RP LeBlanc, BC Chauhan; Comparison of Data Analysis Tools for Detection of Primary Open Angle Glaucoma with the Heidelberg Retinal Tomograph (HRT) . Invest. Ophthalmol. Vis. Sci. 2002;43(13):928.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Purpose: To evaluate the performance of three published Linear Discriminant Functions as well as the Moorfields analysis (MF) in classifying HRT optic disc images from normal controls and patients with mild to moderate open angle glaucoma. Further, to investigate whether the classification of glaucomatous eyes is related to the severity of visual field loss, or to the size of the optic disc. Methods: One eye of 104 patients with a clinical diagnosis of open angle glaucoma (mean age 63.2, range 30-88 yrs) and 48 normal controls (mean age 52.4, range 33-75 yrs) was examined with the HRT. Patients had visual fields with MDs between -2.0 and -10.0 dB. Diagnostic performance was calculated for the LDFs published by Mikelberg et al. (LDF1), Burk et al. (LDF2) and Bathija et al. (LDF3) as well as for the Moorfields analysis (MF). Results: There were large differences in sensitivity and specificity between the different analyses when using the originally suggested cut-off criteria. When specificity was equalised to 90% by adjusting the cut-off criteria, the sensitivities were more similar (LDF1: 61.5%, LDF2: 64.4%, and LDF3: 67.3%). When treating the "borderline" outcomes as test-negatives, the MF analysis gave a sensitivity of 58%, with a specificity of 96%. When treating the "borderline" results as test-positives, sensitivity and specificity were 78% and 81%, respectively. In patients, the correlation between the severity of visual field damage (MD) and the outcome of the classification systems was low to moderate (Spearman's r between 0.26 [LDF1] and 0.39 [LDF2]). When the glaucoma population was stratified into equal thirds based on disc size, medium sized discs (area between 1.84 and 2.19 mm2) tended to be classified most accurately, however, the group differences were not statistically significant. Conclusion: The four methods of analysis performed similarly once specificity was equalised. Their performances were considerably different than those reported in the original populations.
This PDF is available to Subscribers Only