Comparison of data analysis tools for detection of glaucoma with the Heidelberg Retina Tomograph

To evaluate the performance of three linear discriminant functions (LDFs) and the Moorfields Regression Analysis (MRA) in classifying optic disc topography images obtained with the Heidelberg Retina Tomograph (HRT) from patients with open-angle glaucoma and normal controls. Furthermore, to investigate whether the classification of glaucomatous eyes is related to the severity of visual field loss or to optic disc size.

Prospective comparative observational case series.

One eye of 104 patients with a diagnosis of open-angle glaucoma (mean age, 63.2; range, 30-88 years) and 48 normal controls (mean age, 52.4; range, 33-75 years).

Subjects were examined with the HRT.

Diagnostic performance was calculated for the LDFs published by Mikelberg et al (LDF1), Burk (LDF2), and Bathija et al (LDF3) and for the MRA.

Large differences in sensitivity and specificity were found among the different analyses when using the originally suggested cutoff criteria. When specificity was equalized to 90%, the sensitivities were more similar (LDF1, 55%; LDF2, 61%; and LDF3, 67%). At 95% specificity, the sensitivities decreased further (LDF1, 39%; LDF2, 55%; LDF3, 44%). When treating the "borderline" outcomes as test positives, MRA sensitivity and specificity were 78% and 81%, respectively. When treating the "borderline" outcomes as test negatives, the MRA gave a sensitivity of 58%, with a specificity of 96%. In glaucoma patients, the correlation between the severity of visual field damage, as measured by the mean deviation index, and the outcome of the classification systems was low to moderate (Spearman's r between 0.26 and 0.39). When the entire sample was stratified into equal thirds on the basis of optic disc size, large discs (area > 2.10 mm(2)) tended to be classified with a higher sensitivity but lower specificity than small discs (area < 1.73 mm(2)) with all methods except LDF3. Using a multiple regression model controlling for mean deviation, LDF1, and LDF2, but not LDF3, were significantly influenced by disc size.

The 4 methods of analysis had similar sensitivities once their specificities were equalized. In this sample, the LDFs and MRA generally did not discriminate between glaucoma patients and controls as well as reported in the original respective studies that described the methods.