Statistical evaluation of the diagnostic accuracy of methods used to determine the progression of visual field defects in glaucoma.

OBJECTIVE

To evaluate trend-type analyses to judge the progression of visual field damage (VFD) of glaucoma in terms of ratios of making judgments of progression and specificity, and to report a new method.

DESIGN

Retrospective analysis of visual field (VF) results of actual glaucoma cases and those simulated by computer, and receiver operating characteristic curve analysis of performance of methods to judge the progression of VFD.

PARTICIPANTS

One hundred five eyes of 105 open-angle glaucoma (OAG) patients with progressing VFD and 355 eyes of 355 clinically stable OAG patients for VF simulation from 4 university-based referral practices.

METHODS

Methods using regression analysis of total deviation (TD), mean deviation (MD), mean TD of a sectored VF, and original scoring used in the Advanced Glaucoma Intervention Study (AGIS) were compared. A VF test was repeated twice in a short period on the 355 stable OAG eyes, and test-retest fluctuation, including variance at each test point and covariance between 2 test points, was calculated to simulate stable glaucomatous VF series by computer. The sensitivity of each method was calculated with 105 progressing VF series, and specificity was calculated with 10,000 simulated stable glaucomatous VF series.

MAIN OUTCOME MEASURES

Sensitivity (ratios of making judgments of progression), specificity, and diagnostic power.

RESULTS

The methods using the TD slope on one test location showed a sensitivity of 0.848 to 1.000, with a specificity of 0.105 to 0.721, and on 2 adjacent test locations showed a sensitivity of 0.848, with a specificity of 0.722. A significant negative MD slope with P<0.05 showed a sensitivity of 0.524, with a specificity of 0.945. The method using a sectored VF showed a sensitivity of 0.695, with a high specificity of 0.946. The AGIS method showed a sensitivity of 0.305 to 0.467, with a very high specificity of 0.999 to 1.000. The method using previously reported mathematically sectored VFs showed a sensitivity of 0.790, with a specificity of 0.900, and higher diagnostic power (1.69) than the others in this study population.

CONCLUSIONS

Most of the methods using the TD slope were characterized by high sensitivity, the AGIS method had a very high specificity, and those using VF sectors had reasonable sensitivity and specificity.