advertisement
Abstract #8816 Published in IGR 5-1
Sequential classification in glaucoma diagnosis
Stroux A; Martus P; Budde WM; Horn FK; Jünemann AG; Korth M; Jonas JBGraefe's Archive for Clinical and Experimental Ophthalmology 2003; 241: 277-283
BACKGROUND: Large-scale screening in glaucoma diagnosis is expensive and time-consuming. Sequential classification strategies can provide an effective combination of time-efficiency and diagnostic accuracy for glaucoma screening. METHODS: In a cross-sectional clinical study, a sequential diagnostic strategy, based on several psychophysical and electrophysiological tests, was evaluated on measurements from 595 eyes from 310 patients with primary open-angle glaucoma, and 419 eyes from 213 control subjects (age range, 18-70 years in each group). Patients and controls successively underwent up to five psychophysical and electrophysiological diagnostic tests. Optic disc morphometry was taken as gold standard. Adapting group sequential techniques, sensitivity and specificity for the whole diagnostic program were controlled, allotting overall error rates of 10%. The criteria for the diagnostic process were developed in a learning sample (677 eyes) and verified in a validation sample (337 eyes). RESULTS: In the validation sample, 62.0% of the examined eyes could be classified, using a sequential 15-minute two-step program. An overall 13.6% 'gain' of saved time, compared to non-sequential discriminant analysis, was achieved without loss of diagnostic accuracy. A sequential 45-minute five-step program classified 68.8% of the whole sample before morphometry, saving approximately 39% of examination time, compared to taking the complete discriminant score. CONCLUSIONS: Especially in screening, where the use of time-consuming and complicated diagnostic procedures is restricted, the implementation of testing programs based on group sequential strategies might be a promising means of saving personnel resources and reducing inconvenience for patients.
Dr. A. Stroux, Department of Medical Informatics, Biometry and Epidemiology, Free University of Berlin, Hindenburgdamm 30, 12200 Berlin, Germany. andrea.stroux@medizin.fu-berlin.de
Classification:
1.6 Prevention and screening (Part of: 1 General aspects)
