APPLICATION OF NEURAL NETWORKS TO THE CLASSIFICATION OF GIANT-CELL ARTERITIS

Objective. Neural networks are a group of computer-based pattern recog nition methods that have recently been applied to clinical diagnosis a nd classification. In this study, we applied one type of neural networ k, the backpropagation network, to the diagnostic classification of gi ant cell arteritis (GCA). Methods. The analysis was performed on the 8 07 cases in the vasculitis database of the American College of Rheumat ology. Classification was based on the 8 clinical criteria previously used for classification of this data set: 1) age greater than or equal to 50 years, 2) new localized headache, 3) temporal artery tenderness or decrease in temporal artery pulse, 4) polymyalgia rheumatica, 5) a bnormal result on artery biopsy, 6) erythrocyte sedimentation rate gre ater than or equal to 50 mm/hour, 7) scalp tenderness or nodules, and 8) claudication of the jaw, of the tongue, or on swallowing. To avoid overtraining, network training was terminated when the generalization error reached a minimum. True cross-validation classification rates we re obtained. Results. Neural networks correctly classified 94.4% of th e GCA cases (n = 214) and 91.9% of the other vasculitis cases (n = 593 ). In comparison, classification trees correctly classified 91.6% of t he GCA cases and 93.4% of the other vasculitis cases. Neural nets and classification trees were compared by receiver operating characteristi c (ROC) analysis. The ROC curves for the two methods crossed, indicati ng that the better classification method depended on the choice of dec ision threshold. At a decision threshold that gave equal costs to perc entage increases in false-positive and false-negative results, the met hods were not significantly different in their performance (P = 0.45). Conclusion. Neural networks are a potentially useful method for devel oping diagnostic classification rules from clinical data.