Use of artificial neural network to predict esophageal varices in patients with HBV related cirrhosis

BACKGROUND: Prediction of esophageal varices in cirrhotic patients by noninvasive methods is still unsatisfactory. OBJECTIVES: To evaluate the accuracy of an artificial neural network (ANN) in predicting varices in patients with HBV related cirrhosis. PATIENTS AND METHODS: An ANN was constructed with data taken from 197 patients with HBV related cirrhosis. The candidates for input nodes of the ANN were assessed by univariate analysis and sensitivity analysis. Five-fold cross validation was performed to avoid over-fitting. RESULTS: 14 variables were reduced by univariate and sensitivity analysis, and an ANN was developed with three variables (platelet count, spleen width and portal vein diameter). With a cutoff value of 0.5. The ANN model has a sensitivity of 96.5%, specificity of 60.4%, positive predictive value of 86.9%, negative predictive value of 86.5% and a diagnostic accuracy of 86.8% for the prediction of varices. CONCLUSIONS: An ANN may be useful for predicting presence of esophageal varices in patients with HBV related cirrhosis.