BUILDING CLINICAL CLASSIFIERS USING INCOMPLETE OBSERVATIONS - A NEURAL-NETWORK ENSEMBLE FOR HEPATOMA DETECTION IN PATIENTS WITH CIRRHOSIS

One objective of liver transplant evaluation is to identify patients that harbor a hepatoma, but standard screening techniques are not sensitive enough. We trained neural network ensembles to predict the presence of hepatoma in patients with cirrhosis, based on information collected at the time of transplant evaluation. Network architecture and training were modified to handle missing observations. Three ensembles were trained: ensemble A using the subset with no missing observations (528 patients); ensemble B using the complete set, which included missing observations (853 patients); and ensemble C using the smaller subset, originally with complete data, but after a fixed number of observations were deleted (i.e., made ''missing''). Ensemble performance on testing sets was very good. The areas under the ROC curves were 0.91, 0.89, and 0.90, for ensembles A, B, and C, respectively. Neural networks can successfully perform this classification task, and strategies can be developed that allow use of incomplete observations.