This project focused on the usefulness of neural networks in monitoring patient laboratory data for quality control. We developed a neural network by training it with patient data on total serum cholesterol. We also trained the neural network with mathematically simulated data that failed our standard quality control procedures. After training, the neural network was able to recognize the usual pattern of cholesterol results in our laboratory and to differentiate it from incorrect laboratory results produced by a problem with the chemistry analyzer. The neural network was able to recognize problems in bias, as well as precision in the laboratory data, using a sample size as small as 20 patients. Receiver operating characteristic analysis showed that the neural network approach is superior in sensitivity and specificity in error detection compared with other approaches. Future studies will be aimed at extending the neural network approach to other analytes and to interface a neural network with a chemistry analyzer to test the practicality of this approach in monitoring quality control.
