A Neural Network Approach to Predict Acute Allograft Rejection in Liver Transplant Recipients Using Routine Laboratory Data

Background: Discovery of non-invasive methods for acute rejection in liver transplant patients would contribute to preservation of liver function in the graft. Recently, however, outcome prediction based on biostatistical models like artificial neural networks (ANNs) is increasingly becoming impressive in medicine. Objectives: The aim of this study was to obtain a predictive model based on ANN technique and to figure out the best time for early prediction of acute allograft rejection after transplantation in liver transplant recipients. Methods: Feed-forward, back-propagation neural network was developed to predict acute rejection in liver transplant recipients using clinical and biochemical data from 148 liver transplant recipients over days 3, 7, and 14 post-transplantation. Sensitivity and receiver-operating characteristic (ROC) analysis were done to reveal the importance of input variables and the performance of the neural network. Results: The results were compared with a logistic regression (LR) model using the same data. Our results showed that the data related to day 7 gave the best results in terms of ANN performance; and the most important factors in the predictive model were aspartate aminotransferase (AST) and alanine aminotransferase (ALT). The ANN's accuracy was 90%, sensitivity was 87%, specificity was 90% in the testing set, and the performance of the ANN was better than that of the LR model. The ANN recognized correctly eight out of ten acute rejection patients and 34 out of 36 non-rejection ones in the testing set. Conclusions: This study suggests that ANN could be a valuable adjunct to conventional liver function tests for monitoring liver transplant recipients in the early postoperative period.