Application of an artificial network model to predict delayed graft function in paediatric kidney recipients.

Although conventional multivariate models allowed to identify risk factors for delayed graft function in kidney recipients, they cannot predict the outcome in each patient. Artificial neural network technology, that uses non-linear statistics for pattern recognition, was previously applied to outcome prediction in adult kidney recipients. In this study, we have developed a neural network as a predictive tool for delayed graft function in paediatric kidney recipients. The neural network was constructed with data from a training set of paediatric kidney recipients (n=107), by using twenty input variables. Eight input variables with the higher predictive values were retained in the final model. The model was validated in a second set of patients (n=41) by blinding the network for output variable, and further tested it in all patients. The overall accuracy of the neural network for training set, validation set, and whole patient cohort was 89.1%, 76.92%, and 87.14%, respectively. A comparative logistic regression analysis found that only serum creatinine on day 1 post-transplant was an independent predictor for delayed graft function, with an overall accuracy of 79.05%. The sensitivity and specificity of the neural network for validation set and whole patient cohort were 0.8/0.759 and 0.87/0.87, respectively. The logistic regression model showed a sensitivity and specificity of 0.37 and 0.94, respectively. In this study, an effective neural network model to predict delayed graft function in paediatric kidney recipients was constructed. The availability of the source code may allow both the validation of our model in prospective studies and the development of a stand alone neural network.