comparison of artificial neural networks and cox regression models in prediction of kidney transplant survival

Cox regression model serves as a statistical method for analyzing the survival data, which requires some options such as hazard proportionality. in recent decades, artificial neural network model has been increasingly applied to predict survival data. this research was conducted to compare cox regression and artificial neural network models in prediction of kidney transplant survival. the present multi-center retrospective study was conducted on the medical records of 756 kidney transplant recipients undergoing kidney operations at two treatment centers from 2001 through 2012. the data was randomly divided into two educational and experimental (validation) groups. then, kaplan-meier, cox proportional hazard, and three-layer artificial neural network models were used for analyzing the data. to compare the prediction of both models, the area under the curve in the characteristic function was applied. post-operative creatinine and relative family are among the factors of influencing kidney transplant survival. moreover, the survival estimates of the transplanted kidney for periods of six months, one year, three years, and five years were 89, 87.4, 80, and 75 percent, respectively. roc areas under the curve, for multi-layer perceptron neural network model and cox regression, were 81.3% and 71%, respectively. if a structure with high prediction ability is obtained in neural network, we may detect risk patients through the method and consider more treatment resources for them.