Application of Back-propagation Neural Network in Analysis of Influencing Factors for Metabolic Syndrome

Objective To explore the role of utilization of back-propagation(BP) neural network for screening the related factors of disease and constructing predictive models,to compare with the prediction with Logistic regression model, and to provide the scientific evidence for further correct implementation of neural network to solve practical problems in medical practice. Methods We built the predictive model of metabolic syndrome related risk factors by invoking BP neutral network toolbox from MATLAB.We selected the risk factors according to mean impact value,and then compared the performance of predictive models between BP neutral network and Logistic regression by calculating their area under ROC curves respectively. Results The predictive model of BP neutral network with mean impact value algorithm involved showed more reliable and robust results than the one without the algorithm;moreover,the risk factors selected by the former method were practically consistent with the results produced by Logistic regression,and no statistically significant difference was found(AUCBP=0.837,AUCLogistic=0.841,u=0.3310,P=0.7406). Conclusions We can use BP neutral network based on mean impact value to select the influence variables and to construct disease forecast model.The results suggest that it plays as good role as the general Logistic regression in the use of cause exploration in epidemiological studies.