Application of back propagation neural network model in prediction of risk factors of severe hand-foot-mouth disease

ObjectiveTo predicit risk factors of severe progression of hand-foot-mouth disease(HFMD)with back propagation neural network(BPNN)model and to provide a reference for the diagnosis of severe HFMD in children.MethodsClinical data on 233 child HFMD cases hospitalized in Zhengzhou Children's Hospital during April 2013 through June 2013 were selected and surveyed with a questionnaire within 72 hours of the hospitalization.A BPNN model was established with neural network toolbox of MATLAB 7.0 software and mean impact values(MIV)of the risk factors for severe progression of HFMD were obtained and ranked descendingiy according to their absolute values.Then the results of MIV were compared with those of muti-factor logistic regression.ResultsThe final network structure of BPNN was 26→8→1.The top ten risk factors for the severity of HFMD(MIV absolute value)were panic tendency(0.691 4),stiff neck(0.537 3),vomiting(0.435 8),elevated blood sugar(0.429 9),shake of hands and feet(0.381 3),poor spirit(0.328 3),high body temperature above 39 ℃(0.308 6),increase of white blood cell count(0.290 2),fever lasting for more than 3 days(0.262 1),and drowsiness(0.242 7);the risk factors ascertained with multi-factor logistic regression analysis(odds ratio)were stiff neck(183.633),panic tendency(158.868),vomiting(59.347),elevated blood sugar(23.133),increase of white blood cells count(12.243),fever lasting for more than 3 days(7.765),shake of hands and feet(5.738),and poor spirit(4.452).High body temperature above 39 ℃,poor spirit and the increase of white blood cell count interacted with each other based on the results of saturated log linear model.ConclusionBPNN model can well describe the complex nonlinear relationship between risk factors and the severity of HFMD with good fitness and correct discriminaticn and could be used to analyze the risk factors of severe progression of HFMD.