〔Abstract〕 Objective To explore the clinical application value of neural network technology,and to screen the clinical early warning indicators of severe hand-foot-mouth disease(HFMD) by constructing a back propagation neural network(BPNN) model.Methods The clinical data of children with HFMD admitted to the Department of Infectious Diseases and Pediatrics of the First Affiliated Hospital of Xinxiang Medical University from January2019 to January 2023 were collected.The data were divided into 70% training samples and 30 % test samples using SPSS Modeler 18.0.The BPNN model and Logistic model were constructed to compare and evaluate the prediction accuracy and screening effect of the model.Results The clinical data of 589 children were collected and analyzed,including 324 cases in the mild group and 265 cases in the severe group.The prediction accuracy of the test set(n=178) of BP neural network model and Logistic regression model was 82.02% and 84.83%,respectively.The area under the ROC curve(95% CI) was 0.791(0.749-0.834) and 0.625(0.577-0.674),respectively.Among the predictive variables output by the BPNN model,the top five factors that had the greatest impact on the grouping were highest body temperature,duration of fever,glutamyl transpeptidase,aspartate aminotransferase,and globulin.There were 3 of the top 10 overlaps in the importance of the predictive variables output by the two models,which were the highest body temperature,duration of fever and limb shaking.Conelusion The BPNN model and Logistic regression model perform well in screening and verifying the risk factors of severe hand-foot-mouth disease,but the comprehensive prediction performance of BP neural network model is better.The top five influencing factors of severe HFMD screened by the BPNN model are the highest body temperature,duration of fever,glutamyl transpeptidase,aspartate aminotransferase and globulin.