Machine Learning Algorithms For Identifying Risk Factors And Prediction Of Severe Hand,Foot,and Mouth Disease

Hand,foot,and mouth disease(HFMD)is a rash and fever illness caused by enteroviruses in children.There are a small number of children who are prone to progress in severity,and cause various complications involving the nervous system.Such as common viral encephalitis,which can cause intracranial infection,high blood pressure,and vomiting and convulsions.But in rare cases,some children may further progressed in critical HFMD,with poor peripheral circulation and respiratory failure,and even death.Therefore,a model for predicting severe HFMD was established to assist clinicians in the early identification of severe cases,which can accomplish early intervention and reduce the occurrence of serious complications and adverse events.Objective To screen the risk factors of severe HFMD and construct its clinical prediction model.Methods Case data of children with HFMD were collected from the Department of Infectious Diseases of the Affiliated Children’s Hospital of Zhengzhou University from September 2017 to June 2018.Data analysis was performed using SPSS and R software.The risk factors of severe HFMD were screened by single factor analysis and feature selection.From the common algorithms of machine learning,the optimal algorithm was used to construct the machine learning model.At the same time,the traditional logistic model was used as a reference.Various performance evaluation indicators were selected to compare the prediction performance of the model.Results A total of 1292 cases of HFMD were included.There were 933 cases in the mild group and 359 cases in the severe group.The data set was segmented into 70% training samples(904 cases)for screening variables and building predictive models;30% of the test samples(388 cases)were used to evaluate model performance.According to the variable screening process,25 variables were finally selected.The random forest model and logistic model had the accuracy rates of 0.84 and 0.81,respectively;the sensitivity were 0.91 and 0.88,the specificities were 0.67 and 0.63 respectively,and the areas under the ROC curve were 0.88,95% CI(0.85-0.91)and 0.82,95% CI(0.79-0.85)respectively.The random forest model was superior to the logistic model in the comparison of multiple visual performance plots.Conclusions The machine learning algorithm can identify the predictive factors of severe HFMD.Compared with the common logistic model,the machine learning model has higher sensitivity,specificity and predictive accuracy.It has certain reference value for guiding clinicians to identify severe HFMD early.