Epidemiological Characteristics Of Pathogens And Infection Classification Modeling Of Fever Respiratory Syndrome In Gansu Province

Objective: To understand the distribution characteristics of pathogenic positive cases with febrile respiratory syndrome and pathogen detection in Gansu Province,reveal epidemic rules of main pathogens,and identify the high-risk population and peak season.To determine types of pathogenic infection in cases of febrile respiratory syndrome in Gansu Province by classification models based on machine learning algorithm,which provided a theoretical reference for the early diagnosis,monitoring and prevention during the occurrence of respiratory infectious diseases.Methods: According to the surveillance information of febrile respiratory syndrome from January 2012 to November 2018 in Gansu Province,the distribution characteristics of virus-positive cases,bacteria-positive cases and mixed positive cases were described by observational study.The core pathogenic spectrums were analyzed to understand the main pathogens and their distribution in different populations and time.Combined with the data of demography,symptoms and laboratory testing of cases,different classification models of machine learning were established to predict pathogenic infection of cases,and the optimal model was screened out based on comparison and evaluation.Results: 1.The positive rate of pathogens,virus,bacteria and both mixed infection in cases of in febrile respiratory syndrome Gansu Province were 38.32%,17.97%,19.36%,and 7.42%,respectively.In the distribution of virus,bacterial and mixed positive cases,the number of males were more than that of females.The positive rate of virus in children under 5 years old(24.32%)and adolescents under 5-17 years old(22.81%)was higher;the positive rate of bacteria in 18-year-old,45-year-old and 65-year-old groups were all about 25.00%;the positive rate of mixed infection of virus and bacteria in the elderly was higher(12.64%).2.The top three of the virus spectrum were human influenza virus,rhinovirus and respiratory syncytial virus.The positive rate of human influenza virus was higher in the age group of 5 years old,and it was prevalent from January to March and December.The positive rate of rhinovirus was higher in 18-year-old and 65-year-old groups,and the seasonal peak was from March to May and August to November.The positive rate of respiratory syncytial virus was higher in the 0-year-old group,with the epidemic peak in January-February and November-December.3.Among respiratory pathogenic bacteria,Streptococcus pneumoniae and Haemophilus influenzae were dominant.The detection levels of Streptococcus pneumoniae in people of all ages were similar,with seasonal peaks in January-February and October-November.The positive rate of Haemophilus influenza was higher in the group of 5 years old,and it was prevalent in February and August-October.4.The parameters of three models were optimized.The classification accuracy rate of the C5.0 algorithm was higher for decision tree(DT)with 10 times boosting iteration and a 70% pruning purity.The accuracy of the base kernel function(RBF)was higher for support vector machine(SVM)with gamma value of 0.1 and penalty parameter C of 64.The support vector machine decision tree(SVMDT)had higher classification accuracy when classification regression tree combined with RBF-SVM.5.Comparing and evaluating the classification effect of different models,the classification accuracy of C5.0 decision tree,RBF-SVM and SVMDT were 78.67%,76.92% and 79.02% respectively,and the Kappa coefficients were 0.708,0.684 and 0.717 respectively.Conclusions: 1.Among the positive cases of febrile respiratory syndrome in Gansu Province,the number of males was more than that of females.There is a need to prevent and control viral infections in minors,as well as mixed infections of viruses and bacteria in the elderly.2.The main pathogens related to febrile respiratory syndrome were human influenza virus,rhinovirus,respiratory syncytial virus,Streptococcus pneumoniae and Haemophilus influenza,etc.The distribution of pathogens varied among different age groups.In general,viral infections were more common in winter and spring,while bacterial infections had a higher incidence in autumn and winter.3.The classification effect of SVMDT combination model was superior a bit to that of DT and SVM in this study.It can provide a new idea for preliminary estimation of the pathogenic infection in patients of febrile respiratory syndrome in Gansu Province,and seasonable prevention and control of outbreaks of acute respiratory infectious diseases.