The Analysis Of Influenza-like Illness Epidemic Characteristics And The Construction Of Prediction Models In Yangzhou City

Objectives:1.To analyze the epidemiological features of influenza-like illness in Yangzhou City,based on the surveillance data of influenza-like illness in sentinel hospital in Yangzhou City.2.To analyze the etiological characteristics of submitted specimens collected from influenza-like illness in the influenza surveillance network laboratory in Yangzhou City.3.Computer software and programming techniques were used to establish forecast models,based on data of influenza-like illness,so as to provide technical support and scientific basis for the early warning of influenza in Yangzhou City.Methods:1.Data was collected from influenza-like illness case report and pathogen surveillance of influenza virus from influenza sentinel hospitals in Yangzhou from 2014 to 2018.Descriptive epidemiological methods were used to analyze the data.2.Based on the surveillance data of weekly number of influenza-like illness cases and weekly consultation rate of influenza-like illness from 1st week,2014 to 40th week,2018,the ARIMA model was established by using software SPSS20.0 and Eviews8.0.The data of the last 12 weeks of 2018 were used to evaluate the prediction efficiency of the model.By comparing the prediction performance of the two models,the ARIMA model with better prediction performance was selected.3.Three kinds of neural network combined models(ARIMA-BPNN,ARIMA-GRNN,ARIMA-ERNN)were established by using corresponding computer code program which written by the software Matlab8.3 compiler,based on the ARIMA model with better predictive performance to fit and predict the surveillance data of influenza-like illness.The root mean square error,mean absolute percentage error,mean absolute error were used to evaluate fitting effects and predictive effects of the four models.Results:1.There were altogether 137474 influenza-like illness cases reported by influenza sentinel hospitals from 2014 to 2018 and the average consultation rate of influenza-like illness was 6.63%.The consultation rate of influenza-like illness was statistically different among different years(χ2=690.80,P<0.001).There are more influenza-like illness cases in outpatient department of pediatric and pediatric emergency department.The consultation rate of influenza-like illness was statistically different among different monitoring clinics(χ2=87286.33,P<0.001).The age group with the most number of influenza-like illness cases was under 15 age group(71.37%),of which 0～group accounted for 41.53%and 5～group accounted for 29.84%.The consultation rate of influenza-like illness mainly showed two peak seasons which distributed in summer-autumn and winter.2.A total of 11587 influenza-like illness specimens from influenza sentinel hospitals were detected by the influenza network laboratory,of which 1232 were positive for influenza virus,with positive rate of 10.63%.The positive rate of influenza virus was statistically different among different years(χ2=98.78,P<0.001).Pathogen surveillance results found alternating or mixed varieties of influenza virus strains each year.There were at least two kinds of influenza strains that are prevalent every year.The influenza B virus strains were prevalent every year.Except for 2018,the prevalence of seasonal influenza H3 virus strains was similar to influenza B virus strains.However,new influenza A H1 virus strains were prevalent every other year.The consultation rate of influenza-like illness was positively correlated with the positive rate of influenza virus(rs=0.27,P=0.04<0.05).3.The predicted performance of the ARIMA model that was established by using the surveillance data of weekly consultation rate of influenza-like illness was better than the ARIMA model that was established by weekly number of influenza-like illness cases.Three neural network models were combined with the ARIMA model based on weekly consultation rate of influenza-like illness,respectively.In the fitted ARIMA model,RMSE,MAPE and MAE were 0.77%,7.77%,0.53%,respectively.In the fitted ARIMA-BPNN model,RMSE,MAPE and MAE were 0.50%,5.70%,0.37%,respectively.In the fitted ARIMA-GRNN model,RMSE,MAPE and MAE were 0.16%,1.62%,0.10%,respectively.In the fitted ARIMA-ERNN model,RMSE,MAPE and MAE were 0.50%,5.94%,0.38%,respectively.RMSE,MAPE and MAE predicted by the ARIMA model were respectively 0.80%,11.48%,0.69%.RMSE,MAPE and MAE predicted by the ARIMA-BPNN model were respectively 0.64%,9.42%,0.58%.RMSE,MAPE and MAE predicted by the ARIMA-GRNN model were respectively 0.46%,6.02%,0.38%;RMSE,MAPE and MAE predicted by the ARIMA-ERNN model were respectively 0.42%,5.26%,0.31%.It can be found that the best fitting effect model is the ARIMA-GRNN combination model among these forecast models,and the best prediction effect model is the ARIMA-ERNN combination model.Conclusions:1.From 2014 to 2018.the results of influenza surveillance found that children under 15 years old were the high-risk groups.The epidemic peak of consultation rate of influenza-like illness appeared in summer-autumn and winter.2.The pathogen surveillance of influenza-like illness has an important guiding role in grasping the epidemic of influenza virus subtypes.3.The combination models performed better than the single model in fitting effect,and better than the single model in prediction effect.The ARIMA-ERNN combination model has the best prediction effect among the four forecast models,and can better predict the incidence trend of influenza-like illness and provide scientific evidence on the prevention and control of influenza in Yangzhou.