Epidemiology Characteristics Analysis Of Influenza-like Illness And The Construction Of Early Warning Model Based On Neural Network In Huludao City,2011-2017

Objective:1.To analyze the epidemiological characteristics of influenza-like illness in Huludao City,Liaoning Province according to the surveillance data of influenza-like illness in sentinel hospital in Huludao City.2.To analyze the etiological characteristics of influenza-like illness virus samples in the influenza surveillance network laboratory in Huludao City.3.To establish mathematical models for early warning,using the data of influenza-like illness.The models include ARIMA model,ARIMA-BPNN combined model,ARIMA-GRNN combined model.Method:1.The surveillance data of influenza-like illness was collected by the national influenza sentinel hospital,Huludao City Central Hospital.The medical staff of the flu surveillance clinic(including fever clinic,medical clinic,medical emergency,pediatric clinic,pediatric emergency)conducted diagnosis and registration of influenza-like illness according to the definition of influenza-like illness.The data was summarized and recorded into China Influenza Surveillance Information System by the professionals weekly.Describe and analyze the time distribution and age distribution of influenza-like illness in Huludao City.2.The collection of influenza-like illness virus specimens was not less than 20 weekly in the epidemic season(from October to March of the following year)each year and was not less than 20 monthly in the non-epidemic season(from April to September)each year.The virus was isolated and identified by the influenza surveillance network laboratory in Huludao Center for Disease Control and Prevention.The MDCK cells in good growth condition were used for virus isolation.Then the hemagglutination test(HA)was used to determine the presence of influenza virus.If the positive titer was?1:8,hemagglutination inhibition test(HI)was used for virus type identification.When the hemagglutination inhibition titer?20 was judged as positive.The positive rate and type distribution of influenza virus were statistically described and analyzed.3.The early warning model of influenza in Huludao City consisted of three kinds of model.Firstly,the ARIMA model was established by using the weekly number of influenza-like illness cases and the weekly percentage of influenza-like illness from2011 to 2016.Compare the early warning effects of them.Then built two combined models with the ARIMA model fitted values based on the number of influenza-like illness cases adding time information as the network input and actual values as the target output of ARIMA-BPNN combined model and ARIMA-GRNN combined model respectively.The weekly number of 2017 influenza-like illness cases was predicted by the three models.The mean absolute error(MAE),mean absolute percentage error(MAPE),and the coefficient of determination(R~2)were used to evaluate the early warning effects of the three models.Result:1.The total number of influenza-like illness cases reported by sentinel hospital in Huludao City from 2011 to 2017 was 49,625 and the average percentage of influenza-like illness was 2.91%.The percentage of influenza-like illness in each year was statistically different(?~2=759.439,P<0.05).The number of influenza-like illness cases was reported more in pediatric clinic and pediatric emergency.In terms of time distribution,the weekly percentage of influenza-like illness in the seven monitoring years ranged from 1.32%to 6.83%.The peak of the percentage of influenza-like illness in Huludao City mainly concentrated at the end of December to the next February or so,followed by concentrated in August to October or so,and in the other months were relatively mild or peaked sporadically.In the age distribution,influenza-like illness mainly concentrated in the 0 to 4 age group,followed by the 5 to 14 age group.2.From2011 to 2017,3979 samples of influenza virus were collected by influenza surveillance network laboratory of Huludao Center for Disease Control and Prevention,of which 644were positive for influenza virus.The positive rate was 16.18%.The positive rate of influenza virus in each year was statistically different(?~2=263.489,P<0.05).The positive rate of influenza virus was in the peak from December to March every year.Influenza virus subtypes were mixed and alternately epidemic each year,of which H3N2 strains were popular each year.3.The ARIMA model based on the weekly number of influenza-like illness cases warned better than ARIMA model based on the weekly percentage of influenza-like illness.The three early warning models based on the weekly number of influenza-like illness cases,ARIMA model,ARIMA-BPNN combined model and ARIMA-GRNN combined model,predicted the weekly number of influenza-like illness cases in 2017.The MAE were 13.3077,10.8654 and 10.2885,the MAPE were9.73%,7.86%and 7.45%,the R~2 were 0.5007,0.6128 and 0.6223,respectively.The early warning effects of the two combined models were better than the single ARIMA model.In this article,the early warning effects of ARIMA-BPNN combined model and ARIMA-GRNN combined model were similar.Conclusion:1.The peak percentage of influenza-like illness from 2011 to 2017 in Huludao City,Liaoning Province mainly concentrates in the end of December to February of the next year,followed by the epidemic peak in August to October,with a certain seasonality.85.80%of influenza-like illness cases are in the age group below 14years old.The group is a key protected population of influenza.2.Influenza virus isolation and identification is of great significance for grasping of influenza epidemic strains timely.3.Using neural network combined model with ARIMA model which is better to warn influenza in Huludao City has a positive meaning for influenza prevention and control work.