A Supplementary Epidemiological Influenza Epidemic Prediction Model Was Constructed Based On The Information Epidemiology Method

ObjectiveTo build a CAM influenza prediction model combining GFT estimates and CAM search queries and validate it by comparing the estimated results to GFT.Mehtod1)To identify the information-at-hand for Americans when they are searching CAM-related influenza pages,and then determine the commonly-used search queries based on the theory of "equivalent internet search demands and needs".2)Solicit the query dynamics of these commonly-used search queries between October,2004 and April,2015 for correlation and lag-correlation analyses.This is to establish a Query of CAM(QCAM)dataset recording search query dynamics.3)Combing the GFT estimates between October,2004 and April 2010 with QCAM as training set to develop a Nowcasting CAM model for influenza(NCAM),and use it to simulate the nowcasting the influenza outbreaks of the US between October,2010 and April,2014.Then CDC influenza-like illness incidence,as background real-world data,is used to compared with the estimates of NCAM with respect to predicting accuracy and timeliness.4)Using the results from the superior model proven by last step to establish a forcasting CAM model for influenza(FCAM),and then compare the estimates of FCAM with the auto-regression+ GFT model proposed by previous studies to determine which one has a better accuracy.ResultBased on data of search engine demands,the 53 kinds of commonly-used CAM influenza therapies are identified.Pearson correlation analysis found that with the ones with relatively higher correlation coefficient are giant liao(R = 0.4749,p<0.001),chrysanthemum(R =0.4573,p<0.4573),ginger(R = 0.4626,p<0.001),the scythe,shu can(R = 0.7927,p<0.7927),phosphorus(R = 0.7927,p<0.7927),vitamin A(R= 0.6380,p<0.6380),and vitamin D(R = 0.6089,p<0.001).Modeling with these seven variables,the elastic lasso NCAMF model(R2 = 0.54;MAE = 0.221;10.8%MAPE =)to the flu real-time effect is better than that of GFT model(R2 = 0.89;MAE = 0.00381;MAPE = 20.4%).To further introduce regression item after established FCAMF model(R2 = 0.96;MAE = 0.00119;MAPE = 7.3%),for two weeks after the flu prediction effect is better than that of previous reports of GFT modified model(R2 = 0.89;MAE = 0.00326;MAPE = 14.3%),and observed from the scatterplot,found that the newly-proposed model has to some extent corrected the overestimation of GFT.Conclusion1)The inclusion of CAM-related queries into the establishment of infodemiology models significantly improves the prediction accuracy.2)The nowcasting model built by ELASTIC NET using CAM and GFT estimates shows satisfactory nowcasting results that is superior to those of GFT.3)Combing the estimated results of ELSTIC NET with an auto-regression of CDC influenza data gives satisfactory 2-weeks-lead predictions of the dynamics of influenza...