Study On Multi-time Scale Dengue Fever Forecast Based On Climate And Vector As Wellas Baidu Index

Objective This study aims to understand the distribution and influential factors of dengue fever in Guangzhou.Based on the Baidu Search Index(BSI)related dengue fever in Guangzhou combined with climate and vector historical data,a dengue fever forecast model is constructed to provide a decision basis for responding to the dengue fever epidemic in Guangzhou.Methods The data of dengue fever,meteorology and mosquito density in Guangzhou from 2017 to 2019 were collected systematically.BSI of dengue fever related keywords was mined.The descriptive method was used to describe the epidemiological characteristics.Generalized Linear Model(GLM)and Generalized Additive Model(GAM)were applied to analyze the relevant influential factors and effects.Correlation analysis and importance screening of meteorological data,vector data,and BSI of keywords were conducted.Three machine learning algorithms(BackPropagation neural network(BP),random forest regression(RF),support vector regression(SVR))were utilized to forecast the dengue fever epidemic of Guangzhou,and the forecast effect of the models were evaluated by R~2 and RMSE.SPSS22.0,R3.6.0,and MATLAB(2017)were used for statistical analysis or training model.Results1.The local incidence of dengue fever in Guangzhou from 2017 to2019 were 6.12 / 100,000,8.16 / 100,000,and 9.27 / 100,000,showing an upward trend(χ~2= 85.306,P <0.001);High incidence was in summer and autumn,accounting for 99.77%.The peak incidence in October accounted for 38.42%;The cases were mainly distributed in the central urban area(accounting for 82.83%).The proportion of affected streets in the total number of streets has increased year by year(trend χ~2= 13.157,P <0.001);18-60 years old cases(77.21%)accounted for the majority;The incidence rate of males(8.17 / 100,000)was higher than that of females(7.50 /100,000),and the difference was statistically significant(χ~2= 6.367,P<0.012).In the occupational distribution,the proportion of the people including housework and unemployment,business service staffs,workers,retirees,and students accounted for 72.11%.2.GLM model showed that the average temperature and the 24-hourprecipitation of the previous 2 month had a significant impact on the epidemic of dengue fever in Guangzhou.On the premise that other factors remained the same,the monthly risk of dengue fever would increase 1.18 times and 1.06 times for each additional unit of these factors.3.GAM model showed that Bretto Index(BI),average temperature,average air pressure,24-hour precipitation,and average humidity of the previous 2 month had significant effects on the number of dengue fever cases each month in Guangzhou.When other covariates remained the same,for each additional unit of these factors,the monthly risk of dengue fever increased by 1.38 times,2.21 times,1.23 times,1.12 times,and 1.08 times,respectively.There were interactions between average air pressure and sunshine hours(F=4.575,P=0.007).4.Meteorological factors such as air temperature,sunshine hours,and the number of dengue fever had a lag positive correlation,while the air pressure had a lag negative correlation with the number of dengue fever cases.Spearman correlation coefficients between the average temperature as well as the average pressure in the first two months and the number of cases reached 0.934 and-0.949,respectively(all P <0.001).5.There was a lag positive correlation between the vector density index and the number of dengue fever cases.Among them,the Spearmancorrelation coefficients of the number of dengue fever cases and BI,SSI,and MOI in the first two months as well as ADI in the first three months were 0.934,0.931,0.936,and 0.622,respectively(all P <0.001).6.There was a correlation between the keyword BSI and the dengue fever epidemic trend.Among them,"Dengue Fever" BSI has the highest correlation with the number of daily/weekly/monthly dengue fever cases,with Spearman correlation coefficients of 0.793,0.845,and 0.862(P<0.001).7.Forecast the number of daily cases,5 variables("dengue fever" BSI,"dengue fever prevention" BSI,"dengue fever Symptom" BSI,average pressure before 63 days,average temperature before 63 days)were screened,and to construct BP,RF,SVR models trained with the first 70%of the three-year time series data set and verified by the last 30%,R2 reached 0.523,0.709,and 0.637.RMSE was 5.146,3.823,and 3.906,respectively.8.Forecast the number of weekly cases,5 variables("dengue fever Prevention" BSI,average pressure before 9 weeks,maximum pressure before 9 weeks,"dengue fever" BSI,minimum pressure before 9 weeks)were screened,and BP,RF,and SVR models were constructed.It was trained with the first 70% of the three-year time series data set and verifiedby the last 30%.R2 reached 0.783,0.784,and 0.841.RMSE was 30.805,20.049,and 19.246,respectively.9.Forecast the number of monthly cases,4 variables were screened(mean pressure before 2 months,“dengue fever prevention” BSI,BI before 2 months,minimum temperature before 2 months)were screened,BP,RF,SVR models were constructed.It was trained with the first 70% of the three-year time series data set and verified by the last 30%.R2 reached0.863,0.878,and 0.862.RMSE was 67.943,64.138,and 72.164,respectively.Conclusion1.Our research showed that the incidence of dengue fever was increasing in the past three years.Strengthen prevention and control efforts are suggested to control the epidemic.2.There was a lag correlation and lag effect relationship between the meteorological and vector factors and the number of dengue fever cases.The vector control measures to control dengue fever should be taken in advance and continuously.3.Our current analysis based on 2017-2019 data showed that the meteorological factors previously and the near-real-time dengue fever related BSI of the keywords as well as vector index BI could forecast thetrend of dengue fever in Guangzhou and provide auxiliary decision support for dengue fever surveillance early warning.