| Objective:The prevention and control of dengue fever have become a focused work of the World Health Organization.Hangzhou meets the basic conditions for dengue fever transmission,so the prevention and control of dengue fever cannot be ignored.Monitoring and early warning are important components of dengue control.At present,there are few dengue early warning models applicable to Hangzhou,and most of them are early warning models established using cases.In order to promote the multi-stage early warning of dengue fever and supplement the early multi-source early warning,this study divided the early warning according to the occurrence and development process of infectious disease epidemics.By understanding the distribution of dengue fever in Hangzhou,this study selected suitable data sources and methods to establish an early warning model for the sources and an early warning model for the signs.This may provide a basis for early detection of dengue outbreaks in Hangzhou to minimize the harm caused by dengue.Methods:Through theoretical learning,the early warning of dengue fever is divided into three stages: epidemic source early warning,epidemic symptom early warning,and epidemic early warning.This study collected and sorted out dengue disease surveillance data,meteorological data,mosquito monitoring data,and the Baidu index of keywords in Hangzhou from 2017 to 2019,taking day,week,and month as the time dimension.Statistical charts were then used to describe the epidemiological characteristics of the cases.After correlation analysis and independent variable screening,local cases were taken as dependent variables.The meteorological data,mosquito monitoring data,and input cases were included in the independent variables,and the epidemic source early warning model was established using the Poisson regression model,support vector machine,and random forest model.The meteorological data,mosquito monitoring data,input cases,and Baidu index were included in the independent variables,and the support vector machine and random forest model were used to establish the epidemic symptom early warning model.The early warning effect was evaluated with indicators such as coincidence rate,sensitivity,specificity,Yoden index,and AUC value.Finally,a better combination of phased early warning models was selected.Result:(1)The incidence rate of local cases of dengue fever in Hangzhou from 2017 to2019 was 11.92/100,000,0.25/100,000,and 0.45/100,000.It is high in summer and autumn,with peaks mainly concentrated from August to October.The local cases were mainly concentrated in Gongshu District(328 cases,27.31%),Xiacheng District(275cases,22.90%),and Shangcheng District(200 cases,16.65%).The occupations of the local cases were mainly retirees,business services,domestic and unemployed,workers,and cadres and employees,accounting for 80.68% of the total local cases.The imported cases were mainly imported from abroad,coming from Cambodia(63 cases,28.25%)and Thailand(44 cases,19.73%).(2)The main meteorological factors and input cases in this study had a lag correlation with the number of daily reported cases.The correlation coefficient between the average temperature,maximum temperature,and minimum temperature and the number of daily reported cases was the highest,with the lag time of 151 days,138 days,and 151 days respectively.There was a significant cross-correlation between the average air pressure,the maximum air pressure,the minimum air pressure,the average air temperature,the maximum air temperature,the minimum air temperature,the average relative humidity,the minimum humidity,the average wind speed,the input cases,and the number of weekly reported cases.The maximum lag correlation coefficient between the average air temperature,the maximum air temperature,and the minimum air temperature,and the number of cases was the highest,and the lag time was 21 weeks,19 weeks,and 20 weeks respectively.The minimum air pressure,average air temperature,maximum air temperature,minimum air temperature,and average wind speed had a lag correlation with the number of local monthly cases.The lag time is 3 months,4 months,4 months,4 months,and 1 month respectively.(3)The Baidu index of major keywords in this study was mostly correlated with the number of the daily reported case.The Baidu index of "dengue fever symptoms" was most correlated with the number of local cases,and the lagged correlation coefficients with the number of daily,weekly,and monthly local cases were 0.794,0.837,and 0.839.(4)In the epidemic source early warning model,the three early warning models based on the time dimension of the day were not effective.The random forest model with the week as the time dimension had a better effect,with a sensitivity of 0.77,specificity of 0.95,AUC value of 0.860,Yoden index of 0.72,and a coincidence rate of90.38% The model could send the early warning signal 4 weeks in advance.The Poisson regression model established with the month as time dimension had a good effect,with a sensitivity of 1.00,specificity of 1.00,AUC value of 1.000,Yoden index of 1.00,and coincidence rate of 100.00%.This model could send a warning signal 1month in advance.(5)In the epidemic symptom early warning model,the three early warning models based on the time dimension of day were not effective.The random forest model with the week as the time dimension had a better effect,with a sensitivity of 0.85,specificity of 0.92,AUC value of 0.885,Yoden index of 0.77,and a coincidence rate of 90.38%.The model could send out warning signals 1 week in advance.Conclusions:(1)This study showed that there was a lag correlation between meteorological factors,the Baidu index of keywords,and the number of local dengue cases in Hangzhou.The lag time of meteorological factors was mostly 3-5 months,and that of the Baidu index of keywords was mostly 3-6 days.(2)The monthly early warning model of the Poisson regression model and the weekly early warning model of the random forest model showed better early warning effectiveness among epidemic source early warning models.The weekly early warning model with the random forest model showed better early warning effectiveness among epidemic symptom early warning models.(3)The phased early warning of dengue fever was meaningful.The monthly early warning had a long-time span and was relatively rough,so it could be considered a supplement to the weekly early warning.Parallel epidemic source early warning and epidemic symptom early warning can improve the sensitivity of early warning and realize early warning. |
PDF Full Text Request