Comprehensive Risk Zoning,and Forecasting And Early Warning Of Bacillary Dysentery Based On Meteorological Factors

BackgroundAs global warming continues to intensify,the impact of climate change on human health has received widespread attention in recent years.Since meteorological factors can affect the three basic links in the epidemic process of infectious diseases,climate change has a significant impact on intestinal infectious diseases,insect-borne infectious diseases,and so on.Revealing the influence of meteorological factors on meteorologically sensitive infectious diseases,establishing a forecasting and early warning model based on it,and then formulating and adopting targeted intervention strategies and measures will have public health significance for reducing the burden of meteorologically sensitive infectious diseases and improving the people’s adaptability to climate change.Bacillary dysentery(BD)is an intestinal infectious disease caused by Shigella infection.It is highly contagious and is mainly transmitted through the fecal-oral route.People are generally susceptible.In China,BD is one of the national notifiable class B infectious diseases,and its morbidity usually ranked top five among all national notifiable class A and B infectious diseases in recent years.The emergence of drug resistance in Shigella makes the treatment of BD challenging.Meanwhile,the protective effect of the BD vaccine is limited.Therefore,the focus of the prevention and control of BD is to explore the factors affecting the incidence of BD,remove or reduce the promoting factors,establish a forecasting and early warning model,and then take practical and effective measures before the epidemic of BD to prevent the epidemic or reduce adverse effects.Previous studies have found that temperature and relative humidity are important factors affecting BD incidence.However,these studies usually selected different study areas,study periods,data precision,statistical models,and parameters,so their results were different or even contradictory.At present,there is a lack of studies using a unified data format,statistical model,and parameters to obtain the effects of meteorological factors on BD incidence and to explore the sources of heterogeneity of the effects in a large geographic range.In addition,some studies have found an interactive effect between temperature and relative humidity on BD incidence,but these studies still have limitations in study methods and/or study areas.Meanwhile,previous studies usually used relative risk(RR)to report the effects of meteorological factors on BD incidence.RR can’t evaluate the burden of BD caused by meteorological factors.There is still a lack of attributable risk assessment studies and comprehensive risk assessment studies that have more public health significance.In terms of forecasting and early warning research,there is still a lack of practical forecasting and early warning models for BD.Objectives1.To master the population,temporal,and spatial distribution characteristics of BD in China.2.To obtain the effects of meteorological factors on BD incidence at the city,regional,and national levels,respectively,to explore the sources of heterogeneity of effects among cities,and to quantitatively assess the attributable risk of BD caused by meteorological factors.3.To assess the comprehensive risk of BD due to meteorological factors,and to identify the cities with high comprehensive risks.4.To establish a practical forecasting and early warning model of BD for cities with high comprehensive risks.Methods1.Study area and data sources320 cities were selected as the study area,whose BD cases were greater than the 5th percentile(P5)of the BD cases of 334 prefecture-level administrative regions and 4 municipalities.From 2004 to 2018,the yearly morbidity of BD of 31 provincial administrative regions in mainland China was downloaded from the Public Health Science Data Center.Daily reports of BD for the 320 cities from 2014 to 2019 were derived from the National Notifiable Infectious Diseases Reporting Information System.Daily meteorological data for the 320 cities from 2014 to 2019 were obtained from the China Meteorological Data Sharing Service System.Since humidex can reflect the combined effect of temperature and relative humidity,humidex was calculated based on average temperature and average relative humidity.From 2014 to 2019,the demographic characteristics,economic levels,and health resources information of the 320 cities were collected from province-level or city-level statistical yearbooks.The geographic locations of the 320 cities were obtained from the China Meteorological Data Sharing Service System.2.Statistical analysis(1)Combined with statistical figures and tables,the Joinpoint regression model and flexible spatial scan statistic were used to describe the population,temporal,and spatial distribution of BD incidence in China.(2)The two-stage time series analysis was conducted to obtain the effects of meteorological factors on BD incidence at the city,regional,and national levels,respectively.In the first stage,the city-specific effects of meteorological factors on BD incidence were obtained using the distributed lag non-linear model.In the second stage,the multivariate metaanalysis model was applied to obtain the pooled effects at the regional and national levels,respectively,and the multivariate meta-regression model was used to explore the sources of heterogeneity among city-specific effects.The attributable risk assessment method under the framework of the distributed lag non-linear model was adopted to quantitatively evaluate the attributable risks of BD due to meteorological factors.(3)Based on the results of the two-stage time series analysis and attributable risk assessment,the "Hazard-Exposure-Vulnerability" comprehensive risk assessment framework was selected to quantitatively evaluate the comprehensive risks of BD caused by meteorological factors.The natural breakpoint method was applied to classify comprehensive risks and identify the cities with high comprehensive risks.(4)One representative city was selected from cities with high comprehensive risks as the study site for the forecasting and early warning model.The meteorological and BD data of the study site were aggregated into weekly data.The weekly data from 2014 to 2018 was used as the training set to establish the forecasting and early warning model,and the weekly data for 2019 was used as the test set to evaluate the effect of the forecasting and early warning model.The Boruta algorithm was used to screen predictor variables,and then the support vector regression(SVR)was applied to establish the forecasting model using the grid search method to determine its parameters.The forecasting effect was evaluated by the coefficient of determination,root mean square error,and mean absolute percentage error.Then,the early warning model was established based on the forecasted cases,using the percentile method to determine the early warning gold standard for the actual cases and the receiver operating characteristic curve(ROC)to determine the early warning threshold for the forecasted cases.The early warning effect was evaluated by the sensitivity,specificity,positive predictive value,and negative predictive value.Results1.From 2014 to 2019,a total of 682,762 BD cases were reported in the 320 cities,with a male-to-female sex ratio of 1.2:1.In terms of occupational distribution,scattered children had the largest cases(31.52%),followed by farmers(30%).In terms of age distribution,people aged 0～5 years old have the largest cases(33.53%).In terms of temporal distribution,BD incidence has an obvious peak in summer and autumn,and the peak value decreased year by year.The results of the Joinpoint regression model showed that from 2004 to2018 BD incidence showed a significant downward trend in mainland China,with an average annual decline of 11.56%.The BD incidence of people aged 0～5 years old declined the most,but its incidence was consistently higher than that of other age groups.The BD incidence of people aged 15～24 years old declined at the fastest speed.The BD incidence of people aged 60 years and older declined at the slowest speed.In terms of spatial distribution,the results of the flexible spatial scan statistic showed that BD incidence had significant spatial clustering,and the most likely cluster was located in Beijing and Tianjin.2.At the national level,the cumulative exposure-response relationship between average temperature and BD incidence was approximately a straight line.High temperature was a risk factor for BD incidence,with the maximum cumulative RR of 1.25[95%confidence interval(CI):1.15～1.35].At the P95(27.26℃)of average temperature,the hazard effect appeared and reached the maximum on the current day,and disappeared about 11 d later.High relative humidity greater than 76.90%was a risk factor for BD incidence,with the maximum cumulative RR of 1.08(95%CI:1.01～1.14).At the P95(90.23%)of average relative humidity,the hazard effect appeared on the same day,reached the maximum at a lag of 3 d,and disappeared at a lag of 8 d.The cumulative exposure-response relationship between humidex and BD incidence was"J" shaped.High humidex was a risk factor for BD incidence,with the maximum cumulative RR of 1.42(95%CI:1.30～1.56).At the P95(35.81)of humidex,the hazard effect appeared and reached the maximum on the current day,and disappeared at a lag of 11 d.The effects of humidex on BD incidence had significant heterogeneity among regions.The effects of high humidex were larger in regions with higher latitude,but smaller in regions with a higher natural growth rate and a higher number of primary school students per thousand persons.The results of attributable risk assessment of high humidex showed that attributable risks were significant in 41 cities.3.The comprehensive risk assessment identified 7 cities with high comprehensive risks of BD caused by high humidex,including Hulunbuir City,Benxi City,Tacheng Region,Beijing City,Tianjin City,Tangshan City,and Qinhuangdao City.4.Beijing was selected as the study site for the forecasting and early warning model.The coefficient of determination,root mean square error,and mean absolute percentage error of the SVR on the test set were 0.89,18.83,and 15.27%,respectively.Taking the P55 of the weekly BD cases in the training set as the gold standard,the early warning threshold determined by the ROC for the forecasted BD cases was 153 cases.The sensitivity,specificity,positive predictive value,and negative predictive value of the early warning model on the test set were 100%,97.22%,94.12%,and 100%,respectively.Conclusions1.BD usually occurred in males,people aged 0～5 years old,scattered children,and farmers in China.In recent years,remarkable results have been achieved in the prevention and control of BD,with BD morbidity of the total population and each age group showing significant downward trends.BD incidence peaked in summer and autumn and had spatial clustering.2.High temperature,high relative humidity,and high humidex were the risk factors for BD incidence,and high relative humidity could enhance the effect of high temperature.Latitude,natural growth rate,and the number of primary school students per thousand persons were the effect modifiers of the relationship between humidex and BD incidence.3.Cities with high comprehensive risks of BD caused by high humidex were mainly distributed in the northern regions,especially in North China.4.Based on the Boruta algorithm,SVR,percentile method,and ROC,the forecasting and early warning model mainly composed of meteorological factors is simple,practical,and the effect is favorable.