Evaluation And Forecast Of The Effect Of Meteorological Factors On Bacillary Dysentery In Northern China

BackgroundPrevious studies have shown that meteorological factors paly an important role in transmission of infectious diseases by affecting the reproduction ability and survival time of pathogens,changing the geographical distribution range of vectors and changing the susceptibility of people to diseases.In 2016,212,438 people globally died from bacillary dysentery(BD),with 30%of all deaths occurring among children under five years old.The incidence of BD in China is still at high level.The seasonality of BD transmission suggest that BD transmission may be related to meteorological factors.BD transmission presents a complex distribution rule in time and space.Previous studies have quantified the relationship between meteorological factors and BD by using multiple regression analysis,distributed lag nonlinear model(DLNM)and other methods.The relationships exploration from the perspective of time and space were poorly understood.In addition,existing studies have inconsistent results.The effect of meteorological factors on BD transmission varies regionally.Modifiers of the association between them has not been thoroughly discussed.Previous studies used seasonal autoregressive moving average model(SARIMA),generalized additive model(GAM),regression tree and support vector machine regression model(SVR)to forecast the incidence of disease.However,there may be strong correlation and interaction among meteorological factors.When using traditional statistical models,if the independent variables are included too much or not fully,it is likely to lose information,resulting in a decline in the prediction accuracy.Although SVR and neural network models have achieved good results in predicting accuracy,they still have the problem of black box.Recent years,boosted regression tree model(BRT)has achieved good results in machine learning competition.The model could not only fit the nonlinear relationship between independent variables and dependent variables,but also automatically calculate the interaction between variables.The model has high accuracy and strong interpretability.Our previous study showed that BRT was better than GAM and SARIMA in Beijing,China.China Infectious Diseases Automated alert and Response System was used to give early warning to the occurrence of infectious diseases.The early warning system compared the number of the occurrence of an infectious disease in the current observation period with the pre-set early warning threshold.Early warning signal will be issued when the early warning threshold is exceeded.However,the early warning system is carried out after the occurrence of cases,and there is a lag period.If we could send out early warning signals before the occurrence of diseases,it will be more helpful for local health departments to take timely measures to control BD epidemic.Objectives(1)Understanding the epidemiological characteristics of BD in Northern region during 2014 to 2016 on the dimension of time,space and population.(2)Quantifying the effect of meteorological factors on BD transmission.(3)Exploring modifiers of the association between meteorological factors and BD transmission.(4)Forecasting the weekly number of BD cases and detect outbreaks.MethodsData from 1 January 2014 to 31 December 2016 were provided by China CDC.Meteorological variables during the study period were retrieved from the China Meteorological Data Sharing Service System.City-specific social characteristics were collected from statistical yearbooks.Descriptive analysis was used to describe the distribution of the number of BD cases and meteorological factors in Northern China.Spearman correlation was used to analyze the correlation between meteorological factors.From the perspective of time,a GeoDetector was employed to quantify the determinant powers of meteorological factors on BD transmission.A DLNM was then conducted to quantify the effects of meteorological factors on BD transmission.From the perspective of space,a Bayesian spatial and temporal model was used to identify the hot spots.From the perspective of time and space,Bayesian temporal model,Bayesian spatial and temporal model and Bayesian spatial and temporal interaction model were established to quantify the relationship between meteorological factors and BD transmission.The best model was chosen on the basis of deviance information criterion rules.The methodology was compared with DLNM in warm(May to September)A multivariate meta-analysis was used to evaluate the cumulative effects for all cities.Meta-regression models were then to investigate the potential heterogeneity.Finally,data from 2014 to 2015 were used to fit the BRT model.Data of 2016 was used for external validation.Root mean square error(RMSE),mean absolute deviation(MAE),and the squared correlation coefficient(R2)were used to evaluate the model performance.The warning threshold was defined as 75th,80th,85th,90th and 95th percentiles of number of BD cases,respectively.Sensitivity,specificity and Youden index were used to evaluate the predictive accuracy.ResultsDuring 2014 to 2016,there were a total of 101 584 BD cases in Northern China.The times series plot of BD showed an obvious seasonal pattern with a peak in summer(May to October).There were more male cases with a male-to-female sex ratio of 1.22:1.The age group of 0-4 years old had the largest proportion to the numbers of BD.Most cases were farmers and children.Beijing had the largest incidence rate,followed by Tianjin city.There was significant local spatial autocorrelation in Northern China.Beijing,Tianjin,Langfang,Yangquan,Taiyuan,Changzhi,Handan and Xinxiang city are the hot spots of BD.Temperature was the most important meteorological factor contributing to BD transmission regardless of different cities.The largest effect of average temperature on BD was observed at lag day 3.An approximately J-shaped relationships were observed between pooled BD risk and average temperature.Taking the median temperature as a reference,each 1℃ rise average temperature resulted in a 2.42%(95%CI:1.13%,3.73%)increase in BD cases.A positive linear relationship was observed for relative humidity and BD.The largest effect of relative humidity on BD was observed at lag day 0.In terms of cumulative effects from lag 0 to 7 days,each 1%rise relative humidity resulted in a 0.37%(95%CI:0.21%,0.52%)increase in BD cases.In warm season(May to September),the DLNM showed that each 1℃ increase temperature caused a 1.09%(95%CI:0.68%,1.50%)growth in BD cases at lag of 3 days.The Bayesian spatial and temporal interaction model showed that each 1℃increase temperature caused a 1.44%(95%CI:0.62%,2.27%)growth in BD cases at lag of 3 days.The effect of temperature was more obvious in city-level regions at high latitude and longitude,with high natural population growth rate but low urban rate and hospitals.The forecasting model achieved sound forecasting performance with RMSE varying from 29.01 to 35.14,MAE varying from 22.11 to 26.76 and R2 varying from 0.91 to 0.93 for the study areas.The early warning model attained better discriminatory power with Youden index varying from 0.85 to 0.98,sensitivity of 1 and specificity varying from 0.85 to 0.98.When the 80th and 95th percentiles of number of BD cases was defined as early warning thresholds,the Youden index of the test set has the largest value of 0.98.The number of early warning cases was 320 and 382,respectively.ConclusionsTemperature played the most important role in weather-attributable BD transmission.High temperature and high humidity may promote BD incidence in Northern China.The effect of temperature was more obvious in city-level regions at high latitude and longitude,with high natural population growth rate but low urban rate and hospitals.The use of BRT model was a valid approach for monitoring the dynamics of BD and forecasting the outbreaks in the study area.Moreover,the model based on meteorological factors and historical cases had the better performance.