The Influence And Prediction Of Meteorological Factors On Tuberculosis

Backgrounds:Tuberculosis(TB)is a transmissible disease caused by Mycobacterium tuberculosis,and is one of the most widely reported Class B infectious diseases in China.TB is the thirteenth leading cause of death worldwide,posing a severe threat to human health.In 2021,the estimated number of TB cases in China is 780,000,accounting for 7.4%of the total number of cases in the world,ranking the third in the world,next to India and Indonesia,and ranking the second in China’s legally reported class A and B infectious diseases.The overall prevention and control situation in Shandong Province was relatively severe,as it ranked second in the monthly TB of legally reported infectious diseases in 2021.There was no effective TB vaccine for the whole population,it is needed to conduct research on the influencing factors and predictive warning of TB incidence.As an infectious disease with seasonal distribution,meteorological factors and other external environmental factors play an important role in the onset and spread of TB.Several researches have demonstrated the influence of meteorological factors on the occurrence of TB.,but the influencing factors and their effect sizes in different study areas and time scales were inconsistent,and even contrary results appeared,suggesting that there might be spatial heterogeneity in the occurrence and spread of TB.The spatial panel data model has a good performance in processing spatially dependent data.This model allows the consideration of the spatial interaction effect of the cross-sectional dimension to control the spatial dependence and spatial heterogeneity of the data,and then discusses the direct effects and spatial spillover effects of various influencing factors on the risk of disease.The development of TB prediction model and timely and effective intervention can effectively reduce the risk of TB epidemic.However,most of the current prediction models adopted time series models,such as moving average autoregressive model and grey model,which had simple structure and good performance in short-and medium-term prediction of the research object,but had high requirements on data quality.When the data have large fluctuations,the model fitting effect was mediocre,and the inclusion of covariables was limited by collinearity between meteorological factors.Random Forest(RF)model data has good adaptability and is relatively insensitive to collinearity problems,which better improves model accuracy and avoids overfitting problems.With high efficiency of model algorithm,simple and effective practical prediction models can be established.On the basis of describing the epidemiological characteristics of TB in Shandong Province,this study used global spatial autocorrelation analysis and local spatial autocorrelation analysis to explore the temporal and spatial clustering characteristics of TB in Shandong Province,built a spatial panel data model to analyze the meteorological factors affecting the prevalence of TB in Shandong Province and their spatial spillover effects,and built a prediction model to predict the incidence trend of TB.It was helpful to quantify the influence of meteorological factors on TB incidence and provide policy suggestions for local TB prevention and control.Data and methods:In this study,data of TB cases reported in Shandong Province from January 1,2016 to December 31,2020,meteorological data of the same period and vector map data of districts and counties in Shandong Province were collected.Descriptive analysis was made on the epidemiological characteristics of TB in Shandong Province to explore its epidemic rule,and identify the epidemic season,high incidence area and population of the disease.To describe the basic situation of meteorological factors in Shandong Province from 2016 to 2020,and draw corresponding time series maps;Global spatial autocorrelation and local spatial autocorrelation analysis were used to determine the high incidence of TB aggregation areas in Shandong Province.The mixed regression model,panel data model and spatial panel data model were constructed respectively,and the goodness of fit of the model was compared to select the best model.The influence of meteorological factors on TB distribution in Shandong Province was quantitatively analyzed,and the spatial spillover phenomenon was further explored.With the collected data from 2016 to 2019 as the training set and the data from 2020 as the test set,the random forest model was constructed to predict the incidence trend of TB in Shandong Province in 2020,at the same time,the Seasonal Autoregressive Integrated Moving Average(SARIMA)model and Support Vector Machines(SVM)model were constructed.Root Mean Square Error(RMSE),Mean absolute percentage error(MAPE)and correlation coefficient squared(Rsquare,R2)index to evaluate the fitting and prediction effect of the model.Results:1.A total of 136,557 TB cases were reported in Shandong Province from 2016 to 2020,with an overall decreasing trend of TB reporting.There was a clear seasonality and cyclicality in the temporal distribution,with two peaks in March and at the end of each year.More men than women reported TB incidence,with a male to female sex ratio of 2.41:1;TB may occur in all age groups,with fewer cases in the age groups below 15 years and 85 years and above.The highest number of cases were reported in the age group of 55-64 years;most cases were farmers,with a composition ratio of 70.55%.2.The results of spatial autocorrelation showed that there was a spatial autocorrelation of TB in Shandong Province;the results of cold and hot spot analysis showed that the hot spot areas of TB in Shandong Province were mainly distributed in Liaocheng,Qingdao,Linyi and some counties in Heze City,and the cold spot areas were mainly distributed in Dongying,Dezhou and some districts and counties in Binzhou City in the northern part of Shandong Province and Yantai City,and the range of cold spot areas had an increasing tendency.3.The results of the analysis of influencing factors showed that relative humidity,average wind speed and average air pressure were the main meteorological factors influencing the onset of TB.Relative humidity(β=-0.0099,P<0.05)was negatively correlated with the incidence of TB in Shandong Province,while average wind speed(β=0.0738,P<0.05)and average air pressure(β=0.0195,P<0.05)were positively correlated with the incidence of TB in Shandong Province.And there were spatial spillover effects of relative humidity(β=0.0110,P<0.05)and average wind speed(β=-0.0997,P<0.05)on TB incidence in neighboring districts and counties(β=0.0110,P<0.05),and the results of effect decomposition showed that the total effect of relative humidity was 0.0013,the direct effect was-0.0097,and the indirect effect was 0.0110;the total effect of average wind speed was-0.0290,the direct effect was 0.0719,and the indirect effect was-0.1009.4.The RF prediction model had a training set R2 of 0.85 and a test set R2 of 0.84,which was better than the SARIMA(R2=0.66)model and the SVM(training set R2=0.37,test set R2=0.49)model,and the RF prediction model had a smaller RMSE and MAPE,so the RF prediction model fitted better than SARIMA and SVM.Conclusions:1.TB in Shandong Province was well prevalent in adult male farmers,showing a clear seasonal and cyclical distribution,and an overall decreasing trend in the number of reported incidences.2.Spatial autocorrelation and spatio-temporal aggregation existed in the distribution of TB in Shandong Province,and the high-risk areas were mainly concentrated in some counties in the west,southwest and southeast of Shandong Province.3.The results of the spatial panel data model showed that relative humidity,average wind speed,and average air pressure were the main meteorological factors affecting the onset of TB,and the model was able to explore the spatial spillover phenomenon of the influencing factors,and its fitting effect was better than the traditional mixed regression model and panel data model.4.The fitting effect of RF model was better than that of SARIMA model and SVM model,and the prediction effect was better for the number of TB reported incidence in Shandong province.