Application Of Bayesian Spatio-Temporal Modeling In The Analysis Of Hemorrhagic Fever With Renal Syndrome Influencing Factors

Hemorrhagic fever with renal syndrome(HFRS)is a natural foci disease caused by hantavirus(HV),which is widespread all over the world.Humans are primarily infected through contact with infected rodents or the inhalation aerosols of their excrement.According to data from China Center for Disease Control and Prevention,after 2010,the incidence and mortality of HFRS have continuously rise,which is a serious problem threatening the public health.In 2014,the number of HFRS cases in Shandong Province ranked second among all provinces in China.The distribution,population density and HV infection rate of rodent hosts are largely affected by environmental conditions,determining the incidence and geographic distribution of human HFRS.In recent years,there has been more and more regional unit data,which provide spatio-temporal observation data in a certain space unit and continuous time period.It is great significant to analysis these data.With the development of Bayesian statistic,Bayesian spatio-temporal models have been increasingly applied.Compared with traditional modeling methods,the use of prior distributions describes unknown parameters in the model and the random effects represent variations in time and space.The relationship between influencing factors and diseases is modified by effects in time and space,so as to realize the analysis of time,space and related factors,and improve model accuracy.Bayesian analysis of disease spatio-temporal data can study the spatial and temporal distribution characteristics and changing laws of disease.In this study,Bayesian spatio-temporal model was used to analyze the influencing factors of the incidence of HFRS in Shandong Province on the county scale.Compared with the commonly used analysis models,it can provide usable ideas and methods for the research of similar diseases.The dynamic changes of disease in time and space are the basic characteristics of disease epidemics.The time and space distribution of HFRS epidemics are very different.Explore the changes in the risk of disease in time and space to provide suggestions for further research.Materials and MethodsThis study used traditional epidemiological methods to describe the epidemiological characteristics of HFRS cases in Shandong Province from 2009 to 2018,including the epidemic situation and distributions(time,space,and human).Socioeconomic factors,geographic environmental factors,and meteorological factors in Shandong Province from 2009 to 2018 were collected to conduct Bayesian spatio-temporal analysis of the incidence of each county and district to explore the influencing factors of HFRS in Shandong Province.Using the generalized linear model(GLM),including the quasi-Poisson regression model(QP)and the negative binomial hurdle model(NBH)model,the influencing factors of HFRS were analyzed.Under the framework of Bayesian,conditional autoregressive linear spatio-temporal model(CARLSTM)is used,combined with conditional autoregressive priors and spatio-temporal random effects,to conduct spatio-temporal autocorrelation analysis of HFRS in Shandong Province.The relationship between annual HFRS incidence and socioeconomic factors,geographic environment factors,and meteorological factors was analyzed.The relationship between monthly incidence rate and precipitation,average air pressure,average wind speed,temperature,relative humidity,and sunshine duration was analyzed.Finally,the fitting effects of the models are comparedResults1.During the study period,there were 12362 cases in Shandong Province.The average annual incidence was 1.27 per 100,000,and the highest fatality was 3.02%in 2018.The monthly incidence presents obvious periodicity and seasonality There are two peaks each year.The large peak from October to November and the small peak from May to June.In terms of spatial distribution,the counties with higher annual incidence of HFRS in Shandong Province are mostly located in the southeast and central regions.The number of males is higher than that of females in each year,and the number of males is 2.62 times that of females.The 41-60 age group has the largest number of cases.The number of cases in farmer group is significantly higher than that in other groups,accounting for 84.18%of the total number of cases,followed by workers and students.2.The annual HFRS analysis found that population density,gross domestic production(GDP),number of beds in health institutions,normalized difference vegetation index(NDVI),altitude,average air pressure,average wind speed,maximum temperature,relative humidity,and sunshine duration regressed by QP were statistically significant(P<0.05).There are more zero observations in the case data,accounting for 23.56%.Therefore,the NBH model was used.The population density,number of beds in health institutions,altitude,average wind speed,maximum temperature,relative humidity and sunshine duration in the zero-censoring process were statistically significant(P<0.05).These variables affect the incidence of HFRS.In the CARLSTM model,population density,GDP,total power of agricultural machinery,number of beds in health institutions,NDVI,altitude,average air pressure and average wind speed are statistically significant(P<0.05),and precipitation is not statistically significant(P>0.05).It is found that the spatial autocorrelation of the model residuals is reduced compared to the original data3.The monthly HFRS analysis takes the large peak from October to November and the small peak from May to June as the research unit.The QP model and.NBH model fit best when the lag is 4 months,and the CARLSTM model fits best when the lag is 5 months.The CARLSTM model analysis found that counties with more precipitation,high average air pressure,and long sunshine duration from May to June have a high risk of HFRS incidence in large peaks;counties with high average wind speeds and high maximum temperatures have a low risk of HFRS in large peaks.Counties with more precipitation,high relative humidity,and long sunshine duration from December to January have a high risk of HFRS at small peaks;counties with high average wind speeds have a low risk of HFRS at small peaks.Conclusion1.The law of change of HFRS in Shandong Province has not changed significantly.There are two peaks each year.High annual incidence of HFRS are located in the southeastern and central regions of Shandong Province.The 40-69 years old,males,and farmers are the most affected people group.2.Annual population density,GDP,total power of agricultural machinery,number of beds in health institutions,NDVI,altitude,average air pressure and average wind speed affect the incidence of HFRS in all counties in Shandong Province.Among the meteorological factors analyzed on a monthly basis,average wind speed and maximum temperature from May to June were the major factors for the risk of HFRS incidence in large peaks.For small peaks,precipitation and sunshine duration from December to January were the major factors.3.In the model analysis and comparison for Shandong HFRS,the fitting effect and model performance of the CARLSTM are better than the GLM,which indicates the effectiveness of its application.Using Bayesian spatio-temporal model to analyze spatio-temporal data can provide tools and methods for future analysis of infectious disease data and reference for application and expansion in the field of public health research.