Study On The Temporal And Spatial Prediction Of Fever With Thrombocytopenia Syndrome

Background Severe fever with thrombocytopenia syndrome(SFTS)is an important vector-borne communicable disease which was firstly identified in 2009 in China by Chinese scientists.SFTS incidences have increased and its endemic areas have expanded year by year since its identification.Of note,the fatality of SFTS is very high.Increasing incidence and high fatality lead to great damage to human health and life.In recent years,more attentions have been paid to human health and passive measures including early detection,early confirmation and early quarantine can’ t meet the need of communicable disease control and prevention in current situation.It is essential to predict the spatial distribution and temporal distribution accurately and conduct targeted interventions before occurrence of cases.In this study,we will analyze the changes of epidemiological characteristics and spatial-temporal dynamics of SFTS cases,explore the key meteorological factors and environmental factors associated with spatial distribution and temporal distribution of SFTS and constructed predicting models for SFTS.The results will provide scientific information for health resource allocation and precise control and prevention of SFTS.Methods Data on confirmed SFTS cases from 2011 to 2016 were collected from Chinese Information System for Diseases Control and Prevention.Meteorological data including mean temperature,maximum temperature,minimum temperature,minimum relative humidity,mean relative humidity,and precipitation were obtained from Chinese Meteorological Data Sharing Service System.Data on elevation,land cover,NDVI,density of sheep,goats,pigs,poultry were also collected.Then we established database of SFTS cases,meteorological factors and environmental factors.Descriptive analysis was conducted to explore changes of epidemiological characteristics including gender,age,occupation,seasonal distribution,geographic distribution and interval from illness onset to confirmation using SPSS 20.0.Single variable analysis and multivariate analysis were conducted to identify factors associated with SFTS fatal outcome using the Logistic regression method.Global and local spatial autocorrelation analysis were carried out to analyze the spatial patterns of SFTS incidence clusters on the county level during the study period through use of Geographic Information System.Local spatial autocorrelation was used to explore significant hot spots(High-High),cold spots(Low-Low),and spatial outliers(High-Low and Low-High)by calculating local Moran’s I between a given location and the average of neighboring values in the surrounding locations.Temporal clustering of SFTS incidence was identified using SaTScan software.Meteorological factors associates with SFTS temporal distribution were analyzed using a distributed lag non-linear model(DLNM).Forecast models were constructed with autoregressive integrated moving average(ARIMA)model,negative binomial regression model(NBM),and quasi-Poisson generalized additive model(GAM).The dataset from 2011 to 2015 were used for model construction and the dataset in 2016 were used for external validity assessment.Intraclass correlation coefficient(ICC)was applied to verify the consistency between the actual and predicted data.Ecological niche models of SFTS were constructed to explore meteorological factors and environmental factors associated with SFTS geographic distribution using Maxent.Receiver operating characteristic curve(ROC)was used to assess the model and risk of SFTSF occurrence in different areas were predicted using the optimal model.Results From 2011 to 2016,5360 laboratory confirmed cases were identified and reported.Annual case numbers increased year by year,with the highest recorded in 2016.The majority of SFTS cases were farmers and the occupation distribution of SFTS cases in different years were similar(χ2=15.552,P=0.113).Most SFTS cases occurred in individuals aged between 40 years and 80 years(91.57%).During 2011-2016,99.53%of SFTS cases were limited to 7 provinces:2025 cases were reported in Henan Province,1515 cases in Shandong Province,663 cases in Hubei Province,502 cases in Anhui Province,260 cases in Zhejiang Province,224 cases in Liaoning Province,and 146 cases in Jiangsu Province.About 98.00%(5253/5360)of SFTS cases were reported from April to October with a peak in May,June,and July.The seasonal distributions of different years were significantly different(Fisher=276.845,P=0.000).Of note,the seasonal distributions of SFTS cases in different provinces were also significantly different(Fisher=721.157,P=0.000).SFTS cases in Liaoning Province showed shorter epidemic periods than other province.SFTS cases in Zhejiang Province,Hubei Province,Anhui Province,and Henan Province showed earlier peaks than those in Liaoning Province,Shandong Province and Jiangsu Province(April-October vs.May-October).A total of 343 deaths were reported in China from 2011 to 2016 and the average case fatality rate was 6.40%.According to results of multivariable Logistic regression analysis,variables in the final equation included year,province and age.In the spatial clustering analysis,the global autocorrelation suggested that SFTS cases were not of random distribution.Local spatial autocorrelation analysis of SFTS showed that hot spots shifted from 2011 to 2016,but hot spots mainly concentrated in Hubei Province,Henan Province,Anhui Province,Shandong Province,Liaoning Province,and Zhejiang Province.Spatio-temporal cluster analysis indicated that three clusters were observed and the most likely cluster was observed in the central region of China.The cluster consisted of 21 counties in Henan Province and Hubei Province.There were 13,936,345 individuals in the cluster with a radius of 109.94 km.Most importantly,the expected number of SFTS was 27.74,but the observed number was 1659.The relative risk(RR)and the Log Likelihood Ratio for the analysis were 86.15 and 5435.3616.Cluster 2 was located in Jiaodong peninsula,Shangdong Province and consisted of 18 counties.Cluster 3 was located in the central areas of Shandong Province and consisted of 28 counties.Time frames of the three clusters were 2015/4-2016/8,2015/5-2016/10,2015/5-2016/9,respectively.Results of DLNM indicated that a non-linear relationship between weekly temperature and weekly number of SFTS cases.The exposure-response curve was an approximately reversed U-shape and the curve peaked at 23℃.High temperatures had acute and short-term effects,while low temperatures had persistent and long-term effects.The effects of lower temperatures(1.62 ℃ and 6.97 ℃)could last 24 weeks,but the effect of 29.30 ℃ was not significant at lag 8 weeks.According to the results of ACF,PACF,and AIC,ARIMA(2,0,0)X(1,0,0)12 was constructed as the optimal model.Based on the optimal ARIMA model,we predicted monthly number of SFTS cases from January to December,2016 in the study area.When compared with the observed data,the majority of predicted data agreed closely with the observed data.The ICC of the model was 0.924 and the fit of the model was significant.According to the value of AIC,the best model of NBM include the number of SFTS cases in the previous month,monthly maximum temperature,monthly mean relative humidity,and the month ordinal when SFTS cases occurred.The exponentiates of the model coefficients of SFTS cases in the previous month,monthly mean relative humidity,and monthly maximum temperature were 1.0117,1.1031,and 1.2568,respectively.These results indicated that the percent change in the SFTS occurrence is a 1.17%,25.68%,and 10.31%increase for every unit increase in SFTS cases in the previous month,monthly maximum temperature,and monthly mean relative humidity,respectively.According to the values of the R square,deviance explained(%),and GCV principles,the best model of Quasi-Poisson GAM also include the number of SFTS cases in the previous month,monthly maximum temperature,monthly mean relative humidity,and the month ordinal when SFTS cases occurred.Constructed ARIMA,NBM,and Quasi-Poisson GAM fitted the cases reasonably well during the training process and forecast process.However,the NBM model forecasted better than Quasi-GAM and ARIMA model according to the ICC values of the three models.Ecological niche models indicated that elevation,mean temperature and precipitation were significantly associated with SFTS geographic distribution.The model predicted that some areas in Shandong Province,Henan Province,Hubei Provinces,and Liaoning Province were at high risk of SFTS occurrence.Areas under ROC curve(AUC)of ecological niche models from 2011 to 2016 were 0.957,0.958,0.955,0.950,0.956 and 0.942,which indicated that the model forecast well.Conclusion The number of SFTS cases have increased and the affected areas have expanded year by year.Some epidemiological characteristics of SFTS cases including age distribution,seasonal distribution,geographic distribution,and interval from illness onset to confirmation also changed in recent years.SFTS cases were not of random distribution and clusters were identified in Henan Province,Hubei Province and Shandong Province.A U-shape non-linear relationship was revealed between weekly temperature and weekly number of SFTS cases.High temperatures had acute and short-term effects,while low temperatures had persistent and long-term effects.ARIMA,NBM,and GAM can fit the cases reasonably well during the training process and forecast process,while the NBM model forecasted better than other two models.Ecological niche models can assess risk of SFTS cases occurrence in different areas accurately and factors including elevation,mean temperature,and precipitation were associated with geographic distribution of SFTS.