| Objectives:1.To analyze the epidemiological characteristics and reveal the spatial distribution of pulmonary tuberculosis in Hubei Province for the purpose of identifying high-risk areas and time periods and providing guidance on tuberculosis(TB)prevention and control.2.To analyze the relationship between social economic and environmental factors and pulmonary TB from the dimension of time and space for providing scientific evidence to prevent and control effectively.3.To explore the feasibility of ARIMA time series analysis,artificial neural network and combined model in prediction for the incidence of TB in Hubei province and find the optimal forecasting model for the purpose of providing a theoretical reference for the establishment of pulmonary TB forecasting and warning system.Methods:1.We analyzed the long-term trend of reported cases and incidences of TB in Hubei from 2006 to 2015 by Cochran-Armitage Trend Test.The data of TB incidence was used as the basic data to match the GIS geospatial database of Hubei.Arc GIS10.1 software was used to map the annual incidence of pulmonary TB.2.Using Geo Da1.8 software for global Moran ’I spatial autocorrelation analysis,we explored spatial autocorrelation on pulmonary TB in Hubei.Local Moran ’ I spatial autocorrelation analysis was used to analyze the distribution of high risk areas and spatial clustering of TB.Using Sa TScan 9.1 software for simple spatial scanning and spatio-temporal scanning analysis,we analyzed the high-risk regions and time periods for the occurrence of city-level TB in Hubei from 2006 to 2015.3.We collected the data of socioeconomic and environmental information from 2006 to2015.The stepwise regression method was used for initially screening variables.Four Bayesian spatio-temporal models were constructed by Win BUGS1.4 software.The DIC(deviance information criterion)value was used to select the best model.Then,this best spatio-temporal model was used to analyze the impact of socio-economic and environmental factors on the TB from the perspective of space and time.A map was draw to compare the distribution difference between the incidence risk and the actual incidence of pulmonary TB in all regions in Hubei from 2006 to 2015.4.The ARIMA and artificial neural network model were respectively constructed to predict the rates of pulmonary TB incidence from January to December in 2015,based on the monthly incidence rate of pulmonary TB in Hubei from 2011 to 2014.Using the fitting sequence data from ARIMA model,we constructed a new artificial neural network to predict.Compared with the actual incidence of TB in from January to December 2015,we test the predictive value of three models.The mean square error(MSE),the mean absolute error(MAE),the mean absolute percentage error(MAPE)were used to evaluate and compare the accuracy and stability.Results:1.The numbers of TB reported cases and incidence showed a significant downward trend from 2006 to 2015,and the monthly incidence had an obvious seasonal change in Hubei.The top three areas of pulmonary TB incidence in Hubei were Enshi city,Yichang city and Xianning city.The largest decline was Yichang city with 38.7% reduction in TB incidence.The incidence of population increased gradually as ages,on the peak of the 65 to 74 ages old group.The incidence was significantly higher in male.The gender differences became larger when people were older.In all occupational groups,peasants reported the highest number of cases,accounting for 62.04% of all cases.2.The global spatial autocorrelation analysis of pulmonary TB incidence in Hubei during 2011 to 2015 showed that the Moran’I values were 0.42,0.54,0.45,0.41 and 0.53.Moran scatter plots mainly focus on the positive-positive and negative-negative quadrants,which suggested that TB incidence had a significant positive spatial autocorrelation in Hubei Province(P <0.001).Local spatial autocorrelation analysis found that rates of TB incidence in27,28,29,32,and 33 counties has spatial agglomeration,and the numbers of high TB risk areas were 13,14,18,17 and 16 in each year respectively.By simple spatial scanning analysis of TB incidence in Hubei from 2006 to 2015,we found that high-risk aggregations appeared in 2006 to 2009 and 2013 to 2015.The numbers of most likely and second clusters were 7 and3 respectively.We also found that the average incidence and relative risk(RR)in cluster areas in 2006 to 2009 were higher than in 2013 to 2015,which indicated that the high risk in hot spots of TB had decreased in Hubei.The result of spatio-temporal scanning analysis found one most likely clustering area and one second clustering area.The high-risk clustering period was from 2006 to 2008.3.Through stepwise multiple regression analysis,five indicators of per capita disposable income of rural residents,Engel coefficient of rural residents,proportion of elderly population,atmospheric PM10 concentration and atmospheric SO2 concentration were considered in stepwise regression model(P<0.05).Using the analysis of the spatial-temporal interaction effect model to analyze,it was found that TB incidence was positively related with the per capita disposable income of rural residents,and negatively related with the Engel’s coefficient of rural residents,the proportion of the elderly population and the atmospheric SO2 concentration.Through the model calculation,the risk of tuberculosis in each city from2006 to 2015 was obtained.Compared with the actual incidence in areas,we found that the risk of morbidity is not always high in areas with a high TB incidence.4.According to the goodness-of-fit test(BIC=-1.876)and residual white noise test(P>0.05),ARIMA(0,1,0)*(1,1,0)12 was selected as the best time series model to predict TB incidence.The predictive results showed that ARIMA model,artificial neural networks model and combined model all fitted the incidence data of pulmonary tuberculosis well.In these models,MAPE were less than 20%,and predictive incidence curves were close to the actual curve,which indicated that prediction results of the three models were effective.By comparison on the results of three models predicting TB incidence(1/100,000),it was found that the values of MSE(0.934),MAE(0.725),and MAPE(10.255%)in the combined model were smallest,as well as the predicted incidence rate curve closest to the actual curve,the prediction error rate curve closest to the center of zero.These results showed that the effect of the combined model predictions was better than ARIMA and artificial neural network model to predict alone.Conclusions:1.The number of TB reported cases and incidence showed a significant decline trend and seasonally cyclical in Hubei from 2006 to 2015.The incidence was high in male and high-age people.Number S of reported cases were highest in farmers,accounting for 62.04%in the total reported cases.The TB incidence of Hubei province has spatial agglomeration.The hot spot areas were mainly concentrated in western and eastern regions.2.Using the Bayesian spatial-temporal interaction effect model,the variations from time and space were combined together,which effectively fitted the socio-economic and environmental factors and TB incidence and estimated the risk of tuberculosis accurately in different regions.The results showed that TB incidence was related to the disposable income of rural residents per capita,the Engel coefficient of rural residents,the proportion of old age people,and the air SO2 concentrations in Hubei.The risk was not always high in areas,where the actual TB incidence was high.3.The effects of ARIMA time series,artificial neural network and combined models were well in forecasting TB incidence in a short time.Among all predictive models,the accuracy and stability were best in the combined model,which can be used to predict the TB incidence of Hubei province in future.Innovations:1.We used Arc GIS software to describe the spatial distribution of pulmonary tuberculosis,Geo Da software to perform global and local spatial autocorrelation analysis,Sa TScan software to make spatial and spatiotemporal scan analysis,which identified high-risk areas and high-risk periods and dynamic changes effectively in Hubei province.2.We used space-time analysis technology to combine the two dimensions of space and time,and analyzed the influencing factors of pulmonary TB in multiple aspects of population,economy,health and environment.The spatial and temporal information of infectious disease were used effectively in Bayesian spatial-temporal interaction effect model,which can avoid the limitations of traditional statistical analysis and assess the impact of risk factors accurately.From the space-time dimension,The model also estimated the risk of tuberculosis of various regions effectively,which was helpful to understand the influence of social environment-related factors on the risk of pulmonary tuberculosis.3.We built three predictive models to explore the best model for the prediction of TB incidence in Hubei,based on ARIMA time series,artificial neural networks and combination methods from the perspective of linearity and non-linearity.The results can help the establishment of prediction and early warning system of tuberculosis. |
PDF Full Text Request