| Objectives:In order to explore the high-risk clustering areas of tuberculosis(TB),analyze the TB incidence related social environmental factors quantitively,and predict the TB incidence through the meteorological factor,we performed the systematic research of TB in Qinghai province at the ecology level.Considering the surveillance data had time and space attributes,the spatial-temporal statistical analysis and the spatial econometric model was used in this study.The spatial geographic information systems(GIS),space-time clustering analysis and spatial econometric model could be applied to the surveillance data mining of infectious disease,which would provide not only the analysis method and reference for the similar research,but also the scientific basis for the government policy.Methods:In this study,we focused on the cases of pulmonary TB.The TB data were collected from the China information system for disease control and prevention,and the demographics data and related social and meteorological factors data were collected from Statistical Yearbooks of Qinghai province.The study was performed as follows:1.The epidemiological characteristics of TB in Qinghai province from 2009 to 2013 was analyzed by using traditional epidemiological description of three dimension distribution,concentration and circular distribution method,seasonal index number method and three-dimensional trend analysis.2.Considering the spatio-temporal distribution of surveillance data was not independent,the spatial,temporal and spatio-temporal clustering characteristics was analyzed by using the Moran's I,Getis-Ord G spatial autocorrelation and Kulldorff's SaTScan space-time scanning analysis,the high-risk areas of TB in Qinghai province was detected at the county level from 2009 to 2013,and the risk intensity was also evaluated.3.Considering the spatial distribution of cross-section data was not independent,the relationship of TB annual incidence in 2011 and 2013 with six social indicators at the county level in Qinghai province,including the expenditures of health care(thousand yuan per person),education(thousand yuan per person),number of bed in the medical institutions(bed per thousand people),personnel allocation in the medical institutions(person per thousand person),per capita net income of rural residents(thousand yuan),and per capita GDP(ten thousand yuan),was analyzed by using the Moran's I bivariate spatial autocorrelation analysis and spatial cross-section regression model,which could be used to study the spatial aggregation characteristics of social factors,and quantitatively explore the TB-related social factors after deducting the interaction of TB incidence in the adjacent areas.4.TB incidence showed the seasonal periodicity,and the meteorological factors and early TB incidence had the time lag effect on the TB incidence of native area,the monthly TB incidence at the city level of Qinghai province from 2009 to 2013 was set as the dependent variable,and monthly average temperature(MAT,°C),monthly precipitation(MP,mm),monthly total sunshine hours(MSH,hours),monthly average wind speed(MAWS,m/s)and monthly incidence with 0-to 6-month lag were set as the independent variables,which the panel data model was fit to explore the best lag phase of TB incidence influenced by the meteorological factors.After that,the optimal analysis model was determined by using F test,Hausman test,Moran's I test and Lagrange multiplier(LM)test,and the relationship between meteorological factors and TB incidence was analyzed quantitatively.5.Considering the TB incidence had 3-month lag effect influenced by the meteorological factors in the spatial panel data model,which also contained the TB incidence of the adjacent region in the same period,the TB incidence of different cities was predicted by using the expert modelers,then,the monthly modeling data and predicting data were used to construct the spatial panel data model,the regression predicted value was obtained and the predicted accuracy was evaluated.Based the above results,we explored the feasibility of short-term prediction on the incidence by combining expert modelers with the spatial panel data model.Results:1.The annual incidence of TB was 98.26/100,000 in Qinghai province,which was significantly higher than that of the national average rate,and had a slightly upward trend in recent years.The TB incidence was the highest in the middle-aged or elder crowds,and followed by the young adults.The incidence of men was higher than that of women,and the farmers and herdsmen were the high-risk crowds.The time distribution of TB had obvious periodicity and weak seasonal characteristic,and the peaks were mainly concentrated in March to May.The regional distribution of TB had an arc change trend in north-south direction,which the incidence in south was higher than that in north,and had obvious inverted "U" trend in east-west direction.2.The global Moran's I and General G spatial autocorrelation values were greater than the expected values,and the values were ranged from 0.398 to 0.581 and 0.029 to 0.034 respectively,which indicated that the annual incidence of TB had obvious regional distribution of high value clustering tendency.The SaTScan space-time scanning analysis also showed that the TB incidence had obvious high-risk clustering in spatial,temporal,and spatio-temporal distribution in Qinghai province.The most likely spatio-temporal cluster(LLR=1860.09,RR=4.58,P<0.001)was mainly concentrated in the southwest of Qinghai,the center of this area was in Nangqian County,located in 32.17° N and 96.12° E,which was a circular area with the radius of 421.00 Km,including eight counties: Nangqian,Yushu,Zaduo,Chengduo,Qumalai,Dari,Maduo,and Banma,and the incidence peak appeared in January 2012 to June 2013.Combined with the spatial scanning and spatial autocorrelation analysis,the high-risk areas of TB were mainly concentrated in 12 counties,which belonging to Yushu and Guoluo cities,located in the southwest of Qinghai province.3.The Moran's I bivariate spatial autocorrelation analysis showed that four social indicators,such as number of bed in the medical institutions,personnel allocation in the medical institutions,per capita net income of rural residents and per capita GDP,had statistical influence on TB incidence in 2011(P<0.05),which suggested that the above social indicators may affect the local incidence level.As the distribution of the TB incidence rate is highly skewed,log transformation of the TB incidence was used in the analyses,according to the following formula: log incidence=lg(TB incidence).The ordinary least square(OLS)regression showed that the regression residual was not independent(Moran's I=0.16,P<0.05).According to the LM test,the spatial lag model was the best model,the spatial autocorrelation coefficient was 0.4041,which means that a spatial spillover phenomenon existed,the TB incidence of the local county increased 1.54 times when the TB incidence of adjacent counties increased nine times and other influencing factors were kept constant.After adjusting the spatial autocorrelation of TB incidence,the rural per capita net income was the mainly social factor which affected the TB annual incidence.b=-0.0657,which means the TB incidence of the local county decreased for 14% when the rural per capita net income increased for one thousand yuan and the other influencing factors were kept constant.Compared with the OLS,the absolute value of the regression coefficient was decreased,which indicated that the spatial cross-section regression model was more reasonable by considering the spatial autocorrelation of the incidence,and the traditional regression model exaggerated the role of social factors.The same results obtained in 2013 compared with that of 2011.4.The associations between TB incidence and meteorological factors with a 3-month lag were found to have the best goodness of fit by using the panel data model.After the logarithmic transformation of the TB incidence,fixed effect model(F=193.90,H=10.41,P<0.05)showed that the regression residual was not independent(Moran's I=0.20,P<0.05).According to the LM test,spatial lag fixed effect panel data model was the best model,the result showed that the intercept term was different in the different cities,which reflected the spatial heterogeneity of the dependent variable.The spatial autocorrelation coefficient was 0.3017,indicating that a spatial spillover phenomenon existed on monthly incidence in the adjacent regions.The TB monthly incidence of the local city increased one time when the incidence of adjacent cities increased nine times and other influencing factors were kept constant.Compared to the meteorological factors,the influence of the 3-month lag TB incidence on the dependent variable was most obvious.After adjusting the spatial autocorrelation,spatial heterogeneity and the effect of early stage of TB incidence,monthly average temperature and monthly precipitation of 3-month lag were the mainly meteorological factors of TB monthly incidence.With each nine times,10°C and two centimeters increase in the incidence,temperature and precipitation of 3-month lag being associated with 36% increment,9% and 3% decrements in the TB incidence,respectively.Compared with the traditional regression model,the spatial panel data model was more reasonable.The conclusion was the same as the spatial cross-section regression model.5.The results of time series expert modelers showed that the relative error of prediction of monthly incidence from October 2013 to December 2013 were 0.90%-136.14%,the average relative error was 28.99%.And the relative error of prediction of expert modelers combined with the spatial panel data model were 0.17%-94.20%,the average relative error was 21.09%.Similarly,the results showed that the average relative error from January 2014 to March 2014 predicted by two methods were 26.60% and 19.79%,respectively.The combination of expert modelers with spatial panel data model analysis could improve the accuracy of prediction and decrease the relative error obviously.Conclusions:This study made a detailed analysis on the monitoring data of tuberculosis in Qinghai province at the ecology level,which was analyzed by using the spatio-temporal statistical analysis method and the spatial econometric model for the first time.The conclusions were as follows:1.As the spatio-temporal distribution of surveillance data was not independent,the spatial autocorrelation analysis and Kulldorff's spatio-temporal scanning analysis were the optimal analytical methods for the spatio-temporal aggregation and the detection of high-risk clustering areas of disease.And the clustering areas of TB were mainly concentrated in the southwest of Qinghai province,the highest risk region was concentrated in Yushu county(r=259 Km,RR=3.77),which covering Yushu,Nangqian,Chenduo,Zaduo,Maduo,and Qumalai counties.2.Taking the temporal and spatial attributes of the incidence into account,the spatial econometric model,such as spatial cross-section regression model and spatial panel data model,were the optimal models for the analysis of ecological influencing factors,which could be applied in the public health extensively.After deducting the temporal and spatial effects of incidence,the rural per capita net income,temperature and precipitation were the main social and environmental factors which affected the incidence of TB in the local region.3.Compared with the simple time series expert modelers prediction,the predictive strategy of the expert modelers combined with the spatial panel data model has improved when considering the incidence influenced by the adjacent regions and the meteorological factors,which could be applied for early warning.In conclusion,this study would provide the theoretical basis for establishment of prevention and control measures for tuberculosis in Qinghai province,and provide the analysis method and reference for similar research. |
PDF Full Text Request