This article has Open Peer Review reports available.
Local spatial variations analysis of smear-positive tuberculosis in Xinjiang using Geographically Weighted Regression model
© The Author(s). 2016
Received: 4 February 2016
Accepted: 27 September 2016
Published: 6 October 2016
The spatial interplay between socioeconomic factors and tuberculosis (TB) cases contributes to the understanding of regional tuberculosis burdens. Historically, local Poisson Geographically Weighted Regression (GWR) has allowed for the identification of the geographic disparities of TB cases and their relevant socioeconomic determinants, thereby forecasting local regression coefficients for the relations between the incidence of TB and its socioeconomic determinants. Therefore, the aims of this study were to: (1) identify the socioeconomic determinants of geographic disparities of smear positive TB in Xinjiang, China (2) confirm if the incidence of smear positive TB and its associated socioeconomic determinants demonstrate spatial variability (3) compare the performance of two main models: one is Ordinary Least Square Regression (OLS), and the other local GWR model.
Reported smear-positive TB cases in Xinjiang were extracted from the TB surveillance system database during 2004–2010. The average number of smear-positive TB cases notified in Xinjiang was collected from 98 districts/counties. The population density (POPden), proportion of minorities (PROmin), number of infectious disease network reporting agencies (NUMagen), proportion of agricultural population (PROagr), and per capita annual gross domestic product (per capita GDP) were gathered from the Xinjiang Statistical Yearbook covering a period from 2004 to 2010. The OLS model and GWR model were then utilized to investigate socioeconomic determinants of smear-positive TB cases. Geoda 1.6.7, and GWR 4.0 software were used for data analysis.
Our findings indicate that the relations between the average number of smear-positive TB cases notified in Xinjiang and their socioeconomic determinants (POPden, PROmin, NUMagen, PROagr, and per capita GDP) were significantly spatially non-stationary. This means that in some areas more smear-positive TB cases could be related to higher socioeconomic determinant regression coefficients, but in some areas more smear-positive TB cases were found to do with lower socioeconomic determinant regression coefficients. We also found out that the GWR model could be better exploited to geographically differentiate the relationships between the average number of smear-positive TB cases and their socioeconomic determinants, which could interpret the dataset better (adjusted R2 = 0.912, AICc = 1107.22) than the OLS model (adjusted R2 = 0.768, AICc = 1196.74).
POPden, PROmin, NUMagen, PROagr, and per capita GDP are socioeconomic determinants of smear-positive TB cases. Comprehending the spatial heterogeneity of POPden, PROmin, NUMagen, PROagr, per capita GDP, and smear-positive TB cases could provide valuable information for TB precaution and control strategies.
KeywordsTuberculosis Ordinary least square regression model Geographically weighted regression model Local spatial variations
Tuberculosis (TB) now ranks alongside HIV as one dominant cause of death worldwide . Without treatment, the mortality of TB is high, the same is the mortality of sputum smear-positive TB cases, and smear-positive TB cases are highly infectious [1, 2]. Studies from the pre-chemotherapy era indicated that about 70 % of people with sputum smear-positive pulmonary TB died within 10 years , and national tuberculosis programs concentrated on the diagnosis and treatment of sputum smear positive TB cases. In 2013, it was estimated that there were a total of 9.0 million new TB cases and 1.5 million deaths due to TB . The 22 TB high-burden countries accounted for 80 % of the world’s TB cases, and China ranked second on that list, accounting for 12 % of global incidence of TB [2, 3]. Xinjiang has been confirmed as one of the high TB burdened (high TB case notification rates) provinces of China, as reported by the Chinese Center for Disease Control and Prevention (CDC) based on data collected during the period from 2010 to 2013 . Although the notification rates of TB in Xinjiang have shown an absolute downward trend from 172.73/100,000 in 2010 to 164.46/100,000 in 2013, the current rate remains significantly higher than the national average [4-6].
The correlation between the average number of smear-positive TB cases and relative socioeconomic determinants has been well confirmed over the past five years. Some epidemiologic studies have pointed out that the magnitude of the problem varies across settings, possibly due to unfavorable socioeconomic conditions, overcrowding, poverty, socio-cultural barriers and HIV infection [7, 8]. Other spatial studies in China have demonstrated that population density, and economic level are latent risk factors for the spread of TB in China . Xinjiang is a multi-ethnic area, with a minority population of 49.49 million out of a total 72.70 million people. Minorities in Xinjiang refer to Uygurs, Kazaks, Hui, and other 46 minority nationalities, whose total population is less than Han. The lifestyle between most of the minorities and Han differs from each other. Under such a backdrop, a large proportion of the minority population was a special determinant factor for TB prevalence in Xinjiang, compared with other provinces in China. Wubuli A successfully found a correlation between TB incidence and the proportion of minority populations in Xinjiang, but failed in explaining the geographic disparities of sputum smear-positive TB incidences and relative socioeconomic factors .
Historically, researches to understand the relationship between smear-positive TB and relative socioeconomic determinants have meant a great deal to public health personnel and policy makers. In China, the correlation between socioeconomic factors and TB incidences has been shown to vary geographically, especially in Xinjiang province [5, 10]. Local modeling approaches have enabled investigators to much accurately estimate the geographical differences in relations between TB incidences and socioeconomic factors. Local Geographically Weighted Regression (GWR) modeling techniques were utilized to calculate local regression coefficients, which allowed health professors to better assess how the effects of socioeconomic determinants change by geographic location [11, 12]. Thus the objectives of this study were to: (1) identify the socioeconomic determinants of the geographic disparities of smear positive tuberculosis in Xinjiang, China (2) confirm if the average number of smear-positive TB cases and relative risk factors demonstrate spatial variability in Xinjiang, and (3) compare the performances of the two models: the OLS model, and the GWR model.
This study applies the average number of smear-positive TB cases notified in Xinjiang as the measurement of TB prevalence and dependent variable, and takes the five socioeconomic determinants as independent variables. The average number of smear-positive TB cases is calculated from arithmetic mean for the total number of smear-positive TB cases during 2004–2010. Similarly, other five values of the same period are also obtained by this method, including “POPden (population at year-end/land area)”, “PROmin (population of minorities/population at year-end)”, “NUMagen (infectious disease network reporting agencies numbers/population at year-end × 1000)”, “PROagr (agricultural population /population at year-end)”, “per capita GDP”.
Coefficients of OLS results
95 % Confidence Interval
Robust t statistic
Per capita GDP
Diagnostics of OLS results
Number of Observations
Akaike’s Information Criterion (AICc)
Multiple R squared
Joint F statistic
Prob(>F), (5,96) degrees of freedom
Joint Wald statistic
Prob(>chi-squared), (5) degrees of freedom
Prob(>chi-squared), (5) degrees of freedom
Prob(>chi-squared), (2) degrees of freedom
The spatial independency of residuals in OLS model was evaluated by the spatial autocorrelation coefficient, namely Moran’s I, which ranges from −1 (negative autocorrelation) to +1 (positive spatial autocorrelation). Positive spatial autocorrelation (0 < Moran’s I ≤ 1) meant there were similar values in adjacent areas, while negative autocorrelation (−1 ≤ Moran’s I < 0) implied the dissimilar values at the nearby locations. Nevertheless, if there was no spatial autocorrelation (Moran’s I = 0) found, the spatial arrangement would be completely random . If it found the existence of spatial autocorrelation in the residuals (Moran’s I ≠ 0), OLS model would not fit the dataset.
where y i is dependent variable at location i, x i represents kth independent variable at location i, and ε i is the Gaussian error at location i, (u i , v i ) is the x-y coordinate of the ith location; and coefficients β k (u i , v i ) are varying conditions at location i.
The paper examined the spatial variability of an estimated local regression coefficient in order to determine whether there was spatial heterogeneity in this process [16, 13]. Additionally, through corrected Akaike Information Criterion (AICc) and adjusted coefficient of determination (Adjusted R2), it evaluated the comparison of OLS model and GWR model.
The analysis of this article was completed on the basis of smear-positive TB data. OLS model was processed by GeoDa 1.6.7 software with 0.05 significant levels, while GWR model was implemented by GWR 4.0, in which different Kernel Type and estimated Bandwidth Methods were attempted. It turned out the model with the lowest AIC statistic is one with the best model fit , according to the AIC goodness of fit statistic for comparing models. This criterion was also used to compare different GWR models and OLS model. The GWR regression coefficients of five determinants in 98 districts/counties were compared by absolute values.
The VIF for OLS model indicated that OLS estimations were not biased from multicollinearnity. Overall OLS model is trustworthy since Joint Wald Statistic is of significance in statistics (P < 0.05) (Table 1). However, the relations between the average number of smear-positive TB cases and socioeconomic determinants estimated using OLS model were weak and biased (R2 = 0.779, adjusted R2 = 0.7680), with p-value for BP statistic less than 0.05. After examining the residuals of OLS model, it was found that the residuals had positive spatial autocorrelation (Moran’s I = 0.0492, P = 0.0226). The Moran’s I indicated that there was a spatial autocorrelation of the residuals between the average number of smear-positive TB cases and related socioeconomic determinants; meanwhile, local relations couldn’t be estimated by the OLS model. We further employed GWR model to fit the data, since the existence of dependent residuals are violated to the assumptions of OLS model .
Combinations between different Kernel type and Bandwidth method
Gaussian kernel type
Summary of GWR results
Per capita GDP
As indicated by the PROmin coefficients distribution (Fig. 5b), the PROmin coefficients were positive. Higher PROmin coefficients were distributed in Kashi, Hetan, and Bayingolin [Red areas], and lower PROmin coefficients in Tacheng, Yili, and partial Altay [Blue areas]. The regression coefficients of POPagr were negative in Altay, Urumqi, Changji, Turpan, and partial Hami [Blue areas], when POPagr coefficients were mostly positive in other prefectures (Fig. 5c).
The per capita GDP regression coefficients (Fig. 5d) were negative in most areas of Xinjiang. Higher per capita GDP coefficients were distributed in Hetan, Kashi, Kirgiz, and partial Bayingolin [Blue areas], and lower per capita GDP coefficients in East-Xinjiang [Red areas].
Figure 5e showed the evidence of non-stationary of POPden coefficients. The contour map of local coefficients of POPden showed that POPden coefficients were varied spatially. South-West Xinjiang (Kirgiz, Kashi) [Red areas] presented higher POPden coefficients, and North-Xinjiang (Altay, Tacheng, Changji, Urumqi) [Blue areas] lower POPden coefficients.
Figure 5f represented the spatial distribution of intercept term, indicating POPden, PROmin, NUMagen, PROagr, and per capita GDP didn’t cause impact on the average number of smear-positive TB cases. Higher intercept values lied in South-Xinjiang (Hetan, Kashi, Aksu, Kirgiz) [Blue areas]. The distribution implied that besides five socioeconomic determinants, there were still other factors bound up with the average number of smear-positive TB cases (Fig. 5f).
The geographical heterogeneity was detected by GWR model in terms of the relations between the average number of smear-positive TB cases and the corresponding socioeconomic determinants, since the existence of dependent residuals is violated to the assumptions of OLS model. The values of R2, adjusted R2, and AIC statistic all indicated that GWR model was fitted better for the relations. However, we also provided an evidence for the limitation of GWR model. GWR model was very sensitive to the chosen Kernel Type and Bandwidth Method. The similar finding was reported by Lin CH, with regard to the relations of entomology and dengue cases .
Our study also provided further indications that there were non-stationary relations between the average number of smear-positive TB cases and related socioeconomic factors in Xinjiang. It is evident that intensity and direction of NUMagen’s influence were different in Xinjiang. The NUMagen coefficients in Altay, Hami, Changji, Urumqi, Turpan were negative, whereas the coefficients were mostly positive in other areas. In South-Xinjiang, higher PROmin, PROagr, and POPden were shown to contribute to more smear-positive TB cases. The results also demonstrated that areas with higher per capita GDP were in relation to lower average number of smear-positive TB cases. Therefore, the results presented in this study indicate that the spatial interplay between five socioeconomic determinants are of vital significance to the distribution of the average number of smear-positive TB cases in Xinjiang, since a certain distribution trend of five determinants coefficients has shown the corresponding influences on the average number of smear-positive TB cases. And TB prevention and control strategy may intensify in these areas with other complexity factors, which cannot be explained by known determinants.
There are several limitations that deserve to be discussed. First, our spatial regression analysis was focused on socioeconomic determinants of the average number of smear-positive TB cases; however, other determinants of the average number of smear-positive TB cases were also essential, including geographical environment , climate conditions [18–20], cultural customs . In terms of longitude, there were a lack of unified conclusions upon the influence of longitude. A previous finding in China showed that higher TB incidence in some areas was interrelated with higher longitude . However, Vargas MH demonstrated that higher TB incidence in some areas was closely related with lower longitude . And we noticed that, in Xinjiang, some areas with higher longitude (Kunlun Mountains) were related with more smear-positive TB cases, and lower longitude (Junggar Basin) more smear-positive TB cases as well. Therefore, more determinants of the average number of smear-positive TB cases should be taken into account in further studies, such as longitude , and average temperature . Second, we collected smear-positive TB data during the period from 2004 to 2010, and the timeliness and representation of smear positive TB data may be terrible. However, we were purely concerned about the relations between socioeconomic determinants and the average number of smear-positive TB cases, and made a comparison of the fit of OLS and GWR model. Thus, the timeliness would have less impact on the relations. Thirdly, we used the average number of smear-positive TB cases as the dependent variable to estimate the regression coefficients, which would affect the fit of GWR model. Given this, further studies should utilize more TB incidence variables to estimate the regression coefficients in GWR model.
In this study, the spatial heterogeneity of smear-positive TB was analyzed, and GWR model was set up, based on smear-positive TB data during the period from 2004 to 2010, as well as affecting factors. Our findings suggested that GWR model could explain the spatial variation of the dataset, whereas OLS model could not. At the same time, the chosen determinants explained the mostly variation of the average number of smear-positive TB cases during the period from 2004 to 2010. The average number of smear-positive TB cases distributed in South-Xinjiang was higher, and complexity factors affected the TB prevalence in these areas. Therefore, regional strategies aimed at TB prevention and control should be identified in accordance with the relations between socioeconomic determinants and TB.
We are pleased to acknowledge the support of the Natural Science Foundation Projects of Xinjiang Uygur Autonomous Region (2016D01C167). We are also pleased to acknowledge the Supported Disciplines of Health Metrics and Health Economics of Xinjiang Medical University(XYDXK50780308), the Data-Center of China Public Health Science, and the Data Sharing Infrastructure of Earth System Science - Xinjiang & Central Asia Science Data Center.
This study was funded by the Natural Science Foundation Projects of Xinjiang Uygur Autonomous Region (2016D01C167). The sponsor of this study had no role in the collection, analysis, or interpretation of the data.
Availability of data and materials
The data that support the findings of this study are available from the Prevention and the Data-Center of China Public Health Science (www.phsciencedata.cn/), and the Data Sharing Infrastructure of Earth System Science - Xinjiang & Central Asia Science Data Center (http://www.egi.cas.cn/).
WW extracted the data, conducted spatial statistical analysis and drafted the manuscript. JYY helped with the data collection and extraction. YC and AA responded for modification of the figures and tables in accordance with format. CMQ was a tutor and statistical expert. All of authors have read and approved the final manuscript.
The authors declare that they have no competing interests.
Consent for publication
Ethics approval and consent to participate
The study protocol and utilization of smear-positive TB cases were reviewed by the Xinjiang Uygur Autonomous Region center for Disease Control and Prevention and no ethical issues were identified. All patient information was anonymized and de-identified prior to analysis. No tissue samples were used by the authors in this study. And all data used for this study were statistical data. Therefore, no ethics approval was required by our Investigation Review Board.
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.
- World Health Organization. Global tuberculosis report 2014. 2015. Available: http://apps.who.int/iris/bitstream/10665/191102/1/9789241565059. Assessed 21 August.Google Scholar
- WHO report 2014. Global Tuberculosis Control: Surveillance, Planning, Financing. Geneva, Switzerland: World Organization; 2014.Google Scholar
- Wang L, Liu W, Wang L, et al. HIV Prevalence Among Pulmonary Tuberculosis Patients in Guangxi, China. J Acquir Immune Defic Syndr. 2010. doi:10.1097/QAI.0b013e3181c7db2e.Google Scholar
- Chinese Center for Disease Control Prevention. Data of Report of Notifiable Infectious Diseases. 2015. Available: http://www.phsciencedata.cn/Share/en/data.jsp?id=a56cd203-cd11-414d-9efad1583b97476fa&show=0. Accessed July 6th, 2014.Google Scholar
- Wubuli A, Xue F, Jiang D, et al. PLoS ONE. 2015. doi:10.1371/journal.pone.0144010.PubMedPubMed CentralGoogle Scholar
- Zhao RL, Liu YH, Sun JL. Analysis of TB/HIV co-infection epidemic situation in Yining, Xinjiang. Chin J Antitubercul. 2009;31(5):308–9.Google Scholar
- Dangisso MH, Datiko DG, Lindtjørn B. Spatio-Temporal Analysis of Smear-Positive Tuberculosis in the Sidama Zone, Southern Ethiopia. PLoS One. 2015;10(6):e0126369.View ArticlePubMedPubMed CentralGoogle Scholar
- Oxlade O, Murray M. Tuberculosis and poverty: why are the poor at greater risk in India? PLoS One. 2012;7(11):e47533.View ArticlePubMedPubMed CentralGoogle Scholar
- Sun W, Gong J, Zhou J, et al. A Spatial, Social and Environmental Study of Tuberculosis in China Using Statistical and GIS Technology. Int J Environ Res Public Health. 2015;12(2):1425–48.View ArticlePubMedPubMed CentralGoogle Scholar
- Xin-Xu L, Li-Xia W, Hui Z, et al. BMC Public Health. 2014;14:257.View ArticleGoogle Scholar
- Haque U, Scott LM, Hashizume M, Fisher E, Haque R, Yamamoto T, et al. Modelling malaria treatment practices in Bangladesh using spatial statistics. Malar J. 2012;11:63.View ArticlePubMedPubMed CentralGoogle Scholar
- Matthews SA, T-c YANG. Mapping the results of local statistics: Using geographically weighted regression. Malar J. 2012;26:151–66.Google Scholar
- Lin CH, Wen TH. Using geographically weighted regression (GWR) to explore spatial varying relationships of immature mosquitoes and human densities with the incidence of dengue. Int J Environ Res Public Health. 2011;8(7):2798–815. doi:10.3390/ijerph8072798.View ArticlePubMedPubMed CentralGoogle Scholar
- ESRI. Interpreting OLS results. 2015. Available: http://resources.arcgis.com/en/help/main/10.1/005p/005p00000030000000.htm. Assessed 11 July 2016.Google Scholar
- Ge E, Zhang X, Wang X, et al. Spatial and temporal analysis of tuberculosis in Zhejiang Province, China, 2009–2012. Infect Dis Poverty. 2016;5:11. doi:10.1186/s40249-016-0104-2.View ArticlePubMedPubMed CentralGoogle Scholar
- KAUHL B, HEIL J, HOEBE CJ. The Spatial Distribution of Hepatitis C Virus Infections and Associated Determinants-An Application of a Geographically Weighted Poisson Regression for Evidence-Based Screening Interventions in Hotspots. PLoS One. 2015;10:e0135656.View ArticlePubMedPubMed CentralGoogle Scholar
- Weisent J, Rohrbach B, Dunn JR, et al. Socioeconomic determinants of geographic disparities in campylobacteriosis risk: a comparison of global and local modeling approaches. Int J Health Geogr. 2012;11:45. http://doi.org/10.1186/1476-072X-11-45.View ArticlePubMedPubMed CentralGoogle Scholar
- Leung CC, Yew WW, Chan TY, et al. Seasonal pattern of tuberculosis in Hong Kong. Int J Epidemiol. 2005;34(4):924–30.View ArticlePubMedGoogle Scholar
- Beiranvand R, Karimi A, Delpisheh A, et al. Correlation Assessment of Climate and Geographic Distribution of Tuberculosis Using Geographical Information System (GIS). Iran J Publ Health. 2016;45(1):86–93.Google Scholar
- Murray M, Oxlade O, Lin HH. Modeling social, environmental and biological determinants of tuberculosis. Int J Tuberc Lung Dis Suppl. 2011;2:S64–70. doi:10.5588/ijtld.10.0535.View ArticleGoogle Scholar
- Zhao Y, Ehiri J, Li D, et al. A survey of TB knowledge among medical students in Southwest China: is the information reaching the target? BMJ Open. 2013;3(9):e003454. doi:10.1136/bmjopen-2013-003454.View ArticlePubMedPubMed CentralGoogle Scholar
- Vargas MH, Furuya ME, Pérez-Guzmán C. Effect of altitude on the frequency of pulmonary tuberculosis. Int J Tuberc Lung Dis. 2004;8:1321–4.PubMedGoogle Scholar
- Zhang X, Jia H, Liu F, Pan L, Xing A, Gu S, et al. Prevalence and risk factors for latent tuberculosis infection among health care workers in china: a cross-sectional study. PLoS One. 2013;8:e66412.View ArticlePubMedPubMed CentralGoogle Scholar