基于GWR模型的结核病空间流行病学研究

目的 探讨山东省结核病及其影响因素间的局域关系，为制定适宜的结核病防控策略提供依据。 方法 收集山东省2005-2008年各县区结核病登记报告资料和相关影响因素资料；采用全局空间自相关系数Moran’ I检验区域结核病发病的空间自相关性；构建地理加权回归（GWR）模型定量分析结核病登记率与各影响因素间的局域关系，并应用ArcGIS9.0绘制地图。山东省2005-2008年各县（区）活动性结核病登记率分别为12.79/10万～107.35/10万、16.01/10万～86.52/10万、17.36/10万～92.10/10万和17.86/10万～114.86/10万。 结果 空间自相关分析表明2005-2008年各县区结核病登记率在空间分布上具有明显的空间正相关关系（Moran’s I分别为0.3517、0.3505、0.3337和0.3116，P值均<0.05）。GWR模型分析显示其拟合效果优于全局OLS模型［赤池信息准则（akaike information criterion，AIC）下降均大于3，R²均增大］，如2008年GWR模型与OLS模型的AIC和R²分别为1168.8380和1173.5410，0.3537和0.1350）；各模型的R²均具有明显的空间变异性，如2008年R²为0.1162~0.1798。 结论 GWR模型能够揭示影响因素对结核病登记率影响的空间异质性；应根据各因素的空间分布特征及其与结核病登记率间的局域关系制定区域化的结核病防控规划和策略。

Spatial epidemiology study on tuberculosis based on geographical weighted regression model

Objective To explore the local relationship between tuberculosis and influencing factors in Shandong Province, and to provide evidence for appropriate regional TB prevention and control strategy development. Methods The data of TB notification and related influencing factors during 2005 to 2008 in Shandong province were collected. The spatial autocorrelation of tuberculosis was analyzed by Moran’I. Geographical weighted regression （GWR） model was constructed to analyze the local relationship between tuberculosis notification rate and various influencing factors, and mapped by ArcGIS9.0. The notification rates of active tuberculosis of each county during 2005 to 2008 in Shandong were 12.79/100 000～107.35/1 000 000, 16.01/1 000 000～86.52/1 000 000, 17.36/1 000 000～92.10/1 000 000 and 17.86/1 000 000～114.86/1 000 000 respectively. Results The spatial autocorrelation analysis showed that the spatial distribution of tuberculosis notification rate had significant spatial positive correlation （Moran’s I were 0.3517, 0.3505, 0.3337 and 0.3116 respectively, and P value were all less than 0.05） in Shandong between 2005 and 2008. GWR model displayed better fitting effect than global OLS model （the declines of akaike information criterion（AIC） were all higher than 3, and R2 all increased, eg. the AIC and R2 were 1168.8380 and 0.3537 for GWR model, and 1173.5410 and 0.1350 for OLS model）. The local R2 appeared significant spatial variability （eg. the local R2 was 0.1162~0.1798 in 2008）. Conclusion GWR model can reveal the spatial heterogeneity of the effect of influencing factors on tuberculosis notification rate, and regional tuberculosis prevention and control programme and strategy should be developed based on the spatial characteristic of the impact factors and the local relationship with tuberculosis notification rate.