空间混杂偏倚，是空间流行病学研究中普遍且不可回避的问题。混杂因素往往存在空间效应，会严重扭曲暴露与结局之间的关系。故而，探讨空间混杂偏倚的形成机制与控制方法，是十分必要的。. 为此，本课题拟基于反事实理论，从空间混杂偏倚实际特性入手，以区域化随机函数理论为基础，以广义可加模型为框架，结合协同克里金，采用惩罚偏最小二乘，改进向后拟合的参数估计算法，构造倾向指数的空间估计模型；在最优空间抽样准则下，融合广义随机方格分层法与传统的匹配、分层等技术，发展倾向指数的空间均衡技术，从而，建立适用于空间混杂偏倚控制的倾向指数理论与方法；开发相应的统计软件包，并建立基于个体生物－社区空间特征的心脏瓣膜病精准社区干预平台，通过社区干预试验，考查新方法控制空间混杂偏倚的实际效果，为心脏瓣膜病社区干预效果评估与措施改进提供强有力的统计方法学支撑。

Spatial confounding bias is a universal and inevitable problem in spatial epidemiology. The confounding factors can create spatial effects, and hence, bias the relationship between exposure and outcomes. Therefore, it is necessary to explore the law of development and methodology of controlling confounding.. Based on counter-fact theory, this study will develop novel propensity score methods for controlling spatial confounding bias. Aimed at spatial characteristics of confounding, a combination model of generalized additive model and cokriging will be referred to as spatial propensity score estimation model based on regionalized stochastic function theory. And penalized partial least squares and backfitting algorithm will be combined to estimate propensity scores within regionalized space. To balance the distribution of estimated propensity scores between various exposure groups, optimal spatial sampling will be considered as the sampling criteria to integrate generalized random-tessellation stratified sampling with classic matching and stratification methods. Novel statistical package will be developed for a community intervention trial of valvular heart disease. To examine the practical effect of novel propensity score method, we will establish data platform of individual biological and community-based spatial characteristics for precise intervention, and investigate the effect of valvular heart disease interventions, hence throw light on the effect of spatial confounding bias controlling in large sample, non-random spatial epidemiological researches.