基于空间模型的小地域疾病制图研究

发布时间：2018-01-18 16:01

本文关键词：基于空间模型的小地域疾病制图研究　出处：《武汉大学》2014年博士论文　论文类型：学位论文

【摘要】：随着全球经济一体化、全球气候与环境的变化加剧和人类改造自然能力的不断提升,人类健康已成为全世界普遍关注的热点问题,特别是近三十年,我国人口的迅速增长、经济的迅猛发展和环境生态质量的不断恶化等,人们对公共卫生和各类疾病的关注度也在持续增强。疾病制图是空间流行病学的重要研究领域,可对复杂疾病信息进行快速的地理可视化表达,可以识别在表格中难以确定的分布模式。小地域疾病制图是近年的研究热点之一,通过运用空间统计和地理计算方法,识别某种疾病在小尺度范围内的高风险区和爆发源。本文在空间分析和小地域疾病制图研究等关键问题的基础上,结合疾病数据的不同类型,研究了基于空间模型的小地域疾病制图的基本框架,针对小地域疾病制图所遇到的问题采用相应的模型方法,并辅以实例来说明。本文主要以理论和实践两个层面展开研究。 (一)理论研究本文从空间自相关、边界问题、空间关系概念化和统计显著性检验等地理数据空间效应的角度上,详细阐述了空间分析面临的难点和关键问题,针对不同疾病数据,总结了现有疾病聚类分析的方法,探讨了传统统计制图的缺陷,指出由于随机变化,小地域疾病制图往往会导致地图的额外变异,并且传统疾病制图没有考虑空间自相关、随机变化和视觉偏差等因素,难以准确描述空间小概率疾病事件的空间变化,需要引入空间模型移除疾病地图的随机部分,并运用三种空间模型解决小地域疾病制图存在的问题。核密度估计用于疾病点数据,充分考虑了点要素的空间依赖性特征,生成疾病点数据的平滑地图。层次贝叶斯模型用于区域数据,该模型考虑相对风险等制图指标的空间效应,通过引入空间统计单元的空间关系和概率分布,将数据的不确定性和空间自相关关系包含在模型之中。贝叶斯时空模型用于时空数据,将疾病相对风险的空间趋势、时间趋势和时空交互进行统一建模,并可探测疾病风险的热点和冷点及其时空变化趋势。 (二)实际应用本文使用核密度估计对深圳市2011年高血压患者的空间分布进行了探测,采用数字深圳空间基础信息平台的地址匹配服务完成疾病病例的空间化,克服了传统“人工打点”的缺陷。分析和讨论了不同搜索半径对核密度计算过程的影响,并采用局部Moran's I计算每个街道内核密度值均值的局部自相关指数,尝试对核密度估计的性能进行评价。实验结果证明深圳市2011年高血压患者存在显著的空间分布模式,桂园、华强北等街道为高血压的高发区域。针对病例地址信息缺失和定位精度等问题,本文采用层次贝叶斯模型分析深圳市2011年高血压相对风险的空间变化,并讨论了不同结构的空间权重矩阵对模型性能的影响,研究成果有助于深圳市公共卫生部门对高血压患者的防控与管理。基于深圳市2010年-2012年肝癌发病数据,针对空间统计处理时空问题的困难,本文运用贝叶斯时空模型研究肝癌相对风险的时空变化,采用两步分类过程识别对风险的热点和冷点及其时空变化趋势,讨论了不同空间邻域类型对模型性能的影响,使用时空扫描统计探测肝癌患者的时空聚类,研究结果表明三年间深圳市肝癌风险存在明显的东-西划分的分布格局和显著的时空变化趋势,该信息有助于深圳市公共卫生服务和肝癌病因学研究,并可用于其他领域小概率事件的时空建模。论文的最后,本文根据研究过程中所遇到的问题,对整个研究工作进行总结并提出今后研究的重点和方向。
[Abstract]:With the global economic integration, global climate and environment change and continuously improve the ability of the transformation of human nature, human health has become a hot issue all over the world, especially in the past thirty years, the rapid growth of China's population, the rapid development of economy and ecological environment quality worsening, people continue to enhance the public health and various diseases. Attention is also in disease mapping is an important research field of spatial epidemiology, expression of geographic visualization quickly on the information of complex diseases, can be distributed in the form of pattern recognition is uncertain. Small regional disease mapping is one of the research hotspot in recent years, through the use of spatial statistics and geographical calculation method. Identification of a disease in high-risk areas in a small scale and the source of the outbreak. The key issues in spatial analysis and mapping of the small regional disease group Based on the combination of different types of disease data, research the basic frame of the space model of small regional disease mapping based on the corresponding method used for small regional disease mapping problems, and illustrate by examples. This paper mainly focuses on two aspects of theory and practice.
(1) theoretical study
This paper from the spatial autocorrelation, boundary problem, spatial relationship of conceptual and statistical significance test geographical data such as the spatial effect angle, elaborated the key problems and difficulties facing the spatial analysis, according to different disease data, summarizes the method of clustering analysis of existing diseases, discusses the defects of traditional statistical mapping, pointed out that because of random variation, small regional disease mapping often leads to additional variation map, and the traditional disease mapping without considering spatial autocorrelation, random changes and visual bias and other factors, it is difficult to describe the spatial variation of space small probability of disease events, needs to introduce random space model remove disease map, and three space model by solving the existence of small regional disease mapping problem. Kernel density estimation for disease data, considering the dependence of the feature elements of the space, generating disease Smooth map data. Bayesian hierarchical model for the regional data, the model considering the spatial effect of relative risk mapping index, the spatial relation and probability distribution into spatial statistical units, data uncertainty and spatial autocorrelation relations are included in the model. The model for Bayesian spatio-temporal temporal data, spatial trend of relative disease the risk, time trend and temporal interaction of unified modeling and detection of disease risk, hot and cold spots and its change tendency.
(two) practical application
The use of hypertension patients in Shenzhen city in 2011 the spatial distribution of the detection space using kernel density estimation, digital Shenzhen space information platform, service address complete disease cases, overcome the defect of the traditional artificial management ". And discussed different search radius effects on the calculation process of kernel density, local self the correlation index and the local Moran's I kernel density average value calculated for each street, to try to evaluate the performance of kernel density estimation. Experimental results show that there is an obvious spatial pattern of Shenzhen city in 2011, hypertension garden, street Huaqiang North high risk area for hypertension. Aiming at the problem of missing cases address information and the positioning precision in this paper, using a hierarchical Bayesian model analysis of spatial variation in Shenzhen city in 2011, the relative risk of hypertension, and discuss the different structure of space The impact of the weight matrix on the performance of the model is helpful to the prevention and control of the hypertension patients in the public health department of Shenzhen.
2010 -2012 in Shenzhen city based on the data of liver cancer, according to the spatial statistics processing of spatio-temporal problem difficult, using the spatial and temporal variation of liver cancer Bias space-time model of relative risk, the focus of two step classification process in risk identification and cold point and its variation with time and space, the effects of different types of spatial neighborhood model performance is discussed, the use of space and time scan statistics detection temporal clustering of HCC patients, the results showed that there were significant East West Division in Shenzhen during three years the risk of liver cancer the distribution pattern and the temporal and spatial variation of significant trends, the information service and contribute to the study of etiology of public health in Shenzhen, and can be used in other areas of spatio-temporal modeling of small probability events.
At the end of the paper, this paper summarizes the whole research work and puts forward the focus and direction of the future research according to the problems encountered in the study.

【学位授予单位】：武汉大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：R188;P208

【参考文献】