Reseach Of Disease Mapping Over Small-area Based On Spatial Model

With the global economic integration, the exacerbate changes of global climate and environment, and the improvement of ability to transform the nature, human health has become a hot issue of common concern in the world. Especially in the last three decades, as the rapid growth of China's population and economy, and the deteriorating of environmental and ecological quality, people's attention on public health and various diseases has also being increasing continuously. Disease mapping is an important research area of spatial epidemiology, to provide a rapid visual summary of complex geographic information and may identify subtle patterns in the data that are missed in tabular presentations. In recent years, small-area disease mapping has become a hot issue which is aiming to identify the high-risk areas and the outbreak source of a disease in a small scale range by using the methods of spatial statistics and geo-computation. On the basis of discussing the key problems in spatial analysis and small-area disease mapping, the author not only examined the basic framework of spatial model-based small-area disease mapping for different data types, but also tackled some problems in study of small-area disease mapping by applying appropriate model which was adopted to the examples of geographic health issue. The dissertation was composed by the following two parts:(1) Theorectical studyFirstly, the difficulties and critical issues of spatial analysis have been discussed in detail from the perspective of spatial autocorrelation, boundary problem, conceptualization of spatial relationships, and testing statistical significance and other spatial effects, and then the dissertation summarized that the spatial autocorrelation, random variation and visual bias were not considered in traditional statistical mapping, which may failed in describing the spatial variation of disease events with small probability, and applied three spatial models targeted at the problems in small-area disease mapping. Kernel density estimation (KDE) was used to smoothing map for point data by including the spatial dependency of disease point data. For area data, the was applied to tackle the problems due to random variation and spatial effects in relative risk mapping, and the uncertainty of data and spatial dependency were included in Hierarchical Bayesian model by applying the spatial neighborhood and probability distribution. Finally, for spatio-temporal data, the Bayesian space-time model was adopted to identify the hot/cold spot of relative risk by integrating the spatial trends, temporal trends and space-time interaction in modeling process. (2) Practical applicationThe dissertation used the KDE to explore the spatial distribution of hypertension admissions throughout Shenzhen in2011, the address of each patient was located by using the address matching service of Shenzhen geo-spatial information platform, and the impact of different search radius on the KDE's results. In order to evaluate the performance of KDE, the local Moran's I was applied to explore the spatial pattern of density value in each sub-district of Shenzhen. The results indicated that there was a statistical significant spatial pattern on the distribution of hypertension admissions in Shenzhen in2011and identified some sub-district with high risk, such as Guiyuan and Huaqiangbei. Based on the consideration of accuracy in address matching, the hypertension admissions were aggregated to the sub-district level spatial unit, and then the Hierarchical Bayesian model was used to describe the spatial variation of relative risk of hypertension admission in Shenzhen, and the impact of spatial weight matrix with different structure on the model's performance was also discussed. The result was useful for the public health administrators in prevention and management of hypertension patients.The Bayesian space-time model was applied to explore the spatio-temporal variation of liver caner in Shenzhen (2010-2012), and then a novel two-stage method was used to classify the57sub-districts into disease risk hotspots, coldspots or neither and studying the temporal dynamic changes of areas within each risk category, and the impact of different spatial neighborhood was also analyzed and space-time scan statistic was applied to identified the liver cancer cluster. The results indicated that there was a significant East-West division of liver cancer risk in Shenzhen from2010to2012which was helpful for the improvement of public health service and liver cancer etiology study.In conclusion, the main content and the problems were summarized and the further studies in this field were presented at the end of this dissertation.