Study On Methodology And Application Of Spatial Heterogeneity Of Disease Boundary

Geographic boundary can be defined as edge of homogeneous areas or as zones of rapid change in a variable's spatial field. According to the definition for geographic boundary in biogeography, we define spatial heterogeneity of disease boundary as edge of homogeneous ecotope of disease or as zones of rapid change in a disease variable's spatial field. Spatial heterogeneity of disease boundary detection is a key issue in geographic epidemiology and spatial epidemiology. Firstly, the structures and boundaries of spatial heterogeneity of disease identified can display the spatial dynamic characteristics of disease. Secondly, boundary analysis can reveal the distribution of disease between different spatial areas and play an important role in making medical geographic zoning. Thirdly, such method can suggest whether health outcomes are related to environmental exposures. Moreover, boundary analysis can provide the government with policy and measure to prevent and control areal disease.Recently, the development of geographic information system and spatial statsitcs promotes geographic epidemiology to spatial epidemiology. Geographic boundary analysis will be consummated and developed in the future. However, geographic boundary detection methods are not widely used for its limitations and lack of sufficient researches on this field. At present, we do not have special statistical models for spatial heterogeneity of disease boundary. According to the principle of landscape epidemiology and spatial epidemiology and making use of the methods of spatial genetics and landscape genetics, we adopt co-course of studies and combine spatial statistics, graph theory, geology mathematics and boundary analysis together to set up three models for spatial heterogeneity of disease boundary by means of GIS: Contour area multifractal model (CAMM), 2-D graphic clustering model and Monmonier's algorithm. To explore the three models' applicability and epidemiological significance, the structures and boundaries of spatial heterogeneity of Hemorrhagic fever with renal syndrome (HFRS) are identified.For the sake of setting up spatial heterogeneity of disease boundary models, we should focus not only on the geographical epidemiology characteristics of disease, but also on the particularity of the spatial data. Spatial heterogeneity of disease is a function of size. Different size may result in different spatial characteristics of disease. In order to validate those models' applicability in spatial analysis of disease, the structures and boundaries of spatial heterogeneity of hemorrhagic fever with renal syndrome (HFRS) in Shandong province and Linyi city were identified. The models are built on the below aspects in the framework of ArcGIS9.0:1. Spatial heterogeneity of disease boundary model based on spatial interpolated dataThere are spatial interpolated modules in ArcGIS9.0 and geostatistics software for the agricultural and biological sciences, such as spline function models, tendency models, inverse distance-weighted methods, Kriging models. These models are mainly used to estimate and predict the variation of disease and display the gradient of their variation. But the geographic gradient made by the models are arbitrary to some extent and do not have statistical evidence. In this study, we combine multifractal with statistics in the framework of AreGIS9.0 to set up the CAMM in order to overcome the subjectivity and arbitrariness in identifying the interval cut-off vales of mapping interpolated disease.2. Spatial heterogeneity of disease boundary model based on spatial point data for revealing the relationship of spatial ecology and epidemiology between endemic areas or endemic point.In the geographic boundary analysis of spatial genetics, the spatially constrained clustering, such as spatially K-means clustering, based on genetic distance matrix, population spatial locations and geographic areal matrix are connected using a Delaunay triangulation. This kind of geographic boundary analysis models are short of biogeographic significance. In this study, combing graph theory with path analysis of panbiogeography, the 2-D graphic minimal spanning tree model we implemented not only can identify geographic boundary but also can infer the ecological and epidemiological clue of endemic areas or endemic points.3. Spatial heterogeneity of disease boundary model based on spatial point data for detecting the maximum difference between endemic areas or endemic points.Geographic boundaries are hierarchical. Namely, there are many subboundaries in endemic areas or endemic points. In this study, the improved Monmonier's algorithm model based on Monmonier's maximum-distance algorithm and distance matrix between the corresponding data, which are connected using a Delaunay triangulation, is built. This algorithm finds the edges associated to highest rate of change in given distance measure, namely the areas where differences between locations of disease are largest. To offer a more realistic representation of the barriers, we implemented in the software a significance test by meansof bootstrap matrices analysis.To validate the condition, application and geographic epidemiologysignificance of these three models, the structures and boundaries of spatial heterogeneity of Hemorrhagic fever with renal syndrome (HFRS) in Shandong province and Linyi city are identified. Conclusions:1. Using ArcGIS9.0, CAMM combined multifractal with statistics to mark gradient map of disease, avoiding the subjective and arbitrary thresholds marked on the maps. The interval for mapping disease has statistical evidence. The gradient of the model making not only can reflect the nature of spatial heterogeneity of disease, but also can display different geographic epidemiology. More importantly, this statistical gradient can provide policy and measure for the government to prevent and handle areal disease. However, the CAMM also has the similar disadvantage in identification of spatial heterogeneity of disease boundaries, implying the interpolation of the landscape leading to potential artificial continuities or discontinuities.2. 2-D graphic clustering model, which combines the minimal spanning tree of graph theory with two-dimensional constrained clustering theory, not only can identify the comparability between the cases but also can infer the spatio-temporal trace of disease. This model has favorable application in boundary detection. However, it is constrained by sample. The more homogeneity samples the better effective. According to the study, T 2-D graphic clustering model suits to analyze the global boundary of spatial characteristics of disease.3. The improved Monmonier's algorithm model based on Monmonier's maximum-distance algorithm and distance matrix between the corresponding data. This algorithm finds the edges associated to highest rate of change in given distance measure, namely the areas where differences between locations of disease are largest. The bootstrap test was used to assess the robustness of computed boundaries and can infer the spatial connection and arrangement of boundaries. Nevertheless, when identifying the spatial genetic boundaries using software of BARRIER version 2.2, the selection of number of boundaries is subjective and it may affect the results to some extent. At the same time, this model is also constrained by sample.4. The spatio-temporal dynamic characteristics of endemic areas' spatial structure from different aspects of HFRS in Shandong province and Linyi city have been analyzed by these three models: â‘ Contour area multifractal model (CAMM) display the gradient change of HFRS morbility in Shandong province; 2-D graphic clustering model not identified the comparability between the cases but inferred the spatio-temporal trace of disease;"Monmonier's Algorithm" reveal hierarchical and spatial adjunction during the SEO type focus in western part permeating to the HTN type focus in eastern part( 1987-1993) and the consistency after the confluence of the two type focus. â‘¡The structures and boundaries of spatial heterogeneity of disease identified by these three models are concord with the same period of HFRS in Linyi city. Contour area multifractal model (CAMM) and improved "Monmonier's Algorithm" identity the same barriers, although they using different principle. Conclusions yield by the three models are as the same as original research of our study group. Moreover, those three models can more deeply reveal the trace of the evolution of spatio-temporal dynamic in the characteristic of HFRS in Shandong province and Linyi city.5. While the strength of these methods is obvious, their disadvantages should not be ignored. In particular, we suggest combining the three models to do boundary analysis in both local and global areas. Those geographic boundary methods are clearly valuable addition to the spatial statistical toolbox.