Choi Kyung-Mee, Serre Marc L, Christakos George
Center for the Advanced Study of the Environment, School of Public Health, University of North Carolina at Chapel Hill, North Carolina 27599-7431, USA.
J Expo Anal Environ Epidemiol. 2003 Mar;13(2):120-33. doi: 10.1038/sj.jea.7500263.
A meaningful characterization of epidemiologic fields (mortality, incidence rate, etc.) often involves the assessment of their spatiotemporal variation at multiple scales. An adequate analysis should depend on the scale at which the epidemiologic field is considered rather than being limited by the scale at which the data are available. In many studies, for example, data are available at a larger scale (say, counties), whereas the epidemiologist is interested in a smaller-scale analysis (say, residential neighborhoods). We propose a mathematically rigorous and epidemiologically meaningful multiscale approach that uses the well-known BME theory to study important scale effects and generate informative scale-dependent maps. The approach is applied to a real-world case study involving daily mortality counts in the state of California. The approach accounts for scale effects and produces mortality predictions at the zip-code scale by downscaling data from the county scale. The multiscale approach is tested by means of a verification data set with detailed mortality information at the zip-code level for 1 day. A measure of mapping accuracy is used to demonstrate that the multiscale approach offers more accurate mortality predictions at the local scale than existing approaches, which do not account for scale effects.
对流行病学领域（死亡率、发病率等）进行有意义的特征描述通常涉及在多个尺度上评估其时空变化。充分的分析应取决于所考虑的流行病学领域的尺度，而不是受限于数据可用的尺度。例如，在许多研究中，数据是在较大尺度（如县）上可用的，而流行病学家感兴趣的是较小尺度的分析（如居民区）。我们提出一种数学上严谨且在流行病学上有意义的多尺度方法，该方法使用著名的BME理论来研究重要的尺度效应并生成与尺度相关的信息丰富的地图。该方法应用于一个涉及加利福尼亚州每日死亡人数的实际案例研究。该方法考虑了尺度效应，并通过对来自县尺度的数据进行降尺度处理，在邮政编码尺度上生成死亡率预测。多尺度方法通过一个具有邮政编码级详细死亡信息的验证数据集进行了为期1天的测试。使用一种制图精度度量来证明，与不考虑尺度效应的现有方法相比，多尺度方法在局部尺度上提供了更准确的死亡率预测。
