Louie Mary M, Kolaczyk Eric D
National Center for Health Statistics, 3311 Toledo Road, Room 3215, Hyattsville, MD 20782, USA.
Stat Med. 2006 Mar 15;25(5):787-810. doi: 10.1002/sim.2404.
We present a modelling framework for detection of potentially anomalous structure in aggregate spatial disease incidence data in a manner sensitive to localization at multiple scales and/or positions. The key technical contribution is the re-casting of the components of a multiscale disease mapping methodology, recently introduced by the authors in an earlier paper, into a form appropriate for hypothesis testing. In particular, we describe how hypotheses of spatially clustered variations in disease incidence may be linked in one-to-one correspondence with collections of hypotheses on the values of certain multiscale parameters associated with a user-defined hierarchy of nested partitions of an overall spatial region. A Bayesian hypothesis testing methodology is developed in the context of a standard Poisson measurement model, over the collection of possible multiscale hypotheses. We discuss the specification of hyper parameters and prior distributions on the space of models. The methodology is illustrated on both simulated and real data.
我们提出了一个建模框架，用于检测总体空间疾病发病率数据中潜在的异常结构，该框架对多尺度和/或多位置的定位敏感。关键的技术贡献在于，将作者在早期论文中最近引入的多尺度疾病映射方法的组件，重新构建为适合假设检验的形式。具体而言，我们描述了疾病发病率空间聚集变化的假设如何与关于某些多尺度参数值的假设集合一一对应，这些参数与整个空间区域的用户定义嵌套分区层次结构相关。在标准泊松测量模型的背景下，针对可能的多尺度假设集合，开发了一种贝叶斯假设检验方法。我们讨论了模型空间上超参数和先验分布的设定。该方法在模拟数据和真实数据上均得到了验证。
