Evaluation and Development of Cancer Cluster Diagnostics

描述(申请人提供)：随着癌症集群检测中空间和时空模型的丰富，迫切需要对这些模型进行比较评估，以检查它们在集群检测中的相对性能。在这个项目中，我们将在癌症集群建模方面实现三个具体目标。第一个目标将通过对在贝叶斯范式内癌症测绘中常用的各种空间模型进行比较评估来实现。将考虑计数数据的一系列空间模型，包括随机目标建模、数据依赖建模、随机效果建模、混合建模和地质统计建模。在这次对比评估中，我们将在实现复杂性和性能收益之间做出平衡判断。将通过Hossain和Lawson(2006)开发的一套集群检测诊断方法进行比较。还将尝试扩展这些集群检测诊断。第二个目标将通过将集群检测诊断扩展到时空模型来实现。首先，我们建议将Hossain和Lawson(2006)提出的空间模型的所有诊断方法推广到时空模型。为了展示这些时空集群检测诊断的应用，我们将考虑有限数量的时空模型进行比较。最后，第三个目标将通过开发灵活的软件来实现，该软件将包括各种癌症集群诊断，以便负责癌症集群建模任务的研究人员或公共卫生工作者可以检查首选模型在检测癌症集群方面的能力。该软件将在免费提供的R编程环境中开发。将开发与此包一起使用的库和图形用户界面。通过使用易于获得的软件，我们希望确保我们的方法调查的结果可以轻易地提供给其他人。

DESCRIPTION (provided by applicant): With the abundance of spatial and spatio-temporal models in cancer cluster detection, there is an urgent need to make a comparative evaluation among these models to check their relative performances in cluster detection. In this project, we will achieve three specific goals in cancer cluster modeling. The first goal will be achieved by making a comparative evaluation of a variety of spatial models that have been commonly used in cancer mapping within a Bayesian paradigm. A range of spatial models for count data will be considered ranging from random object modeling, data-dependent modeling, random effect modeling, mixture modeling and geostatistical modeling. In this comparative evaluation, we will make a balance judgment between implementation complexity and performance gain. The comparison will be made by a set of cluster detection diagnostics developed in Hossain and Lawson (2006). An extension of these cluster detection diagnostics will also be attempted. The second goal will be achieved by extending cluster detection diagnostics to space-time models. Initially, we propose to extend all the diagnostics proposed in Hossain and Lawson (2006) for spatial models to space-time models. To show application of these space-time cluster detection diagnostics, we will consider a limited number of space-time models in comparison. Finally, the third goal will be achieved by developing flexible software, which will include a variety of cancer cluster diagnostics so that researchers or public health workers assigned with the task of cancer cluster modeling can check the ability of a preferred model in detecting cancer cluster. The software will be developed within the freely available R programming environment. Libraries and GUIs will be developed for use with this package. By using readily accessible software we hope to ensure that the results of our methodological investigations can be made readily available to others.