Estimating and communicating spatial certainty when childhood cancers co-cluster

点击翻译按钮获取中文摘要

PROJECT SUMMARY/ABSTRACT There are several recent advances, in disease cluster evaluation that if used collectively could avoid a myriad of statistical faults, including the Texas Sharp Shooter Fallacy. These advances include models that estimate the exceedance probability (EP); defined as the Bayesian probability that the relative risk at a specific location is greater than 1. When applied across continuous space, the EP provides a sensitive identification of disease clusters with varying cluster boundaries and sizes and with explicit, spatially-varying certainty. Furthermore, extending the model to multiple disorders can objectively combine disorders with common spatial patterns thereby enhancing the effective sample size. The problem is that childhood cancer is so rare that the prior distributions for the spatial parameters may unduly influence the results. What we need most is an objective way to combine CC when specific CC have common spatial risk patterns. The long-term goal is to prevent diseases caused by environmental exposures. The overall objective of this application, which is the next step in our long-term goal, is to find the most objective way to pool CC subgroups when they share common spatial risk patterns. Our central hypothesis is that CC have common spatial patterns near some environmental hazards. The hypothesis is formulated based on our preliminary findings. The rationale that underlies the proposed research is that recently developed Bayesian multivariate spatial modeling is the link that we need to mitigate spatial uncertainty and restore public faith in cluster investigations. The central hypothesis will be tested and the objective of this application attained by pursuing the following specific aims: 1. Evaluate case-excess for single CC using univariate geostatistical modeling of EP. We postulate, based on our current studies, that geostatistical modeling of the EP will provide an improved sensitivity for cluster detection by allowing flexible cluster shapes, sizes and statistical certainty. 2, Evaluate case-excess for multiple CC using multivariate geostatistical modeling of EP. We postulate, based on our preliminary studies, that multiple CC share common geographic patterns near some toxic sites and multivariate modeling of the CC will enhance the sensitivity of cluster detection. With respect to expected outcomes, the work proposed in aim 1 will identify significant risk patterns of individual CC near some Texas Superfund Sites. Aim 2 will identify CC with common geographic risk patterns at these locations. This contribution is significant because we live in an era in which the public's vigilance is sought for all environmental risks and the public's input must be encouraged and validated and, most importantly, addressed, objectively. The contribution is innovative because the proposed research combines recent advances to resolve issues in the modeling and reporting of spatial uncertainty in disease cluster investigation.