Dreassi Emanuela, Lagazio Corrado, Maule Milena M, Magnani Corrado, Biggeri Annibale
Dipartimento di Statistica G. Parenti, Università di Firenze, Firenze, Italy.
Geospat Health. 2008 May;2(2):263-71. doi: 10.4081/gh.2008.249.
The relation between disease risk and a point source of pollution is usually investigated using distance from the source as a proxy of exposure. The analysis may be based on case-control data or on aggregated data. The definition of the function relating risk of disease and distance is critical, both in a classical and in a Bayesian framework, because the likelihood is usually very flat, even with large amounts of data. In this paper we investigate how the specification of the function relating risk of disease with distance from the source and of the prior distributions on the parameters of the function affects the results when case-control data and Bayesian methods are used. We consider different popular parametric models for the risk distance function in a Bayesian approach, comparing estimates with those derived by maximum likelihood. As an example we have analyzed the relationship between a putative source of environmental pollution (an asbestos cement plant) and the occurrence of pleural malignant mesothelioma in the area of Casale Monferrato (Italy) in 1987-1993. Risk of pleural malignant mesothelioma turns out to be strongly related to distance from the asbestos cement plant. However, as the models appeared to be sensitive to modeling choices, we suggest that any analysis of disease risk around a putative source should be integrated with a careful sensitivity analysis and possibly with prior knowledge. The choice of prior distribution is extremely important and should be based on epidemiological considerations.
疾病风险与点源污染之间的关系通常通过使用到污染源的距离作为暴露的替代指标来进行研究。该分析可以基于病例对照数据或汇总数据。在经典框架和贝叶斯框架中，定义疾病风险与距离之间的函数都至关重要，因为即使有大量数据，似然函数通常也很平坦。在本文中，我们研究了在使用病例对照数据和贝叶斯方法时，疾病风险与到污染源距离之间函数的设定以及该函数参数的先验分布如何影响结果。我们在贝叶斯方法中考虑了不同的流行风险距离函数参数模型，并将估计值与通过最大似然法得出的估计值进行比较。作为一个例子，我们分析了一个假定的环境污染源（一家石棉水泥厂）与1987 - 1993年意大利卡萨莱蒙费拉托地区胸膜恶性间皮瘤发生之间的关系。结果表明，胸膜恶性间皮瘤的风险与到石棉水泥厂的距离密切相关。然而，由于模型似乎对建模选择敏感，我们建议对假定污染源周围的疾病风险进行任何分析时，都应结合仔细的敏感性分析以及可能的先验知识。先验分布的选择极其重要，应该基于流行病学考虑。
