Jerrett Michael, Burnett Richard T, Willis Alette, Krewski Daniel, Goldberg Mark S, DeLuca Patrick, Finkelstein Norm
School of Geography and Geology, McMaster University, Hamilton, Ontario, Canada.
J Toxicol Environ Health A. 2003;66(16-19):1735-77. doi: 10.1080/15287390306438.
Lack of control for confounding by ecological covariates that may relate to sulfate air pollution and mortality was a key criticism of the two studies that were the focus of the Particle Reanalysis Project. To assess the validity of this criticism, we address the question: "Does sulfate air pollution exert health effects when the impact of other individual and ecologic variables thought to influence health is taken into account?" A related question arises from the possibility of autocorrelation in the mortality risks and ecologic covariates. Failure to control for autocorrelation can lead to false positive significance tests and may indicate bias resulting from a missing variable or group of variables. We control for more than 25 individual risk factors and for 20 ecologic variables representing environmental, socioeconomic, demographic, health- care, and lifestyle determinants of health in a two-stage multilevel analysis. Four modeling strategies are used to control for spatial autocorrelation. Of the 20 ecologic variables tested, only sulfate and sulfur dioxide are significant in models that incorporate spatial autocorrelation. Accounting for autocorrelation also reduces the size and certainty of the sulfate effect on mortality when compared to results generated from Cox models where independent observations are assumed. Confidence limits for the sulfate relative risk include unity in models that simultaneously control for sulfur dioxide and autocorrelation.
缺乏对可能与硫酸盐空气污染和死亡率相关的生态协变量混杂因素的控制，是粒子再分析项目所关注的两项研究的关键批评点。为了评估这一批评的有效性，我们提出问题：“在考虑到其他被认为会影响健康的个体和生态变量的影响后，硫酸盐空气污染是否会产生健康影响？”一个相关问题源于死亡率风险和生态协变量中自相关的可能性。未能控制自相关可能导致假阳性显著性检验，并可能表明由一个或一组缺失变量导致的偏差。我们在两阶段多级分析中控制了超过25个个体风险因素以及代表健康的环境、社会经济、人口统计学、医疗保健和生活方式决定因素的20个生态变量。使用四种建模策略来控制空间自相关。在测试的20个生态变量中，只有硫酸盐和二氧化硫在纳入空间自相关的模型中具有显著性。与假设独立观察的Cox模型产生的结果相比，考虑自相关也会降低硫酸盐对死亡率影响的大小和确定性。在同时控制二氧化硫和自相关的模型中，硫酸盐相对风险的置信区间包括1。
