Flexible modeling of exposure-response relationship between long-term average levels of particulate air pollution and mortality in the American Cancer Society Study
Flexible modeling of exposure-response relationship between long-term average levels of particulate air pollution and mortality in the American Cancer Society Study
复制标题
DOI：
10.1080/15287390306426
复制
发表时间：
2003-08-01
影响因子：
2.6
通讯作者：
Krewski, D
中科院分区：
文献类型：
--
作者：
Abrahamowicz, M;Schopflocher, T;Krewski, D
从病因学和监管的角度来看，准确估计环境颗粒物空气污染与死亡率之间的因果关系是很重要的。然而，很少有人知道的实际形状，这些应力响应曲线。本研究的目的是估计死亡率和长期平均城市特定水平的硫酸盐和细颗粒物（PM2.5）之间的因果关系。我们重新分析了来自美国癌症协会（ACS）癌症预防研究II的数据，这是一项1982年至1989年在美国进行的大型前瞻性研究。在进入队列之前，对颗粒空气污染的暴露进行评估。1980年，151个城市的平均硫酸盐浓度可用，1979年至1983年期间，50个城市的PM2.5水平中位数可用。两种抽样策略，以减少计算负担。改良的病例队列方法将1200名个体的随机子队列与另外1300例病例（即，死亡）。第二种策略是将10个不相交的随机子集的单独分析结果汇总起来，每个子集约有2200名参与者。为了评估颗粒物水平对全因死亡率的独立影响，我们依赖于灵活的非参数生存分析方法。为了消除传统模型潜在的限制性假设，我们采用了灵活的回归样条概括的考克斯比例风险（PH）模型。回归样条方法使我们能够同时模拟随时间变化的颗粒物对危害的影响和可能的非线性响应关系。使用似然比检验检验PH和线性假设。在所有分析中，我们按年龄和5岁年龄组分层，并调整受试者的年龄、终生吸烟暴露、肥胖和教育。对于细颗粒物（PM2.5）和硫酸盐，有一个统计学显着（0.05水平）偏离传统的线性假设。细颗粒物对死亡率的调整效应表明，在较低的范围内（高达约16 μ g/m3）比在较高的范围内更强的关系。在较低的范围内增加硫酸盐水平（高达约12 μ g/m3）对死亡率几乎没有影响，这表明可能存在“无效阈值”。“对于体重指数（BMI），风险在中等范围内最低，对于非常肥胖和非常瘦的人来说都有所增加。得出的结论是，灵活的建模产生了关于长期空气污染对死亡率影响的新见解。
Accurate estimation of the exposure-response relationship between environmental particulate air pollution and mortality is important from both an etiologic and regulatory perspective. However, little is known about the actual shapes of these exposure-response curves. The objective of this study was to estimate the exposure-response relationships between mortality and long-term average city-specific levels of sulfates and fine particulate matter (PM2.5). We reanalyzed the data derived from the American Cancer Society (ACS) Cancer Prevention Study II, a large prospective study conducted in the United States between 1982 and 1989. Exposure to particulate air pollution was assessed prior to entry into the cohort. Mean sulfate concentrations for 1980 were available in 151 cities, and median PM2.5 levels between 1979 and 1983 were available in 50 cities. Two sampling strategies were employed to reduce the computational burden. The modified case-cohort approach combined a random subcohort of 1200 individuals with an additional 1300 cases (i.e., deaths). The second strategy involved pooling the results of separate analyses of 10 disjoint random subsets, each with about 2200 participants. To assess the independent effect of the particulate levels on all-causes mortality, we relied on flexible, nonparametric survival analytical methods. To eliminate potentially restrictive assumptions underlying the conventional models, we employed a flexible regression spline generalization of the Cox proportional-hazards (PH) model. The regression spline method allowed us to model simultaneously the time-dependent changes in the effect of particulate matter on the hazard and a possibly nonlinear exposure-response relationship. The PH and linearity hypotheses were tested using likelihood ratio tests. In all analyses, we stratified by age and 5-yr age groups and adjusted for the subject's age, lifetime smoking exposure, obesity, and education. For both fine particles (PM2.5) and sulfates, there was a statistically significant (at .05 level) departure from the conventional linearity assumption. The adjusted effect of fine particles on mortality indicated a stronger relationship in the lower (up to about 16 mug/m(3)) than in the higher range of their values. increasing levels of sulfates in the lower range (up to about 12 mug/m(3)) had little impact on mortality, suggesting a possible "no-effect threshold." For body mass index (BMI), the risks were lowest in the middle range and increased for both very obese and very lean individuals. It was concluded that flexible modeling yields new insights about the effect of long-term air pollution on mortality.