Exposure misclassification and threshold concentrations in time series analyses of air pollution health effects
Exposure misclassification and threshold concentrations in time series analyses of air pollution health effects
复制标题
DOI：
10.1111/1539-6924.00282
复制
发表时间：
2002-12-01
期刊：
影响因子：
3.8
通讯作者：
Petkau, AJ
中科院分区：
文献类型：
--
作者：
Brauer, M;Brumm, J;Petkau, AJ
空气污染和死亡率的时间序列研究通常报告线性、无阈值关系。由于监管标准和经济估值通常假设一些阈值水平，我们评估了暴露错误分类对潜在的个人层面阈值的持久性的影响的基本问题时，个人数据被汇总到人口水平的评估的风险-响应关系。作为一个例子，我们测量了个人暴露于两个颗粒指标，PM2.5和硫酸盐（SO 42-），为肺部疾病患者的样本，并将这些与从环境测量估计的暴露进行比较。以前的工作表明，PM2.5的环境：个人相关性远低于SO 42-，这表明环境PM2.5测量错误分类暴露于PM2.5。然后，我们开发了一种方法，通过该方法，这些患者的测量：估计暴露关系用于模拟更大人群的个人暴露，然后在不同的阈值假设下估计个人水平的死亡风险。将这些个体风险合并，以获得群体死亡风险，从而显示阈值在风险和估计暴露之间的关系中的显著性（和值）。我们的研究结果表明，对于分类不好的暴露（本例中为PM2.5），人口水平阈值在环境浓度低于规定的常见个人阈值时是明显的，而对于分类良好的暴露（例如，SO 42-），表观阈值与这些潜在的个人阈值相似。这些结果表明，替代指标，不高度相关的个人暴露掩盖了存在的阈值在流行病学研究的较大人群，而暴露指标，是高度相关的个人暴露可以准确地反映潜在的个人阈值。
Linear, no-threshold relationships are typically reported for time series studies of air pollution and mortality. Since regulatory standards and economic valuations typically assume some threshold level, we evaluated the fundamental question of the impact of exposure misclassification on the persistence of underlying personal-level thresholds when personal data are aggregated to the population level in the assessment of exposure-response relationships. As an example, we measured personal exposures to two particle metrics, PM2.5 and sulfate (SO42-), for a sample of lung disease patients and compared these with exposures estimated from ambient measurements. Previous work has shown that ambient:personal correlations for PM2.5 are much lower than for SO42-, suggesting that ambient PM2.5 measurements misclassify exposures to PM2.5. We then developed a method by which the measured:estimated exposure relationships for these patients were used to simulate personal exposures for a larger population and then to estimate individual level mortality risks under different threshold assumptions. These individual risks were combined to obtain the population risk of death, thereby exhibiting the prominence (and the value) of the threshold in the relationship between risk and estimated exposure. Our results indicated that for poorly classified exposures (PM2.5 in this example) population-level thresholds were apparent at lower ambient concentrations than specified common personal thresholds, while for well-classified exposures (e.g., SO42-), the apparent thresholds were similar to these underlying personal thresholds. These results demonstrate that surrogate metrics that ire not highly correlated with personal exposures obscure the presence of thresholds in epidemiological studies of larger populations, while exposure indicators that are highly correlated with personal exposures can accurately reflect underlying personal thresholds.