Department of Biostatistics, School of Public Health, University of Michigan, Ann Arbor, MI 48109-2029, USA.
Environ Res. 2011 Nov;111(8):1137-47. doi: 10.1016/j.envres.2011.06.002. Epub 2011 Jul 20.
Asthma morbidity has been associated with ambient air pollutants in time-series and case-crossover studies. In such study designs, threshold effects of air pollutants on asthma outcomes have been relatively unexplored, which are of potential interest for exploring concentration-response relationships.
This study analyzes daily data on the asthma morbidity experienced by the pediatric Medicaid population (ages 2-18 years) of Detroit, Michigan and concentrations of pollutants fine particles (PM2.5), CO, NO2 and SO2 for the 2004-2006 period, using both time-series and case-crossover designs. We use a simple, testable and readily implementable profile likelihood-based approach to estimate threshold parameters in both designs.
Evidence of significant increases in daily acute asthma events was found for SO2 and PM2.5, and a significant threshold effect was estimated for PM2.5 at 13 and 11 μg m(-3) using generalized additive models and conditional logistic regression models, respectively. Stronger effect sizes above the threshold were typically noted compared to standard linear relationship, e.g., in the time series analysis, an interquartile range increase (9.2 μg m(-3)) in PM2.5 (5-day-moving average) had a risk ratio of 1.030 (95% CI: 1.001, 1.061) in the generalized additive models, and 1.066 (95% CI: 1.031, 1.102) in the threshold generalized additive models. The corresponding estimates for the case-crossover design were 1.039 (95% CI: 1.013, 1.066) in the conditional logistic regression, and 1.054 (95% CI: 1.023, 1.086) in the threshold conditional logistic regression.
This study indicates that the associations of SO2 and PM2.5 concentrations with asthma emergency department visits and hospitalizations, as well as the estimated PM2.5 threshold were fairly consistent across time-series and case-crossover analyses, and suggests that effect estimates based on linear models (without thresholds) may underestimate the true risk.
时间序列和病例交叉研究表明，哮喘发病率与环境空气污染物有关。在这些研究设计中，空气污染物对哮喘结果的阈值效应相对较少被探讨，而这对于探索浓度-反应关系具有潜在的意义。
本研究分析了 2004 年至 2006 年期间密歇根州底特律市儿科医疗补助人群（2-18 岁）的哮喘发病率和污染物细颗粒物（PM2.5）、一氧化碳（CO）、二氧化氮（NO2）和二氧化硫（SO2）浓度的日数据，使用时间序列和病例交叉设计。我们使用一种简单、可测试且易于实施的基于似然比的方法来估计两种设计中的阈值参数。
发现 SO2 和 PM2.5 的每日急性哮喘事件有显著增加的证据，并分别使用广义加性模型和条件逻辑回归模型估计 PM2.5 的阈值效应为 13 和 11μg/m3。与标准线性关系相比，通常在阈值之上观察到更强的效应大小，例如，在时间序列分析中，PM2.5（五日移动平均值）的四分位距增加（9.2μg/m3）在广义加性模型中的风险比为 1.030（95%置信区间：1.001，1.061），而在阈值广义加性模型中为 1.066（95%置信区间：1.031，1.102）。病例交叉设计的相应估计值在条件逻辑回归中为 1.039（95%置信区间：1.013，1.066），在阈值条件逻辑回归中为 1.054（95%置信区间：1.023，1.086）。
本研究表明，SO2 和 PM2.5 浓度与哮喘急诊就诊和住院的关联以及估计的 PM2.5 阈值在时间序列和病例交叉分析中相当一致，并表明基于线性模型（无阈值）的效应估计可能低估了真实风险。
