基于工具变量法的PM_(2.5)浓度对每日死亡影响的效应估计  

Using Instrumental Approach to Estimate the Effect of PM_(2.5)Concentrations on Daily Deaths

作　　者：祝贵明 颜豪森 王彤[2] 吕鹏[1] 孙红卫[1] Zhu Guiming;Yan Haosen;Wang Tong(College of Public Health and Management,Binzhou Medical University,Yantai 264000)

机构地区：[1]滨州医学院公共卫生学院,264000 [2]山西医科大学公共卫生学院

出　　处：《中国卫生统计》2024年第4期526-531,538,共7页Chinese Journal of Health Statistics

基　　金：国家自然科学基金(81502891)。

摘　　要：目的大量环境流行病学研究表明PM_(2.5)浓度短期变化与人群每日死亡相关,但是大多数关于PM_(2.5)对每日死亡的急性效应的研究是通过回归来控制混杂因素,而通常可用的人群资料里仅包含少量测量的混杂因素,这就面临着大量未观测的混杂因素未纳入模型的问题从而导致估计有偏,而工具变量法可以较好地解决未观测混杂带来的效应估计问题,本文通过工具变量法估计PM_(2.5)对每日死亡的急性效应。方法收集中国某市2016—2019年PM_(2.5)日均浓度、气象数据及每日非意外死亡人数,采用边界层高度和风速作为工具变量,分析该地PM_(2.5)浓度对每日非意外死亡的影响;采用阴性暴露对照法检验工具变量假设;采用时间序列的bootstrap方法估计置信区间。并与广义相加模型的效应估计进行比较。结果工具变量法得出PM_(2.5)浓度与人群每日非意外死亡存在相关,PM_(2.5)浓度每升高10μg/m^(3),人群每日非意外死亡增加0.94%(95%CI:0.39%~1.55%);阴性暴露对照结果显示阴性暴露与人群日非意外死亡不相关(P=0.19),说明前述工具变量模型不受未测量且未控制混杂的影响。传统广义相加模型估计PM_(2.5)浓度每升高10μg/m^(3),人群每日非意外死亡增加0.24%(95%CI:0.01%~0.47%)。结论经工具变量法估计,该地PM_(2.5)浓度与居民每日非意外死亡存在相关,边界层高度和风速可以作为工具变量估计PM_(2.5)浓度对人群每日非意外死亡的急性效应。Objective Many environmental epidemiological studies have shown the associations between short-term exposure of air pollution and daily deaths.However,the generally available population data only contain a small number of measured confounding factors,which is faced with the problem that a large number of unobserved confounding factors are not included in the model,resulting in biased estimates.The instrumental variable method can solve the problem of estimating the effects caused by unobserved confounders.In this paper,We used the instrumental variable method to estimate the effects of PM_(2.5)on daily mortality.Methods We collected daily PM_(2.5)concentrations,meteorological data,and nonaccidental daily deaths in a Chinese city from 2016 to 2019.We used boundary layer height and wind speed as instrumental variables to estimate the effects of PM_(2.5)on nonaccidental daily mortality.Negative exposure control was used to test the hypothesis of instrumental variables.Meanwhile time series bootstrap method was used to estimate confidence interval.We compared the results of the generalized additive model and instrumental variable method.Results The instrumental variable method showed that PM_(2.5)was significantly related to daily deaths.For every 10μg/m^(3)increase of PM_(2.5)concentrations,the nonaccidental daily deaths increased by 0.94%(95%CI:0.39%~1.55%).Negative exposure control results showed no correlation between negative exposure and nonaccidental deaths(P=0.19),indicated that the aforementioned instrumental variable model was not affected by unmeasured and uncontrolled confounders.The traditional generalized additive model estimated that for every 10μg/m^(3)increase in PM_(2.5)concentrations,the nonaccidental deaths would increase by 0.24%(95%CI:0.01%~0.47%).Conclusion The instrumental variable method estimated that PM_(2.5)concentrations were significantly correlated with the nonaccidental daily deaths.Boundary layer height and wind speed can be used as instrumental variables to estimate the effects of

关 键 词：PM_(2.5) 工具变量法 风速 边界层高度 非意外死亡 阴性暴露对照 BOOTSTRAP 

分 类 号：R195.1[医药卫生—卫生统计学]