Brunekreef B, Noy D, Clausing P
Am J Epidemiol. 1987 May;125(5):892-8. doi: 10.1093/oxfordjournals.aje.a114606.
In studies in environmental epidemiology, exposure to harmful agents is often highly variable in time and space. It is not usually possible to measure the relevant, personal exposure of study subjects to these agents directly. Instead, exposure measurements are performed at fixed sites and/or for limited periods of time in many cases. Such measurements are imperfect in the sense that they only approximate the "true" personal exposure of study subjects. When measures of exposure are highly variable in time and space, single measurements approximate the true exposure only to a limited extent. The variability of measures of exposure can be investigated by repetition of the measurements in time and space. Analysis of variance techniques can be used to separate the within-subject or "error" variance from the between-subjects or true variance. Computation of the ratio between the error variance and the true variance is a useful technique to evaluate the potential bias in correlation and regression coefficients calculated with these measures of exposure. Using data from a number of different studies, the authors have estimated the variance ratio of lead exposure and nitrogen dioxide exposure variables. The results suggest that these ratios may be large. Empirical illustrations are given of bias in regression coefficients of childhood blood levels on different lead exposure variables. It is recommended that pilot studies be performed more routinely to estimate the magnitude of the variance ratios of exposure variables of interest in studies in environmental epidemiology.
在环境流行病学研究中，接触有害因子的情况在时间和空间上往往高度可变。通常无法直接测量研究对象对这些因子的相关个人接触情况。相反，在许多情况下，接触测量是在固定地点和/或有限时间段内进行的。从只能近似研究对象“真实”个人接触情况的意义上来说，此类测量并不完美。当接触测量在时间和空间上高度可变时，单次测量仅在有限程度上近似真实接触情况。接触测量的变异性可通过在时间和空间上重复测量来研究。方差分析技术可用于将个体内部或“误差”方差与个体间或真实方差区分开来。计算误差方差与真实方差之间的比率是评估用这些接触测量值计算的相关系数和回归系数中潜在偏差的一种有用技术。作者利用来自多项不同研究的数据，估算了铅接触和二氧化氮接触变量的方差比。结果表明这些比率可能很大。文中给出了儿童血铅水平对不同铅接触变量的回归系数偏差的实证示例。建议更常规地开展预试验研究，以估算环境流行病学研究中感兴趣的接触变量的方差比大小。
