Xue Xiaonan, Kim Mimi Y, Shore Roy E
Division of Biostatistics, Department of Epidemiology & Population Health, Albert Einstein College of Medicine, 1300 Morris Park Avenue, Bronx, NY 10461, USA.
Health Phys. 2006 Dec;91(6):582-91. doi: 10.1097/01.HP.0000225466.45659.08.
Occupational exposures are subject to several types of measurement errors. This paper considers two of the most common types of measurement error associated with occupational exposures: the error due to below minimum detection level and doses due to random measurement error. Doses are often recorded as zero when the exposure is below the minimum detection level. Values that are below the minimum detection level and are entered as zero lead to underestimation of the true exposure and can result in either an overestimate or underestimate of risk associated with the exposure. Random measurement error leads to an inefficient and attenuated estimate of risk associated with exposure. However, the levels of bias and inefficiency that can result from the simultaneous presence of both types of measurement error have not previously been studied. In addition, the impact of these measurement errors on the type I error and type II error for an exposure-response effect is unclear. Since the magnitude of the random error associated with cumulative exposure may vary with individuals and across time within an individual, traditional methods to correct for random measurement errors are not applicable here. Further, correcting errors for minimum detectable levels and random errors simultaneously is too complex for analytical solutions. Therefore, this paper uses simulation studies to quantitatively evaluate the magnitude of the bias, inefficiency, and type I and type II errors associated with them. The simulation results are applied to a sample of historical occupational radiation exposure data from the Oak Ridge National Laboratory. We conclude that after taking into consideration random measurement error and missed doses due to falling below the minimum detection level, radiation exposure is not significantly associated with all-cause mortality in that cohort.
职业暴露存在几种类型的测量误差。本文考虑了与职业暴露相关的两种最常见的测量误差类型：因低于最低检测水平导致的误差以及随机测量误差导致的剂量。当暴露低于最低检测水平时，剂量通常记录为零。低于最低检测水平并记为零的值会导致对真实暴露的低估，并可能导致对与该暴露相关风险的高估或低估。随机测量误差会导致对与暴露相关风险的估计效率低下且有偏差。然而，此前尚未研究过这两种测量误差同时存在可能导致的偏差和效率低下的程度。此外，这些测量误差对暴露 - 反应效应的I型错误和II型错误的影响尚不清楚。由于与累积暴露相关的随机误差大小可能因个体而异，且在个体内部随时间变化，传统的校正随机测量误差的方法在此处不适用。此外，同时校正最低可检测水平的误差和随机误差对于解析解来说过于复杂。因此，本文使用模拟研究来定量评估与之相关的偏差、效率低下以及I型和II型错误的大小。模拟结果应用于橡树岭国家实验室的一组历史职业辐射暴露数据样本。我们得出结论，在考虑随机测量误差和因低于最低检测水平而漏记的剂量后，该队列中的辐射暴露与全因死亡率无显著关联。
