Terrell Metrecia L, Manatunga Amita K, Small Chanley M, Cameron Lorraine L, Wirth Julie, Blanck Heidi Michels, Lyles Robert H, Marcus Michele
Department of Epidemiology, Rollins School of Public Health, Emory University, Atlanta, Georgia 30322, USA.
J Expo Sci Environ Epidemiol. 2008 Jul;18(4):410-20. doi: 10.1038/sj.jes.7500633. Epub 2008 Jan 9.
The Michigan Long-Term PBB Study was established following exposure to polybrominated biphenyls (PBBs) in the early 1970s. Serum samples from cohort members were analyzed for PBB during 1976-1993. More than 20 years following this industrial incident, some participants still had measurable serum PBB concentration levels. Thus, there is continuing interest in understanding the elimination of PBB from the body. In the present study, we estimated serum PBB decay and investigated the effects of covariates on serum PBB decay rates among 406 female cohort members. We developed a decay model using a general linear mixed model, which attributes unique intercept and slope estimates for each individual while borrowing information across individuals for predicting these quantities. Age at exposure and body mass index (BMI) at the initial measurement were time-independent covariates. Time since exposure, smoking history, pregnancy status, and breast-feeding status were time-dependent covariates. Higher BMI was associated with a slower decay rate; smokers had a faster decay rate than nonsmokers; and increasing age at exposure was marginally associated with a slower decay rate. Our results suggest a faster serum PBB decay rate for women who breast-fed during the interval between serum PBB measurements. To evaluate the predictive performance of our modeling approach, we compared the results from this model with those from a previously developed ordinary least squares (OLS) two-stage decay model. The mixed-effects decay model predicted the observed serum PBB concentration levels significantly better than the OLS two-stage decay model (mixed-effects model, r=0.93; OLS two-stage model, r=0.86; P<0.0001).
密歇根长期多溴联苯研究始于20世纪70年代初人群接触多溴联苯（PBBs）之后。在1976 - 1993年期间对队列成员的血清样本进行了PBB分析。在这次工业事故发生20多年后，一些参与者的血清PBB浓度水平仍可检测到。因此，人们一直对了解PBB从体内的消除情况很感兴趣。在本研究中，我们估计了406名女性队列成员血清PBB的衰减情况，并研究了协变量对血清PBB衰减率的影响。我们使用一般线性混合模型开发了一个衰减模型，该模型为每个个体赋予独特的截距和斜率估计值，同时在个体间借用信息来预测这些量。接触时的年龄和初始测量时的体重指数（BMI）是与时间无关的协变量。接触后的时间、吸烟史、怀孕状况和母乳喂养状况是与时间有关的协变量。较高的BMI与较慢的衰减率相关；吸烟者的衰减率比不吸烟者快；接触时年龄的增加与较慢的衰减率有一定关联。我们的结果表明，在两次血清PBB测量间隔期间进行母乳喂养的女性血清PBB衰减率更快。为了评估我们建模方法的预测性能，我们将该模型的结果与之前开发的普通最小二乘法（OLS）两阶段衰减模型的结果进行了比较。混合效应衰减模型预测的观察到的血清PBB浓度水平明显优于OLS两阶段衰减模型（混合效应模型，r = 0.93；OLS两阶段模型，r = 0.86；P < 0.0001）。
