Norwegian Institute of Public Health, Oslo, Norway.
Environ Int. 2012 Dec 1;50:15-21. doi: 10.1016/j.envint.2012.09.003. Epub 2012 Sep 29.
Dioxins and PCBs accumulate in the food chain and might exert toxic effects in animals and humans. In large epidemiologic studies, exposure estimates of these compounds based on analyses of biological material might not be available or affordable.
To develop and then validate models for predicting concentrations of dioxins and PCBs in blood using a comprehensive food frequency questionnaire and blood concentrations.
Prediction models were built on data from one study (n=195), and validated in an independent study group (n=66). We used linear regression to develop predictive models for dioxins and PCBs, both sums of congeners and 33 single congeners (7 and 10 polychlorinated dibenzo-p-dioxins and furans (PCDDs/PCDFs), 12 dioxin-like polychlorinated biphenyls (PCBs: 4 non-ortho and 8 mono-ortho), sum of all the 29 dioxin-like compounds (total TEQ) and sum of 4 non dioxin-like PCBs (∑ CB-101, 138, 153, 183=PCB(4)). We used the blood concentration and dietary intake of each of the above as dependent and independent variables, while sex, parity, age, place of living, smoking status, energy intake and education were covariates. We validated the models in a new study population comparing the predicted blood concentrations with the measured blood concentrations using correlation coefficients and Weighted Kappa (К(W)) as measures of agreement, considering К(W)>0.40 as successful prediction.
The models explained 78% (sum dioxin-like compounds), 76% (PCDDs), 76% (PCDFs), 74% (no-PCBs), 69% (mo-PCBs), 68% (PCB(4)) and 63% (CB-153) of the variance. In addition to dietary intake, age and sex were the most important covariates. The predicted blood concentrations were highly correlated with the measured values, with r=0.75 for dl-compounds 0.70 for PCB(4), (p<0.001) and 0.66 (p<0.001) for CB-153. К(W) was 0.68 for sum dl-compounds 0.65 for both PCB(4) and CB-153. Out of 33 congeners 16 (13dl-compounds and 3 ndl PCBs) had К(W)>0.40.
The models developed had high power to predict blood levels of dioxins and PCBs and to correctly rank subjects according to high or low exposure based on dietary intake and demographic information. These models underline the value of dietary intake data for use in investigations of associations between dioxin and PCB exposure and health outcomes in large epidemiological studies with limited biomaterial for chemical analysis.
二恶英和多氯联苯会在食物链中积累，可能对动物和人类产生毒性作用。在大型流行病学研究中，基于生物材料分析的这些化合物的暴露估计可能无法获得或无法承受。
开发并验证使用综合食物频率问卷和血液浓度预测血液中二恶英和多氯联苯浓度的模型。
预测模型建立在一项研究（n=195）的数据上，并在独立的研究组（n=66）中进行了验证。我们使用线性回归来建立二恶英和多氯联苯的预测模型，包括同系物和 33 种单同系物（7 种和 10 种多氯二苯并对二恶英/呋喃（PCDD/Fs）、12 种类二恶英多氯联苯（PCBs：4 种非邻位和 8 种单邻位）、所有 29 种类二恶英化合物的总和（总 TEQ）和 4 种非类二恶英多氯联苯的总和（∑ CB-101、138、153、183=PCB(4)）。我们将血液浓度和每种上述物质的饮食摄入量作为因变量和自变量，同时考虑性别、生育次数、年龄、居住地点、吸烟状况、能量摄入和教育程度作为协变量。我们在新的研究人群中验证了这些模型，通过比较预测的血液浓度和测量的血液浓度，使用相关系数和加权 Kappa（K(W)）作为一致性的衡量标准，考虑到 K(W)>0.40 表示成功预测。
该模型解释了 78%（类二恶英化合物总和）、76%（PCDDs）、76%（PCDFs）、74%（非多氯联苯）、69%（单多氯联苯）、68%（PCB(4)）和 63%（CB-153）的方差。除了饮食摄入量外，年龄和性别是最重要的协变量。预测的血液浓度与测量值高度相关，dl-化合物的 r=0.75，PCB(4)的 r=0.70（p<0.001），CB-153 的 r=0.66（p<0.001）。K(W)值为类二恶英化合物总和的 0.68，PCB(4)和 CB-153 的 K(W)值分别为 0.65 和 0.65。在 33 种同系物中，有 16 种（13 种类二恶英化合物和 3 种非类二恶英多氯联苯）的 K(W)>0.40。
所开发的模型具有较高的能力来预测血液中二恶英和多氯联苯的水平，并根据饮食摄入和人口统计学信息正确地对高暴露和低暴露人群进行分类。这些模型强调了饮食摄入数据在大型流行病学研究中用于研究二恶英和多氯联苯暴露与健康结果之间关联的价值，这些研究的生物材料化学分析有限。
