Predicting Occupational Exposures to Carbon Nanotubes and Nanofibers Based on Workplace Determinants Modeling
Predicting Occupational Exposures to Carbon Nanotubes and Nanofibers Based on Workplace Determinants Modeling
复制标题
DOI：
10.1093/annweh/wxy102
复制
发表时间：
2019-03-01
影响因子：
2.6
通讯作者：
Schubauer-Berigan, Mary K.
中科院分区：
文献类型：
--
作者：
Dahm, Matthew M.;Bertke, Stephen;Schubauer-Berigan, Mary K.
背景：最近的横断面流行病学研究调查了人类健康影响与碳纳米管和纳米纤维（CNT/F）工作场所暴露之间的关系。然而，由于许多健康效应的潜伏期，可能需要进行足够随访的队列研究。本研究的目的是确定影响暴露的工作场所决定因素，并开发预测模型来估计CNT/F暴露，以便将来在流行病学研究中使用。方法：通过15个独特的设备对可吸入和可吸入气溶胶大小部分的元素碳（EC）质量进行了暴露测量，并通过透射电子显微镜（TEM）进行了定量分析。这些指标在模型开发中充当依赖变量。从127名CNT/F工人参与者中收集了重复的个人样本，共进行了252次观察。决定因素被分类为公司级别或工人级别，并用于描述因变量内的暴露关系。利用混合线性模型探讨了决定因素对方差成分的影响，该模型利用了降低模型决定因素选择AIC的反向逐步选择过程。建立了额外的脊回归模型，检查了有和没有所有双向交互的预测性能。对每个模型进行交叉验证，以评估其预测能力的泛化性，并根据相应的R-2值和均方根误差（RMSE）评估预测性能。结果：在公司层面，增加暴露的决定因素包括不充分或半充分的工程控制等级、碳纳米管/碳纳米管平均直径/长度的增加、每天处理的材料数量从101克增加到110公斤和110公斤、碳纳米管材料的使用、混合生产商/用户的行业类型以及专家对高暴露潜力的评估。与高暴露相关的工人水平决定因素包括处理碳纳米管/碳纳米管的干粉状形式，每天处理的材料数量为100 - 100公斤，直接/间接暴露，拥有工程师的职称，使用呼吸器，使用通风或不通风的外壳，以及处理粉末的工作任务。当使用所有公司和员工层面的决定因素来创建三个暴露模型时，混合线性模型解释了60%的总方差。三种混合模型的交叉验证RMSE值范围为2.50 ~ 4.23。与此同时，脊回归模型在没有双向交互作用的情况下，可吸入性EC、可呼吸性EC和TEM预测模型的交叉验证RMSE值分别为2.85、2.23和2.76。结论：脊回归模型在预测CNT/F暴露方面表现最佳，尽管它们只提供了适度的预测能力。因此，得出的结论是，模型本身不足以预测工作场所的暴露，需要与其他方法相结合。
Background: Recent cross-sectional epidemiologic studies have examined the association between human health effects and carbon nanotube and nanofiber (CNT/F) workplace exposures. However, due to the latency of many health effects of interest, cohort studies with sufficient follow-up will likely be needed. The objective of this study was to identify workplace determinants that contribute to exposure and develop predictive models to estimate CNT/F exposures for future use in epidemiologic studies.Methods: Exposure measurements were compiled from 15 unique facilities for the metrics of elemental carbon (EC) mass at both the respirable and inhalable aerosol size fractions as well as a quantitative analysis performed by transmission electron microscopy (TEM). These metrics served as the dependent variables in model development. Repeated personal samples were collected from most of the 127 CNT/F worker participants for 252 total observations. Determinants were categorized as company-level or worker-level and used to describe the exposure relationship within the dependent variables. The influence of determinants on variance components was explored using mixed linear models that utilized a backwards stepwise selection process with a lowering of the AIC for model determinant selection. Additional ridge regression models were created that examined predictive performance with and without all two-way interactions. Cross-validation was performed on each model to evaluate the generalizability of its predictive capabilities while predictive performance was evaluated according to the corresponding R-2 value and root mean square error (RMSE).Results: Determinants at the company-level that increased exposure included an inadequate or semi-adequate engineering control rating, increasing average CNT/F diameter/length, daily quantities of material handled from 101 g to >1 kg and >1 kg, the use of CNF materials, the industry type of hybrid producer/user, and the expert assessment of a high exposure potential. Worker-level determinants associated with higher exposure included handling the dry-powdered form of CNT/F, handling daily quantities of material > 1 kg, direct/indirect exposure, having the job title of engineer, using a respirator, using a ventilated or unventilated enclosure, and the job task of powder handling. The mixed linear models explained >60% of the total variance when using all company- and worker-level determinants to create the three exposure models. The cross-validated RMSE values for each of the three mixed models ranged from 2.50 to 4.23. Meanwhile, the ridge regression models, without all two-way interactions, estimated cross-validated RMSE values of 2.85, 2.23, and 2.76 for the predictive models of inhalable EC, respirable EC, and TEM, respectively.Conclusions: The ridge regression models demonstrated the best performance for predicting exposures to CNT/F for each exposure metric, although they only provided a modest predictive capability. Therefore, it was concluded that the models alone would not be adequate in predicting workplace exposures and would need to be integrated with other methods.