Distinguishing petroleum (crude oil and fuel) from smoke exposure within populations based on the relative levels of benzene toluene ethylbenzene and xylenes (BTEX) styrene and 25-dimethylfuran by pattern recognition using artificial neural networks

摘要

Studies of human exposure to petroleum (crude oil and fuel) often involve monitoring volatile monoaromatic compounds because of their toxicity and prevalence. Monoaromatic compounds such as benzene, toluene, ethylbenzene, and xylenes (BTEX) associated with these sources have been well studied and have established reference concentrations (RfC) and reference doses (RfD). However, BTEX exposure levels for the general population are primarily from tobacco smoke, where smokers have blood levels up to 8 times higher on average than nonsmokers. Therefore, in assessing petroleum exposure, it is essential to identify exposure to tobacco smoke as well as other types of smoke exposure (e.g., cannabis, wood) because many smoke volatile organic compounds are also found in petroleum products such as crude oil, and fuel. This work describes a method using partition theory and artificial neural network (ANN) pattern recognition to accurately categorize exposure source based on BTEX and 2,5-dimethylfuran blood levels. For this evaluation three categories were created and include crude oil/fuel, otheronsmoker, and smoker. A method for using surrogate signatures (i.e., relative VOC levels derived from the source material) to train the ANN was investigated where blood levels among cigarette smokers from the National Health and Nutrition Examination Survey (NHANES) were compared with signatures derived from machine-generated cigarette smoke. Use of surrogate signatures derived from machine-generated cigarette smoke did provide a sufficient means with which to train the ANN. As a result, surrogate signatures were used for assessing crude oil/fuel exposure because there is limited blood level data on individuals exposed to either crude oil or fuel. Classification agreement between using an ANN model trained with relative VOC levels and using the 2,5-dimethylfuran smoking biomarker cutpoint blood level of 0.014 ng/mL was up to 99.8 % for nonsmokers and 100.0% for smokers. For the NHANES 2007–08 data, the ANN model using a probability cutpoint above 0.5 assigned 7 samples out of 1998 (0.35%) to the crude oil/fuel signature category. For the NHANES 2013–14 data, 12 out of 2906 samples (0.41%) were assigned to the crude oil/fuel signature category. This approach using ANN makes it possible to quickly identify individuals with blood levels consistent with a crude oil/fuel surrogate among thousands of results while minimizing confounding from smoke. Use of an ANN fixed algorithm makes it possible to objectively compare across populations eliminating classification inconsistency that can result from relying on visual evaluation.