Background: Polychlorinated biphenyls (PCBs) have been associated with a variety of health outcomes. Enhanced laboratory techniques can provide a relatively large number of individual PCB congeners for investigation. However, to date there are no established frameworks for grouping a large number of PCB congeners into meaningful analytic units.
Methods: In a case-control study of serum PCB levels on breast cancer risk, measured levels of 56 PCB congener peaks were available for analysis. We considered several approaches for grouping these compounds based on 1) chlorination, 2) factor analysis, 3) enzyme induction, 4) enzyme induction and occurrence, and 5) enzyme induction, occurrence, and other toxicological aspects. The utility of a framework was based on the mechanism of biologic actions within each framework, lack of collinearity among congener groups, and frequency of detection of PCB congener groups in measured serum levels of 192 healthy postmenopausal women.
Results: Most participants had detectable levels for the proposed PCB congeners groups, using degree of chlorination as a grouping framework. In addition, the previously proposed grouping approach based on enzyme induction, occurrence, and other toxicological aspects was an applicable alternative to the crude approach of grouping by degree of chlorination. Grouping these congeners with respect to P450 enzyme induction activity, and the previously proposed framework based on enzyme induction and occurrence, did not fit these data as well, because only a small proportion of participants had detectable levels for the congener groups with the greatest toxicological potential. Statistical grouping did not result in an interpretable and meaningful clustering of these exposures.
Conclusions: In these data, grouping with respect to degree of chlorination and the previously proposed framework based on enzyme induction, occurrence, and other toxicological aspects were the most useful approaches to reducing a large number of PCB congeners into meaningful analytic units. Factors affecting the utility of the proposed grouping frameworks are discussed.