Institutional members access full text with Ovid®

Share this article on:

Identifying Subsets of Complex Mixtures Most Associated With Complex Diseases: Polychlorinated Biphenyls and Endometriosis as a Case Study

Gennings, Chrisa; Sabo, Roya; Carney, Edb

doi: 10.1097/EDE.0b013e3181ce946c
Methodologic Issues in Environmental Exposures: Mixtures and Limits of Detection: Original Article

Background: Exploratory statistical analyses have been conducted on an epidemiologic data set in which the relationship was examined between exposure to polychlorinated biphenyl (PCB) mixtures and risk of endometriosis in women. In that study, the association between endometriosis and the sum of 4 antiestrogenic PCBs (PCBs 105, 114, 126, and 169) was borderline significant (P = 0.079), whereas an association was not found (P = 0.681) with the sum of 12 estrogenic PCBs. This finding was inconsistent with the widely held notion that endometriosis is an estrogen-dependent disease, prompting further statistical analyses to explore these associations in more detail.

Methods: As an alternative method of data reduction, an optimization algorithm was developed to determine weights in a linear combination of scaled PCB levels that has the strongest possible association with the risk of endometriosis.

Results: Application of this method to the antiestrogenic PCB subgroup revealed that PCB 114 was responsible for nearly 100% of the association. The fact that PCB 114 is neither the most potent nor abundant antiestrogen in the mixture suggests that PCB 114 might be estrogenic or that the association may be driven by a different mechanism. Use of this statistical weighting method for further analyses of 12 estrogenic PCBs showed that any association with endometriosis was driven mainly by PCBs 99 and 188 and possibly a few others.

Conclusion: Although the role of PCB mixtures in endometriosis remains unclear, these results demonstrate how the integration of refined statistical methods coupled with toxicologic and biologic interpretation can generate testable hypotheses that might not otherwise have been generated.

From the aDepartment of Biostatistics, Virginia Commonwealth University, Richmond, VA; and bToxicology and Environmental Research and Consulting, The Dow Chemical Company, Midland, MI.

Submitted 15 December 2008; accepted 14 October 2009; posted 23 March 2010.

Supported in part by the American Chemistry Council and the Intramural Research Program of the Eunice Kennedy Shriver National Institute of Child Health and Human Development.

Correspondence: Chris Gennings, Department of Biostatistics, Virginia Commonwealth University, 730 East Broad St., No. 3026, Richmond, VA 23298-0032. E-mail:

© 2010 Lippincott Williams & Wilkins, Inc.