Introduction: Individuals are exposed to many diverse chemicals and it is important to estimate the relationship between many chemical exposures and cancer risk. The method of weighted quantile sum (WQS) regression estimates a single weighted linear index of correlated components and its association with a health outcome. WQS performs well in simulation studies in identifying the important components of a chemical mixture, however, the chemicals in the index are constrained to have either effects in the same direction or no effect and the application of WQS is limited to settings where it is reasonable to combine all chemicals into a single index. In order to allow chemicals to have associations in different directions with the outcome, we present the method of grouped weighted quantile sum (GWQS) regression.

Methods: We develop an approach to model many chemical exposures from different chemical classes jointly and apply it to a case-control study of non-Hodgkin lymphoma (NHL) in four geographically distinct National Cancer Institute-Surveillance, Epidemiology, and End Results centers (Detroit; Iowa; Los Angeles; Seattle). The study contains concentrations of 5 polychlorinated biphenyls (PCBs), 7 polycyclic aromatic hydrocarbons (PAHs), and 15 pesticides measured in residential house dust. We consider different approaches to specifying the group memberships for chemicals and compare the goodness-of-fit of the approaches.

Results: We found evidence of a directional change in the association between NHL and select pesticides across study sites (e.g., negative for 2,4-D and dicamba in Iowa and Seattle) and performed grouped WQS in each study center using site-specific groupings of pesticides to create both a positive and negative pesticide index along with an index for PCBs and PAHs. There was a positive association between PCBs and NHL.

Conclusions: GWQS regression extends WQS regression to have a more flexible and realistic specification of chemical mixture effects.