Defense Date


Document Type


Degree Name

Doctor of Philosophy



First Advisor

David C. Wheeler

Second Advisor

Chris Gennings


Cancer incidence is associated with exposures to multiple environmental chemicals, and geographic variation in cancer rates suggests the importance of accommodating spatially varying effects in the analysis of environmental chemical mixtures and disease risk. Traditional regression methods are challenged by the complex correlation patterns inherent among co-occurring chemicals, and the applicability of geographically weighted regression models is limited in the setting of environmental chemical risk analysis. In comparison to traditional methods, weighted quantile sum (WQS) regression performs well in the identification of important environmental exposures, but is limited by the assumption that effects are fixed over space. We present an extension of the WQS method that models spatially varying chemical mixture effects called local weighted quantile sum (LWQS) regression, and assess through a simulation study its ability to identify important environmental risk factors over space. We use two different approaches to estimate the LWQS model based on variable subspaces. One uses an ensemble of variable subsets of the same size, and the other selects the best subset over a range of candidate subset sizes according to the model goodness-of-fit. We assess the performance of both estimation methods in simulated scenarios that incorporate increasingly complex levels of spatial dependency in the model, and consider correlation patterns from observed exposure data. The results demonstrate that LWQS has the ability to replicate spatially dependent mixture effects and can correctly identify important exposures in a mixture of environmental chemicals. In all scenarios, the best subset approach correctly chose an index containing only the important chemicals and improved on the accuracy of the chemical importance weights in comparison with the ensemble solutions. Future work will evaluate if the ensemble subset approach has better relative performance with larger chemical mixtures of highly correlated components.


© The Author

Is Part Of

VCU University Archives

Is Part Of

VCU Theses and Dissertations

Date of Submission


Included in

Biostatistics Commons