Method | Research Question | Results |
---|---|---|
Unsupervised Methods | ||
K-means Clustering Hierarchical Clustering | Are there population subgroups that share similar exposure profiles? | The study population is clustered by level of exposure: high, average, and low. High exposure is associated with longer log-LTL. |
PCA | Are there specific patterns in POP exposure? | Three patterns of exposure were identified. Exposure to all POPs (PC1) is associated with longer log-LTL. |
EFA | Â | Four patterns of exposure were identified. Exposure to all four furans (FA2) and to PCBs 118 and 126 (FA3) is associated with longer log-LTL. |
Supervised Methods | ||
Lasso | Which congeners are associated with changes in log-LTL? | PCB 99, PCB 118, PCB 126, and furan 2,3,4,7,8-pncdf are associated with longer log-LTL. |
Elastic Net | Â | PCB 99, PCB 118, PCB 126, furan 2,3,4,7,8-pncdf, and furan 1,2,3,4,6,7,8 hxcdf are associated with longer log-LTL. |
Group Lasso | Which a priori defined congener groups are associated with changes in log-LTL and what is the magnitude of the association with congeners within those groups? | All mPFD congeners are associated with log-LTL, with variability in direction—five mPFDs with longer log-LTL, and three mPFDs with shorter log-LTL. Non-ortho PCBs (PCBs 126 and 169) are associated with longer log-LTL. |
WQS | What is the overall effect of the mixture on log-LTL? What congeners are most important? | The mixture index is associated with longer log-LTL. Three furans and both non-ortho PCBs are important mixture members. Furans 1,2,3,4,6,7,8-hxcdf and 2,3,4, 7,8-pncdf has the largest weights. |
BKMR | Is there an association between the mixture and log-LTL? What is the exposure-response relationship between each congener and log-LTL? Are there interactions between congeners? | The overall mixture is associated with longer log-LTL. Furan 2,3,4,7,8-pncdf, PCB 126, and PCB 169 are independently associated with longer log-LTL. No interactions or nonlinearities were found. |