As the primary goal of the present study is to introduce this novel approach rather than to estimate and generalize actual risks in the U.S. population, and as some of the statistical procedures used in our approach are not equipped with automated handling of survey weights, we did not account for the complex sampling design and used conventional regression modeling. Biomonitoring data in NHANES were not measured in all participants; some pollutants were measured only in a subset (e.g., one third) and different kinds (classes) of pollutants were measured in different subsets in order to reduce the burden of examinations, which limits the sample size for this multi-pollutant model. To maximize the power of the proposed approach, we imputed unmeasured or missing pollutant data. For these reasons, our findings should be cautiously interpreted as potential associations. Another new feature of the present study is that we examined 4 lipid outcomes separately (single-phenotype approach) as well as all 4 lipid outcomes together as a whole (multi-phenotype approach). This multi-phenotype approach can also help improve the power to detect modest individual effects of environmental pollutants and reduce the burden of multiple testing [30]–[32].
