Our study aimed to investigate dietary and non-dietary predictors of exposure to pyrethroids, organophosphates pesticides and 2,4-D herbicide in two cohorts of pregnant women in New York City: 153 women from the Thyroid Disruption and Infant Development (TDID) cohort and 121 from the Sibling/Hermanos Cohort(S/H). Baseline data on predictors were collected from the women at time of recruitment. We used three different modeling strategies to address missing data due to biomarker values below the limit of detection (<LOD): (1) logistic regression models with biomarkers categorized as (<median, ≥median); (2) linear regression models, imputing the <LOD values with (LOD/√2); (3) regression models, considering <LOD values as left-censored. Generally, all three models identified similar predictors of exposure. We found that ethnicity, higher income and education predicted higher concentrations of most of the biomarkers in both cohorts. Mothers who consumed processed meat in the TDID cohort, and broiled, barbequed food or burgers in the S/H cohort, tended to have lower concentrations of organophosphates and 2,4-D. The choice of modeling led to a few different predictors identified, and the selection of modeling strategy should be based on the study question.
Keywords: limit of detection; modeling strategies; organophosphate; pesticides; predictors; pyrethroid.