Statistical analysis was performed using STATA version 12 [49 ]. Samples with analyte concentrations below the LOD were imputed with LOD/√2. For analyses that compared concentrations of BPA, BPA glucuronide and BPA sulfate within samples, we converted the concentrations of BPA glucuronide and BPA sulfate to the concentrations of BPA in glucuronide and sulfate forms, respectively. Specifically, we multiplied the concentrations of BPA glucuronide by the ratio of the molecular weight of BPA to that of BPA glucuronide (0.5645), and the concentrations of BPA sulfate by the ratio of the molecular weight of BPA to that of BPA sulfate (0.7404). We also summed BPA, BPA in glucuronide form and BPA in sulfate form to calculate Total BPA and the percent of Total BPA in original, glucuronide and sulfate form. In addition to permitting a comparison of the relative levels of the analytes within a sample, BPA in glucuronide form, BPA in sulfate form and Total BPA are the best metrics for comparing concentrations in this study to all previous studies of BPA exposure in umbilical cord serum, which made indirect measurements of conjugated BPA.
We log-transformed analyte concentrations due to their right-skewed distributions; only BPA sulfate and Total BPA levels were approximately normal after transformation and therefore we used non-parametric tests (Kruskal-Wallis, Spearman’s, Wilcoxon signed rank) for univariate analyses. To compare concentrations of analytes within samples, we defined three groups based on the dominant analyte in the sample, compared levels of Total BPA between these groups and, using Pearson’s χ2 test, evaluated the association between these groups and having Total BPA levels above the median. Sex differences in BPA metabolic pathways have been reported in the literature on adult humans; therefore we examined sex differences in analyte levels and in dominant BPA analyte using Fisher’s Exact test. Lastly, we evaluated the association between gestational age and analyte levels through correlation and regression models, adjusting for sex. Due to the frequencies of non-detect BPA and BPA glucuronide, for these outcomes we used tobit regression [50 -51 ], which treats