Results
Table 1 presents the baseline characteristics of the study population. Among 2,200 children 6–18 years of age included in the current analysis, approximately half were girls (48.5%), and 62.4% were non-Hispanic whites. Over half (55.4%) of the parents/guardians had an education above high school, and over a third of the sample (34.4%) were overweight or obese.
Table 2 presents the association between increasing urinary BPA levels and mean change in body mass index in children. Overall, there was a positive association between increasing levels of urinary BPA and body mass index in both the age- and sex-adjusted model and the multivariable-adjusted model. Corresponding models evaluating linear trend in this association were statistically significant. We also found a similar positive association between urinary BPA levels and body mass index when BPA was analyzed as a continuous variable, with logarithmic transformation.
Table 3 presents the association of increasing levels of urinary BPA with obesity in children. Overall, there was a positive association between the increasing levels of urinary BPA and obesity in both the age- and sex-adjusted models and in the multivariable adjusted model. Models evaluating linear trend in this association were statistically significant.
Table 4 and Table 5 present the association of increasing levels of urinary BPA with obesity in children, within the subgroups of gender and race/ethnicity. In Table 4, we found that the association between increasing urinary BPA levels and obesity was of strong magnitude and statistically significant among boys, but weak and statistically nonsignificant in girls (Pinteraction = 0.07). In Table 5, we found that the association between increasing urinary BPA levels and obesity was strongly present among non-Hispanic whites, but it was weak and statistically nonsignificant in nonwhites (Pinteraction = 0.05).
In order to further clarify these observed differences in Table 4 and Table 5 and to examine any potential synergistic interaction, we created joint exposure categories of sex and race/ethnicity on the association between increasing urinary BPA levels and obesity in Table 6. We found that the observed positive association between BPA quartiles and obesity in children was predominantly present among non-Hispanic white boys (odds ratio = 18.89, 95% confidence interval (CI): 3.97, 89.89). In contrast, the BPA–obesity association was weak and statistically nonsignificant in the other subgroups.
We also performed several supplementary analyses. First, to examine the influence of adding higher order polynomial terms for age, we included a quadratic term for age in the multivariable model; the results were found to be essentially similar (Web Table 1 available at http://aje.oxfordjournals.org/). Second, to examine the possibility that sex differences in maturation may influence the association between BPA and obesity, we incorporated an age (continuous)–sex interaction term in the multivariable logistic regression model. The P value for the age–sex interaction was 0.2709. Third, to examine age differences in the BPA–obesity association in more detail, we performed a stratified analysis by age group; the BPA–obesity association was found to be essentially similar in prepubertal (6–11 years) and postpubertal (12–18 years) children (Web Table 2).
Fourth, to further examine the potential of selection bias, we compared the demographic characteristics of the entire NHANES sample of 6,559 children with the sample of 2,200 children included in the current study (who had BPA data available); the demographic characteristics were found to be essentially the same in the 2 samples (Web Table 3). Fifth, we performed a sensitivity analysis to examine the BPA–obesity association in 464 children who were excluded from the analyses because of missing data on covariates. The results showed a similar magnitude of association as the main findings, albeit not statistically significant because of the small sample size, where the odds ratio of obesity associated with log-transformed BPA was 1.12 (95% CI: 0.90, 1.39).
Sixth, to examine whether the use of gender-specific and race/ethnicity-specific BPA quartiles made any difference to our findings, we repeated the main analyses using gender-specific (Web Table 4) and race/ethnicity-specific BPA quartiles (Web Table 5); the odds ratios for the BPA–obesity association were essentially the same as those from the main analyses using full-sample BPA quartiles. Seventh, we calculated prevalence proportion ratios instead of odds ratios as a measure of the magnitude of the BPA–obesity association in this cross-sectional study. This analysis used a Cox proportional hazards model (which has complex survey options in SAS/SUDAAN) by assuming that the risk period is constant (by assigning an equal follow-up period for each observation). The resultant prevalence proportion ratios were essentially similar in magnitude to the odds ratios in the main analysis, but with wider standard errors (Web Table 6). In a final supplementary analysis, we examined whether there were differences in demographic factors between gender and race/ethnicity subgroups. We found that age and urinary BPA levels were similar across these subgroups, but that body mass index was different; it was higher in nonwhite girls, but similar in the other subgroups (Web Table 7).