Race/Ethnicity–Specific Associations of Urinary Phthalates with Childhood Body Mass in a Nationally Representative Sample

Background: Phthalates have antiandrogenic effects and may disrupt lipid and carbohydrate metabolism. Racial/ethnic subpopulations have been documented to have varying urinary phthalate concentrations and prevalences of childhood obesity. Objective: We examined associations between urinary phthalate metabolites and body mass outcomes in a nationally representative sample of U.S. children and adolescents. Methods: We performed stratified and whole-sample cross-sectional analyses of 2,884 children 6–19 years of age who participated in the 2003–2008 National Health and Nutrition Examination Survey. Multivariable linear and logistic analyses of body mass index z-score, overweight, and obesity were performed against molar concentrations of low-molecular-weight (LMW), high-molecular-weight (HMW), and di-2-ethylhexylphthalate (DEHP) metabolites, controlling for sex, television watching, caregiver education, caloric intake, poverty–income ratio, race/ethnicity, serum cotinine, and age group. We used sensitivity analysis to examine robustness of results to removing sample weighting, normalizing phthalate concentrations for molecular weight, and examining different dietary intake covariates. Results: In stratified, multivariable models, each log unit (roughly 3-fold) increase in LMW metabolites was associated with 21% and 22% increases in odds (95% CI: 1.05–1.39 and 1.07–1.39, respectively) of overweight and obesity, and a 0.090-SD unit increase in BMI z-score (95% CI: 0.003–0.18), among non-Hispanic blacks. Significant associations were not identified in any other racial/ethnic subgroup or in the study sample as a whole after controlling for potential confounders, associations were not significant for HMW or DEHP metabolites, and results did not change substantially with sensitivity analysis. Conclusions: We identified a race/ethnicity–specific association of phthalates with childhood obesity in a nationally representative sample. Further study is needed to corroborate the association and evaluate genetic/epigenomic predisposition and/or increased phthalate exposure as possible explanations for differences among racial/ethnic subgroups.


Research | Children's Health
Phthalates are diesters of phthalic acid that are used to produce an array of consumer products and have been associated with abnormalities in end points related to endo crine processes, including reproductive and neuro developmental outcomes (Bornehag and Nanberg 2010;Colon et al. 2000;Engel et al. 2010;Miodovnik et al. 2011;Sathyanarayana 2008;Swan 2008;Swan et al. 2005). A causal role for phthalates in obesity is biologically plausible, based in part on evidence of high molecularweight (HMW) phthalate interactions with three peroxisome proliferatoractivated receptors (PPARs) that are members of the nuclear receptor superfamily (Desvergne et al. 2009). PPARs are involved in lipid and carbohydrate metabolism; therefore, their activation may represent a mechanism by which phthalates could produce metabolic derangements that might contribute to obesity (Grün and Blumberg 2007).
Three analyses have examined urinary phthalates and obesity in humans to date. The first reported that urinary phthalates were associated with abdominal obesity and insulin resistance in adults in the 1999-2002 U.S. National Health and Nutrition Examination Survey (NHANES) (Stahlhut et al. 2007). In contrast, quartiled urinary phthalate con centrations in 1,209 children and adolescents in the 1999-2002 NHANES were not asso ciated with unstandardized measurements of body mass index (BMI) or waist circum ference (WC) (Hatch et al. 2008), although patterns of association varied by age and sex. Finally, in a population of largely Latino, New York City children (Teitelbaum et al. 2012), urinary phthalates measured in 6 to 8yearolds were not associated with BMI and WC 1 year later in the population as a whole, but body mass measures were posi tively associated with logtransformed phtha late metabolites for monoethyl phthalate (MEP) and with the sum of LMW phthalate metabolites [MEP, mononbutylphthalate (MBP), monoisobutyl phthalate (MiBP), and mono(3carboxypropyl) phthalate (MCPP)] among overweight children.
Racial/ethnic differences in creatinine excretion are well documented (Barr et al. 2005), as are differences in urinary phthalate concentrations Wolff et al. 2007). The prevalence of obesity also varies by race/ethnicity (Freedman et al. 2006;Ogden et al. 2012;Strauss and Pollack 2001), raising the possibility that racial/ethnic dif ferences in phthalate exposures may partly explain racial/ethnic differences in obesity. Yet despite the possibility that differences in obesogenic exposures or genetic/epigenomic predisposition may contribute to differences in obesity, and the potentially substantial implications this would have for targeting prevention efforts, few studies have examined racial/ethnic differences in environmental risk factors for childhood obesity.
We therefore analyzed data from the 2003-2008 NHANES to study associations of urinary phthalate metabolites with body mass and examine differences according to race/ethnicity and other factors.

Methods
Data source and sample. NHANES is a continuous, multicomponent, nation ally representative survey of the noninstitu tionalized U.S. population administered by the National Centers for Health Statistics (NCHS) of the Centers for Disease Control and Prevention (CDC 2012b). We used data from the questionnaire, laboratory, diet, and physical examination components in the present analysis, for which data are available in biennial groupings. Our analytic sample comprised 2,884 nonpregnant participants 6-19 years of age with urinary phthalate measurements. NHANES is approved by the NCHS Research Ethics Review Board, and written informed consent and child assent (as appropriate) was obtained from participants. The NYU School of Medicine Institutional Review Board exempted the present study from review because it is based on previously collected and deidentified data.
volume 121 | number 4 | April 2013 • Environmental Health Perspectives Measurement of urinary phthalates. Phthalates were measured in a spot urine sample collected from a randomly selected subsample of NHANES participants using highperformance liquid chromatography and tandem mass spectroscopy, as previously described . Phthalate concen trations below the level of detection [5.1% for mono(2ethylhexyl) phthalate (MEHP), < 1% for all other metabolites studied] were assigned the limit of detection divided by the square root of 2, as recommended by NHANES. All models included urinary creatinine to adjust for urine dilution, following usual practice (Barr et al. 2005;Stahlhut et al. 2009).
Our primary exposure variables were the natural logtransformed total molar concentra tions of LMW, HMW, and DEHP metabo lites. In addition, we estimated associations with metabolite groups categorized into tertiles, and with selected individual phthalate metabolites.
Body mass outcomes. In the NHANES, trained health technicians assessed body mea surements, following published, standardized measurement procedures (Lohman et al. 1998). BMI was calculated by dividing the weight in kilograms by the height in meters squared, and zscores were derived from 2000 CDC refer ence growth curves  using the zanthro command in Stata 12.0 (StataCorp., College Station, TX). Overweight and obese were classified as BMI zscore ≥ 85th percentile for age and sex and ≥ 95th percentile, respectively (GrummerStrawn et al. 2010;Ogden et al. 2002). Study outcomes were obesity (BMI zscore ≥ 95th percentile vs. < 95th percentile), overweight (BMI zscore ≥ 85th percentile vs. < 85th percentile), and BMI zscore (as continuous variable).
Potential confounders and other covariates. Trained interviewers fluent in Spanish and English elicited two total 24hr dietary recalls using standard measuring guides to assist reporting of volumes and dimensions of food items, and responses were converted to energy and nutrients by appropriate nutri tional software (CDC 2012a). We used the first of the two 24hour recalls in the pres ent analysis. Because the measurement of physical activity changed during the study period, we were unable to categorize physical activity into low, medium, and high groups normally used to derive caloric needs based on age and sexspecific U.S. Department of Agriculture (USDA) guidelines (USDA 2010). Therefore, as a conservative measure, we categorized participants into "normal" or "excessive" caloric intake groups based upon daily caloric guidelines for high physical activ ity children, recognizing that this probably underestimates the proportion who exceeded USDA calorie intake guidelines. We dichoto mized selfreported television watching as < 2 or ≥ 2 hr/day, in light of previous associations with obesity (American Academy of Pediatrics Committee on Public Education 2001), and associations with urinary phthalates in our study sample. Because exposure to tobacco smoke is a risk factor for metabolic syndrome in adolescence (Weitzman et al. 2005), and because serum cotinine was positively associ ated with urinary phthalate metabolites in our study population, we included serum cotinine in multivariable models categorized as low (< 0.015 ng/mL), medium (0.015-2 ng/mL), or high (≥ 2 ng/mL).
Race/ethnicity was categorized as Mexican American, other Hispanic, nonHispanic white, nonHispanic black, and other, based on selfreport by 17 to 19yearolds and care giver report for 6 to 16yearolds. Caregiver education was categorized as less than 9th grade, 9th-12th grade, high school/graduate equivalency diploma, some college, and col lege or greater. Poverty-income ratio (annual household income/poverty level) was catego rized into quartiles. Age was categorized as 6-11 or 12-19 years. To maximize sample sizes in multivariable analyses, we included "missing" categories for all potential con founders. Television watching was missing for 24.4%, and serum cotinine was missing for 9.6%. Otherwise, < 5% of values were miss ing for any covariate. Recognizing concerns raised about potential bias due to the use of missing data categories in regression models (Jones 1996), we repeated our main model as a complete case analysis, omitting participants with missing values for any of the covariates.
Statistical analysis. We conducted uni variable, bivariable, and multivariable analyses using statistical techniques that account for the complex survey sampling design, using Stata 12.0, and following NCHS guidelines (CDC 2012b). We used multivariable linear regression analysis to model BMI zscore, and logistic regression to model categorical over weight and obesity in separate models.
We used logtransformed LMW, HMW, and DEHP urinary metabolite concentrations in our analyses to account for skew in the dis tribution of urinary phthalates. We performed separate univariate regressions of each expo sure against BMI zscore, overweight, obesity, and covariates. We adjusted all multivariable models for urinary creatinine (model A). Next we added demographic and exposure characteristics (race/ethnicity, age, caregiver education, poverty-income ratio, sex, serum cotinine) (model B), and then lifestyle charac teristics (measures of caloric intake, television watching) (model C).
We also developed univariate and multi variable regression models of the phtha lateobesity association stratified by sex, for which differences in urinary phthalates have also been noted, age (6-11 or 12-19 years), poverty-income ratio (< 1.6 or ≥ 1.6), cotinine level (< 2.0 or ≥ 2.0 ng/mL), parent education (no college or at least some college), caloric intake (excessive or appropriate), and tele vision watching (< 2 or ≥ 2 hr/day). In addi tion, we stratified on race/ethnicity, classified as non Hispanic black, Hispanic (Mexican American and other Hispanic combined), and non Hispanic white to maintain large stratum specific samples. As a test of robustness, we esti mated associations according to race/ ethnicity by modeling product interaction terms for the exposure and potential modifier, in addi tion to lowerorder terms and covariates, in wholesample regression models controlling for all covariates. These models did not combine Hispanics into one group, maintaining the subgroupings of other Hispanics and Mexican Americans used by NHANES. In secondary analyses, we analyzed individual phthalate metabolites according to race/ethnicity.
To ensure that our results were not an arti fact of statistical weighting, we also repeated our analysis of race/ethnicity-stratified models in unweighted modeling. We also reprised our models substituting continuous kilocalories in lieu of categorized excessive/ appropriate caloric intake for age and sex. Finally, we repeated our analyses, substituting continuous for categorized age, and recalculated LMW, HMW, and DEHP concentrations by weight ing molar concentrations using each metabo lite's molecular weight, following published practice (Teitelbaum et al. 2012).

Results
Geometric mean concentrations of LMW phthalates metabolites were higher among girls than boys and among adolescents than children 6-11 years of age (Table 1). HMW phthalates were also significantly higher among adolescents. NonHispanic blacks had higher concentrations of all metabolites than Mexican Americans, whereas nonHispanic whites had lower concentrations of urinary LMW metabo lites and higher concentrations of HMW and DEHP metabolites. The highest half of the income distribution had lower LMW phthalates than the lower half. Children with medium and high urinary cotinine also had steadily higher LMW concentrations than children with low concentrations. Excessive caloric intake was associated with significantly lower LMW and significantly higher HMW and DEHP con centrations, whereas ≥ 2 hours/day of television watching was associated with higher LMW concentrations. Anthropometric measurements were not available in 22 participants, leaving a final sample size of 2,862 for regression analysis of biomarkeroutcome associations. Table 2 presents associations of LMW, HMW and DEHP metabolites in relationship to body mass outcomes in the entire study population. Controlling for urinary creatinine (model A), a 1unit increase in logLMW metabolites was associated with a 0.07SD unit increase in BMI zscore (95% CI: 0.02, 0.13; p = 0.012), and with a nonsignificant increase in obesity [odds ratio (OR) = 1.12; 95% CI: 0.99, 1.25; p = 0.063]. However, associations were consistent with the null after further adjustment for confounding, and nei ther HMW nor DEHP metabolites were sig nificantly associated with body mass outcomes in any regression model. LMW metabolites were significantly asso ciated with BMI zscore and overweight in children with high caregiver education and appropriate caloric intake subpopulations in models adjusted for urinary creatinine, but associations were consistent with the null in both population subgroups after adjustment for other covariates (data not shown). LMW metabolites were not associated with the out comes when stratified according to age group, sex, poverty-income ratio, television watch ing, or serum cotinine, and there were no significant associations of HMW or DEHP metabolites with body mass outcomes in any models stratified on these characteristics (data not shown). Table 3 presents full, stratified multi variable models (model C) for associations of LMW, HMW, and DEHP metabolites with body mass outcomes within each of the a All percentages are weighted using population weights for the sample in which phthalates were measured. Total number of subjects from some variables (e.g., obesity status) do not total to 2,884 due to missing data. b Derived using regression of log-molar concentration of urinary phthalate against characteristic. Geometric mean urinary phthalate represents retransformed mean from log base.
volume 121 | number 4 | April 2013 • Environmental Health Perspectives major racial/ethnic groups studied within NHANES (Hispanics as a combined group, nonHispanic whites, and nonHispanic blacks). Within Hispanic and white popula tions, no significant associations were iden tified. However, LMW metabolites were significantly associated with all three of the body mass outcomes among nonHispanic blacks, such that each log unit increase in LMW metabolites was associated with a 0.090SD unit increase in BMI zscore (95% CI: 0.003, 0.18; p = 0.043) and ORs of 1.21 (95% CI: 1.05, 1.39; p = 0.008) and 1.22 (95% CI: 1.07, 1.39; p = 0.003) for over weight and obese, respectively. Estimates from fully adjusted models that included race/ethnicity-LMW interaction terms also indicated significant differences in associations in the nonHispanic black sub group for obesity (OR = 1.14; 95% CI: 1.004, 1.29; p = 0.044) and BMI zscore (0.081SD unit increase, 95% CI: 0.01, 0.15; p = 0.027) and a near significant association with over weight (OR = 1.13; 95% CI: 0.998, 1.27; p = 0.053), consistent with findings from sepa rate models for nonHispanic blacks. Estimates for other race/ethnicity groups based on the interaction models also were consistent with stratumspecific models.
Different sample weighting produced simi lar point estimates (0.07SD unit increase in BMI zscore, OR = 1.29 for obesity, and OR = 1.13 for overweight among nonHispanic blacks), confirming the robustness of the race/ethnicitystratified multivariable results. Complete case analysis (n = 1,892) produced more modest point estimates (0.05SD unit increase in BMI zscore, OR = 1.11 for obesity, and OR = 1.05 for overweight among non Hispanic blacks). When we substituted continu ous kilocalorie intake for the categorical normal/ excessive caloric intake variable, the point esti mates were also similar (0.09SD unit increase in BMI zscore, OR = 1.16 for obesity, and OR = 1.14 for overweight among non Hispanic blacks). Substitution of molecular weight edconcentrations also failed to significantly change the point estimates (data not shown). Substitution of continuous for categorized age also did not distort the point estimates.

Discussion
We have found associations of urinary phtha late biomarkers with increased body mass among nonHispanic black children in a nationally representative sample. Translated for the average 12yearold (with weight 40 kg and height 150 cm), an increase in LMW phthalate concentration from the 25th to the 75th percentile among nonHispanic blacks (a 1.5 logunit difference) would be associated with a 0.8kg increase in body weight, a 1.5% increment in prevalence of obesity, and a 3.3% increment in prevalence of overweight.
Although our results differ from those for children and adolescents in NHANES 1999(Hatch et al. 2008, our findings are consistent with evidence from a conve nience sample of New York City children (Teitelbaum et al. 2012). No specific mecha nism has been put forth for LMW phthalates, Model A adjusts only for urinary creatinine; model B also includes sex, poverty-income ratio, parental education, serum cotinine, age, and race/ethnicity category; model C includes model B covariates plus caloric intake and television watching. All associations are with the natural log of the phthalate metabolite concentration. Results represent increase in odds of BMI z-score per unit log increase in phthalate metabolite. *p < 0.05.  All models were controlled for sex, caloric intake, television watching, poverty-income ratio, parental education, serum cotinine, urinary creatinine, and age categories. *p < 0.05. **p < 0.01. but other studies have identified similar asso ciations (Stahlhut et al. 2007;Teitelbaum et al. 2012), suggesting the need for further studies of possible mechanisms for LMW phthalates to influence body mass in early life. And although the New York City study estimated associations of urinary phthalates with body mass 1 year later, our design was crosssectional. The association we find is sub ject to concerns about reverse causality-for example, obese children may eat more pack aged foods that contain phthalates, and thus have higher urinary levels. However, in our data, HMW and DEHP metabolites were not associated with body mass outcomes, whereas LMW metabolites, which derive from applica tion of cosmetic and other personal care prod ucts, were. Our models included a variety of demographic, exposure, and lifestyle variables, thus providing more convincing evidence for nonspuriousness of the association. However, many other environmental exposures were not measured, including bisphenol A (BPA), which has been associated with childhood obesity in NHANES 2003(Trasande et al. 2012. Because children are likely to be exposed to multiple environmental chemicals, and because these chemical exposures may be correlated with each other, it is challenging to isolate associations. Simultaneous analysis of BPA with phthalates was not possible due to their collection in different subsamples of NHANES 2003NHANES -2004 Phthalates are a heterogeneous group. Using best evidence from prior practice, we aggregated biomarkers into three groups (LMW, HMW, and DEHP metabolites). Our strategy was to examine associations within the three groups and then to look for impacts within population strata, given prior reports of differences in exposure across groups. In our analysis, then, associations were subject to multiple comparisons. It is possible that associations between body mass outcomes and LMW exposure among nonHispanic blacks were driven by chance or uncontrolled bias within this population subgroup. Animal studies support an association between phtha lates and obesity, but the association is most plausible for metabolites of DEHP, rather than the LMW metabolites reported here. LMW phthalates have been linked to obesity in other human studies (Stahlhut et al. 2007;Teitelbaum et al. 2012). One explanation for the consistent LMW association across this study and the other two human studies is that MBP, an LMW metabolite with potential antiandrogenic effects, is metabolized readily to MEP (Committee on the Health Risks of Phthalates-National Research Council 2008). Moreover, we found associations with all three outcomes among nonHispanic blacks, and the association was robust to several sensitivity analyses.
It is unclear why the race/ethnicityspecific association would occur. One possi bility is that different racial/ethnic groups use phthalatecontaining shampoos and lotions differently, or use products containing differ ent mixes of LMW compounds. Supporting this hypothesis is a study in a population of Mexican women, in which the number of personal care products used was positively correlated with urinary MEP, and increases in multiple LMW phthalates were associ ated with use of multiple types of personal care products (RomeroFranco et al. 2011). Higher concentrations among nonHispanic blacks such as those identified in cross sectional NHANES surveys may represent the ranges in which LMW phthalate effects on body mass occur.
Relationships between phthalate (includ ing DEHP) intake and urinary metabolites are complex (Frederiksen et al. 2007). We know of no pharmacokinetic studies in chil dren or adolescents, and population studies are limited in their capacity to evaluate exposureexcretion relationships. In animal studies, MBP is more plausibly associated with antiandrogenic activity and body mass effects than MEP (Committee on the Health Risks of Phthalates-National Research Council 2008). Indeed, logtransformed MEP and MBP were moderately correlated (r = 0.40) in our sample. Dinbutyl phthalate (DBP; typically metabo lized to MBP) and diethyl phthalate (typically metabolized to MEP) may coexist in products, and MBP metabolism to MEP (Frederiksen et al. 2007) may explain the association of body mass with MEP identified in our study and the one from a New York City sample (Teitelbaum et al. 2012).
Other authors have derived indices of LMW, HMW, and DEHP metabolite expo sures by weighting concentrations of indi vidual metabolites according to the molecular weight of individual phthalates, under the presumptive hypothesis that molecular weight is related to potency (Teitelbaum et al. 2012;Wolff et al. 2007). We followed this approach initially, and we found nearly identical results. Absent a potencyweighted scaling that accounts for androgen antagonism, PPAR activation, and other pathways by which phthalates might affect body mass, we chose to use unweighted, aggregate molar concen trations and present alternative weighting of concentrations as a sensitivity analysis.
Phthalate exposure was measured at one time point for this analysis, and monoesters of phthalates typically have halflives of 12-48 hr (Hoppin et al. 2002), yet fat depo sition of phthalates may also lengthen the halflife and may contribute both to obesity and increments in urinary phthalate metabo lites (Mes et al. 1974). Urinary phthalates do represent current exposure better than chronic exposure. Yet one study has suggested that a single urine sample may classify exposure over the previous 3 months with higher sensitivity (63%) and specificity (87%) for MEP (an LMW metabolite), than HMW and DEHP metabolites (Hauser et al. 2004). This sug gests that a single urine sample is not poor at estimating urine MEP chronically, and MEP accounted for the vast majority of the LMW metabolites in our study population. Even if current urinary phthalates are weak indices of chronic exposure, our estimates of asso ciation should be biased toward the null for dichotomous outcomes (Carroll 1998;Fleiss and Shrout 1977;Fuller 1987).
In contrast with adult studies (Stahlhut et al. 2007), we did not observe differences according to sex, thus suggesting a mechanism that does not involve antiandrogenic effects. The linear relationship with the logarithm sug gests increasing incremental effects at lower ranges of exposure, consistent with studies of lead (Canfield et al. 2003;Lanphear et al. 2000) and methylmercury (Grandjean et al. 1997) exposures.

Conclusion
We identified a race/ethnicity-specific asso ciation of phthalates with childhood obesity in a nationally representative sample. Further study is needed to corroborate the association and evaluate genetic/epigenomic predispo sition and increased phthalate exposure as possible explanations for differences among racial/ethnic subgroups.