The present analysis included measurements from 2 years of NHANES data, 2007–2008. NHANES is an ongoing cross-sectional study designed to collect nationally representative data on dietary intake and disease. Methods for demographic and survey data collection are described in detail elsewhere [National Center for Health Statistics (NCHS) 2010b].
Urinary phthalate metabolites and BPA.
Urine samples collected at a mobile examination center collection were stored at 4°C or frozen at –20°C and then shipped to the Division of Environmental Health Laboratory Sciences, CDC, for analysis. Measurement of urinary phthalate metabolite concentrations was conducted using online solid-phase extraction (SPE), isotope dilution, and high-performance liquid chromatography (HPLC) separation, followed by electrospray ionization and tandem mass spectrometry (MS/MS), as described in detail elsewhere (NCHS 2010d; Silva et al. 2004
). Urinary BPA concentrations were measured using online SPE coupled to isotope dilution, HPLC, and atmospheric pressure chemical ionization–MS/MS (NCHS 2010c; Ye et al. 2005
). Quality control procedures for all analytes followed those described by Westgard et al. (1981)
, and values below the limit of detection (LOD) were replaced with a value of the LOD divided by the square root of 2 (Hornung and Reed 1990
In addition to urinary BPA, we focused our analyses on primary and secondary metabolites of DEHP [MEHP and three oxidized DEHP metabolites: mono(2-ethyl-5-hydroxyhexyl) phthalate (MEHHP), mono(2-ethyl-5-oxohexyl) phthalate (MEOHP), and mono(2-ethyl-5-carboxypentyl) phthalate (MECPP)] and DBP [mono-n-butyl phthalate (MnBP), mono-isobutyl phthalate (MiBP), and mono(3-carboxypropyl) phthalate (MCPP), an oxidized metabolite of both DBP and di-n-octyl phthalate (DOP)], because previous studies have suggested that exposure to these phthalates or their metabolites may be associated with altered thyroid hormones.
Because all urinary biomarker concentrations were right-skewed, the data were transformed using the natural logarithm (ln) prior to analysis. In addition, metabolite measurements were creatinine standardized for presentation of descriptive statistics and calculation of simple correlations by dividing metabolite concentrations by urinary creatinine (micrograms per gram creatinine). For regression analysis, we used unadjusted metabolite levels; urinary creatinine was included as a covariate in all models (Barr et al. 2005
Serum thyroid measures. The NHANES thyroid panel, including measures of free and total T3 and T4, TSH, and thyroglobulin, was measured in serum collected at the same time as urine samples and analyzed using various immunoenzymatic assays as described elsewhere (NCHS 2009). Because of limited serum available for thyroid hormone analysis, this aspect of the study excluded all children under 12 years of age. All distributions except total T3 and total T4 were right-skewed and ln-transformed for analysis.
Covariates. Demographic data were collected in an in-home survey component of NHANES. From these data, we examined age, sex, race and ethnicity, and education level as potential confounding variables. From examination and laboratory data we considered body mass index (BMI), serum cotinine (log-transformed) as a measure of exposure to tobacco smoke, and urinary iodine (log-transformed). Variables were evaluated for inclusion by examining bivariate relationships with urinary biomarkers and serum thyroid measures and by their significance in full models. Variables that were significantly associated with one or more urinary biomarkers or serum thyroid measures or that were statistically significant (α = 0.05) in more than one multivariable model were included in the final models. All models were adjusted for the same covariates for consistency.
Statistical analysis. There were 2,035 subjects ≥ 12 years of age available who had data for one or more of the urinary phthalate metabolites, urinary BPA, urinary creatinine, and one or more of the thyroid measures. We excluded from our analysis 164 subjects with a reported history of thyroid disease, 20 women who were pregnant, and 88 subjects with data missing for age. We also excluded 3 subjects with outlying/influential values, which included 2 subjects with outlying high levels of free and total T3 and T4 and low levels of TSH, as well as 1 subject with outlying high levels of total T4. This resulted in 1,760 total subjects available for analysis. An additional 85 participants were missing data on covariates (14 missing BMI, 3 missing serum cotinine, and 68 missing urinary iodine) and were not included in the multivariable models, leaving 1,675 participants (1,346 adults and 329 adolescents 12–19 years of age).
Data analysis was performed using SAS 9.2 (SAS Institute Inc., Cary, NC). NHANES data are collected using a complex, multistage study design. For our analysis, we used 2-year weights for individual probabilities drawn from the urinary biomarker data sets according to the NCHS web tutorial (NCHS 2010a) to account for the sampling method. In addition, stratum and cluster weights were included in regression models to correct for the study design. For comparison, we also constructed models that did not include the sample weights, because the weighted method may result in an inefficient analysis due to the large variability in assigned weights, and because adjustments for variables used in the creation of weights (such as age, sex, and ethnicity) are already included in the full models (Korn and Graubard 1991
In descriptive analyses, we explored differences in urinary biomarker or serum thyroid measures between categories using Wilcoxon rank-sum or Kruskal-Wallis tests. Spearman rank correlations were calculated to assess relationships between continuous variables. Because of significant differences in thyroid hormone levels in adolescents compared with adults in these data, we separated our data set into two groups (ages 12–19 years and > 20 years) in subsequent analyses.
We then constructed full multivariable linear regression models with serum thyroid measures as the dependent variable and individual ln-transformed urinary phthalate metabolite or BPA concentration as a predictor along with age (continuous variable), sex (dichotomous), race and ethnicity (categorical), ln-transformed serum cotinine (continuous), BMI (continuous), ln-transformed iodine (continuous), and ln-transformed urinary creatinine (continuous) as covariates. Analyses were performed both with and without including the sample weights to examine the effects of weighting. Secondary analyses were conducted by repeating all models when stratifying on sex. Regression results are presented as the change in thyroid measure associated with a unit increase in ln-transformed phthalate metabolite concentration. To improve interpretability, we also provide several examples in the “Results” where we back-transformed the regression coefficient to represent a change in hormone measure associated with an interquartile range (IQR) increase in urinary phthalate metabolite concentration. Finally, for urinary exposure biomarkers detected in > 80% of samples, we explored evidence for nonlinear relationships by assessing relationships between urinary phthalate metabolite or BPA quintiles and serum thyroid measures. Tests for trend were conducted for ordinal urinary biomarker quintiles in regression models using integer values (0–4).