NHANES, conducted annually since 1999 by the Centers for Disease Control and Prevention (CDC), is an ongoing survey designed to measure the health and nutritional status of the civilian noninstitutionalized U.S. population (CDC 2003
). The surveys include household interviews, standardized physical examinations, and collection of medical histories and biologic specimens. Some of these specimens are used to assess exposure to environmental chemicals (CDC 2005
). NHANES 2003–2004 included examinations of 9,643 people (CDC 2006a
). Urine specimens for analyses of triclosan were collected from a random one-third subset of 2,517 participants ≥ 6 years of age. Because the subset was random, the representative design of the survey was maintained. The National Centers for Health Statistics Institutional Review Board reviewed and approved the study protocol. Informed written consent was obtained from all participants.
A single spot urine sample per participant was collected during one of three daily examination session periods (i.e., morning, afternoon, evening). The samples were shipped on dry ice to CDC’s National Center for Environmental Health and stored at or below −20°C until analyzed. We measured the concentrations of free plus conjugated triclosan, in 100 μL of urine, by online solid-phase extraction coupled to high-performance liquid chromatography–isotope dilution–tandem mass spectrometry, as described in detail elsewhere (Ye et al. 2005
). Briefly, the conjugated species of triclosan were hydrolyzed by use of 50 μL of a solution (4,000 μg/mL) of β-glucuronidase/sulfatase (Helix pomatia
, 463,000U/g solid; Sigma Chemical Co., St. Louis, MO) in 1 M ammonium acetate pH 5 buffer (Sigma Chemical Co.). After hydrolysis, samples were acidified with 0.1M formic acid; triclosan was preconcentrated by online solid-phase extraction, separated from other urine components by reversed-phase high-performance liquid chromatography, and detected by atmospheric pressure chemical ionization–tandem mass spectrometry. The limit of detection (LOD)—calculated as 3 S0
, where S0
is the standard deviation as the concentration approaches zero (Taylor 1987
)—was 2.3 μg/L; the precision ranged from 14.3% to 23.2%. To minimize potential contamination with triclosan during the laboratory operations, we avoided the use of triclosan-containing soaps and detergents. In addition, low-concentration (~ 40 μg/L) and high-concentration (~ 230 μg/L) quality control materials, prepared with pooled human urine, and reagent blanks (to monitor and control for the potential contamination arising from the reagents and apparatus used) were analyzed with analytical standards and NHANES samples.
We performed statistical analyses using SAS (version 9.1.3; SAS Institute Inc., Cary, NC) and SUDAAN (version 9.0.1; RTI International, Research Triangle Park, NC). SUDAAN incorporates sample weights and design variables to account for the complex sample design of NHANES. We calculated the percentage of detection, the geometric mean, and distribution percentiles for both the volume-based (in micrograms per liter) and creatinine-corrected (in micrograms per gram creatinine) concentrations. For concentrations below the LOD, as recommended for the analysis of NHANES data (CDC 2006b
), we used a value equal to the LOD divided by the square root of 2 (Hornung and Reed 1990
We used analysis of covariance to examine the influence of several variables, selected on the basis of statistical, demographic, and biologic considerations, on the concentrations of triclosan. For the multiple regression models, we used the variables described below and all possible two-way interactions to calculate the adjusted least square geometric mean (LSGM) concentrations (in micrograms per liter), which provide geometric mean estimates for a variable after adjustment for the model covariates. Because the distribution of the triclosan concentrations was skewed, triclosan concentrations were log-transformed. A variable based on self-reported data defined three major racial/ethnic groups: non-Hispanic black, non-Hispanic white, and Mexican American. Self-reported annual household income was available in $5,000 increments (ranging from < $5,000 to > $75,000). To obtain comparable numbers of participants in each income group, we categorized income as < $20,000, $20,000–$45,000, and > $45,000. Those participants who had serum cotinine concentrations (the biomarker used to define smoking status) > 10 μg/L were classified as smokers. Creatinine concentrations were log-transformed for the data analysis because of their skewed distribution. Age was reported in years at the previous birthday. Because body mass index (BMI) is age- and sex-specific for people < 19 years of age, CDC recommends for children and teens the use of BMI-for-age percentile (BMIPCT) instead of BMI (Kuczmarski et al. 2002
). Therefore, we conducted two separate models: one for adults (≥ 20 years of age) and one for children and teenagers (6–19 years of age). We could not include only children (6–11 years of age) in the model because of the small sample size for some strata. We considered age (continuous), sex, race/ethnicity, creatinine concentration (Barr et al. 2005
), and income for both models. Additionally, for the adult model, we included smoking status and BMI, and for the children and teens model, we included BMIPCT. When both age and age-squared were in the model, to avoid multicollinearity we centered age by subtracting the mean age from each participant’s age (Bradley and Srivastava 1979
). To evaluate the relation between the log-transformed concentration of triclosan and age, we estimated the weighted geometric mean and LSGM concentrations after adjusting by the other covariates in the model, and we generated a bar chart of triclosan concentrations by age group.
To reach the final model, we used backward elimination, with a threshold of p < 0.05 for retaining the variable in the model, using Satterwaite-adjusted F statistics. We evaluated for potential confounding by adding each of the excluded variables back into the final model one by one and examining changes in the β coefficients of the statistically significant main effects. If addition of one of these excluded variables caused a change in a β coefficient by ≥ 10%, the variable was re-added to the model.