. NHANES is a cross-sectional, multistage, stratified, cluster sampling survey conducted by the Centers for Disease Control and Prevention’s (CDC) National Center for Health Statistics. NHANES uses a complex sampling design that includes questionnaires about demographics and health-related behaviors, as well as laboratory and clinical measurements (CDC 2011
). We limited our analysis to participants between 12 and 16 years of age to capture exposures close to the age of menarche among those who completed reproductive heath questionnaires (administered beginning at age 12). Reproductive health questionnaires were administered through a proxy of parent or guardian to females 12 years of age, the lower age limit in our analysis, to 16 years. All procedures were approved by the NCHS Institutional Review Board, and all participants provided written informed consent. During 2003–2004, NHANES sampled urinary phenols and phthalates in two separate one-third subsets, and a single one-third subset was analyzed for phenols, phthalates, and parabens from 2005 through 2008. In each case, subsets were a representative sample of NHANES participants ≥ 6 years of age from each 2-year study cycle. Therefore, data presented for phenols and phthalates are from years 2003–2008, whereas data from parabens are from years 2005–2008 only. Laboratory samples were collected on the same day the questionnaire was administered.
Measurement of urinary phenols, parabens, and phthalates.
Phenols and parabens. Environmental phenols considered for inclusion were bisphenol A (BPA), 4-tert
-octyl phenol, 2,4,4´-trichloro-2´-hydroxyphenyl ether (triclosan), 2-hydroxy-4-metoxybenzophenone (benzophenone-3). Parabens considered for inclusion were propyl-, butyl-, ethyl-, and methyl-paraben, 2,4-dichlorophenol (2,4-DCP), o
-phenyl phenol, 2,5-dichlorophenol (2,5-DCP), 2,4,5-trichlorophenol, and 2,4,6-trichlorophenol (CDC 2011
). These substances were quantified in urine by use of solid phase extraction (SPE) coupled on-line to high performance liquid chromatography–tandem mass spectrometry (HPLC-MS/MS) (Ye et al. 2005
). In addition we evaluated the phenol metabolites 2,4-DCP, o
-phenyl phenol, 2,5-DCP, 2,4,5-trichlorophenol, and 2,4,6-trichlorophenol, which were measured in urine by use of SPE-HPLC followed by atmospheric pressure chemical ionization–high-performance liquid chromatography–isotope dilution tandem mass spectrometry (APCI-MS/MS) (Ye et al. 2005
Phthalate metabolites. We used HPLC-MS/MS with electrospray ionization to quantify the following phthalate metabolites in urine: monomethyl phthalate, monoethyl phthalate, monobutyl phthalate, mono-isobutyl phthalate, mono(3-carboxypropyl) phthalate, monocyclohexyl phthalate, mono(2-ethylhexyl) phthalate, monooctyl phthalate, monobenzyl phthalate, monoisononyl phthalate, mono(2-ethyl-5-oxohexyl) phthalate, mono(2-ethyl-5-hydroxyhexyl) phthalate, mono(2-ethyl-5-carboxypentyl) phthalate, monocarboxyoctyl phthalate, and monocarboxynonyl phthalate (CDC 2011
).Before analysis by HPLC-MS/MS, the urine samples were processed through use of enzymatic deconjugation of the glucuronidated phthalate monoesters, followed by on-line SPE (Kato et al. 2005
; Silva et al. 2007
We included female 2003–2008 NHANES study participants 12–16 years of age who had completed the reproductive health questionnaire and physical examination, and for whom data regarding age of menarche, defined by NHANES as age of first menstruation, were available. Of the 1,598 individuals 12–16 years of age who had completed the reproductive health questionnaire, 1,420 participants had complete data on body mass index (BMI) and age of menarche (age or not yet reached). Of these, 461 were included in NHANES subsamples with urinary phenol and phthalate measurements. We used the one-third subsample weighting variables for each 2-year subset of data. Nonmissing values for urine concentrations below the limit of detection (LOD) were replaced with the value of the LOD divided by the square root of 2. In our analysis, all urinary compounds and metabolites were creatinine-corrected by dividing urine concentrations by creatinine concentrations to give micrograms per gram of creatinine as the final units. Urine samples with creatinine levels > 300 mg/dL or < 30 mg/dL were excluded because they were too dilute or too concentrated for accurate analysis (n
= 11) (Sata et al. 1995
We used SAS 9.2 for data analysis, and calculated means and percentiles of the EDCs and demographic factors by use of the PROC SURVEYMEANS (weighted geometric means) procedure to account for the complex sampling design of NHANES. We calculated the weighted mean self-reported age of menarche using PROC LIFETEST (Kaplan–Meier censored survival estimates) to account for censoring at the age of participation among 43 individuals who had not reached menarche at the time of participation (all programs from SAS Institute Inc., Cary, NC). We used the Taylor series (linearization) method to estimate standard errors and confidence intervals. We calculated BMI percentile for age in months using the standardized CDC growth charts (Grummer-Strawn et al. 2010
). We set significance at α = 0.05 for two-sided p
To estimate associations between EDCs and age of menarche, we used the PROC SURVEYPHREG (Cox-proportional hazards model with censored data) procedure and the efron method of ties handling, after confirming the proportional hazards assumption. We modeled the natural log of creatinine-corrected urine analyte concentrations to normalize distributions in our data and, in accordance with NHANES analytic guidelines, we excluded 10 observations with outlier values > 3 SDs for deviance residuals of the log-transformed weighted values, as determined by a probability plot of residuals (CDC 2011
). These adjustments left a sample size of 440 for our phenol analysis, 437 for phthalate analysis, and 287 individuals for parabens analysis.
We assessed the urinary concentration of 18 EDCs found above the LOD in at least 75% of study participants as exposures in our model, out of a total of 27 phthalate, phenol, and pesticide compounds or metabolites evaluated in urine in NHANES. These compounds were then analyzed either as single compounds (benzophenone-3, triclosan, BPA, 2,4-DCP, and 2,5-DCP) or as the sum of urinary analyte concentrations within a class of compounds (parabens, phthalates, and environmental phenols), again including only those compounds found above the LOD in at least 75% of study participants. We converted compounds to molar weights for summing to adjust for the different molecular weights of compounds and used micrograms per gram of creatinine as the unit in our models. We identified potential confounders from the literature (mother’s age at birth of girl, mother’s smoking status during pregnancy, birth weight < 5.5 pounds (yes or no), birth weight > 9.0 pounds (yes or no), family income (1–12 categories based on NHANES categories of $0–4,999 to ≥ $75,000), family income-to-poverty ratio (0–5), self-defined race/ethnicity, BMI (continuous) as calculated by measurements taken during the NHANES examination, and BMI percentile for sex and age in months (0–100) as calculated using CDC standardized growth charts) (Grummer-Strawn et al. 2010
) and included them in our final models if they either predicted the outcome with p
< 0.05 or there was a > 10% change in the hazard ratio for the exposure–menarche association when they were removed from the model. Insufficient observations existed for birth weight to include in model building. Backward and forward selection resulted in the same final model.
We evaluated BMI and race/ethnicity as effect modifiers by evaluating stratum-specific hazard ratios because differences in age of menarche among ethnicities are well established (Himes et al. 2009
). Interactions between race/ethnicity and urinary phenol concentrations were evaluated using interaction terms in the model, with p
< 0.10 for the interaction term used to evaluate significance of the interaction.