From October through November 2005, a group of four male and four female CDC employees in Atlanta, Georgia, provided urine samples. The group participated in a study designed to examine temporal variability in the urinary concentrations of polycyclic aromatic hydrocarbon (PAH) metabolites (Li et al. 2010
). The study participants were nonsmoking volunteers between 26 and 58 years of age who had no documented occupational PAH or BPA exposure. The institutional review board of CDC approved the study, and all participants signed an informed consent.
For a 1-week study period, the eight participants collected a total of 427 samples, including 56 first morning voids, and missed collecting 23 samples [see Supplemental Material, Table 1
(doi:10.1289/ehp.1002701)]. Each participant collected every urine void in a commercial, nonvinyl, nonpolycarbonate plastic specimen collection cup. After recording the volume of each void, the urine was decanted in a prelabeled, sterile urine cup and stored in an ice cooler. The urine samples were retrieved from participants daily (or after the weekend), aliquoted into polypropylene cryovials or glass jars, and stored at –70°C until analysis. During the study week, participants also were asked to record food and drink intake, medications consumed (if any), driving (all participants commuted by car 10–50 miles during the work week), and other activities.
Analytical method for measuring BPA.
The total BPA concentration was measured using a mass spectrometry method described previously (Ye et al. 2005a
). The limit of detection (LOD) was 0.4 µg/L. To ensure data accuracy and precision, each batch of samples included quality control (QC) samples, standards, and reagent blanks. The QC concentrations were evaluated using standard statistical probability rules (Caudill et al. 2008
). Urinary creatinine, used to correct the dilution of the urine, was measured at CDC using a Roche Hitachi 912 Chemistry Analyzer (Hitachi, Pleasanton, CA).
We performed the statistical analyses using SAS software (SAS Institute Inc., Cary, NC). For concentrations below the LOD, we used a value equal to the LOD divided by the square root of 2 (Hornung and Reed 1990
). The urinary BPA concentration followed a log-normal distribution. Thus, before statistical analysis, all data were log10
The first morning void was defined as the first sample collected by each person at or after 5:00 a.m. each day. The simulated 24-hr void concentration was calculated based on the volume-weighted average of all urine samples collected (missed collections were not accounted for) during a 24-hr period starting after midnight. To assess the impact of creatinine adjustment on the total variance when exposure is categorized from the BPA concentrations of spot urine samples, we built three different models. For model A, BPA was not creatinine-corrected (log10
noncorrected concentration in micrograms per liter), whereas for model B, BPA was creatinine-corrected to account for urinary dilution (log10
creatinine-corrected concentration in micrograms per gram creatinine). Model C adjusted for urine dilution by including creatinine as a model covariate (log10
-adjusted concentration in micrograms per liter). We ranked these models based on their Akaike information criterion (AIC; Akaike 1974
) (the lower the AIC, the better the model). To assess the temporal variability in BPA concentrations of spot samples, first morning voids, and simulated 24-hr voids, we calculated intraclass correlation coefficients (ICCs) using a three-level model. Level 1 is the time (i
), which is irregular, unequal, and interval nested within the day (level 2, j
= 7), which is nested within the participants (level 3, k
= 8). The equation for models A and B was Yijk
) + (V00k
), where Yijk
(the dependent variable) is the log10
(BPA) (for model A) or log10
(creatinine-corrected BPA) (for model B) for participant k
on day j
at time i
. The equation for model C includes creatinine as an additional independent variable, with log10
(BPA) as the dependent variable. The intercept Y000
is the grand mean (i.e., the average value across all observations), and V00k
, and γijk
are the random errors for level 3, level 2, and level 1 residual, respectively (Singer 1998
). The ICC indicates the temporal reproducibility of repeated measures and is computed by dividing the estimate of the between-subject variance by the estimated total variance. ICC ranges from 0 (poor reproducibility) to 1 (perfect reproducibility).
For the spot urine samples, we also compared the variance apportionment of BPA and creatinine concentrations by constructing a model in which creatinine concentration was the outcome. We also checked the effect of the missed collections on the variation pattern of urinary BPA in spot samples by comparing the variance apportionment of urinary BPA concentrations in spot samples with and without participant 2, who had the largest number of missed collections [see Supplemental Material, Table 1