The urine samples analyzed for this study were selected from the Third National Health and Nutrition Examination Survey (NHANES III) callback cohort, a nonrepresentative subset of NHANES III composed of approximately 1,000 adults. The urine samples were all spot-urine samples, collected at different times throughout the day and were not necessarily first-morning voids. Creatinine adjustment was used to correct for urine dilution (Jackson 1966 (link)).
BPA and 4-n-nonylphenol (nNP), the linear chain NP isomer, were measured using a method based on an automated solid-phase extraction (SPE) coupled to isotope dilution-GC/MS (Kuklenyik et al. 2003 (link)). First, the urine samples were treated with β-glucuronidase to hydrolyze the glucuronide conjugates. Then, during the automated SPE process, BPA and nNP were both extracted from the deconjugated urine matrix and derivatized, using pentafluorobenzyl bromide, on commercial styrene-divinylbenzene copolymer-based SPE cartridges. After elution from the SPE column, the derivatized phenols in the SPE eluate were measured by isotope-dilution GC/MS. The limits of detection (LODs) for BPA and nNP in a 1-mL urine sample were 0.1 μg/L.
Quality control (QC) materials were analyzed along with the samples to assure the accuracy and reliability of the data. Low-concentration (QCL, 2–5 ng/mL) and high-concentration (QCH, 12–20 ng/mL) QC materials were prepared from a base urine pool—obtained from multiple anonymous donors as described previously (Kuklenyik et al. 2003 (link))—dispensed in 5-mL aliquots and stored at −20°C. Each QC material was characterized by repeated measurements, spanned over at least 4 weeks, to define the mean concentrations and the 95% and 99% control limits of BPA and nNP. Each analytical run consisted of 40 (2 QCH, 2 QCL, 4 blanks, and 32 unknown) samples. The concentrations of the two QCH and the two QCL, averaged to obtain one measurement of QCH and QCL for each run, were evaluated using standard statistical probability rules.
The samples used for this study were stored securely at −70°C and may have been subject to repeated thaw/freeze cycles. Before analysis, the samples and QC materials were left to thaw overnight at 5°C. The concentrations of the analytes in the QCs remained essentially constant under these experimental conditions. Furthermore, QC materials reanalyzed after the initial characterization showed that BPA and nNP remained stable in the QC materials at −20°C for at least 1 year. Although the long-term stability of the analytes in the urine samples stored for > 1 year is not known, the QC data suggest that the integrity of the specimens is likely maintained and that chemical degradation of the phenols was undetectable.
To estimate total sample size, we used a standard formula n = t2p(1 − p)/d2, where n is the estimated sample size, t is the critical value associated with the desired statistical confidence level, and d is the maximum allowable error above or below the estimate of the true proportion (p) of the target population with measurable levels of the analyte(s) of interest (Peavy 1996 ). Using a confidence level of 99% (t = 2.6), d = 0.065, and a 50% percentage of the population with measurable BPA and nNP levels (p = 0.5), the estimated total sample size was 400. Participants in this study were 20–59 years of age, of both sexes, and urban and rural residents. An arbitrary cutoff of 100,000 inhabitants per county was used to distinguish rural from urban areas. Each sample, defined by age (< 50 years or ≥50 years), residence (rural or urban), and sex (male or female), was categorized in eight subpopulation groups (e.g., < 50-year-old rural female).
Because samples were obtained from the NHANES III callback cohort, a nonrepresentative subset of NHANES III samples, the summary statistics are not representative of the U.S. population but serve as reference ranges for the three population breakdowns specified above (i.e., persons < 50 or ≥50 years of age; rural or urban residents; male or female). To improve the extent to which the results represent the U.S. population, we used sample weights. We developed our own weights for demographic groups, not for individual subjects. This approach is different from that used by the National Center for Health Statistics (NCHS) of the Centers for Disease Control and Prevention (CDC). The NCHS assigns a unique weight to each subject based on demographics, geographical data, and oversampling of certain population groups. Because we only had information on age group, sex, and residence (i.e., rural and urban), we could not assign weights to individual subjects, only to the demographic groups. We determined the weights by relating the sample sizes in each of the eight groups to the total numbers of persons in the U.S. population in the same groups defined by sex, residence, and age. From within these eight groups, we randomly selected 394 samples. The institutional review board of the NCHS approved the study.
We analyzed the weighted data using SAS software, version 8.2 (SAS Institute, Cary, NC). Because the base-10 logarithm of the concentrations (log-transformed concentrations) was less skewed than the nontransformed values, we used the log-transformed values in the analyses. We calculated GMs and distribution percentiles for both volume-based (micrograms per liter) and creatinine-corrected concentrations (micrograms per gram creatinine). The GMs were exponentiated results obtained from the means of the log-transformed concentrations. GMs were calculated when the frequency of detection of the analyte was > 60%. We did not use weights to obtain GM or percentile estimates for the various demographic groups because each subject in a demographic group had the same weight.
For exploratory purposes only, we compared BPA and NP levels among subgroups (by age, sex, and place of residence) even though we did not design the study to assure adequate statistical power for this type of hypothesis testing (i.e., the sample size was determined to answer only the question about the percentage of population with measurable urinary BPA and/or NP levels). We used weighted analysis of covariance models to study the effects of residence, sex, age group, and urinary creatinine on the urinary log-transformed concentrations of BPA and NP. The analyses were performed using SAS Proc GENMOD (SAS Institute) to model the log-transformed concentrations (dependent variable) as a function of sex, residence, age group (categorical covariates), and urinary creatinine (continuous covariate used to adjust for urine dilution). The purpose of our model adjustment was not to apply an individual adjustment to BPA and NP concentrations, but rather to enable us to determine whether there are differences in average BPA or NP urinary levels between individuals in the same demographic groups (e.g., men vs. women) after accounting for the differences due to urinary dilution. By adjusting for creatinine, we obtained a comparison that was not influenced by differences in creatinine levels. We also considered all possible two-way interactions between covariates. Type 3 equivalent sums of squares from the model were used to form likelihood ratio tests of model effects and various tests of hypotheses. Statistical significance was set at p < 0.05. We dealt with results < LOD by using a multiple imputation method (Lynn 2001 ) along with the SAS procedure PROC MIANALYZE, which summarizes parameter estimates and incorporates the resulting uncertainty associated with the multiple imputations used to obtain them.
BPA and 4-n-nonylphenol (nNP), the linear chain NP isomer, were measured using a method based on an automated solid-phase extraction (SPE) coupled to isotope dilution-GC/MS (Kuklenyik et al. 2003 (link)). First, the urine samples were treated with β-glucuronidase to hydrolyze the glucuronide conjugates. Then, during the automated SPE process, BPA and nNP were both extracted from the deconjugated urine matrix and derivatized, using pentafluorobenzyl bromide, on commercial styrene-divinylbenzene copolymer-based SPE cartridges. After elution from the SPE column, the derivatized phenols in the SPE eluate were measured by isotope-dilution GC/MS. The limits of detection (LODs) for BPA and nNP in a 1-mL urine sample were 0.1 μg/L.
Quality control (QC) materials were analyzed along with the samples to assure the accuracy and reliability of the data. Low-concentration (QCL, 2–5 ng/mL) and high-concentration (QCH, 12–20 ng/mL) QC materials were prepared from a base urine pool—obtained from multiple anonymous donors as described previously (Kuklenyik et al. 2003 (link))—dispensed in 5-mL aliquots and stored at −20°C. Each QC material was characterized by repeated measurements, spanned over at least 4 weeks, to define the mean concentrations and the 95% and 99% control limits of BPA and nNP. Each analytical run consisted of 40 (2 QCH, 2 QCL, 4 blanks, and 32 unknown) samples. The concentrations of the two QCH and the two QCL, averaged to obtain one measurement of QCH and QCL for each run, were evaluated using standard statistical probability rules.
The samples used for this study were stored securely at −70°C and may have been subject to repeated thaw/freeze cycles. Before analysis, the samples and QC materials were left to thaw overnight at 5°C. The concentrations of the analytes in the QCs remained essentially constant under these experimental conditions. Furthermore, QC materials reanalyzed after the initial characterization showed that BPA and nNP remained stable in the QC materials at −20°C for at least 1 year. Although the long-term stability of the analytes in the urine samples stored for > 1 year is not known, the QC data suggest that the integrity of the specimens is likely maintained and that chemical degradation of the phenols was undetectable.
To estimate total sample size, we used a standard formula n = t2p(1 − p)/d2, where n is the estimated sample size, t is the critical value associated with the desired statistical confidence level, and d is the maximum allowable error above or below the estimate of the true proportion (p) of the target population with measurable levels of the analyte(s) of interest (Peavy 1996 ). Using a confidence level of 99% (t = 2.6), d = 0.065, and a 50% percentage of the population with measurable BPA and nNP levels (p = 0.5), the estimated total sample size was 400. Participants in this study were 20–59 years of age, of both sexes, and urban and rural residents. An arbitrary cutoff of 100,000 inhabitants per county was used to distinguish rural from urban areas. Each sample, defined by age (< 50 years or ≥50 years), residence (rural or urban), and sex (male or female), was categorized in eight subpopulation groups (e.g., < 50-year-old rural female).
Because samples were obtained from the NHANES III callback cohort, a nonrepresentative subset of NHANES III samples, the summary statistics are not representative of the U.S. population but serve as reference ranges for the three population breakdowns specified above (i.e., persons < 50 or ≥50 years of age; rural or urban residents; male or female). To improve the extent to which the results represent the U.S. population, we used sample weights. We developed our own weights for demographic groups, not for individual subjects. This approach is different from that used by the National Center for Health Statistics (NCHS) of the Centers for Disease Control and Prevention (CDC). The NCHS assigns a unique weight to each subject based on demographics, geographical data, and oversampling of certain population groups. Because we only had information on age group, sex, and residence (i.e., rural and urban), we could not assign weights to individual subjects, only to the demographic groups. We determined the weights by relating the sample sizes in each of the eight groups to the total numbers of persons in the U.S. population in the same groups defined by sex, residence, and age. From within these eight groups, we randomly selected 394 samples. The institutional review board of the NCHS approved the study.
We analyzed the weighted data using SAS software, version 8.2 (SAS Institute, Cary, NC). Because the base-10 logarithm of the concentrations (log-transformed concentrations) was less skewed than the nontransformed values, we used the log-transformed values in the analyses. We calculated GMs and distribution percentiles for both volume-based (micrograms per liter) and creatinine-corrected concentrations (micrograms per gram creatinine). The GMs were exponentiated results obtained from the means of the log-transformed concentrations. GMs were calculated when the frequency of detection of the analyte was > 60%. We did not use weights to obtain GM or percentile estimates for the various demographic groups because each subject in a demographic group had the same weight.
For exploratory purposes only, we compared BPA and NP levels among subgroups (by age, sex, and place of residence) even though we did not design the study to assure adequate statistical power for this type of hypothesis testing (i.e., the sample size was determined to answer only the question about the percentage of population with measurable urinary BPA and/or NP levels). We used weighted analysis of covariance models to study the effects of residence, sex, age group, and urinary creatinine on the urinary log-transformed concentrations of BPA and NP. The analyses were performed using SAS Proc GENMOD (SAS Institute) to model the log-transformed concentrations (dependent variable) as a function of sex, residence, age group (categorical covariates), and urinary creatinine (continuous covariate used to adjust for urine dilution). The purpose of our model adjustment was not to apply an individual adjustment to BPA and NP concentrations, but rather to enable us to determine whether there are differences in average BPA or NP urinary levels between individuals in the same demographic groups (e.g., men vs. women) after accounting for the differences due to urinary dilution. By adjusting for creatinine, we obtained a comparison that was not influenced by differences in creatinine levels. We also considered all possible two-way interactions between covariates. Type 3 equivalent sums of squares from the model were used to form likelihood ratio tests of model effects and various tests of hypotheses. Statistical significance was set at p < 0.05. We dealt with results < LOD by using a multiple imputation method (Lynn 2001 ) along with the SAS procedure PROC MIANALYZE, which summarizes parameter estimates and incorporates the resulting uncertainty associated with the multiple imputations used to obtain them.
Full text: Click here