Descriptive statistics and distributions of urinary BPA concentrations were tabulated and compared among demographic categories. We constructed graphs to visually and qualitatively compare BPA concentrations within and between subjects over time. Using SAS software (version 9.1; SAS Institute Inc., Cary, NC), mixed effects models were fit to determine the association of urinary BPA concentrations (log10) with age, BMI, sex, and pregnancy status. Each model included the predictor of interest and SG as fixed effects. To account for possible correlation of measurements, random effects were included initially for subject, couple, and within-couple specimen collection date. The variance component for couple was estimated to be zero and was dropped from the models, but the random effects for subject and within-couple collection date were retained. Full covariate data were available for all specimens, so information from all samples was used in the analyses.
To explore the nature of within-couple correlation, we regressed the geometric mean of SG-adjusted BPA measurements from the female partner on that from the male partner. Another analysis compared BPA measurements (SG-adjusted and log10-transformed) of the female to those of the male when the specimens were collected on the same date. In this model, a random effect was included to account for a possible correlation due to repeated measurements from the same couple.
We calculated the sensitivity, specificity, and positive predictive value of a single urine sample for predicting high BPA tertile by comparing predicted and observed classifications for agreement (Hauser et al. 2004 (link); Meeker et al. 2005 (link)). Positive predictive value is the probability that a person is actually in the exposure group of interest, given that they were classified in that group based on the single urinary BPA value that was available. BPA tertiles were determined for the SG-adjusted BPA concentrations among all 82 subjects. For those with more than one sample, we used the subject’s geometric mean value in the tertile classification. A contingency table was then constructed to display the level of agreement between each subject’s “true” tertile classification as determined by the geometric mean value of their repeated samples and their tertile classification predicted by each of their single repeat samples. Only subjects with three or more urine samples (31 subjects with a total of 149 urine samples) were included in the contingency tables. After combining contingency tables for all subjects with three or more samples, we calculated sensitivity, specificity, and positive predictive value for the ability of a given single urine sample to classify a subject in the highest BPA tertile.
We then calculated the sensitivity, specificity, and positive predictive value of two urine samples from an individual to classify subjects in the highest BPA tertile. This analysis was limited to subjects with at least six repeat urine samples (n = 8 subjects who contributed a total of 67 samples). Geometric means of all possible combinations of sample pairings from each individual were used in the analysis as the predicted value. The geometric mean value from all possible within-subject combinations was then compared with the geometric mean value of their repeated samples, used to determine their “true” BPA tertile classification. As above, contingency tables were then constructed. The goal of the validity analysis was to simulate and compare the ability of exposure assessments that involve one or two urine samples to predict a subject’s “true” longer-term exposure.
In both of these analyses, the single or two urine samples used to predict a subject’s tertile classification are not independent predictors of their overall geometric mean value because they are also used to calculate the subject’s geometric mean. For instance, when we calculated the geometric mean for subjects with three or more urine samples, the single urine sample evaluated for its predictive ability was included in the geometric mean calculation for that subject. Because of this, we limited these analyses to subjects with three (or six) or more urine samples to minimize the structural dependence between the predicted value based on the single sample and the “true” observed values based on the geometric mean of that subject’s full complement of samples.
To explore the nature of within-couple correlation, we regressed the geometric mean of SG-adjusted BPA measurements from the female partner on that from the male partner. Another analysis compared BPA measurements (SG-adjusted and log10-transformed) of the female to those of the male when the specimens were collected on the same date. In this model, a random effect was included to account for a possible correlation due to repeated measurements from the same couple.
We calculated the sensitivity, specificity, and positive predictive value of a single urine sample for predicting high BPA tertile by comparing predicted and observed classifications for agreement (Hauser et al. 2004 (link); Meeker et al. 2005 (link)). Positive predictive value is the probability that a person is actually in the exposure group of interest, given that they were classified in that group based on the single urinary BPA value that was available. BPA tertiles were determined for the SG-adjusted BPA concentrations among all 82 subjects. For those with more than one sample, we used the subject’s geometric mean value in the tertile classification. A contingency table was then constructed to display the level of agreement between each subject’s “true” tertile classification as determined by the geometric mean value of their repeated samples and their tertile classification predicted by each of their single repeat samples. Only subjects with three or more urine samples (31 subjects with a total of 149 urine samples) were included in the contingency tables. After combining contingency tables for all subjects with three or more samples, we calculated sensitivity, specificity, and positive predictive value for the ability of a given single urine sample to classify a subject in the highest BPA tertile.
We then calculated the sensitivity, specificity, and positive predictive value of two urine samples from an individual to classify subjects in the highest BPA tertile. This analysis was limited to subjects with at least six repeat urine samples (n = 8 subjects who contributed a total of 67 samples). Geometric means of all possible combinations of sample pairings from each individual were used in the analysis as the predicted value. The geometric mean value from all possible within-subject combinations was then compared with the geometric mean value of their repeated samples, used to determine their “true” BPA tertile classification. As above, contingency tables were then constructed. The goal of the validity analysis was to simulate and compare the ability of exposure assessments that involve one or two urine samples to predict a subject’s “true” longer-term exposure.
In both of these analyses, the single or two urine samples used to predict a subject’s tertile classification are not independent predictors of their overall geometric mean value because they are also used to calculate the subject’s geometric mean. For instance, when we calculated the geometric mean for subjects with three or more urine samples, the single urine sample evaluated for its predictive ability was included in the geometric mean calculation for that subject. Because of this, we limited these analyses to subjects with three (or six) or more urine samples to minimize the structural dependence between the predicted value based on the single sample and the “true” observed values based on the geometric mean of that subject’s full complement of samples.
Free full text: Click here
