Subjects. Participants were male and female patients (some of whom were couples) from the Fertility Center at the Massachusetts General Hospital (MGH) who were recruited into a prospective cohort study on environmental risk factors for reproductive health and contributed at least one urine sample for measuring environmental chemicals, including parabens. All patients > 18 years of age seeking infertility evaluation or treatment at the MGH Fertility Center were eligible to participate, and approximately 60% consented. We recruited participants between December 2004 and October 2010 and followed them from study entry until discontinuation of fertility treatment, a live birth, or loss to follow-up. The human studies institutional review boards of the MGH, Harvard School of Public Health (HSPH), and the Centers for Disease Control and Prevention (CDC) approved the study. Participants signed an informed consent after the study procedures were explained by a research nurse and all questions were answered.
Urine sample collection.
We collected a spot urine sample from study participants at the time of recruitment, at subsequent visits during infertility treatment cycles, and, if applicable, during pregnancy. Convenience (spot) samples were collected between August 2005 and November 2010. Urine samples collected before August 2005 were not analyzed for parabens because these chemicals were added to the study protocol after that date. Urine was collected in a sterile polypropylene cup. After specific gravity (SG) was measured using a handheld refractometer (National Instrument Company, Inc., Baltimore, MD), the urine was divided into aliquots and frozen at –80°C. Samples were shipped on dry ice overnight to the CDC (Atlanta, GA), where concentrations of total (free + conjugated) MP, PP, and BP were measured using on-line solid phase extraction-high performance liquid chromatography–isotope dilution tandem mass spectrometry as previously reported (Ye et al. 2006b
). The limits of detection (LOD) were 1.0 μg/L for MP and 0.2 μg/L for PP and BP.
Demographic predictors of paraben concentrations.
Information on demographic factors of interest, collected through nurse-administered and take-home questionnaires, included sex, race, and age, which were previously shown to be associated with urinary paraben concentrations in the general population (Calafat et al. 2010
). Race was categorized as Caucasian, African American, Asian, and other. We also examined urinary concentrations of parabens according to body mass index (BMI) categorized as underweight (< 18.5 kg/m2
), normal (18.5–24.9 kg/m2
), overweight (25–29.9 kg/m2
), and obese (≥ 30 kg/m2
). Weight and height were measured by a research nurse at study entry.
Statistical analysis. We evaluated demographic characteristics of male and female study participants (means and percentages). We report the distribution of urinary paraben concentrations for all individual samples, and also report the distribution of within-person geometric mean (GM) values because the number of urine samples from each participant varied and within-person concentrations were log-normally distributed. These data are uncorrected for SG to allow comparison with other studies.
We replaced paraben concentrations less than the LOD with LOD divided by the square root of 2 (Hornung and Reed 1990
). We calculated the Spearman correlation between the different parabens. We corrected the urinary paraben concentrations for SG using a modification of a previously described formula: Pc
[(1.016 – 1)/SG
– 1], where Pc
is the SG
-corrected paraben concentration (micrograms per liter), 1.016 is the mean SG
for the samples examined, and P
is the measured paraben concentration (micrograms per liter) (Duty et al. 2005
). Natural log-transformed SG-corrected paraben concentrations were used as the outcome in all statistical models. We excluded BP from further statistical analyses including both males and females due to a low detection frequency (65% detected).
We fit linear mixed-effects models to estimate associations of urinary MP and PP concentrations (micrograms per liter urine) with age, sex, race, and BMI, with each paraben modeled separately. We included a random effect for subject in the models to account for correlation among repeat samples collected on the same individual over time. Sex, race, and BMI were included as fixed effects, whereas age at urine collection was included as a time varying factor. Using step-wise backward elimination, we retained covariates with a p-value < 0.1. Final models included sex, race, and BMI. The parameter estimates were exponentiated to estimate the difference in paraben concentrations relative to the reference category of each predictor variable.
To determine whether couples have similar patterns of paraben exposure, we calculated Spearman correlation coefficients for within-person GM paraben concentrations between partners, as well as for paraben concentrations between partners with urine samples that were collected on the same day (time matched).
To examine the reproducibility of urinary MP and PP concentrations, we calculated intraclass correlation coefficients (ICCs) using SAS PROC MIXED (SAS Institute Inc., Cary, NC) with a random effect for subject for participants who provided at least two urine samples. The ICC is calculated as the ratio of between-person variability to total variability (total variability = between-person + within-person variability). ICCs closer to zero indicate less reproducibility (large within-person variability) and ICCs closer to one indicate higher reproducibility (low within-person variability). Rosner (1995)
defined an ICC < 0.4 as indicating poor reproducibility, an ICC between 0.4 and < 0.75 as indicating fair to good reproducibility, and an ICC ≥ 0.75 as indicating excellent reproducibility.
Subset analysis of pregnant women. To compare the variability of urinary paraben concentrations before and during pregnancy, we evaluated a subset of women who became pregnant during follow-up and had provided at least two prepregnancy and at least two pregnancy spot urine samples. An intrauterine pregnancy was defined by the presence of a fetal heart beat detected by transvaginal ultrasound. We assigned urine samples to a trimester based on the following definition: first trimester: 0–13.9 weeks gestation; second trimester: 14.0–28.0 weeks; and third trimester: ≥ 28.1 weeks. We assigned the gestational week of the urine sample collection using the estimated date of conception, which was defined as the expected date of delivery minus 266 days. We estimated the delivery date using three dating methods (in order of preference if more than one was available): a) oocyte retrieval date as recorded from medical records; b) crown–rump length as measured by first trimester ultrasound; or c) women’s reported date of last menstrual period.
We calculated the within-woman GM for prepregnancy and pregnancy urinary paraben concentrations and report the median and 25th–75th percentiles [interquartile range (IQR)]. We also report urinary paraben concentrations (median and IQR) for samples collected in each trimester of pregnancy. We estimated the Spearman correlation between the GM paraben concentrations before and during pregnancy.
We fit linear mixed-effects models with a random effect for subject to estimate the change in urinary paraben concentrations before and during pregnancy. First, we used pregnancy status (before vs. during) to estimate the difference in urinary paraben concentrations during pregnancy as compared with before pregnancy. Second, restricting to urine samples collected during pregnancy, we estimated the change in urinary paraben concentrations over continuous time in weeks since conception. We exponentiated the parameter estimates to estimate the percent change in the paraben concentration per week since conception. We evaluated the reproducibility of urinary MP, PP, and BP concentrations before and during pregnancy by calculating the ICCs for samples collected during each time period.
Finally, we conducted a classification analysis (Hauser et al. 2004
; Mahalingaiah et al. 2008
) using the GM of the two or three urine samples collected during pregnancy as the gold-standard exposure measure. We divided this GM summary exposure measure into tertiles, as well as each trimester-specific concentration (using trimester-specific tertile cut points). We calculated the sensitivity, specificity, and positive predictive value (PPV) of each trimester-specific paraben concentration to correctly classify a woman into the highest exposure tertile (based on the gold standard). To minimize bias in this analysis we excluded women with all of their urine samples collected in the same trimester (n
= 3 women). Among women with two samples collected in the same trimester (who also had one other sample collected in another trimester) we included only the first of the two samples collected in the same trimester (n
= 4 women). In a second analysis restricted to women with one urine sample collected in each trimester, we counted the number of women who remained in the same exposure tertile over the course of pregnancy (Braun et al. 2012
). Each woman could have either zero, two, or all three urine samples remaining in the same tertile during pregnancy. We conducted all statistical analyses using SAS version 9.2 (SAS Institute Inc.). We made no adjustment for multiple comparisons.