We used data from the Health Outcomes and Measures of the Environment (HOME) Study for this project (Braun et al. 2009
; Phelan et al. 2009
). The HOME Study is a prospective birth cohort with the primary goal of investigating the effects of exposure to environmental toxicants on child growth, neurobehavioral development, and respiratory health. Between March 2003 and January 2006, the HOME Study enrolled English-speaking women at a mean ± SD of approximately 16 ± 2 weeks gestation who were ≥ 18 years and lived in a home built before 1979. We tracked the women through pregnancy and are continuing to follow their children through 5 years of age. Women resided within five Ohio counties surrounding Cincinnati, received prenatal care from one of nine participating obstetrical clinics, and delivered at one of three participating hospitals. The obstetric clinics provided us with lists of new patients, and we prescreened women for eligibility criteria. We sent study information to eligible women and placed a follow up phone call to assess interest in the study and complete screening. Approximately 37% (468 of 1,263) of eligible women agreed to participate and provided written informed consent. We excluded potential participants if they were HIV positive, taking antiepileptic medication, or were diabetic (not gestational) or had bipolar disorder or schizophrenia. Infants of participating mothers were eligible for the longitudinal study. The longitudinal study included an embedded randomized control trial of a lead hazard reduction intervention and injury hazard reduction control.
We included only children with prenatal urinary BPA concentration and respiratory outcome data for this study. These data were available for 365 children (92% of HOME Study participants: 398 live-born infants, randomly excluding one child for twin births). The institutional review boards of Cincinnati Children’s Hospital Medical Center, the Centers for Disease Control and Prevention (CDC), and the involved birth hospitals approved the HOME Study and this project.
We assessed prenatal BPA exposure by measuring the concentrations of BPA in serial maternal spot urine samples. We collected maternal urine and serum at enrollment (15.9 ± 1.9 weeks gestation), 26 weeks gestation, and birth. Urine and serum samples were sent to the Division of Laboratory Sciences at the CDC for analysis. Urinary BPA concentrations were quantified using online solid-phase extraction coupled with high-performance liquid chromatography/isotope-dilution tandem mass spectrometry (Ye et al. 2005a
). The limit of detection (LOD) was 0.4 μg/L. When BPA concentrations were below the LOD (< 13% of samples), we replaced these values with the LOD divided by √–
2 (Hornung and Reed 1990
We assessed tobacco exposure using serum concentrations of cotinine, a metabolite of nicotine. We used serum cotinine, a biomarker of tobacco exposure, as a covariate in analyses because of the established association of prenatal tobacco exposure with wheeze in childhood (Sherriff et al. 2001
; Spanier et al. 2011
). Analyses of serum for cotinine were performed using high-performance liquid chromatography/atmospheric-pressure chemical ionization tandem mass spectrometry (Bernert et al. 1997
). The LOD was 0.015 ng/mL. When serum cotinine values were below the LOD (~ 35% of samples), we imputed values by sampling randomly from the left tail of a lognormal distribution. We collected other environmental exposure information (i.e., pet ownership, cockroach exposure, and house characteristics) by survey as described below.
After birth, we employed trained research assistants to survey parents every 6 months through the child’s age 3 years to collect study data. The baseline, 12-, 24-, and 36-month surveys were conducted via home visits, and the 6-, 18-, and 30-month surveys were conducted over the phone. We designed these survey questions to parallel the NHANES wheeze question (National Center for Health Statistics 1994
). We asked “Has [child’s name] had wheezing or whistling in his/her chest in the last 6 months?” We also asked about the number of wheeze attacks, and we dichotomized the number of wheeze attacks at each time point (no wheeze vs. any wheeze) to minimize effects of extreme values. We used this dichotomized value as our outcome measure. We also asked if wheezing usually occurred with or followed a cold or viral illness.
Trained research assistants conducted extensive surveys at baseline and every 6 months after the child was born to collect data on potential covariates, including demographic characteristics, socioeconomic status, and health status. Demographic and socioeconomic characteristics such as maternal education, race/ethnicity, occupation, income, housing volume, and health insurance status were considered as possible covariates in all models. We also considered other factors potentially associated with wheeze by evaluating them as possible covariates [e.g., prenatal tobacco exposure (cotinine), season, history and duration of breast-feeding, family history of asthma, family history of allergy, child eczema, child allergy, neonatal characteristics, pet ownership, and cockroach exposure] (Ball et al. 2000
; Fleming et al. 1987
; Gold et al. 1999
; Oddy et al. 2002
; Simoes 2003
; Taussig et al. 2003
; Young et al. 2000
We categorized residence location/type into three categories based on census tract information. Family residence was categorized as urban if their census tract was located in a city defined as a metropolitan area by U.S. Census 2000 (Cincinnati, OH; Newport, KY; and Covington, KY) (unpublished data), rural if their census tract had a 2000 Census population density < 500 persons/square mile, and suburban if their census tract did not fit into urban or rural categories.
Statistical analysis. We first calculated descriptive statistics for all demographic, exposure, and outcome data. We used a natural log transformation for all BPA data because urinary BPA concentrations were approximately log-normal, and to minimize the influence of outlying values. We used the geometric mean and 95% confidence interval (CI) as the primary descriptors of central tendency and dispersion for BPA data.
We used the mean of creatinine standardized prenatal urinary BPA concentrations as our primary independent variable to approximate exposure over pregnancy. To calculate this variable, we first standardized BPA concentrations for urinary creatinine (micrograms of BPA per gram of creatinine) at each time point. We then calculated the mean of all available standardized BPA concentrations and log transformed the mean of prenatal BPA concentrations for each participant. We analyzed the association of prenatal urinary BPA concentration with the dichotomous outcome variable wheeze (wheeze over the previous 6 months) using generalized estimating equations (GEEs) with a logit link. GEE is an extension of logistic regression that accounts for within-subject correlation resulting from repeated outcome measurements in longitudinal studies. Therefore, all data collected during the six time points (months 6, 12, 18, 24, 30, and 36) were used in the analysis. Interpretation of the association between the exposure variable and outcome (i.e., wheeze) using odds ratios (ORs) and 95% CIs from GEEs is similar to ordinary logistic regression but represents the association with wheeze over the entire 3-year period rather than at a single point in time. We first conducted bivariate analyses to evaluate the association of prenatal BPA concentration and potential covariates with wheeze. After bivariate analysis, we conducted a multivariable analysis using forward selection techniques. In the initial multivariable analysis, we included covariates that predicted wheeze with a p
-value ≤ 0.2 in bivariate analysis. Covariates were retained if they were significantly associated with wheeze (p
< 0.05) or if their addition caused a > 10% change in the estimate of the association between wheeze and prenatal BPA. In all analyses (including the bivariate analyses), we included a variable for survey time point (6, 12, 18, 24, 30, or 36 months) and a variable for intervention arm to account for any potential design effects of the embedded randomized trial. We also adjusted for log-transformed maternal serum cotinine (continuous) because previous studies have demonstrated an association of cotinine with wheeze (Spanier et al. 2011
). After conducting the multivariable analysis, we tested for potential multiplicative interactions of prenatal BPA concentration with all covariates on the logit scale.
We conducted secondary analyses to explore potential windows of vulnerability by replacing the mean prenatal urinary BPA concentration variable with BPA concentration measured at each of the prenatal time points (i.e., 16 weeks, 26 weeks, and at birth) in three separate analyses, each standardized for same time urinary creatinine. GEE analyses were used for these secondary analyses. SAS (version 9.2; SAS Institute Inc., Cary, NC, USA) was used for all data analyses.