Study design and sample. The Committee for the Protection of Human Subjects at the University of California, Berkeley, and at the Centers for Disease Control and Prevention (CDC) approved all study activities. The study sample consisted of participants in the Center for the Health Assessment of Mothers and Children of Salinas (CHAMACOS), a longitudinal cohort study of environmental factors and children’s growth and development. We enrolled pregnant mothers in 1999 and 2000 from prenatal clinics serving the farmworker population in the Salinas Valley, California. Eligible women were at least 18 years of age, spoke English or Spanish, qualified for low-income health insurance, were at < 20 weeks gestation, and were planning to deliver at the county hospital. Mothers provided written informed consent for themselves and their children to participate in the study.
Of 601 pregnant women enrolled in the study, a total of 527 were followed through the birth of a singleton, live-born infant. BPA measurements in spot urine collected during pregnancy were available for 498 mothers, and 402 children had at least one measure of BMI between 2 and 9 years of age.
Data collection. We interviewed mothers twice during pregnancy, after delivery, and when their children were 2, 3.5, 5, 7, and 9 years of age to obtain information about demographic characteristics, diet, and behaviors. All interviews were conducted in English or Spanish using structured questionnaires. At the baseline interview, we asked mothers about their race/ethnicity, education, income, marital status, and number of years they had lived in the United States, as well as information about soda consumption, smoking, and alcohol and drug use during pregnancy. We calculated prepregnancy BMI from self-reported prepregnancy weight and measured height. If self-reported prepregnancy weight was unavailable or invalid, we used measured weight at first prenatal visit (n = 23) if the first prenatal visit occurred at or before 13 weeks gestation or used regression models to impute prepregnancy weight based on weight at all prenatal visits if the first prenatal visit occurred after 13 weeks (n = 16).
We collected information about child behaviors from the mother during follow-up interviews, including asking about her child’s soda consumption, time spent watching television on weekdays and weekend days, and how often her child ate fast food (“from restaurants like McDonalds, Burger King, or KFC”) and sweet snacks (“like candy, cookies, cake, or other sweet snacks”).
When the child was 9 years of age, clinical Tanner staging was conducted by trained research staff to determine stages of pubertal development. Children were considered to have entered puberty if they were stage 2 or above for breast development for girls or stage 2 or above for pubic hair or genital development for boys.
BPA measurements. We collected spot urine samples from mothers at two time points during pregnancy: near the end of the first (mean ± SD, 13.8 ± 5.0 weeks gestation) and second (mean ± SD, 26.4 ± 2.4 weeks gestation) trimester of pregnancy and from the children when they were 5 (mean ± SD, 5.1 ± 0.2 years) and 9 (mean ± SD, 9.4 ± 0.4 years) years of age. Urine samples were collected in polypropylene urine cups, aliquotted into glass vials, and frozen at –80°C until shipment to the CDC for analysis. Analysis of field blanks showed no detectable contamination by BPA using this collection protocol.
We used solid-phase extraction coupled to high performance liquid chromatography–isotope dilution tandem mass spectrometry (Ye et al. 2005
) to measure total urinary BPA concentration (conjugated plus unconjugated). The limit of detection (LOD) was 0.4 μg/L. Concentrations < LOD for which a signal was detected were reported as measured. Concentrations < LOD with no signal detected were randomly imputed based on a log-normal probability distribution using maximum likelihood estimation (Lubin et al. 2004
Specific gravity was measured with a refractometer (National Instrument Company Inc., Baltimore, MD) for the maternal urine samples, but was unavailable for the children’s samples. Thus, maternal concentrations were normalized for urinary dilution using urine specific gravity (Mahalingaiah et al. 2008
), and child BPA concentrations were normalized by dividing by urinary creatinine concentration.
For maternal BPA, two urinary measures were available for 307 women and were averaged to better approximate exposure over the course of pregnancy. In the 95 women for whom only one BPA measurement was available, the single measurement was used in lieu of the average. Child concentrations were used as separate variables representing exposure at 5 and 9 years of age.
Anthropometric measurements. At each assessment point (age 2, 3.5, 5, 7, and 9 years), children were weighed and measured without jackets or shoes by trained study staff. We measured weight using a digital scale (Tanita Mother-Baby Scale model 1582; Tanita Corp., Arlington Heights, IL)and rounded to the nearest 0.1 kg. We measured height using a stadiometer (Seca 222; Seca, Chino, CA) and rounded to the nearest 0.1 cm. Starting at 5 years of age, we measured waist circumference at each visit by placing a measuring tape around the abdomen at the level of the iliac crest, parallel to the floor. Height and waist circumference measurements were conducted in triplicate and averaged for analysis. When the children were 9 years of age, we measured fat percentage using “foot-to-foot” bio-impedence technology with a Tanita TBF-300A body composition analyzer (Tanita Corp.).
BMI was calculated as weight (kilograms) divided by height squared (square meters) and compared with the sex-specific BMI-for-age percentile data issued by CDC in 2000 (National Center for Health Statistics 2005
). Children who were ≥ 85th but < 95th percentile for their age and sex were classified as overweight, children ≥ 95th percentile were classified as obese. Age- and sex-standardized BMI z
-scores were also generated using the CDC norms (Grummer-Strawn et al. 2010
Statistical analysis. All analyses were conducted using Stata 11 (StataCorp, College Station, TX). We used multivariable linear regression to examine the association of BPA concentrations with continuous outcomes (BMI z-score, waist circumference, and percent body fat) and multivariable logistic regression for categorical outcomes (e.g., overweight/obese).
We first examined the association of prenatal, age 5, and age 9 urinary BPA concentrations with outcomes at age 9 years. We also examined the association of prenatal BPA concentrations with repeated measures of child growth between age 2 and 9 years, using generalized estimating equations (GEE). GEE models included an interaction term for the BPA and age (continuous in months) variables to allow the association to differ by child age. Because the p-value on the age interaction term was < 0.1, we kept the interaction term in the model and estimated BPA effect coefficients for each age using the Stata “lincom” postestimation command. Because animal studies indicate that associations of BPA and BMI may differ by sex, and that differences in body weight may not become apparent until sexual maturity, we tested for effect modification by sex and by puberty status at age 9 using α = 0.1.
Urinary BPA concentrations were analyzed as continuous and categorical variables. Continuous BPA variables were log-transformed to reduce the influence of outliers. Because of the relatively narrow range of exposure, log base 2 was used. During pregnancy and at 5 years, categorical BPA concentrations were classified by tertiles because very few individuals had concentrations < LOD. At 9 years of age, categorical BPA concentrations were classified as < LOD, ≥ LOD but below the median, and above the median. Both visual inspection of the associations using lowess plots and regression models using categorical exposure variables gave no indication that associations were not linear.
We identified potential confounders a priori using directed acyclic graphs. Potential confounders included maternal prepregnancy BMI, age, education, years of residence in the United States, smoking during pregnancy, soda consumption during pregnancy, and family income. Time-varying covariates considered were child consumption of soda, fast food, and sweets, television watching, environmental tobacco smoke exposure, and time spent playing outdoors, assessed at multiple times during childhood. We included covariates in the final models if they were associated with both exposure and any of the growth outcomes at p-value < 0.2 or if removing them changed the coefficient for the main BPA exposure variable by > 10%. Maternal age and prepregnancy BMI were analyzed as continuous variables. Other variables were categorized as shown in .
Geometric mean (GM) and SD (GSD) of urinary BPA concentrations (µg/L) by demographic characteristics of the study population, CHAMACOS study, Salinas, California, 2000–2010.
In sensitivity analyses we a
) analyzed BPA concentrations unadjusted for urinary dilution (specific gravity or creatinine); b
) reanalyzed our models controlling separately for three important prenatal exposures in this population: organochlorine pesticides [using prenatal serum concentrations of dichlorodiphenyldichloroethylene (DDE)], organophosphate pesticides (using prenatal urinary metabolites of organophosphate pesticides), and brominated flame retardants [using prenatal serum concentrations of polybrominated diphenyl ethers (PBDEs)]; c
) analyzed the two prenatal BPA measurements separately, as early pregnancy (< 20 weeks) or late pregnancy (≥ 20 weeks) measurements; and d
) used inverse probability weights to account for potential bias due to loss to follow-up (Hernan et al. 2004