We selected 319 pregnant inner-city women who delivered between 1999 and 2006 from the longitudinal birth cohort of 727 mothers and newborns being conducted by the Columbia Center for Children’s Environmental Health (CCCEH). Enrollment and exclusion criteria have been described previously (Perera et al. 2003
). The CCCEH cohort was restricted to nonsmoking women 18–35 years of age who self-identified as either African American or Dominican and who had resided in northern Manhattan or the South Bronx in New York City for at least 1 year before pregnancy. Women were excluded if they used illicit drugs, had diabetes, hypertension, or known HIV, or had their first prenatal visit after the 20th week of pregnancy. The study was approved by the Columbia University Medical Center and CDC institutional review boards (IRBs). Study procedures were explained at enrollment, and each woman signed an IRB-approved consent form. We selected women (n
= 319) for participation in the present study if phthalate metabolite concentrations had been measured in spot urine samples collected during pregnancy, if the child had completed the Bayley Scales of Infant Development II (BSID-II; n
= 297) or the mother had completed the Child Behavior Checklist (CBCL; n
= 286) at the child age 3–year visit, and data were available on model covariates. The 319 subjects did not differ significantly from the remaining subjects in the CCCEH cohort in terms of basic demographics (race/ethnicity, maternal prenatal marital status and education level, household income, proportion on Medicaid or other public assistance) or on child sex, gestational age, and birth weight (all p
-values > 0.05).
Questionnaire and medical record data. A trained bilingual interviewer administered a 45-min questionnaire to each woman in her home during the third trimester of pregnancy, collecting information on demographics, race/ethnicity, home characteristics and residential history, history of active and passive smoking, occupational history, marital status, education and income level, prenatal alcohol and drug use, and maternal psychosocial conditions. We also abstracted information from the mothers’ and infants’ medical records after delivery, including gestational age, infant sex, birth weight, length and head circumference, complications of pregnancy, medication use, and delivery method.
Urine sample collection and phthalate measurements.
We collected a spot urine sample from the women during the third trimester of pregnancy (mean, 33.1 ± 3.0 weeks of gestation; median, 33 weeks). Samples were stored at Columbia University at –80°C, shipped to the CDC on dry ice, and stored at –70°C until analysis. The urinary phthalate metabolite concentrations were measured at the CDC as previously described (Kato et al. 2005
). Each analytical run included calibration standards, reagent blanks, and quality control samples. We used specific gravity to correct for urinary dilution as recommended for phthalates (Hauser et al. 2004
). Specific gravity was measured using a handheld refractometer (Atago PAL 10-S; Atago U.S.A. Inc., Bellevue, WA). As a measure of reliability, we calculated intraclass correlation coefficients (ICCs) for the phthalate metabolites in serial spot urine samples collected biweekly from 48 women in the CCCEH cohort over 6–8 weeks late in pregnancy (n
= 135 samples, 2–4 repeats per woman). Adjusting for specific gravity, ICCs were 0.77 for MBzP, 0.65 for mono-n
-butyl phthalate (MnBP), and 0.60 for monoisobutyl phthalate (MiBP) and ranged from 0.27 to 0.42 for the DEHP metabolites.
Measures of child mental, psychomotor, and behavioral development.
The BSID-II (Bayley 1993
) provides a developmental quotient (raw score/child chronologic age) from which a continuous Mental Development Index (MDI) and a Psychomotor Development Index (PDI) are generated. The raw scores are converted to a normalized scale with a mean of 100 and standard deviation of 15. Scores can be analyzed continuously (with higher scores indicating better development), or children can be classified as normal or at risk of delay (scores ≤ 85). The child age at the administration of the BSID-II averaged 36.4 ± 1.7 (range, 27–42) months. Cohort children were tested under controlled conditions by trained bilingual research assistants; interrater reliability has been previously described (Rauh et al. 2006
). Behavioral problems were measured through maternal report of the 99-item CBCL for children 1.5–5 years of age, which provides an early indicator of potential behavioral problems in young children (Achenbach 2000
). The 99 items are summed into seven syndrome scales, with four scales (emotionally reactive, anxious/depressed, somatic complaints, withdrawn) subsequently summed into internalizing behaviors and two (attention and aggressive behaviors) summed into externalizing behaviors. The CBCL scales can be analyzed as continuous scores, or the children can be classified in the normal, borderline, or clinical range based on predetermined cut-points (Achenbach 2000
). The child age at the administration of the CBCL averaged 36.6 ± 2.8 (range, 33–48) months.
Prenatal psychosocial factors included maternal self-report of hardship during pregnancy (lack of food, clothing, housing, gas or electricity, or medicines) and satisfaction with overall living conditions. Maternal demoralization was measured by the 27-item Psychiatric Epidemiology Research Instrument-Demoralization Scale (Dohrenwend et al. 1978
). Maternal intelligence was assessed postnatally by the Test of Non-Verbal Intelligence, third edition (Brown et al. 1990
), a 15-min, language-free measure of general intelligence, which is relatively stable and free of cultural bias. The quality of proximal care-taking environment was measured by the Caldwell and Bradley’s (1979)
Home Observation for Measurement of the Environment (HOME) scale at child age 38.4 ± 6.2 months. Prenatal alcohol consumption and exposure to environmental tobacco smoke were measured by maternal self-report. Eight polycyclic aromatic hydrocarbons (PAHs) were quantified in 48-hr maternal third-trimester personal air samples and summed (Perera et al. 2003
). Bisphenol A (BPA) was measured in the maternal prenatal spot urine samples at CDC as previously described (Ye et al. 2005
To examine the relationship between prenatal exposure to the four phthalates (assessed from the urinary metabolite concentrations) and BSID-III and/or CBCL outcomes, we used linear models for the continuous outcomes and logistic models for categorical outcomes. The few phthalate metabolite concentrations below the limit of detection (LOD) were assigned a value of LOD/2. Metabolite concentrations were right skewed and were transformed using the natural logarithm. From a pool of covariates known or suspected of being associated with the phthalate concentrations or BSID-II or CBCL outcomes (Eskenazi et al. 2007
; Rauh et al. 2006
; Wasserman et al. 2003
; Whyatt et al. 2009
), we selected those remaining significant or marginally significant (p
< 0.10) in the regression model for at least one of the outcome variables in the same set. Model covariates for the BSID-II outcomes were child sex (boy vs. girl), race/ethnicity (Dominican vs. African American), the quality of proximal care-taking environment (continuous HOME scale), gestational age (in weeks), maternal marital status (never vs. ever married), maternal prenatal alcohol use (yes/no), and urine specific gravity. Maternal IQ was not controlled because it was not significant once the HOME scale had been added to the model and did not appreciably change the magnitude of the exposure–BSID-II outcome relationships. Covariates in the linear models for CBCL outcomes were child age in months at the time of test administration, child sex, race/ethnicity, maternal IQ (categorized as described below), maternal satisfaction with overall living conditions (yes/no), maternal perceived hardship (yes/no), maternal demoralization (continuous scale), maternal prenatal PAH exposure (categorized as described below), maternal prenatal urinary loge
BPA concentrations, and specific gravity. Multinominal logistic regression was used to analyze the association between the urinary phthalate metabolite concentrations and whether the child fell in the normal, borderline, or clinical range on the CBCL scales. The analyses were conducted only on the four scales that had at least 15 subjects (≥ 5% of the sample) in each cell. The control variables were child’s age in months at test administration, child sex, mother’s satisfaction, maternal demoralization, and urine specific gravity. Urine specific gravity was standardized [(individual subject value – mean value)/SD] before inclusion in all models. Additional variables assessed as confounders but not controlled were maternal education, prenatal environmental tobacco smoke, year and season of urine collection, and umbilical cord lead and chlorpyrifos. To evaluate whether language of test administrations (Spanish vs. English) acted as an effect modifier, we conducted stratified analyses and tested the interaction terms between the phthalate metabolites and language of administration; results were comparable to those presented here, and none of the interaction terms was significant. Missing values for the following five covariates were imputed: a
) three missing values for maternal demoralization were imputed by a linear regression model with maternal education, maternal satisfaction, and maternal hardship as predictors (model R2
= 0.15, n
= 285); b
) 12 missing observations from the HOME scale were imputed by linear regression with race/ethnicity, maternal education and IQ, and household income as predictors (model R2
= 0.18, n
= 294); c
) five missing observations for maternal hardship were imputed based on a logistic regression with race/ethnicity, maternal demoralization, maternal satisfaction, and prenatal PAH as predictors (model R2
= 0.18; n
= 285); d
) where there were > 5% of missing values (maternal IQ and PAH exposure), we categorized the observed data and added an additional category for missing values. We excluded subjects with missing values for gestational age (n
= 4) and BPA concentrations (n
= 9), because these could not be imputed. We also conducted analyses a
) before imputation of the missing values and b
) after removing subjects with very dilute urine (specific gravity < 1.007, n
= 15) or concentrated urine (specific gravity > 1.03, n
= 7); results were comparable to those presented here. Results from the linear models and the logistic model of BSID-II outcomes are presented for the total cohort and also after stratifying by child sex. Because of small sample sizes in at-risk categories, logistic regression models were adjusted to remove covariates not significantly related to the outcome. Sex differences in the effect of the exposure variable were detected by the Wald test. Sample sizes were too small to stratify by child sex in the multinominal logistic regression analyses for CBCL outcomes. Results were considered significant at p
< 0.05. Analyses were conducted using SAS (version 9.2; SAS Institute Inc., Cary, NC).