Study population. Children participating in this longitudinal cohort study were enrolled at birth. English- or Portuguese-speaking mothers ≥ 18 years of age residing in one of four towns (New Bedford, Acushnet, Fairhaven, Dartmouth) near a PCB-contaminated harbor in New Bedford, Massachusetts, for at least the duration of pregnancy were recruited from a local hospital with approximately 2,000 births per year. Approximately 10% of mothers met study eligibility criteria and were available for recruitment during times when a study examiner was on site. Infants too ill to undergo neonatal examination or born by cesarean section were excluded from the study. Of the 788 mother–infant pairs enrolled at birth, 607 were followed up for neurodevelopmental testing when the child was approximately 8 years of age (78% of those eligible). Multiple births (n = 3 children) were excluded from the present analysis.
Cord blood samples for organochlorine analyses were collected at the infant’s birth; the serum fraction was removed after centrifugation and stored at –20°C. All sample analyses were performed by the Harvard School of Public Health Organic Chemistry Laboratory (Boston, MA). Laboratory personnel were blinded to health outcomes. Cord serum samples were analyzed for 51 individual PCB congeners and p
´-DDE. Laboratory analytic methods and quality control procedures are described elsewhere (Korrick et al. 2000
). Briefly, liquid-liquid extraction and column chromatography cleanup were used, and the extracts were analyzed by gas chromatography with electron capture detection using an internal standard. Primary and confirmatory capillary columns were used, and where results differed, the lower value was reported.
PCB concentrations were reported as individual congeners in units of nanograms per gram serum after the amount of analyte in the procedural blank was subtracted. Lipid content could not be determined for study subjects because of insufficient sample volume (1.5–4 mL) and was therefore measured for 12 randomly selected cord bloods from discarded, anonymous samples collected at the study recruitment site; values were reproducible (1.7 g/L ± 0.3).
The method detection limits (MDLs) for individual PCBs ranged from 0.002 to 0.04 ng/g serum, with most MDLs < 0.01 ng/g; the MDL for p
´-DDE in serum was 0.07 ng/g (Korrick et al. 2000
). The frequency of values below the MDL for each congener for this study cohort was previously reported (Korrick et al. 2000
). We retained measured values below the detection limit to optimize statistical power and avoid biased exposure estimates associated with censoring at the MDL (Kim et al. 1995
). Reproducibility of serum analyses was good; the within-batch coefficient of variation (CV) for the sum of four prevalent PCB congeners [118, 138, 153, and 180 (ΣPCB4
)] was 3% and the between-batch CV was 20% over the 5 years of analysis, with similar performance for p
We analyzed components of a continuous performance test (CPT) and a psychometric test of intelligence that reflect attention and impulse control; both tests were administered at the 8-year exam. The Neurobehavioral Examination System 2 (NES2) CPT is a computer-assisted examination that measures response to visual cues in a continuous performance task (Letz 1998
). A random series of animal silhouettes are displayed on the screen, and the child is instructed to press a button on a joystick only upon the appearance of a cat, to respond as quickly as possible, and to refrain from pressing the button for any other animal. The computer records response time (recorded only for correct responses), the number of nonresponses or errors of omission (button not pressed when cat appeared; the maximum time given was 1,200 msec), and the number of false-positive errors or errors of commission (button pressed for an animal other than the cat). Four blocks of trials were completed, with the first block designated as a practice block that was not scored. We analyzed four components of the CPT: a
) mean response time, b
) response time variability (standard deviation of mean response time), c
) total errors of omission, and d
) total errors of commission. Outcomes were summed across the last three test blocks. Inattention was interpreted as a higher number of errors of omission and longer reaction time, and poor response inhibition was interpreted as a higher number of errors of commission. Higher reaction time variability or performance inconsistency is also thought to indicate fluctuations or lapses in attention (Van de Molen 1996
The Wechsler Intelligence Scale for Children, 3rd edition (WISC-III), is a test that evaluates intellectual abilities (Wechsler 1991
). We focused on the two specific age-standardized subscales for which children with ADHD are found to score lowest: Processing Speed (includes coding and symbol search)and Freedom from Distractibility (includes digit span and arithmetic) (Wechsler 1991
We analyzed associations between attention and impulse control and two PCB congener groups: a
, the sum of four prevalent PCB congeners with relatively high levels that were measured with less error, and b
) the computed toxic equivalent (TEQ) for the sum of the five dioxin-like mono-ortho
PCB congeners measured (105, 118, 156, 167, and 189), computed on a lipid basis (1.7 g/L) and weighted with toxic equivalency factors (Van den Berg et al. 2006
). The TEQ group was included to investigate the potential for an aryl hydrocarbon receptor–mediated mechanism for the effect of dioxin-like congeners on neurodevelopment. We also investigated associations with p
CPT mean reaction time and reaction time variability and WISC-III outcomes were approximately normally distributed, met regression model assumptions, and were modeled with linear regression. Processing Speed and Freedom from Distractibility scores were standardized (mean ± SD = 100 ± 15). CPT errors of omission and commission were considered count data and initially modeled as Poisson distributed variables with log risk models; to correct for overdispersion (variance exceeded the mean), our final models were fitted using negative binomial regression.
Covariate data came from multiple sources, including maternal and pediatric medical records (including lead screening) and questionnaires administered 2 weeks after birth and at the 8-year follow-up examiantion. The 8-year exam included an assessment of maternal intelligence and depression using the Kaufman Brief Intelligence Test (Kaufman and Kaufman 1990
) and the Beck Depression Inventory (Beck et al. 1996
), respectively, and family and home characteristics using the Home Observation for Measurement of the Environment (HOME) (Caldwell and Bradley 1984
). Potential confounders considered were characteristics of the mother [age at child’s birth, prenatal smoking and alcohol consumption, prenatal diet (local and overall fish consumption), illicit drug use in the year before the child’s birth, breast-feeding, parity, and, at the 8-year examination, education, marital status, IQ, and depression] and the child (age at exam, birth year, sex, race/ethnicity, ADHD medication use, and peak and mean 12- to 36-month blood lead levels). Examiner (there were two possible examiners), household income, and quality of the home environment (HOME score) assessed at the time of the 8-year examination were also considered.
Inclusion of covariates in multivariable models was based on a priori considerations of covariate associations with exposure and outcome, model fit (statistically significant partial F-test at α < 0.10), and whether covariate inclusion materially affected the organochlorine exposure effect estimate. To check the sensitivity of our estimates to covariates that were not included in the model, we added each back to the final model to make sure they did not substantially change our final effect estimates. We also assessed differences in exposure–outcome associations across sex using stratified analyses and by including an interaction term between sex and exposure in the model.
Based on recent literature suggesting that performance on the CPT may vary over the course of the test session (Julvez et al. 2010
) (scores averaged over the entire testing period could mask these differences), block-specific effect estimates were also investigated. The Julvez et al. (2010)
paradigm divides performance over a 10-min testing period into three sequential stages: a
) orientation, learning, and habituation (first 2 min), b
) processing speed and selective focused attention (next 4 min), and c
) sustained attention (last 4 min). Because our children were younger than those in the Julvez et al. (2010)
study, we used a 4-min CPT, where block 1 is omitted as a training block and blocks 2–4 (1 min each) correspond to Julvez’s three sequential stages. We investigated block-specific results for all four CPT outcomes.
Sensitivity of our results to ADHD medication use was explored by excluding children with parent-reported medication use and recomputing organochlorine exposure–outcome associations. In addition, we investigated the influence of missing covariate data on our results by comparing unadjusted exposure–outcome associations for all children (regardless of whether covariate data was missing) with unadjusted exposure–outcome associations in the subset of children with nonmissing covariate data.
The study protocol was reviewed and approved by the Human Subjects Committees of Harvard School of Public Health and Brigham and Women’s Hospital (Boston, MA) and Southcoast Hospitals Group (New Bedford, MA). Written informed consent was obtained from all participating families before study evaluations.