Abstract

Polychlorinated biphenyls (PCBs) are ubiquitous environmental contaminants. In utero exposure to background levels of PCBs has been associated with intellectual impairment among children in most, but not all, studies. The authors evaluated prenatal PCB exposure in relation to cognitive test (intelligence quotient (IQ)) scores on the Wechsler Intelligence Scale for Children at age 7 years. Pregnant women were recruited from 12 US study centers from 1959 to 1965, and their children were followed until age 7 years (the Collaborative Perinatal Project). Third trimester serum was analyzed for PCBs in 1997–1999 for 732 women selected at random and for an additional 162 women whose children had either a low or a high IQ score. The PCB–IQ association was examined in multivariate models. Among those in the lowest exposure category (<1.25 μg of PCB/liter of serum), the fully adjusted mean IQ score was 93.6 (standard error: 1.8); among those in the highest category (≥5 μg of PCB/liter), the mean IQ was 97.6 (standard error: 1.2); and overall the increase in IQ per unit increase in PCB level (μg/liter) was 0.22 (95% confidence interval: −0.28, 0.71). In these data, in utero exposure to background levels of PCBs was not associated with lower IQ at age 7 years.

Polychlorinated biphenyls (PCBs) are ubiquitous, persistent environmental pollutants. Early life exposure, within the range reported in developed countries, has been associated with cognitive deficits among children (1). Moreover, exposure to PCBs in utero has been implicated as neurotoxic more often than has subsequent exposure (24), even though smaller amounts of PCBs are transferred transplacentally than through breastfeeding. In vitro and in vivo models demonstrate that PCBs cause altered synaptic transmission in the central nervous system, reduction in the striatum dopamine concentration, and decreased thyroxine levels (59). Furthermore, in rodent and primate studies, perinatal PCB exposure is usually associated with impairment of learning and cognition (1012).

Epidemiologic studies of perinatal exposure to background-level PCBs in relation to cognitive functioning in children have given inconsistent results. For example, among Michigan children followed from birth, prenatal PCB exposure was associated with poorer performance on the verbal and memory scales of the McCarthy Scales of Children's Abilities at 4 years of age (n = 133) and with lower full-scale and verbal intelligence quotient (IQ) scores on the Weschler Intelligence Scale for Children-Revised at age 11 years (n = 178) (3, 13). In a birth cohort from the Netherlands, the findings were similar on the Kaufman Assessment Battery for Children at 42 months of age. Children who had the highest level of in utero exposure to PCBs performed less well on cognitive tests (n = 395) (4). On subsequent testing at age 6.5 years, however, the relation was no longer present. In a cohort from upstate New York, higher umbilical cord PCB levels were associated with lower scores on the McCarthy Scales of Children's Abilities at 38 months of age but not at 54 months (n ∼ 195) (14). In a North Carolina birth cohort (15), performance on these same scales at ages 3, 4, and 5 years was unrelated to PCB exposure. A clearer understanding of the relation between low-level PCB exposure and performance on examinations of cognitive function would improve assessment of risks associated with exposure.

Our study was designed to evaluate the association between maternal third trimester serum PCB levels and offspring cognitive test scores at 7 years of age by use of the Wechsler Intelligence Scale for Children (WISC). Subjects were participants in the Collaborative Perinatal Project.

MATERIALS AND METHODS

The Collaborative Perinatal Project was a prospective study designed to identify determinants of neurodevelopmental deficits in children (16). Pregnant women were recruited from 1959 to 1965 from 12 US study centers (Baltimore, Maryland; Boston, Massachusetts; Buffalo, New York; Memphis, Tennessee; Minneapolis, Minnesota; New Orleans, Louisiana; New York, New York (two); Philadelphia, Pennsylvania; Portland, Oregon; Providence, Rhode Island; and Richmond, Virginia). Prenatal clinics at university hospitals comprised the majority of the study centers. The method of systematic sampling used to selected subjects varied across center. More details about the centers and sampling have appeared previously (17). Women were ineligible if they were incarcerated, if they were planning to leave the area or to give up the child for adoption, or if they gave birth on the day they were recruited into the study.

Study data were collected at each prenatal visit, at delivery, and when the child's age was 24 hours, 4 months, 8 months, and 1, 3, 4, and 7 years. Data were collected by physicians or personnel, for example, psychologists, who were specifically trained for the job. They administered age-appropriate standardized tests that assessed cognitive, neurologic, and motor development, as well as psychological and behavioral functioning.

Of the 55,908 pregnancies recruited into the study, 46.2 percent were Black, 46.0 percent were White, 6.8 percent were Puerto Rican, and the remaining 1 percent were Asian or other race. At registration, the subjects' median socioeconomic index score was 4.5, slightly lower than the median socioeconomic index score of 5.0 for the US general population (18). The socioeconomic index calculated for subjects in the Collaborative Perinatal Project was the mean of three percentile scores (for education, occupation, and family income), where education was that of the head of the household, occupation was that of the head of the household or the chief wage earner, and the score used to calculate the percentile for an occupation was based on the percentiles of education and income among those with the same occupation. By delivery, 4 percent of the mothers were lost to follow-up. Among those women not lost to follow-up at delivery, approximately 80 percent were recruited after 20 weeks of gestation; all had at least one prenatal visit recorded, and 9,161 women had more than one pregnancy included in the study. Of the liveborn children in the Collaborative Perinatal Project, 71 percent were followed to age 7 years.

Selection criteria and sampling scheme

For the present study, eligible children met the following criteria: 1) they were liveborn, 2) they were singletons, and 3) a 3-ml third trimester maternal serum specimen was available. From the 43,628 eligible children, 1,256 subjects were selected at random. An additional 207 children were randomly selected from eligible children whose 7-year full-scale IQ score was either one or more standard deviations below the mean or one or more standard deviations above the mean. By increasing the probability of inclusion for subjects with “extreme values” of an outcome (outcome-dependent sampling), we increased the study power relative to a study of equivalent size that used simple random sampling alone (19). The institutional review board of the National Institute of Environmental Health Sciences approved the study protocol.

Exposure assessment

Beginning at registration, nonfasting maternal blood was collected every 8 weeks during pregnancy. Serum samples were stored at −20°C in glass with no recorded thaw. The variation of PCB levels across trimesters of pregnancy and after delivery has been examined using Collaborative Perinatal Project sera (20). Between any pair of serial measures of lipid-adjusted PCBs, such as the first trimester and the third trimester specimens, the correlations were high (Pearson's r = 0.77 or above). Because more subjects had a third trimester serum specimen available than for any other period, third trimester samples were analyzed. Levels of PCBs in maternal serum and in umbilical cord serum correlate fairly well; for example, Spearman's rs ranged from 0.52 to 0.74 for four congeners (21). The 3-ml serum samples were analyzed for 11 specific PCB congeners (denoted by their International Union of Pure and Applied Chemistry (IUPAC) (22) designation: 28, 52, 74, 105, 118, 138, 153, 170, 180, 194, and 203) and nine other organochlorines at the Centers for Disease Control and Prevention by use of solid-phase extraction, followed by dual-column gas chromatographic separation with electron capture detection (23). Measured concentrations that were below the detection limit were reported by the laboratory and included in the analyses; in such instances, imputation was not done (2426). In the primary analysis, PCB exposure was characterized using the sum of the 11 measured PCB congeners, expressed as μg/liter of serum (“wet weight”). Serum cholesterol and triglycerides were measured, using standard enzymatic assays.

We included in every batch of specimens analyzed (10 specimens/batch) an aliquot from a single large pool of serum to allow an independent calculation of a coefficient of variation between batches (19.0 percent for total PCBs; mean, 3.54 μg/liter). Fifty-three percent of batches contained specimens for both subjects selected at random and those selected because of their IQ at age 7 years; all batches that included a specimen for a subject selected because of his/her IQ score had at least one specimen for a subject selected at random.

Outcome measures

Wechsler Intelligence Scale for Children.

Children were given seven of the 11 components of the 1949 WISC (information, comprehension, vocabulary, digit span, picture arrangement, block design, and coding) (27) at the 7-year visit. A verbal IQ score was based on results from the information, comprehension, vocabulary, and digit span tests, and a performance score was based on results from the picture arrangement, block design, and coding tests. The full-scale score was based on the results from all seven of the components. For this analysis, the age-adjusted standardized test scores calculated by the original investigators were used.

The Collaborative Perinatal Project investigators used great care to ensure the comparability of testing procedures among study centers. For each battery of tests, a few examiners from each center traveled to another center and readministered the test to a randomly selected subset of children (28). The correlation between the scores of two IQ tests administered 3 months apart by examiners from different study centers was high (r = 0.83). Additional information regarding the validity and reliability of the WISC has been reported elsewhere (29).

Assessment of related outcomes.

Although our analyses focused on the WISC IQ results, other related outcome data were also available. At the 4-year examination, the Stanford-Binet IQ test was administered (28). At the 7-year visit, children were also given the Wide Range Achievement Test (WRAT) (30). The WRAT has three subtests: spelling, reading, and arithmetic. Age-standardized WRAT scores were used in the analysis.

Data analyses

We focused on the 7-year cognitive measures, as compared with similar measures at age 4 years or achievement at age 7 years, because the persistent effects of PCB exposure on cognition were of the greatest interest to us, and because cognition has been assessed in most of the other studies of early PCB effects. For PCB congeners 28, 74, 118, 138, and 180, there were two, 23, 41, 20, and three subjects, respectively, for whom the level of the congener was missing, mainly because the measured value did not meet the quality control standards for acceptance (22). Using regression models fit to subjects with complete data, we imputed values for the missing congeners. The levels of specific congeners are highly correlated in background-exposed populations (31, 32).

Subjects were divided into five categories on the basis of in utero PCB exposure, as reflected by the maternal serum total PCB level. Cutpoints at 1.25-μg/liter intervals gave at least 80 subjects in the category with the fewest subjects in the crude analyses. The mean IQ showed an essentially monotonic change across categories, suggesting that representing PCBs as a continuous, linear variable adequately reflected the PCB–IQ relation. We also fit quadratic models to assess whether a linear model was sufficient; all results indicated that it was.

All models in the analysis used weights. The weights were the inverse of the sampling probabilities (33). Subjects in the random sample had a weight of 35; those in the special sample with low IQ (n = 61) were given a weight of 57, and those in the high-IQ sample were given a weight of 43 (n = 101). (The numbers of subjects presented here are the number included in the final model, described below.) Use of the weights avoided bias that would have occurred had the outcome-dependent sampling been ignored in the analysis (19). Models were fit using Proc GLM and Proc REG in SAS software (34), with normalized weights.

Study center (12 categories), maternal age, serum cholesterol, and triglycerides (all continuous), as well as child's sex, were a priori considered to be potentially confounding factors and were included in all multivariate models. Many other factors were considered as potentially confounding, and these are listed in appendix table 1. To evaluate confounding in a model of IQ, we compared the coefficient for PCB as a continuous variable, adjusted for the a priori confounders, with the PCB coefficient from a model that additionally included each of the other potential confounders one at a time. Any variable whose inclusion changed the PCB coefficient by 10 percent or more was considered a confounder. The confounders so identified were maternal race, parity, education, socioeconomic index, housing density, smoking status, serum level of heptachlor epoxide, socioeconomic index at the 7-year follow-up, maternal or caregiver education at the 7-year follow-up, maternal or family income at the 7-year follow-up, whether the home emotional environment at the 4-year follow-up was favorable, age in months at the time of the 7-year follow-up, whether meconium was present at birth, and whether the child was breastfed during the hospitalization for delivery. In the fully adjusted model, the coefficient for PCBs changed by less than 3 percent when we deleted maternal education at registration, maternal/caregiver income at 7 years, and housing density; thus, these variables were not included. Effect modification was evaluated similarly, but by examining model fit statistics before and after the addition of the cross-product terms for all the variables listed above. Any term(s) giving an F test with a p value of ≤0.10 were considered potentially noteworthy. Although our focus was on PCB levels expressed on a wet-weight basis (17, p. 321), we also fitted models with PCB levels expressed on a per unit of serum lipid basis (without cholesterol or triglycerides as covariates).

Of the 1,463 total subjects selected for inclusion in this study (selected at random: n = 1,256; IQ sample: n = 207), 321 of those selected at random had no IQ data, 39 had a missing PCB value, one had a missing value for an a priori confounder, and 208 had a missing value for a covariate included in the fully adjusted model. Thus, data for 894 subjects were included in the final IQ analysis; slightly fewer were in the Stanford-Binet and WRAT analyses because of missing data for those outcomes. Confounding and effect modification were evaluated in models with equal numbers of subjects. The 894 subjects included six children (three pairs of siblings) who were from separate pregnancies from three mothers.

RESULTS

Of the 894 subjects in the final analysis, 49 percent were White, 47 percent were Black, and 4 percent were of other race. The latter subjects were grouped with non-Whites in table 1, although grouping them with Whites did not alter the interpretation.

TABLE 1.

Selected characteristics of the women and children in the low-IQ* sample (child's IQ ≤ 82), high-IQ sample (child's IQ ≥ 111), and random sample, by racial group, Collaborative Perinatal Project, United States, 1959–1972,


 

Low-IQ sample (n = 61)
 

Random sample
 
   
High-IQ sample (n = 101)
 

 
 All subjects (n = 732)
 
Low IQ (n = 111)
 
Moderate IQ (n = 520)
 
High IQ (n = 101)
 
 
Maternal characteristics at registration       
    Racial group (no., %)       
        Non-White 53 (86.9) 391 (53.4) 90 (81.1) 285 (54.8) 16 (15.8) 13 (12.9) 
        White 8 (13.1) 341 (46.6) 21 (18.9) 235 (45.2) 85 (84.2) 88 (87.1) 
    Age (years)       
        Non-White 24.0 (20, 29) 22 (19, 27) 22 (19, 27) 22 (19, 27) 25 (22, 31) 26 (22, 30) 
        White 22.5 (20, 25.5) 24 (21, 30) 24 (20, 32) 24 (20, 30) 24 (21, 29) 23 (21, 27) 
    Parity       
        Non-White 2 (1, 4) 2 (0, 3) 1 (0, 3) 1 (0, 3) 3 (1, 4) 1 (0, 2) 
        White 1.5 (0, 2.5) 1 (0, 3) 3.0 (1, 5) 1 (0, 3) 1 (0, 2) 1 (0, 2) 
    Socioeconomic index       
        Non-White 3.0 (2.3, 4.5) 4.0 (2.7, 5.3) 3.7 (2.7, 4.7) 4.0 (3.0, 5.3) 3.4 (2.0, 5.7) 6.3 (5.0, 8.0) 
        White 4.9 (2.8, 7.0) 5.7 (4.0, 7.3) 4.3 (3.0, 6.0) 5.3 (4.0, 7.0) 7.0 (5.7, 8.7) 7.7 (5.7, 9.0) 
    Current smoker (%)       
        Non-White 41.5 42.7 44.4 41.8 50.0 53.9 
        White 75.0 51.0 47.6 57.5 34.1 38.6 
    Triglycerides (mg/dl)       
        Non-White 194 (146, 224) 182 (147, 234) 183 (144, 228) 182 (148, 235) 176 (140, 214) 175 (137, 214) 
        White 233 (215, 275) 222 (175, 284) 226 (157, 318) 223 (174, 283) 218 (180, 284) 198 (168, 264) 
    Cholesterol (mg/dl)       
        Non-White 203 (172, 254) 217 (189, 255) 216 (195, 254) 217 (187, 255) 223 (199, 259) 234 (217, 277) 
        White 250 (205, 287) 250 (212, 288) 245 (217, 288) 245 (207, 286) 262 (233, 297) 247 (217, 278) 
    Total imputed PCBs* (μg/liter)       
        Non-White 2.49 (1.74, 3.43) 2.87 (1.97, 4.03) 2.72 (1.77, 4.00) 2.87 (2.00, 3.98) 3.37 (2.10, 4.87) 4.19 (2.02, 4.73) 
        White 3.27 (2.61, 4.52) 2.78 (1.99, 3.75) 1.92 (1.45, 2.53) 2.65 (1.94, 3.46) 3.30 (2.45, 4.57) 3.09 (2.26, 4.81) 
    Heptachlor epoxide (μg/liter)       
        Non-White 0.43 (0.23, 0.59) 0.56 (0.32, 0.86) 0.54 (0.31, 0.83) 0.56 (0.33, 0.86) 0.73 (0.33, 1.01) 0.46 (0.29, 0.51) 
        White 0.23 (0.20, 0.36) 0.29 (0.21, 0.40) 0.27 (0.22, 0.39) 0.28 (0.21, 0.38) 0.31 (0.22, 0.47) 0.32 (0.22, 0.43) 
Maternal characteristics at 7-year follow-up       
    Socioeconomic index       
        Non-White 2.9 (2.0, 4.1) 4.0 (2.6, 5.5) 3.4 (2.3, 4.7) 4.3 (2.8, 5.6) 4.6 (2.6, 6.2) 5.8 (3.3, 8.1) 
        White 2.7 (1.3, 6.6) 5.9 (4.1, 7.9) 4.0 (2.8, 4.7) 5.5 (3.7, 7.1) 8.0 (6.2, 9.4) 8.4 (6.1, 9.3) 
    Maternal/caregiver education (years)       
        Non-White 10 (8, 12) 11 (10, 12) 10 (9, 11) 11 (10, 12) 12 (10, 12) 12 (12, 12) 
        White 9 (8, 11) 12 (10, 12) 10 (9, 12) 12 (10, 12) 12 (12, 16) 13 (12, 16) 
Child's characteristics at delivery       
    Sex: male (%)       
        Non-White 54.7 47.8 54.4 46.0 43.8 46.2 
        White 37.5 49.9 61.9 47.7 52.9 59.1 
    Breastfed 1 or more days (%)       
        Non-White 3.8 6.7 2.2 8.1 6.3 23.1 
        White 12.5 26.1 14.3 19.6 47.1 52.3 
Child's characteristics at 4-year follow-up       
    Home environment unfavorable (%)       
        Non-White 9.4 5.4 3.3 5.6 12.5 7.7 
        White 12.5 10.9 14.3 11.5 8.2 9.1 
Child's characteristics at 7-year follow-up       
    Age (years)       
        Non-White 7 (6.92, 7.17) 7 (6.92, 7.17) 7 (6.92, 7.17) 7 (6.92, 7.08) 7 (7.00, 7.17) 7 (6.92, 7.00) 
        White 7 (6.96, 7.00) 7 (6.92, 7.08) 7 (6.92, 7.08) 7 (6.92, 7.08) 7 (7.00, 7.17) 7 (6.92, 7.08) 
Delivery complications       
    Meconium condition present (%)       
        Non-White 15.1 20.5 23.3 19.7 18.8 23.1 
        White
 
0.0
 
19.7
 
23.8
 
21.3
 
14.1
 
17.1
 

 

Low-IQ sample (n = 61)
 

Random sample
 
   
High-IQ sample (n = 101)
 

 
 All subjects (n = 732)
 
Low IQ (n = 111)
 
Moderate IQ (n = 520)
 
High IQ (n = 101)
 
 
Maternal characteristics at registration       
    Racial group (no., %)       
        Non-White 53 (86.9) 391 (53.4) 90 (81.1) 285 (54.8) 16 (15.8) 13 (12.9) 
        White 8 (13.1) 341 (46.6) 21 (18.9) 235 (45.2) 85 (84.2) 88 (87.1) 
    Age (years)       
        Non-White 24.0 (20, 29) 22 (19, 27) 22 (19, 27) 22 (19, 27) 25 (22, 31) 26 (22, 30) 
        White 22.5 (20, 25.5) 24 (21, 30) 24 (20, 32) 24 (20, 30) 24 (21, 29) 23 (21, 27) 
    Parity       
        Non-White 2 (1, 4) 2 (0, 3) 1 (0, 3) 1 (0, 3) 3 (1, 4) 1 (0, 2) 
        White 1.5 (0, 2.5) 1 (0, 3) 3.0 (1, 5) 1 (0, 3) 1 (0, 2) 1 (0, 2) 
    Socioeconomic index       
        Non-White 3.0 (2.3, 4.5) 4.0 (2.7, 5.3) 3.7 (2.7, 4.7) 4.0 (3.0, 5.3) 3.4 (2.0, 5.7) 6.3 (5.0, 8.0) 
        White 4.9 (2.8, 7.0) 5.7 (4.0, 7.3) 4.3 (3.0, 6.0) 5.3 (4.0, 7.0) 7.0 (5.7, 8.7) 7.7 (5.7, 9.0) 
    Current smoker (%)       
        Non-White 41.5 42.7 44.4 41.8 50.0 53.9 
        White 75.0 51.0 47.6 57.5 34.1 38.6 
    Triglycerides (mg/dl)       
        Non-White 194 (146, 224) 182 (147, 234) 183 (144, 228) 182 (148, 235) 176 (140, 214) 175 (137, 214) 
        White 233 (215, 275) 222 (175, 284) 226 (157, 318) 223 (174, 283) 218 (180, 284) 198 (168, 264) 
    Cholesterol (mg/dl)       
        Non-White 203 (172, 254) 217 (189, 255) 216 (195, 254) 217 (187, 255) 223 (199, 259) 234 (217, 277) 
        White 250 (205, 287) 250 (212, 288) 245 (217, 288) 245 (207, 286) 262 (233, 297) 247 (217, 278) 
    Total imputed PCBs* (μg/liter)       
        Non-White 2.49 (1.74, 3.43) 2.87 (1.97, 4.03) 2.72 (1.77, 4.00) 2.87 (2.00, 3.98) 3.37 (2.10, 4.87) 4.19 (2.02, 4.73) 
        White 3.27 (2.61, 4.52) 2.78 (1.99, 3.75) 1.92 (1.45, 2.53) 2.65 (1.94, 3.46) 3.30 (2.45, 4.57) 3.09 (2.26, 4.81) 
    Heptachlor epoxide (μg/liter)       
        Non-White 0.43 (0.23, 0.59) 0.56 (0.32, 0.86) 0.54 (0.31, 0.83) 0.56 (0.33, 0.86) 0.73 (0.33, 1.01) 0.46 (0.29, 0.51) 
        White 0.23 (0.20, 0.36) 0.29 (0.21, 0.40) 0.27 (0.22, 0.39) 0.28 (0.21, 0.38) 0.31 (0.22, 0.47) 0.32 (0.22, 0.43) 
Maternal characteristics at 7-year follow-up       
    Socioeconomic index       
        Non-White 2.9 (2.0, 4.1) 4.0 (2.6, 5.5) 3.4 (2.3, 4.7) 4.3 (2.8, 5.6) 4.6 (2.6, 6.2) 5.8 (3.3, 8.1) 
        White 2.7 (1.3, 6.6) 5.9 (4.1, 7.9) 4.0 (2.8, 4.7) 5.5 (3.7, 7.1) 8.0 (6.2, 9.4) 8.4 (6.1, 9.3) 
    Maternal/caregiver education (years)       
        Non-White 10 (8, 12) 11 (10, 12) 10 (9, 11) 11 (10, 12) 12 (10, 12) 12 (12, 12) 
        White 9 (8, 11) 12 (10, 12) 10 (9, 12) 12 (10, 12) 12 (12, 16) 13 (12, 16) 
Child's characteristics at delivery       
    Sex: male (%)       
        Non-White 54.7 47.8 54.4 46.0 43.8 46.2 
        White 37.5 49.9 61.9 47.7 52.9 59.1 
    Breastfed 1 or more days (%)       
        Non-White 3.8 6.7 2.2 8.1 6.3 23.1 
        White 12.5 26.1 14.3 19.6 47.1 52.3 
Child's characteristics at 4-year follow-up       
    Home environment unfavorable (%)       
        Non-White 9.4 5.4 3.3 5.6 12.5 7.7 
        White 12.5 10.9 14.3 11.5 8.2 9.1 
Child's characteristics at 7-year follow-up       
    Age (years)       
        Non-White 7 (6.92, 7.17) 7 (6.92, 7.17) 7 (6.92, 7.17) 7 (6.92, 7.08) 7 (7.00, 7.17) 7 (6.92, 7.00) 
        White 7 (6.96, 7.00) 7 (6.92, 7.08) 7 (6.92, 7.08) 7 (6.92, 7.08) 7 (7.00, 7.17) 7 (6.92, 7.08) 
Delivery complications       
    Meconium condition present (%)       
        Non-White 15.1 20.5 23.3 19.7 18.8 23.1 
        White
 
0.0
 
19.7
 
23.8
 
21.3
 
14.1
 
17.1
 
*

IQ, intelligence quotient; PCB, polychlorinated biphenyl.

Total number of subjects = 894.

Values are expressed as the median and quartiles, unless otherwise noted. The level of IQ used to assign subjects to the “low” groups was the same for the low-IQ sample and for the random-sample, low-IQ group. The same was true for those in the corresponding “high” groups.

The majority of subjects in our study are described in the column labeled “all subjects” under “random sample” in table 1. The median age at registration was in the early twenties, and the median education was less than high school. The median socioeconomic index among participants was close to the US median of 5.0 at the time of the Collaborative Perinatal Project. Thirty-one percent of mothers were nulliparous at registration. Roughly half of the mothers were current smokers at registration. At delivery, the presence of meconium was noted for nearly 20 percent of infants. A small proportion of children were breastfed during the hospitalization for delivery, and thus the proportion that was breastfed subsequently would have been even less. The home environment was deemed unfavorable for about 5–10 percent of the children at the 4-year follow-up. At the 7-year follow-up, the median age of the children was 7.0 years (quartiles: 6.9, 7.2), and the median grade in school was close to second. The values for maternal socioeconomic status and education at the 7-year follow-up were similar to the values at registration.

The distribution of race across the “low IQ sample,” “random sample,” and “high IQ sample” reflects the previously described association between race and IQ score (29). Because of that association, we stratified by race for all the other factors in table 1, to allow a more straightforward evaluation of the relation of other factors to IQ in our data. Within each racial group, IQ showed clear relations with maternal education, socioeconomic index, and breastfeeding by the child.

For PCB congeners 118, 138, and 153, 90 percent or more of the subjects had levels that were above the detection limit for the assay (table 2). The overall median PCB level was 2.85 μg/liter (quartiles: 2.00, 4.02), similar to that in most other studies of PCBs and cognitive function (35). Women from Richmond had the highest median total PCB level (3.6 μg/liter), while those from Portland had the lowest (1.7 μg/liter).

TABLE 2.

Third trimester maternal PCB* serum levels among a sample of enrolled women (n = 894), Collaborative Perinatal Project, United States, 1959–1972


IUPAC*-designated PCB congener no.
 

Limit of detection (μg/liter)
 

PCB serum levels ≥ limit of detection
 
 
Concentration (μg/liter)
 
   
  %
 
No.
 
25th
 
Median
 
75th
 
95th
 
28 0.27 24.0 892 0.12 0.18 0.27 0.45 
52 0.27 2.0 894 0.00 0.00 0.00 0.15 
74 0.20 65.1 871 0.17 0.26 0.35 0.66 
105 0.21 18.1 894 0.05 0.11 0.18 0.38 
118 0.25 90.0 853 0.38 0.56 0.83 1.84 
138 0.25 94.9 874 0.43 0.60 0.85 1.39 
153 0.33 91.4 894 0.47 0.65 0.92 1.47 
170 0.22 8.2 894 0.00 0.10 0.16 0.26 
180 0.23 49.9 891 0.15 0.23 0.33 0.54 
194 0.21 1.3 894 0.00 0.00 0.09 0.16 
203 0.20 2.8 894 0.00 0.00 0.09 0.18 
Total PCBs§  45.3 894 2.00 2.85 4.02 6.75 

 

 
44.1#
 
894#
 
1.97
 
2.83
 
3.97
 
6.75
 

IUPAC*-designated PCB congener no.
 

Limit of detection (μg/liter)
 

PCB serum levels ≥ limit of detection
 
 
Concentration (μg/liter)
 
   
  %
 
No.
 
25th
 
Median
 
75th
 
95th
 
28 0.27 24.0 892 0.12 0.18 0.27 0.45 
52 0.27 2.0 894 0.00 0.00 0.00 0.15 
74 0.20 65.1 871 0.17 0.26 0.35 0.66 
105 0.21 18.1 894 0.05 0.11 0.18 0.38 
118 0.25 90.0 853 0.38 0.56 0.83 1.84 
138 0.25 94.9 874 0.43 0.60 0.85 1.39 
153 0.33 91.4 894 0.47 0.65 0.92 1.47 
170 0.22 8.2 894 0.00 0.10 0.16 0.26 
180 0.23 49.9 891 0.15 0.23 0.33 0.54 
194 0.21 1.3 894 0.00 0.00 0.09 0.16 
203 0.20 2.8 894 0.00 0.00 0.09 0.18 
Total PCBs§  45.3 894 2.00 2.85 4.02 6.75 

 

 
44.1#
 
894#
 
1.97
 
2.83
 
3.97
 
6.75
 
*

PCB, polychlorinated biphenyl; IUPAC, International Union of Pure and Applied Chemistry.

The values shown for individual congeners are before imputation. Measured concentrations that were below the detection limit were reported by the laboratory and included in the analyses; in such instances, imputation was not done. When the measured concentration was missing, however, imputation was done.

Number of subjects with nonmissing value for that congener before imputation.

§

Sum of the 11 specific PCB congeners shown for total PCBs.

Imputed.

#

Unimputed.

Across categories of PCB level, the fully adjusted mean IQ increased (figure 1). Results for the verbal and performance subscales and their individual components (not shown) showed a comparable association. Accordingly, coefficients from models with PCB as a continuous variable were positive, though they were near zero (table 3). The confounders that changed the association the most were center, race, and grade level at 7 years. Results for the WRAT were similar to those for the WISC. When PCB levels were expressed on a lipid-weight basis, the results were similar and reflected the change in exposure scale. We repeated these analyses using the 4-year Stanford-Binet IQ test score as the outcome. The findings were similar to those shown for age 7 years, but the association was slightly stronger and was statistically significant. Specifically, the 4-year IQ–PCB fully adjusted coefficient was 1.03 IQ units/μg of PCB/liter (standard error: 0.29; n = 758); lipid-weight basis results were similar (7.84 IQ units/μg of PCB/g of lipid; standard error: 2.15).

FIGURE 1.

Fully adjusted mean (standard error) Wechsler Intelligence Scale for Children (WISC) full-scale intelligence quotient (IQ) score at age 7 years according to maternal serum polychlorinated biphenyl (PCB) level during pregnancy, Collaborative Perinatal Project, United States, 1959–1972. The number of subjects in each category of PCB level is 55, 299, 287, 128, and 125, respectively (total n = 894).

FIGURE 1.

Fully adjusted mean (standard error) Wechsler Intelligence Scale for Children (WISC) full-scale intelligence quotient (IQ) score at age 7 years according to maternal serum polychlorinated biphenyl (PCB) level during pregnancy, Collaborative Perinatal Project, United States, 1959–1972. The number of subjects in each category of PCB level is 55, 299, 287, 128, and 125, respectively (total n = 894).

TABLE 3.

Change in cognitive test scores at age 7 years per unit increase in maternal serum PCB level, by methods of expressing PCBs, by type of test, and according to degree of adjustment, Collaborative Perinatal Project, United States, 1959–1972


Cognitive test
 

Score (mean (SD))
 

Crude estimate (SE)
 

Adjusted estimate (SE)
 

Fully adjusted estimate (SE)§
 
PCBs expressed on a wet-weight basis     
    WISC (7 year)     
        Full scale 97.0 (16.3) 1.31 (0.30)* 0.61 (0.27)* 0.22 (0.25) 
        Verbal 95.5 (15.6) 1.24 (0.28)* 0.50 (0.26) 0.15 (0.24) 
        Performance 99.3 (16.8) 1.16 (0.30)* 0.61 (0.30)* 0.24 (0.28) 
    WRAT (7 year) standard score     
        Spelling 96.9 (13.7) 1.12 (0.25)* 0.43 (0.24) 0.18 (0.23) 
        Reading 99.6 (15.9) 1.32 (0.28)* 0.68 (0.28)* 0.40 (0.26) 
        Arithmetic 96.6 (12.0) 1.07 (0.22)* 0.45 (0.22)* 0.22 (0.21) 
PCBs expressed on a lipid-weight basis#     
    WISC (7 year)     
        Full scale 97.0 (16.3) 6.81 (2.34)* 4.80 (2.09)* 1.90 (1.92) 
        Verbal 95.5 (15.6) 6.28 (2.22)* 3.71 (2.01) 1.20 (1.84) 
        Performance 99.3 (16.8) 6.12 (2.41)* 4.99 (2.25)* 2.17 (2.15) 
    WRAT (7 year)     
        Spelling 96.9 (13.7) 6.31 (1.95)* 3.21 (1.85) 1.30 (1.75) 
        Reading 99.6 (15.9) 8.09 (2.23)* 5.13 (2.12)* 3.25 (1.99) 
        Arithmetic
 
96.6 (12.0)
 
6.72 (1.71)*
 
3.15 (1.68)
 
1.57 (1.59)
 

Cognitive test
 

Score (mean (SD))
 

Crude estimate (SE)
 

Adjusted estimate (SE)
 

Fully adjusted estimate (SE)§
 
PCBs expressed on a wet-weight basis     
    WISC (7 year)     
        Full scale 97.0 (16.3) 1.31 (0.30)* 0.61 (0.27)* 0.22 (0.25) 
        Verbal 95.5 (15.6) 1.24 (0.28)* 0.50 (0.26) 0.15 (0.24) 
        Performance 99.3 (16.8) 1.16 (0.30)* 0.61 (0.30)* 0.24 (0.28) 
    WRAT (7 year) standard score     
        Spelling 96.9 (13.7) 1.12 (0.25)* 0.43 (0.24) 0.18 (0.23) 
        Reading 99.6 (15.9) 1.32 (0.28)* 0.68 (0.28)* 0.40 (0.26) 
        Arithmetic 96.6 (12.0) 1.07 (0.22)* 0.45 (0.22)* 0.22 (0.21) 
PCBs expressed on a lipid-weight basis#     
    WISC (7 year)     
        Full scale 97.0 (16.3) 6.81 (2.34)* 4.80 (2.09)* 1.90 (1.92) 
        Verbal 95.5 (15.6) 6.28 (2.22)* 3.71 (2.01) 1.20 (1.84) 
        Performance 99.3 (16.8) 6.12 (2.41)* 4.99 (2.25)* 2.17 (2.15) 
    WRAT (7 year)     
        Spelling 96.9 (13.7) 6.31 (1.95)* 3.21 (1.85) 1.30 (1.75) 
        Reading 99.6 (15.9) 8.09 (2.23)* 5.13 (2.12)* 3.25 (1.99) 
        Arithmetic
 
96.6 (12.0)
 
6.72 (1.71)*
 
3.15 (1.68)
 
1.57 (1.59)
 
*

p ≤ 0.05.

PCB, polychlorinated biphenyl; SD, standard deviation; SE, standard error; WISC, Wechsler Intelligence Scale for Children (n = 894); WRAT, Wide Range Achievement Test: for spelling (n = 890); for reading (n = 883); for arithmetic (n = 890).

Regressions on wet-weight PCBs were adjusted for study center, serum total cholesterol and triglycerides, maternal age at registration, and gender of the child. Regressions on lipid-basis PCBs were adjusted for study center, maternal age at registration, and gender of the child.

§

Regressions on wet-weight PCBs were adjusted for study center, serum total cholesterol and triglycerides, maternal age at registration, gender of the child, race, parity, socioeconomic index at registration, smoking status at registration, number of days breastfed, presence of meconium, socioeconomic index at 7-year follow-up, educational level of mother/caregiver at 7-year follow-up, presence of an unfavorable emotional environment at 7-year follow-up, age of child at 7-year follow-up, and heptachlor epoxide level. Regressions on lipid-based PCBs were adjusted for study center, maternal age at registration, gender of the child, race, parity, socioeconomic index at registration, smoking status at registration, number of days breastfed, presence of meconium, socioeconomic index at 7-year follow-up, educational level of mother/caregiver at 7-year follow-up, presence of an unfavorable emotional environment at 7-year follow-up, age of child at 7-year follow-up, and heptachlor epoxide level.

Units for PCBs are μg/liter.

#

Units for PCBs are μg/g of lipid. The median was 0.35 μg/g, and the quartiles were 0.25 and 0.75.

Our evaluation of effect modification identified no cross-product terms of potential interest. Results stratified by race were essentially the same as those shown. The study center-specific coefficients for PCBs (continuous) were negative in some centers, but the precision was low (data not shown).

The studies of in utero PCB exposure showing an inverse relation with cognitive function (3, 4) were done in populations with higher mean IQ and socioeconomic status than in our population. Therefore, we reevaluated our results after excluding subjects 1) with an IQ less than 82 and 2) with a socioeconomic status less than the median. In both cases, the results were essentially unchanged from those shown.

The association of IQ with the level of each PCB congener (continuous, unimputed values only) was examined in separate, fully adjusted models. In addition, after summing the levels of the most highly correlated PCBs (congeners 74, 105, 118, 138, 153, 170, and 180; all rs > 0.6), we included the sum and the levels of the remaining four congeners in one model. In no instance was there an indication that any congener (or group) was especially associated with IQ or that the association differed among congeners.

DISCUSSION

In these data, the background-range prenatal PCB exposure was not associated with lower IQ test scores at age 7 years. Contrary to the hypothesis, the PCB–IQ coefficient we found at age 4 years was positive, possibly because of residual confounding by socioeconomic status.

As noted earlier, Gladen and Rogan (15), who examined McCarthy scores in 712 children aged 3–5 years, found no association between prenatal PCB exposure and cognitive testing results. In another study of 435 subjects from a population with PCB serum levels about three- to fourfold higher than in the Collaborative Perinatal Project (and most other studies of this relation), in utero PCB exposure was not related to WISC scores at age 7 years (36). In the cohort of 353 Dutch children, prenatal PCB exposure was associated with decreased cognitive scores at 3.5 years of age (4), but subsequent testing at age 6.5 years no longer supported this relation (37). In the cohort from upstate New York, higher umbilical cord PCB levels were associated with lower scores on the McCarthy Scales of Children's Abilities at 38 months of age but not at 54 months (n ∼ 195) (14). Only in the two remaining smaller studies of cognition was an unequivocal adverse association with PCB exposure found among the children tested after the longest period of follow-up, at ages 11 (3) and 3 (38) years. Although different tests were used across studies, scores among the tests are highly correlated (29). The ages of subjects also varied across studies, but test results are correlated, especially at age 7 years or more (29). An explanation for why results vary was not obvious, though differences in adjustment for confounding factors may have played a role (39).

The limitations of the present data and their potential impact on our findings merit consideration. First, as a result of loss to follow-up, IQ data were missing for 321 of 1,463 subjects (22 percent). The median PCB level among those with missing IQ data (2.92 μg/liter) was similar to that for the subjects included in the final analysis (2.85 μg/liter), suggesting that follow-up was unrelated to exposure. In addition, the 208 subjects missing data for covariates were excluded. The β coefficient for the adjusted IQ–PCB relation was similar in those with and without the covariate information (not shown), suggesting that limiting the analysis to those with complete covariate data did not substantially alter the findings.

Second, the sera used to ascertain exposure had been stored since the early 1960s. Despite the widely held view that PCBs resist degradation in most settings (40), few data exist regarding stability in serum during long-term frozen storage. However, pooled milk specimens from Swedish mothers, stored in 1972 at −20°C, were analyzed for PCBs after 15 and 25 years; levels showed no decline over the 10-year period between analyses (41, 42).

Third, we reported a slightly higher degree of measurement error, with a between-assay coefficient of variation of 19 percent for total PCBs, as compared with the value of 12 percent in one of the few epidemiologic studies for which these types of data were calculated using blinded samples and presented (40). The precision of the measurement of exposure, however, was nonetheless sufficient to detect a modest association with IQ at age 4 years, albeit in the direction opposite of that hypothesized. Additionally, if third trimester serum PCB levels were not the optimal biomarker, as compared with levels measured in cord serum or maternal milk, the cascading effect of relatively small errors at various steps affecting exposure measurement could have attenuated the measure of an effect, if any. Finally, use of a subset of the WISC components could have decreased our power.

In our data, women with higher PCB exposure had higher socioeconomic status (Spearman's r = 0.16) and were more likely to breastfeed their children (r = 0.11). If socioeconomic status and breastfeeding protected children against the adverse effect of PCBs, then failure to adequately adjust for socioeconomic status and breastfeeding could have caused a true adverse effect of PCBs on neurodevelopment to be masked. In additional analyses (not shown) that included only children who were not breastfed (n = 727), adjustment for socioeconomic status (baseline and age 7 years) and maternal education caused the “fully” adjusted estimate (change in full-scale WISC IQ per unit increased in PCB level in μg/liter) to decrease from 0.56 to 0.36. But stratification of those not breastfed according to whether the family's socioeconomic index (at child's age of 7 years) was above or below the median revealed essentially no difference in the PCB–IQ relation, suggesting no adverse effect of PCBs even in the most disadvantaged children.

In developing animals, including primates, PCBs are usually neurotoxic, even at environmental exposure levels (43). While exposure to PCBs themselves could account for the inverse PCB–IQ associations observed in some epidemiologic studies, the possibility exists that PCB levels serve as a biomarker for the level of another neurotoxicant or for an altogether different determinant of IQ.

As noted above, the level of total PCBs among mothers in our study was similar to those in studies in which an adverse effect was found (3, 4, 38). On the other hand, the mixture of PCBs in the Collaborative Perinatal Project specimens was unusual compared with that in other studies (35); for example, the level of PCB 118 was higher than in any other study, and the level of PCB 180 may well have been proportionately lower (20, 36, 44). Levels of unquantitated PCBs may also have varied in biologically significant ways. Thus, if the composition of PCBs affects neurotoxicity, we may have had a relatively benign mixture. However, our examination of congener-specific effects, to the extent that this was possible, did not suggest that any specific PCBs were related to the outcomes.

Manufacture of PCBs was banned in the United States in 1977, and background-level exposure levels appear to be decreasing (35). However, understanding the risks associated with low-level PCB exposure is nonetheless an important ongoing process, because of the possibility that present-day exposure levels have adverse effects (1, 14) and because of environmental management and legal questions in areas of local contamination (45, 46). Recently, several scientists have made the following comments: “The literature on prenatal exposure to PCBs and developmental toxicity is one of the strongest bodies of evidence of human health effects in low-level environmental exposures” (45, p. A187), and “the weight of evidence for PCB effects on neurodevelopment is growing” (1, p. 376). Low-level PCB exposure early in life may well have adverse effects, but the present data indicate that the long-term effects on cognition, as reflected by performance on IQ-type tests, are not so clear.

APPENDIX TABLE 1.

List of variables considered as potentially confounding the polychlorinated biphenyl–intelligence quotient relation, Collaborative Perinatal Project, United States, 1959–1972


Variables ascertained at the time the mother enrolled in the Collaborative Perinatal Project* 
    Race 
    Parity 
    Prepregnancy weight 
    Prepregnancy body mass index 
    Smoking status 
    Education 
    Employment status 
    Total family income 
    Socioeconomic index 
    Housing density 
    Length of gestation at the time of blood draw 
    Serum level of 1,1-dichloro-2,2-bis(p-chlorophenyl)ethylene 
    Serum level of heptachlor epoxide 
Variables ascertained at birth or shortly thereafter 
    Prenatal care index 
    Infection during pregnancy 
    Eclampsia 
    Prolonged rupture of the membranes 
    Arrested labor 
    Ruptured uterus 
    Abruptio placenta 
    Placenta previa 
    Cord pathology 
    Presence of meconium 
    Heart abnormality (infant) 
    Apnea (infant) 
    Hyperbilirubinemia 
    Seizures (infant) 
    Apgar score at 1 minute 
    Polyhydraminos 
    Breastfeeding in nursery 
Variables ascertained when the child was aged 4 years 
    Maternal intelligence quotient 
    Whether the home emotional environment was favorable 
    Preschool attendance 
    Age at 4-year psychological examination, in months 
Variables ascertained when the child was aged 7 years 
    Maternal or caregiver education 
    Maternal or caregiver employment status 
    Maternal marital status 
    Socioeconomic index 
    Maternal or family income 
    Age at 7-year psychological examination, in months
 

Variables ascertained at the time the mother enrolled in the Collaborative Perinatal Project* 
    Race 
    Parity 
    Prepregnancy weight 
    Prepregnancy body mass index 
    Smoking status 
    Education 
    Employment status 
    Total family income 
    Socioeconomic index 
    Housing density 
    Length of gestation at the time of blood draw 
    Serum level of 1,1-dichloro-2,2-bis(p-chlorophenyl)ethylene 
    Serum level of heptachlor epoxide 
Variables ascertained at birth or shortly thereafter 
    Prenatal care index 
    Infection during pregnancy 
    Eclampsia 
    Prolonged rupture of the membranes 
    Arrested labor 
    Ruptured uterus 
    Abruptio placenta 
    Placenta previa 
    Cord pathology 
    Presence of meconium 
    Heart abnormality (infant) 
    Apnea (infant) 
    Hyperbilirubinemia 
    Seizures (infant) 
    Apgar score at 1 minute 
    Polyhydraminos 
    Breastfeeding in nursery 
Variables ascertained when the child was aged 4 years 
    Maternal intelligence quotient 
    Whether the home emotional environment was favorable 
    Preschool attendance 
    Age at 4-year psychological examination, in months 
Variables ascertained when the child was aged 7 years 
    Maternal or caregiver education 
    Maternal or caregiver employment status 
    Maternal marital status 
    Socioeconomic index 
    Maternal or family income 
    Age at 7-year psychological examination, in months
 
*

Or later in pregnancy, where noted.

Conflict of interest: none declared.

References

1.
Schantz SL, Widholm JJ, Rice DC. Effects of PCB exposure on neuropsychological function in children.
Environ Health Perspect
 
2003
;
111
:
357
–76.
2.
Rogan WJ, Gladen BC. PCBs, DDE, and child development at 18 and 24 months.
Ann Epidemiol
 
1991
;
1
:
407
–13.
3.
Jacobson JL, Jacobson SW. Intellectual impairment in children exposed to polychlorinated biphenyls in utero.
N Engl J Med
 
1996
;
335
:
783
–9.
4.
Patandin S, Lanting CI, Mulder PG, et al. Effects of environmental exposure to polychlorinated biphenyls and dioxins on cognitive abilities in Dutch children at 42 months of age.
J Pediatr
 
1999
;
134
:
33
–41.
5.
Gilbert ME, Liang D. Alterations in synaptic transmission and plasticity in hippocampus by a complex PCB mixture, Aroclor 1254.
Neurotoxicol Teratol
 
1998
;
20
:
383
–9.
6.
Niemi WD, Audi J, Bush B, et al. PCBs reduce long-term potentiation in the CA1 region of rat hippocampus.
Exp Neurol
 
1998
;
151
:
26
–34.
7.
Bemis JC, Seegal RF. Polychlorinated biphenyls and methylmercury act synergistically to reduce rat brain dopamine content in vitro.
Environ Health Perspect
 
1999
;
107
:
879
–85.
8.
Morse DC, Seegal RF, Borsch KO, et al. Long-term alterations in regional brain serotonin metabolism following maternal polychlorinated biphenyl exposure in the rat.
Neurotoxicology
 
1996
;
17
:
631
–8.
9.
Morse DC, Wehler EK, Wesseling W, et al. Alterations in rat brain thyroid hormone status following pre- and postnatal exposure to polychlorinated biphenyls (Aroclor 1254).
Toxicol Appl Pharmacol
 
1996
;
136
:
269
–79.
10.
Schantz SL, Moshtaghian J, Ness DK. Spatial learning deficits in adult rats exposed to ortho-substituted PCB congeners during gestation and lactation.
Fundam Appl Toxicol
 
1995
;
26
:
117
–26.
11.
Levin ED, Schantz SL, Bowman RE. Delayed spatial alternation deficits resulting from perinatal PCB exposure in monkeys.
Arch Toxicol
 
1988
;
62
:
267
–73.
12.
Bushnell PJ, Moser VC, MacPhail RC, et al. Neurobehavioral assessments of rats perinatally exposed to a commercial mixture of polychlorinated biphenyls.
Toxicol Sci
 
2002
;
68
:
109
–20.
13.
Jacobson JL, Jacobson SW, Humphrey HE. Effects of in utero exposure to polychlorinated biphenyls and related contaminants on cognitive functioning in young children.
J Pediatr
 
1990
;
116
:
38
–45.
14.
Stewart PW, Reihman J, Lonky EI, et al. Cognitive development in preschool children prenatally exposed to PCBs and MeHg.
Neurotoxicol Teratol
 
2003
;
25
:
11
–22.
15.
Gladen BC, Rogan WJ. Effects of perinatal polychlorinated biphenyls and dichlorodiphenyl dichloroethene on later development.
J Pediatr
 
1991
;
119
:
58
–63.
16.
Niswander KR, Gordon M. The women and their pregnancies. Washington, DC: US Department of Health, Education, and Welfare,
1972
. (DHEW publication no. (NIH) 73-379).
17.
Longnecker MP, Klebanoff MA, Brock JW, et al. Maternal serum level of 1,1-dichloro-2,2-bis(p-chlorophenyl)ethylene and risk of cryptorchidism, hypospadias, and polythelia among male offspring.
Am J Epidemiol
 
2002
;
155
:
313
–22.
18.
Myrianthopoulos NC, French KS. An application of the U.S. Bureau of Census socioeconomic index to a large, diversified patient population.
Soc Sci Med
 
1968
;
2
:
283
–99.
19.
Zhou H, Weaver MA, Qin J, et al. A semiparametric empirical likelihood method for data from an outcome-dependent sampling scheme with a continuous outcome.
Biometrics
 
2002
;
58
:
413
–21.
20.
Longnecker MP, Klebanoff MA, Gladen BC, et al. Serial levels of serum organochlorines during pregnancy and postpartum.
Arch Environ Health
 
1999
;
54
:
110
–14.
21.
Koopman-Esseboom C, Huisman M, Weisglas-Kuperus N, et al. Dioxin and PCB levels in blood and human milk in relation to living areas in the Netherlands.
Chemosphere
 
1994
;
29
:
2327
–38.
22.
Ballschmiter K, Bacher R, Mennel A, et al. The determination of chlorinated biphenyls, chlorinated dibenzodioxins, and chlorinated dibenzofurans by GC-MS.
J High Resolut Chromatogr
 
1992
;
15
:
260
–70.
23.
Brock JW, Burse VW, Ashley DL, et al. An improved analysis for chlorinated pesticides and polychlorinated biphenyls (PCBs) in human and bovine sera using solid-phase extraction.
J Anal Toxicol
 
1996
;
20
:
528
–36.
24.
Rothman N, Cantor KP, Blair A, et al. A nested case-control study of non-Hodgkin lymphoma and serum organochlorine residues.
Lancet
 
1997
;
350
:
240
–4.
25.
D4210-89 (1996). Standard practice for intralaboratory quality control procedures and a discussion on reporting low-level data (withdrawn 2002). West Conshohocken, PA: ASTM International,
1989
.
26.
Fitzgerald EF, Hwang SA, Langguth K, et al. Fish consumption and other environmental exposures and their associations with serum PCB concentrations among Mohawk women at Akwesasne.
Environ Res
 
2004
;
94
:
160
–70.
27.
Weschler D. Manual for the Weschler Intelligence Scale for Children. San Antonio, TX: Psychological Corporation,
1949
.
28.
Broman SH, Nichols PL, Kennedy WA. Preschool IQ: prenatal and early developmental correlates. Hillsdale, NJ: L Erlbaum Associates,
1975
.
29.
Sattler JM. Assessment of children. 3rd ed. San Diego, CA: Jerome M Sattler, Publisher, Inc,
1992
:125.
30.
Jastak JF, Jastak SR. The Wide Range Achievement Test manual of instructions. Rev ed. Wilmington, DE: Guidance Associates of Delaware,
1965
.
31.
DeVoto E, Fiore BJ, Millikan R, et al. Correlations among human blood levels of specific PCB congeners and implications for epidemiologic studies.
Am J Ind Med
 
1997
;
32
:
606
–13.
32.
Longnecker MP, Ryan JJ, Gladen BC, et al. Correlations among human plasma levels of dioxin-like compounds and polychlorinated biphenyls (PCBs) and implications for epidemiologic studies.
Arch Environ Health
 
2000
;
55
:
195
–200.
33.
Horvitz DG, Thompson DJ. A generalization of sampling without replacement from a finite universe.
J Am Stat Assoc
 
1952
;
47
:
663
–85.
34.
SAS Institute, Inc. SAS (computer program), version 8.0 ed. Cary, NC: SAS Institute, Inc,
1999
.
35.
Longnecker MP, Wolff MS, Gladen BC, et al. Comparison of polychlorinated biphenyl levels across studies of human neurodevelopment.
Environ Health Perspect
 
2003
;
111
:
65
–70.
36.
Grandjean P, Weihe P, Burse VW, et al. Neurobehavioral deficits associated with PCB in 7-year-old children prenatally exposed to seafood neurotoxicants.
Neurotoxicol Teratol
 
2001
;
23
:
305
–17.
37.
Vreugdenhil HJ, Lanting CI, Mulder PG, et al. Effects of prenatal PCB and dioxin background exposure on cognitive and motor abilities in Dutch children at school age.
J Pediatr
 
2002
;
140
:
48
–56.
38.
Walkowiak J, Wiener JA, Fastabend A, et al. Environmental exposure to polychlorinated biphenyls and quality of the home environment: effects on psychodevelopment in early childhood.
Lancet
 
2001
;
358
:
1602
–7.
39.
Mink PJ, Goodman M, Barraj LM, et al. Evaluation of uncontrolled confounding in studies of environmental exposures and neurobehavioral testing in children.
Epidemiology
 
2004
;
15
:
385
–93.
40.
Laden F, Hankinson SE, Wolff MS, et al. Plasma organochlorine levels and the risk of breast cancer: an extended follow-up in the Nurses' Health Study.
Int J Cancer
 
2001
;
91
:
568
–74.
41.
Noren K. Changes in the levels of organochlorine pesticides, polychlorinated biphenyls, dibenzo-p-dioxins and dibenzofurans in human milk from Stockholm, 1972–1985.
Chemosphere
 
1988
;
17
:
39
–49.
42.
Lunden A, Noren K. Polychlorinated naphthalenes and other organochlorine contaminants in Swedish human milk, 1972–1992.
Arch Environ Contam Toxicol
 
1998
;
34
:
414
–23.
43.
Rice DC, Hayward S. Effects of postnatal exposure of monkeys to a PCB mixture on concurrent random interval-random interval and progressive ratio performance.
Neurotoxicol Teratol
 
1999
;
21
:
47
–58.
44.
Jacobson JL, Humphrey HE, Jacobson SW, et al. Determinants of polychlorinated biphenyls (PCBs), polybrominated biphenyls (PBBs), and dichlorodiphenyl trichloroethane (DDT) levels in the sera of young children.
Am J Public Health
 
1989
;
79
:
1401
–4.
45.
Claudio L. The Hudson: a river runs through an environmental controversy.
Environ Health Perspect
 
2002
;
110
:
A184
–7.
46.
Hermanson MH, Scholten CA, Compher K. Variable air temperature response of gas-phase atmospheric polychlorinated biphenyls near a former manufacturing facility.
Environ Sci Technol
 
2003
;
37
:
4038
–42.