Study setting and design. The CHAMACOS study is a community/university partnership. We conducted this longitudinal birth cohort study of predominantly Mexican-American low-income families living in the Salinas Valley, California. This agricultural region is located southeast of the San Francisco Bay Area. Common crops include lettuce, strawberries, artichokes, broccoli, and grapes. About 235,000 kg of OP pesticides were used in this region in 1999–2000, when study participants were pregnant (CDPR 2001). Current data show that use of OP pesticides in California is decreasing overall; however, use of OP pesticides in Monterey County remained steady between 2001 and 2008 but declined 18% from 2008 to 2009 (CDPR 2010).
Detailed methods for the CHAMACOS study have been described elsewhere (Eskenazi et al. 2004
). Briefly, pregnant women were recruited in six community clinics, serving primarily farmworker families, between October 1999 and 2000. Eligible women were ≥ 18 years old, < 20 weeks of gestation, Spanish or English speaking, eligible for low-income health insurance, and planning to deliver at the local public hospital. All study activities were approved by the University of California–Berkeley Committee for the Protection of Human Subjects. Written, informed consent was obtained from the mothers, and child assent was obtained at 7 years of age.
The initial cohort included 601 women who delivered 526 live-born surviving singletons. For the present study, we excluded two children with missing prenatal DAP concentration measurements, four children with a medical condition that would affect assessment (autism, Down syndrome, hydrocephalus, deafness), children who were lost to follow-up and/or did not participate at the 7-year study visit (n = 72 moved, n = 59 refused, n = 24 unable to trace, n = 21 unable to schedule, n = 2 deceased), and children missing the cognitive assessment at the 7-year visit (n = 13). Families included in this analysis (n = 329) did not differ significantly from the original full cohort on most attributes, including urinary DAP concentrations during pregnancy, maternal measures of cognitive ability, maternal education, marital status, poverty category, and child’s birth weight. However, mothers of children included in the present study were slightly older (mean age, 26.7 vs. 26.0 years, p = 0.07) and breast-fed longer (8.7 months vs. 7.2, p = 0.01) than those from the initial cohort.
We used the Wechsler Intelligence Scale for Children, 4th edition (WISC-IV), to assess cognitive abilities at the 7-year study visit (Wechsler 2003
). All assessments were completed by a single experienced bilingual psychometrician, who was trained and supervised by a pediatric neuropsychologist. Quality assurance measures included review of videotaped assessments. For subtests in which a ceiling was not achieved (< 10%), missing values were imputed based on scores obtained by other children with similar score patterns. Scores for four domains were calculated based on the following subtests: Verbal Comprehension (composed of the Vocabulary and Similarities subtests), Perceptual Reasoning (Block Design and Matrix Reasoning subtests), Working Memory (Digit Span and Letter-Number Sequencing subtests), and Processing Speed (Coding and Symbol Search subtests). All subtests were administered in the dominant language of the child, which was determined through administration of the oral vocabulary subtest of the Woodcock–Johnson/Woodcock–Muñoz Tests of Cognitive Ability in both English and Spanish (Woodcock and Johnson 1990
). Ultimately, 67% of children were tested in Spanish and 33% in English. WISC-IV scores are standardized against U.S. population–based norms for English- and Spanish-speaking children.
The numbers of children with available scores were 329 for Perceptual Reasoning and Verbal Comprehension and 298 for Processing Speed and Working Memory (because we did not administer letter-number sequencing and symbol search for the first 3 months of assessments). A Full-Scale intelligence quotient (IQ) was available for 297 children.
Maternal interviews and assessments.
Women were interviewed twice during pregnancy (median gestation, 13 and 26 weeks), after delivery, and when children were 6 months and 1, 2, 3.5, 5, and 7 years of age. Interviews were conducted in Spanish or English by bilingual, bicultural interviewers. At the 6-month visit, mothers were administered the Peabody Picture Vocabulary Test (PPVT) to assess verbal intelligence (Dunn and Dunn 1981
). The Infant-Toddler HOME (Home Observation for Measurement of the Environment) inventory was completed at 6 months and 1 and 2 years of age, and a short version was completed at 3.5 and 5 years (Caldwell and Bradley 1984
). Additional information was obtained from prenatal and delivery medical records, which was abstracted by a registered nurse.
Urinary OP metabolite measurements. Urine was collected at two time points during pregnancy. The first urine sample was collected at enrollment into the study, between 5 and 27 weeks of gestation (median, 13 weeks). The second urine sample was collected between 18 and 39 weeks (median, 26 weeks). Urine was collected from the children at 6 months and 1, 2, 3.5, and 5 years of age; no urine was collected at the 7-year visit.
Urine specimens were aliquoted and stored at −80°C until shipment on dry ice to the Centers for Disease Control and Prevention (CDC; Atlanta, GA) for analysis. Six nonspecific OP DAP metabolites were measured in maternal and child urine: three dimethyl (DM) phosphate metabolites (dimethylphosphate, dimethylthiophosphate, dimethyldithiophosphate) and three diethyl (DE) phosphate metabolites (diethylphosphate, diethylthiophosphate, and diethyldithiophosphate). These six metabolites cannot be traced back to individual pesticides but together represent the breakdown products of about 80% of the total OP pesticides used in the Salinas Valley (CDC 2009
). The most commonly used OP pesticides in the Salinas Valley are chlorpyrifos and diazinon (which devolve to DE), as well as malathion and oxydemeton-methyl (which devolve to DM). DAP metabolite concentrations were measured using gas chromatography/tandem mass spectrometry and quantified using isotope dilution calibration (Bravo et al. 2002
). Details of urine collection, analysis, detection frequencies, and quality control procedures are described elsewhere (Bradman et al. 2005
). Concentrations below the limit of detection (LOD) were randomly imputed based on a log-normal probability distribution whose parameters were estimated using maximum likelihood estimation. This method has been shown to perform better than simple substitution methods such as LOD/2 or LOD/√–
2 (Lubin et al. 2004
). The DAP metabolite concentrations were expressed on a molar basis and summed to yield total DE, DM, and DAP concentrations.
Other environmental contaminants.
We also considered the potential confounding effects of other known or suspected neurotoxicants: polybrominated diphenyl ethers (PBDEs), polychlorinated biphenyls (PCBs), p
´-dichlorodiphenyltrichloroethane (DDT), p
´-dichlorodiphenyldichloroethylene (DDE), and lead. Lead was measured in maternal blood at 26 weeks of gestation, in cord blood for a subset of the participants by the California Department of Public Health (Richmond, CA, USA), and in children’s blood at 2 years of age by the Monterey County Public Health Laboratory (Salinas, CA, USA), using graphite furnace atomic absorption spectrophotometry. PBDEs, PCBs, DDT, and DDE were measured by the CDC (Atlanta, GA) in maternal serum samples collected at 26 weeks of gestation, on average, using gas chromatography/isotope-dilution high-resolution mass spectrometry and were expressed on a serum lipid basis. Total lipids were determined based on the measurement of triglyceride and total cholesterol in serum using standard enzymatic methods (Roche Chemicals, Indianapolis, IN) (Phillips et al. 1989
Data analysis. Nonspecific total DAP, DE, and DM metabolites (nanomoles per liter) were transformed to the log10 scale. All analyses were conducted on non-creatinine-adjusted values; models were rerun with creatinine-adjusted values (nanomoles per gram of creatinine) in sensitivity analyses. We examined the association between urinary DAP concentrations and cognitive scores using multiple linear regression, with point estimates representing the change in cognitive scores for each 10-fold increase in DAP concentrations. For prenatal exposure, we examined associations with the DAP concentrations measured separately for urine collected during the first and second half of pregnancy (≤ 20 vs. > 20 weeks of gestation). Because we found similar relations between cognitive scores and DAPs measured earlier or later than 20 weeks of gestation, we averaged the two DAP measures for further analyses; for 20 children (6%) only one prenatal measure was available for the analyses. Because DM and DE metabolites might have different relationships to the outcomes, they were examined separately.
For postnatal exposure, we examined the cross-sectional association of cognitive scores with DAP concentrations measured in children’s urine collected at different ages in separate models. We also calculated the cumulative DAP level between 6 and 60 months using the area under the curve (AUC), calculated using the trapezoidal method. For 46 children with one missing DAP measurement at 1, 2, or 3.5 years, we imputed the mean of the two measures closest in time for the AUC calculation. Forty-nine children who were missing DAP measures at either the 6-month or 5-year visit, or missing more than one DAP measure from the three other time points, were excluded from the AUC analysis. For comparison with prenatal exposure, we calculated the mean urinary DAP concentrations measured during childhood for children with at least three of five measures (taken at 6 months and 1, 2, 3.5, and 5 years); this excluded 20 children.
To explore possible synergistic effects between pre- and postnatal DAP concentrations, we included an interaction term for mean prenatal DAP concentrations × AUC. However, this term was not statistically significant (p > 0.15) and thus was not included in the final models.
We retained the following variables as covariates for all analyses: maternal intellectual abilities (PPVT score, continuous), maternal education (three categories), and continuous HOME score at 6 months. Maternal intellectual abilities and HOME score were included in models because they were associated with both DAP concentrations and IQ scores in univariate analyses (p
< 0.2), and maternal education was included because it is an important determinant of children’s cognitive development. Language of testing was also included in models for Verbal Comprehension and Full-Scale IQ because of observed language-related differences in scores for these scales. We conducted additional analyses to evaluate the confounding effect of other factors associated with neurodevelopment in the literature: breast-feeding duration (in weeks, continuous), maternal age (continuous), birth order (continuous), HOME score at 1, 2, 3.5, and 5 years (continuous), poverty category (coded as in ), marital status (married or living as married/not married), children’s age at WISC-IV testing (in months, continuous), and maternal levels of PBDEs, PCBs, DDE, DDT, and lead during pregnancy. Each of these variables was added individually to the final model, but none was retained because none changed the magnitude of the coefficient for urinary DAP concentrations by > 10%. In separate analyses, we also investigated potential confounding and effect modification by variables possibly on the causal pathway (i.e., birth weight and gestational age, assessed continuously). Because most children (67%) were tested in Spanish, we reran the analyses restricted to this subset. Finally, we examined the interaction between sex and DAP concentrations, based on previous findings in this cohort (Marks et al. 2010
Study cohort characteristics and maternal urinary DAP concentrations (mean of two measures taken during pregnancy), CHAMACOS (n = 329).
We compared effect estimates for urinary DAPs measured in early versus late pregnancy and in the prenatal versus postnatal periods using seemingly unrelated estimation (Weesie 1999
); we used the mean postnatal DAP concentrations (as opposed to AUC) for these analyses in order to compare metrics with similar units. We used generalized additive models with 3-degree-of-freedom cubic splines to evaluate the shape of dose–response curves, test the linearity assumption, and investigate potential thresholds while controlling for covariates. We did not observe evidence of departure from linearity or threshold for effect, so we retained the simpler models based on linear regression. For illustration, we grouped DAP concentrations into quintiles, entered this categorical variable in the multiple regression model with the same covariables described above, and obtained the mean IQ score for each quintile.
Univariate and multiple linear regression analyses were conducted with SPSS (version 19.0; IBM Corp., Somers, NY), and generalized additive model and seemingly unrelated estimation (“suest” command) were performed with STATA (version 10.1; StataCorp, College Station, TX).