Pesticide Exposure Patterns in Children

Study design. This study is part of the Children’s Pesticide Exposure Study–Washington (CPES-WA) in which 23 children, 3–11 years of age and living in suburban Seattle, Washington, participated in an organic diet substitution study from 2003 to 2004; details have been reported elsewhere (Lu et al. 2008 (link), 2009 (link)). Briefly, the children were recruited from two local public elementary schools and one Montessori preschool. The children participated in consecutive day urine sampling periods in July/August 2003 (median, 15 days; range, 15–16 days); October/November 2003 (median, 12 days; range, 11–13 days); January/February 2004 (median, 7 days; range, 7–8 days); and April/May 2004 (median, 7 days; range, 5–9 days). In the summer and fall sampling periods, an organic diet substitution phase (from day 4 to day 8) was incorporated into the study design to assess the dietary pesticide exposures. For the present study we included only samples collected during the conventional diet portions of the study so that metabolites measured in the urine samples would be representative of typical exposures in the children. Participating children in each session numbered 23 in the summer, 21 in the fall, 20 in the winter, and 19 in the spring. Written informed consent was provided by older children and by the parents of all participants, and oral assent was provided by younger children. The study was approved by the University of Washington Human Subjects Division.
Urine collection and laboratory analysis. Each child provided two urine samples per day: the last void before bedtime and the following first morning void. Previous studies have demonstrated that first voids are good predictors of overall daily exposure for OPs (Kissel et al. 2005 (link)), and in combination with last voids, a large portion of the daily exposure is represented. Additional spot urine samples collected at different times of the day during the study were excluded from the present analysis to increase the comparability of the samples evaluated.
After collection, urine samples were stored on ice or refrigerated before processing in the laboratory and then stored at –20^oC. Samples were analyzed at the National Center for Environmental Health at the Centers for Disease Control and Prevention (CDC; Atlanta, GA) using high performance liquid chromatography–tandem mass spectrometry (Olsson et al. 2004 (link)). Target OP metabolites were malathion dicarboxylic acid (MDA), TCPy, 2-isopropyl-4-methyl-6-hydroxypyrimidinol (IMPy), and 2-diethylamino-6-methyl-pyrmidin-4-ol (DEAMPy). Target pyrethroid insecticide metabolites were 3-phenoxybenzoic acid (PBA), 4-fluoro-3-phenoxybenzoic acid (4F3PBA), cis-2,2-(dichloro)-2-dimethylvinylcyclopropane carboxylic acid (cis-DCCA), trans-2,2-(dichloro)-2-dimethylvinylcyclopropane carboxylic acid (trans-DCCA), and cis-2,2-(dibromo)-2-dimethylvinyl-cyclopropane carboxylic acid (DBCA).
Data analysis. Metabolite concentrations were adjusted for specific gravity to control for dilution using a reference specific gravity of 1.019 g/cm³, the 2007–2008 National Health and Nutrition Examination Survey (NHANES) mean for children 6–11 years of age (CDC 2009 ; Levine and Fahy 1945 ). Because carryover from a previous day could be expected due to the biological half-life (hours to a couple of days) of these compounds, we confirmed that the conventional diet days following the end of the organic diet portions of the original study were not significantly different from other conventional diet days before including them in the analyses (t-test p-value > 0.05).
Intraclass correlation coefficients (ICC), defined as the ratio of between-subject variance to total variance, were calculated as a measure of the reproducibility of measurements over time within individuals. ICCs can range from 0 to 1; ≥ 0.75 indicates excellent reproducibility and ≤ 0.4 indicates poor reproducibility (Rosner 2006 ). Between- and within-subject variances were calculated with a linear mixed effects model using maximum likelihood estimation (MLE) modified to account for values below the limit of detection (LOD) and repeated measurements as implemented in SAS 9.3 (SAS Institute Inc., Cary, NC) using PROC NLMIXED, assuming a compound symmetry covariance structure (Jin et al. 2011 (link)). Age (3–6, 7–11 years), sex, and season were included as covariates in the model:
Ln(Y) = β₀ + β₁(age) + β₂(sex) + β₃(season) + b₁ + ε, [1]
where Y is the metabolite concentration adjusted for specific gravity, b₁ is the between-subject random effect, and ε is the within-subject error.
For metabolites with high percentages of nondetects, the model’s stipulation of a normal distribution of the data was difficult to evaluate. Therefore, we restricted the ICC analyses to the four metabolites (MDA, TCPy, PBA, and trans-DCCA) that were > LOD in > 50% of samples. Data were natural log-transformed before analysis. We performed a sensitivity analysis on the calculation of the ICCs by substituting LOD/2 for values < LOD and using an unmodified NLMIXED procedure. A further sensitivity analysis was used to test the model’s sensitivity to the assumption of equal covariance. The ICC calculations were repeated using a subset of the data with an equal covariance pattern, where only samples with at least 2 intervening days were included in the subset per season.
To address how much exposure misclassification may develop when participants are categorized into exposure groups and how many samples may be necessary to improve the categorization, we performed surrogate category analyses for the first and last voids separately (Hauser et al. 2004 (link); Willet 1998 ) with an additional scoring step (Figure 1). We calculated the geometric mean value of a metabolite across all samples collected from each participant, resulting in 23 participant mean values. Next, we assigned each participant to an exposure quartile (a “surrogate category”) based on the metabolite concentration of a single sample selected at random from each participant’s pool of samples. Then we populated each surrogate category with the children’s geometric means and calculated the group grand means. Then we evaluated the performance of the category assignment. It would not be possible to directly determine whether individuals were correctly assigned according to their “true” and unobserved distribution in the population; however, if surrogate categories were correctly assigned, the mean value of each category should increase monotonically from the lowest to the highest exposure category. If this were the case, we assigned the run a score of 1, and 0 otherwise. Then we repeated the sampling and classification steps 1,000 times and used the mean value of the 1,000 scores (expressed as a percentage) to indicate an average “success rate.” We performed this process three additional times based on the mean value of two, three, and four randomly selected samples for each participant. For this analysis, we substituted instrument-read values (when available), or the LOD/2, as the concentration for all samples with measurements < LOD.