Study population and follow-up.
Pregnant women were recruited for this study at the time of their admittance to the hospital for delivery (Hertz-Picciotto et al. 2003
). Recruitment occurred between 2002 and 2004 in two districts of eastern Slovakia: Michalovce, a region with substantial environmental contamination due to past manufacturing (Kocan et al. 2001
); and Svidnik, a region located approximately 70 km to the northwest, with less environmental contamination and lower serum PCB concentrations among inhabitants (Petrik et al. 2006
). Because each district has only one hospital, the vast majority of births to residents in these two districts occur in these hospitals. We excluded mothers who had more than four previous births, who were < 18 years of age at the time of delivery, who reported residing < 5 years in either district, or who had a major illness during pregnancy. Infants with severe birth defects were also excluded. Trained nursing staff at each hospital explained the study procedures, and all participants gave written informed consent.
All forms of communication were in the Slovak language. In total, 1,134 women were enrolled (811 in Michalovce and 323 in Svidnik). During the hospital stay, each infant underwent an ultrasound examination of the thymus. The study protocol was approved by institutional review boards at the University of California–Davis and the Slovak Medical University in Bratislava.
When the infants were approximately 6 and 16 months of age, mother–infant pairs were invited back to the Department of Pediatrics at their local hospital in either Michalovce or Svidnik, where blood was collected for PCB analysis and infants again underwent ultrasound examinations (described below). At the infants’ age 6 months, 971 mother–infant pairs were still participating in the study (86%), and at age 16 months, 887 (78%) were still participating.
At the time of delivery, two 9-mL Vacutainer tubes were used to collect maternal blood, and at the 6- and 16-month follow-ups, approximately 9 mL of blood was collected from the infant. Details on the processing, transport, and storage of specimens and the isolation of serum have been published elsewhere (Jusko et al. 2010
PCB and lipid measurement.
The concentrations of 15 PCB congeners [International Union of Pure and Applied Chemistry (IUPAC) PCB congeners 28, 52, 101, 105, 114, 118, 123+149, 138+163, 153, 156+171, 157, 167, 170, 180, and 189] were determined in maternal and 6- and 16-month serum samples. The procedure for determination of PCB concentrations involved extraction, cleanup, and quantitation by high-resolution gas chromatography with electron capture detection; specifics of the analysis are detailed elsewhere (Conka et al. 2005
; Jusko et al. 2010
). PCB congeners 123+149, 138+163, and 156+171 coeluted on the gas chromatograph column. The Department of Toxic Organic Pollutants at the Slovak Medical University in Bratislava performed the laboratory analyses. This laboratory has participated in intercalibration studies organized by the World Health Organization (WHO) and the German Agency for Occupational and Environmental Medicine (Deutsche Gesellschaft fuer Arbeitsmedizine und Umweltmedizin e.V.). Total serum lipids were measured at a biochemical laboratory accredited by the Slovak National Accreditation Service located at the Ministry of Defense military hospital in Bratislava, and total lipid concentrations were estimated using the enzymatic summation method (Akins et al. 1989
Maternal PCBs were analyzed for 1,104 women. At age 6 months, 249 infant specimens were selected for analysis based on their corresponding maternal PCB concentration. Specifically, 6-month infant specimens were randomly sampled within strata defined by total maternal PCB concentrations: a
) < 75th percentile, b
) between 75th and 85th percentiles, c
) between 85th and 95th percentiles, and d
) > 95th percentile (Jusko et al. 2010
). At 16 months of age, 831 infant serum specimens were analyzed (all available specimens).
Thymus volume was measured by four radiologists from Michalovce and two radiologists from Svidnik using a sonographic scanner [Esaote 580 FD Caris plus (convex probe 7.5 MHz) in Michalovce; Esaote AU 5 Harmonic (convex probe 7.5 MHz) in Svidnik; both from Esaote, Genoa, Italy]. The radiologists were unaware of maternal and infant PCB concentrations. Measurements of the thymus derived from sonographs are regarded as reliable, especially in young children (Hasselbalch et al. 1996
). A transsternal approach was used to measure the maximal transverse diameter (width) of the thymus, and in the plane perpendicular to this width, the largest sagittal area (longitudinal scan plan) was also measured. These two measurements were multiplied to obtain a “thymic index,” a proxy of thymus volume. The thymic index has been used as a surrogate for thymus volume, because it is strongly correlated (r
≥ 0.80) with the actual volume and weight of the thymus (Hasselbalch et al. 1996
) (hereafter, the thymic index is referred to as thymus volume). Thymus volume was measured in 1,074 newborns, 940 children 6 months of age, and 820 children 16 months of age, for whom 1,020 (95%), 241 (26%), and 806 (98%) had PCB measurements.
Other data collection. Maternal questionnaire. During the 5-day hospital stay after delivery, trained nursing staff administered a questionnaire that obtained information on lifestyle and living environment, past pregnancies and medical conditions, medication use during and before pregnancy, and sociodemographic data, including ethnicity. Ethnicity was categorized into two groups: Slovakian/other eastern European or Romani. Romani ethnicity was attributed to the mother if the ethnic origin of either of her parents was Romani, the Romani language was spoken at the home of the child, or the mother was planning to raise her child with the Romani language. Romani ethnicity was subsequently independently confirmed by a Slovak member of the research team who used additional information such as the family’s last name to confirm ethnicity. Data on maternal smoking and alcohol use during and in the 3 months before pregnancy were also collected. History of maternal illness was also obtained, which included respiratory symptoms, asthma, or allergy during pregnancy or in the 3 months before pregnancy. At the 6- and 16-month follow-up visits, mothers again completed a questionnaire that updated demographic and lifestyle information and data on current maternal smoking and collected data on the child’s day care attendance and breast-feeding.
Infant medical records. The infant’s medical records were abstracted at birth and 6 and 16 months to obtain birth weight, gestational age (weeks), and the child’s weight taken at well- and sick-child visits with their pediatricians. Because these office visits did not necessarily coincide with the dates of their thymus measurements at 6 and 16 months, we interpolated linearly the two weights closest in time both before and after the thymus measurement date to estimate the child’s weight at the time of thymus measurement.
Selection of PCBs for statistical models. We selected maternal and infant PCB congeners for inclusion in our statistical models if at least 80% of the measurements were above the limit of detection (LOD) in maternal and 6- and 16-month serum samples. These were congeners 138+163, 153, 170, and 180. When an individual congener value was below the LOD, we assigned the value as the LOD divided by the square root of 2 (Lubin et al. 2004
). The sum PCB variable is therefore the arithmetic sum of these four congeners.
Multivariate methods. To eliminate differences in thymic volume measures attributable to different examiners, we standardized thymic volume by examiner and within each time point (birth and 6 and 16 months of age). We fit repeated measures models that included a repeated measure of standardized thymus volume at birth and 6 and 16 months of age as the dependent variable. One set of models focused on maternal PCB concentration in relation to all three thymus volume measurements; the other examined time-varying PCB exposure based on maternal and 6- and 16-month infant PCB concentrations as the exposure of interest. Continuous PCB concentrations (per lipid basis, in nanograms per gram of lipid) were modeled as natural log values to reduce the influence of extreme values, and preliminary, age-specific models showed that the PCB–thymus volume association was adequately modeled with a linear term. To allow for age to modify the associations between PCB concentrations and thymus volume, we included a PCB × age interaction term, where age was a categorical variable (0, 6, and 16 months, corresponding to the month of assessment). We used directed acyclic graphs to select covariates for our model (Greenland et al. 1999
), which included ethnicity (Romani/other), infant sex, district of residence (Svidnik/Michalovce), infant weight (kilograms), and duration of exclusive breast-feeding (months). Infant weight was modeled as a time-varying covariate using the recorded birth weight and the interpolated weight at 6 and 16 months of age, as described above. Duration of exclusive breast-feeding was also parameterized as a time-varying covariate and included an interaction with categorical time, so that separate effects could be estimated at 6 and 16 months of age. Duration of exclusive breast-feeding values at birth were set equal to 0 (because women were not yet breast-feeding), and values at 6 and 16 months corresponded to the duration of exclusive breast-feeding up until that point in time. All models were fit using the MIXED procedure in SAS (version 9.2; SAS Institute Inc., Cary, NC, USA) with random effects specified for intercept and age at assessment (weeks), with no explicit structure assumed for the covariance among repeated measures (“unstructured covariance”). Because only a portion (n
= 249) of 6-month infant PCB samples were analyzed, and the specimens selected for analysis were selected based on maternal PCB concentration (stratified random sampling), sampling weights were added for the 6-month time point (weights were set to 1 at birth and 16 months of age). Finally, to facilitate interpretability and for consistency with a previous study of PCBs and infant thymus size from this cohort (Park et al. 2008
), we present the results for regression models as the difference in standardized thymus volume [and associated 95% confidence intervals (CIs)] for an increase in PCB concentration from the 10th to the 90th percentile. In addition to our primary model (described above), we limited our study sample to those infants who had thymus measures conducted within the first postnatal week and within 2 weeks of their 6- and 16-month birthdays. Results using the truncated age range are presented alongside our primary model for comparison.
We also conducted two sensitivity analyses to evaluate the role of exclusive breast-feeding duration on PCB concentrations and thymus development. First, we fit 6- and 16-month PCB models without adjustment for breast-feeding to evaluate the role of confounding, and second, we also examined potential heterogeneity by exclusive breast-feeding duration at the 6- and 16-month time points. Results of the sensitivity analyses are reported in text.