Study population
Between May 2002 and February 2004, 1,441 pregnant women were enrolled and provided a blood sample in the INUENDO (Biopersistent organochlorines in diet and human fertility) cohort from Greenland, Kharkiv (Ukraine) and Warsaw (Poland). To be eligible at baseline, the women had to be born in the country of the study, be pregnant as well as at least 18 years of age. At baseline, 2,478 women were eligible in Ukraine of whom 612 (25%) participated. In Greenland, 665 were eligible and 588 (88%) participated. All Greenlandic women were Inuits. In Poland, 690 were eligible and 258 (37%) participated. With few exceptions, the antenatal health programs covered all pregnant women in the localities.
At follow-up, 493 (81%) singleton children with measured exposure information were accessible and willing to participate in Ukraine, 525 (89%) in Greenland and 92 (36%) in Poland. A total of 1,110 children were followed-up. Further details on distribution of the study population are provided elsewhere [27].
After recording of child birth anthropometrics shortly after birth, the first follow-up of the cohort was conducted between January 2010 and May 2012 when the mean age of the children was 8 years in Greenland and Poland and 7 years in Ukraine. The retrospective reports of the developmental milestones in infancy, the children’s motor development at follow-up, and other characteristics were assessed by the parents (usually (95%) the mother), through an interview-based questionnaire. Mother-child-pairs were eligible for the study if the woman had a live-born singleton baby who was still alive at follow-up.
Ethics statement
The study was approved by local ethical committees; Polish Bioethical Committee (approval no. 6/2002 of 3.07.2002), Ethical Committee for Human Research in Greenland (approval no. 2010–13) and the Commission on Ethics and Bioethics Kharkiv National Medical University in Ukraine (protocol number 7, October 7 2009) and all participating parents signed informed consent.
Exposure assessment
At baseline, the women were interviewed and had a blood sample drawn. In Greenland and Kharkiv, the blood samples were drawn when the women were on average 24 weeks pregnant and in Warsaw when the women were on average 33 weeks pregnant. Ten ml cubital vein blood samples were drawn into vacuum tubes for serum collection without additives (Becton Dickinson, Maylan, France). The sera were analysed for PCBs, measured as CB-153, and DDE, measured as p,p′-DDE, and used as bio-markers of the prenatal exposure to the compounds. All chemical analyses were performed at The Department of Occupational and Environmental Medicine in Lund, Sweden. The sera were analysed by gas chromatography-mass-spectrometry following solid phase extraction. All samples were analysed twice at different days and the mean concentration of these two determinations was used. For an estimation of the imprecision in the method, the results of the analysed samples were divided into three equal sized groups, one with low levels, one with medium levels and one with the highest levels and the mean concentration for each group were calculated. Then the relative standard deviations were calculated from the duplicate determinations [28]. These were 18% at 0.1 ng/mL (n = 990), 10% at 0.5 ng/mL (n = 990) and 10% at 2 ng/mL (n = 990) for CB-153 and 11% at 1 ng/mL (n = 1058), 8% at 3 ng/mL (n = 1058) and 7% at 8 ng/mL (n = 1058) for p,p′-DDE. The detection limits were 0.05 ng/mL for CB-153 and 0.1 ng/mL for p,p′-DDE. For CB-153 there was 85 samples below the detection limit (LOD) and for p,p′-DDE there was 10 samples below LOD. If the concentration was less than LOD, the concentration was set to half the LOD based on fresh weight concentration. Sera were stored at – 20°C until analysis. CB-153 and p,p′-DDE levels were adjusted for serum concentrations of cholesterol and triglycerides which were determined by enzymatic methods. The inter-assay coefficients of variation for cholesterol and triglycerides were 1.5-2.0%. Further details are described elsewhere [29]. The laboratory was blinded concerning the measured outcomes and the mothers’ lifestyle etc. during pregnancy.
Outcome assessment
To evaluate the motor development of the children in the early school age, we applied the country specific version of the Developmental Coordination Disorder Questionnaire 2007 (DCDQ’07). The DCDQ’07 is a parent report measure developed to assist in the identification of developmental coordination disorder in children between five and 15 years, using a five-point Likert scale. It consists of 15 items which are grouped in three factors: 1) motor control, 2) fine motor and handwriting and 3) general coordination. The sum of the age specific scores of the 15 items give an indication of whether the child suffers from a developmental coordination disorder [30]. The sum of the scores ranges from 15 to 75, lower scores indicating motor problems.
At follow-up, the parents retrospectively stated the age at which their child first crawled, stood-up with support and walked without support, using the definitions of child developmental milestones (crawling, standing-up and walking) from the Multicentre Growth Reference Study by the World Health Organization [31]. Interviewers and parents were unaware of the level of prenatal exposure of the compounds.
Assessment of co-variates
Variables that might influence child motor development and developmental milestones and could be associated with the exposures were a priori identified in the literature, and data on these were harvested from questionnaires at baseline and follow-up. Baseline variables were: maternal pre-pregnancy smoking (yes/no), maternal pre-pregnancy alcohol intake (≤7, >7 servings of alcohol per week), maternal education (finished education before age 15 years, at 16–17 years, at or above 18 years), parity (1, 2–3, ≥4 child births), maternal age at the baseline interview (<30, 30–35, >35 years) and gestational age at blood drawing in weeks (continuous). Follow-up variables from immediately after birth were: sex of the child and gestational age (<37, ≥ 37 gestational week) and current follow-up variables were: breastfeeding (<6, 6–12, > 12 months) and child age at interview in years (continuous).
Data collection
The interviews were primarily conducted face-to-face at the participants’ residence or at the local hospitals in Greenland. A medical doctor was the main interviewer in Greenland, assisted by local health workers. A telephone interview was performed when families lived in remote areas (n = 130) or participants had moved to Denmark (n = 34). In Poland, four interviewers conducted the interviews at the participant’s residence or other local meeting points. In Ukraine, all interviews were conducted at eight paediatric polyclinics by a team of 59 paediatricians. A proportion of the questionnaires were filled in by the parents without the face-to-face interview in Greenland. Also, one of the 15 items of the DCDQ was erroneously lacking in the Greenlandic version of the Questionnaire in Greenland (but not in the Danish version used for some participants in Greenland).
Statistical analyses
Missing information
A relatively large amount of the parents did not fill in all items in the questionnaire, leading to missing data on outcome and co-variates. If parents left one item unanswered, it was impossible to calculate a total DCDQ score. To overcome this and take into account that complete case analysis may cause biased estimates of the association under study [32], we performed chained multiple imputations on the dataset. This approach will result in more unbiased estimates if the data are missing at random (the missingness is based on other observed characteristics such as the observed outcome or covariates) [32].
Briefly, chained multiple imputation is a statistical method that creates several new complete datasets (m > 1), based on known subject characteristics and other predictors in the complete dataset, incorporating the appropriate variability across the m datasets. The new m complete datasets are analysed, producing a single set of results accounting for the variability of the missing data [32].
We excluded subjects who had not filled in any of the 15 items in the DCDQ’07 from the chained multiple imputation model (n = 7), leaving a final study population of 1,103 women and children. We made 100 imputed datasets (m = 100) based on the following predictors: CB-153, p,p′-DDE, maternal education, maternal age at baseline interview, maternal pre-pregnancy smoking status, maternal pre-pregnancy alcohol consumption, parity, preterm birth, child sex, gestational age at blood drawing, child age at interview, the 15 items in the DCDQ’07, age at crawling, age at standing-up and age at walking. We performed different sensitivity analyses of the chained imputation models, including fewer and more predictors and generating less (m = 20) and more (m = 150) samples to check the robustness of the final imputation model. We present the results of the multiple imputation based analyses. Results of the complete-case analyses are available as supplementary material.
Data analyses
Study subjects were divided into tertiles of CB-153 and p,p′-DDE exposures. CB-153 and p,p′-DDE concentrations were adjusted for serum lipids and thus expressed as ng/g lipid. As a first step in the data analysis, the correlation between CB-153 and p,p′-DDE was checked by use of Spearman’s correlation. Secondly, the crude relation between CB-153 and p,p′-DDE and motor development and developmental milestones were examined with lowest exposure tertile as the reference category. Thirdly, the adjusted relations of prenatal exposure to CB-153 and p,p′-DDE on motor development were examined by means of multiple linear regression analyses with lowest exposure tertile as the reference category. Furthermore, analyses were performed using log-transformed continuous exposures as a test for trend. The relation between in utero exposure to the compounds and developmental milestones was studied, using multiple linear regression analyses, with referents as above. In a sub-analysis, we excluded children born before gestational week 37. A priori set variables that could affect child motor development, based on literature studies, were included in the regression model and kept in the model throughout the analyses. The possible interaction between exposures and country were tested by adding an interaction term (exposures*country) to the regression models. All analyses were stratified by country as well as pooled (additionally adjusted for country).
A p-value of 0.05 or less was considered statistically significant. For all analyses the Stata statistical package was used (Version 12.1, StataCorp, College Station, Texas, USA).