The Chapaevsk study is an ongoing prospective study of 499 generally healthy boys (Hauser et al. 2008
). All male residents 8–9 years of age (n
= 623) were identified between 2003 and 2005 from health insurance records and the town’s clinic system. Of these, 572 met eligibility criteria and 516 (90%) agreed to participate. Children were ineligible if their address was unavailable, if they were likely to move during the study, or if they had severe cerebral palsy. After enrollment, 17 children living in orphanages were excluded because of missing birth or family history. For this analysis, 10 additional boys were excluded for chronic illnesses that could affect growth (e.g., severe asthma or malignancy), leaving 489 boys.
Once enrolled, each boy underwent a physical examination, provided a blood sample, and together with his mother or guardian completed health, lifestyle, and dietary questionnaires. Annual follow up examinations were conducted on or close to each boy’s birthday and questionnaires were updated. For this analysis, 3 years of follow-up data were available, with each boy observed up to four times between ages 8–11 or 9–12 years.
The study was approved by the Human Studies Institutional Review Boards of the Chapaevsk Medical Association (Chapaevsk, Russia); Harvard School of Public Health and Brigham and Women’s Hospital (Boston, MA, USA), and University of Massachusetts Medical School (Worcester, MA, USA). The parent or guardian signed an informed consent, and each boy signed an assent before participation.
Growth and pubertal assessment.
At annual visits, an endocrinologist (O.S., with a nurse present) conducted standardized physical examinations without knowledge of the boy’s exposure. Examination included measurement of height in stocking feet (to the nearest 0.1 cm) using a fixed arm stadiometer and weight in underclothes (to the nearest 100 gm) using a balance scale. Body mass index (BMI; kilograms per square meter) was calculated from measured height and weight. Pubertal maturation was graded from 1 to 5 by visual inspection according to established criteria (Tanner and Whitehouse 1976
). Testicular volume (TV) was measured using an orchidometer. Pubertal onset was defined as stage 2 or higher for genitalia (G2+) or TV > 3 mL for either testis.
At enrollment, each mother or guardian completed a nurse-administered questionnaire ascertaining a
) the child’s birth and medical history, breast-feeding status, physical activity, b
) family demographics, income, residential history, and c
) parental reproductive and medical history, occupation, education, smoking, and alcohol consumption. Birth weight and gestational age were obtained from medical record review. Diet was ascertained using a food frequency questionnaire modified from a validated Russian Institute of Nutrition instrument (Burns et al. 2009
; Martinchik et al. 1998
Organochlorine exposure assessment.
Fasting blood samples were collected before baseline examination, and the serum fraction was stored at –35°C until shipment for analysis at the National Center for Environmental Health at the Centers for Disease Control and Prevention (Atlanta, GA, USA). Analytes included 7 polychlorinated dibenzo-p
-dioxins (PCDDs, or dioxins), 10 polychlorinated dibenzofurans (PCDFs, or furans), 4 co-planar PCBs (co-PCBs), 6 mono-ortho
–substituted PCBs, and 31 other PCBs (non-dioxin-like PCBs) described in Burns et al. (2009)
For dioxin-like analytes, sera, method blanks, and quality control samples (aliquots of pooled bovine sera) were spiked with a mixture of 13
-labeled PCDDs/PCDFs and co-PCBs as internal standards, and serum analytes were isolated by solid phase extraction (SPE) followed by a multicolumn automated cleanup and enrichment procedure (Turner et al. 1997
). Analytes were separated on a DB-5 MS capillary column (Phenomenex, Torrance, CA, USA) and quantified using selected-ion-monitoring (SIM) high-resolution (10,000 resolving power) mass spectrometry (HRGC-ID/HRMS; Thermo Electron North America, LLC, West Palm Beach, FL, USA) (Patterson et al. 1987
). Quantification by isotope dilution MS used calibration standards containing 13
-labeled and unlabeled analytes.
For all analytes, quality control sample coefficients of variation combining between-run and within-run reproducibility were generally < 15%. All concentrations were expressed on a per-lipid basis, with serum total cholesterol and triglycerides measured enzymatically, and total lipids were calculated using the Phillips equation (Phillips et al. 1989
). Congener concentrations below the limit of detection (LOD) were assigned the LOD divided by the square root of 2.
Dioxin toxic equivalents (TEQs) were computed on a lipid basis using the 2005 World Health Organization (WHO) toxic equivalency factors to weight the potency of each congener relative to 2,3,7,8-tetrachlorodibenzo-p
-dioxin (TCDD) before summation (Van den Berg et al. 2006
). Nine different exposure measures were considered: (1) total (summed) TEQ measures (picograms per gram lipid) for combined dioxin, furan, co-PCB, and mono-ortho
PCB congeners; (2) TCDD (picograms per gram lipid); (3–5) total (summed) TEQs (picograms per gram lipid) for each of the dioxins, furans, and co-PCBs; (6–8) total (summed) concentrations (picograms per gram lipid) for each of the dioxins, furans, and co-PCBs; and (9) total (summed) concentrations of non-co-planar PCBs, including mono-ortho
–substituted PCBs (ΣPCBs) (nanograms per gram lipid). Organochlorine measures were categorized into quartiles because of potential nonlinear associations. Analyses were repeated using a quartile indicator (1, 2, 3, 4) for exposure to test for trend across quartiles. Statistical significance was defined as a p
-value < 0.05.
We used standard Cox proportional hazards models to assess time to pubertal onset as a function of exposure adjusted for potential confounders. Age of pubertal onset was assigned to the midpoint between age at the previous visit and age at the visit at which onset was noted. For boys in puberty at study enrollment (n = 141 by G2+; n = 66 by TV > 3 mL), age at onset was defined as 6 months before age at enrollment. Observations were censored at the last visit for boys not yet in puberty.
Sensitivity analyses were performed using both interval-censored likelihood-based models and repeated measures generalized estimating equation (GEE) models. The interval-censored approach does not assign a specific time of onset, but instead assumes that pubertal onset occurred in the interval between study visits. This approach was used to estimate overall mean age of pubertal onset, assuming a normally distributed age at onset, and mean age at pubertal onset for each exposure quartile, adjusted for confounders. The GEE approach was used to fit a logistic regression model for pubertal onset at each visit as a function of age at visit, with adjustment for potential confounders and correlation among multiple visits via an autoregressive structure. GEEs were also used to evaluate the impact of clustering within household for twins (four pairs) and siblings (three pairs). To account for possible examiner and laboratory drift over time, uncertainty regarding age of pubertal onset, and the potential for reverse causation (due to dilution of dioxin concentrations in larger, more mature boys), additional sensitivity analyses were performed excluding boys with pubertal onset at study entry and adjusting for year of organochlorine analysis.
Covariates considered in models included potential determinants of pubertal onset: age of child at examination, birth weight, gestational age, breast-feeding; nutrition, height, weight, and BMI at enrollment; household income; maternal age at birth and parity; prenatal smoking (active and secondhand) and alcohol intake; parental education; and blood lead (Williams et al. 2010
). A core model was developed by first assessing the univariate relation of covariates to each pubertal onset measure and retaining those with a p
-value < 0.20. Covariates meeting this criterion were included in a full model; backward selection (likelihood ratio test) was then used to iteratively exclude the least important covariates (retain p
< 0.15). Covariates were retained if they were significant for at least one pubertal onset measure or if they resulted in a ≥ 10% change in exposure effect estimates when added, one at a time, into our final model. Because height and BMI at enrollment may be proxies for pubertal onset or on the causal pathway relating dioxins with onset, we performed sensitivity analyses excluding these covariates from the final model. Because age of mother at menarche was missing for 8% of participants, this covariate was added to the final models in sensitivity analyses.
The association of pubertal onset with each of the nine different exposure measures was assessed, one exposure at a time. These nine measures were moderately to strongly correlated (Spearman r = 0.44–0.90); therefore, secondary analyses were performed to assess the independent relation of dioxin-like versus non-dioxin-like exposures with pubertal onset. Specifically, final models for the relation of each dioxin-like measure (total TEQ, TCDD, as well as dioxin, furan, and co-PCB TEQ and concentration measures) with pubertal onset were rerun with non-co-planar PCB concentrations (ΣPCBs) added to the models.