The study cohort was recruited from a referent cohort comprising 2,637 women aged 18 – 44 years who participated in a population-based angler cohort study whose aim was to assess species specific fish consumption and knowledge of consumption advisories. Specifically, women who stated in 1991 that they were either considering or undecided about future pregnancies were re-contacted in 1996 – 1997 and asked to participate in a prospective pregnancy study, providing they were ≤40 years of age, had no physician-diagnosed infertility and were planning to begin trying to become pregnant in the next six months. Among the 244 eligible women, 113 (46%) were recruited, of which 14 were found to be pregnant at or immediately following enrollment, leaving 99 women in the study cohort. An additional 20 women withdrew from the study without an observed pregnancy. Of the 79 women completing follow-up, 69 (87%) became pregnant, of which 55 (80%) had live births and 14 (20%) experienced pregnancy losses. Ten (13%) women did not achieve pregnancy (i.e., infertile). The study was approved by the Institutional Review Board for the School of Medicine and Biomedical Sciences at the University at Buffalo, State of New York; all women provided informed consent prior to their participation.
Women participated in a baseline interview administered by a research registered nurse and were instructed in the accurate use of the Clearblue Easy™
home pregnancy test, capable of detecting ≤50 mIU/ml of hCG (per manufacturer) in urine on the date of expected menses. Clearblue Easy is a digital pregnancy test kit that displays “pregnant”, “not pregnant”, or “error” for simple interpretation. This pregnancy test is one of the most sensitive and accurate presently available [21
]. Participants were instructed to use pregnancy tests on the first day menses was expected and one week later, regardless of the first result, and to complete daily diaries while trying to become pregnant to obtain information on menstruation, sexual intercourse and lifestyle exposures that may adversely affect fecundity, including pregnancy loss (i.e.
, cigarette smoking and consumption of caffeine and alcohol). Reminder telephone calls were placed by the research nurse when diary cards were a week late. A non-fasting blood sample (10 ml) was collected at the baseline interview by the research nurse utilizing venipuncture collection equipment determined to be free of the chemical exposures of interest in the study.
Serum samples were analyzed by the Toxicology Research Center at the University at Buffalo (Buffalo, NY, USA), for 76 congeners (64 single and 12 di-eluting congeners) using gas chromatography with electron capture detection as previously described [22
]. Briefly, ten samples were run in each batch, including four controls consisting of one reagent blank, one matrix blank, one quality control sample and one duplicate sample [22
]. PCB congeners were categorized a priori
by purported biologic activity and summed (Σ) into four groupings: 1) total PCBs or the Σ of all measured congeners; 2) Σ estrogenic congeners (# 4_10, 5_8, 15_17, 18, 31, 44, 47, 48, 52, 70, 99, 101, 136, 153, 188); 3) Σ anti-estrogenic congeners (# 77_110, 105, 114, 126, 156_171, 169); and 4) Σ other PCB congeners (#6, 7_9, 16_32, 19, 22, 24_27, 25, 28, 33, 40, 42, 45, 50, 55, 59, 60, 64, 66_95, 74, 81_87, 82, 94, 97, 118, 128, 129, 132, 134, 135, 138, 141, 147, 149, 151, 157_200, 163, 167, 170, 172, 174, 176, 177, 179, 180, 181, 183, 185, 187, 189, 190, 194, 195, 196_203, 205, 206) [24
]. PCB and DDE values were corrected only for recovery; no substitution of values below the laboratory limits of detection were made to avoid introducing potential bias [26
]. We did not adjust PCBs or DDE for serum lipids, given the absence of evidence supporting a causal relation between serum lipids and pregnancy loss [28
]. Serum PCB and DDE concentrations are expressed as nanogram/gram (ng/g) serum, which is equivalent to parts per billion. Total lipids (TL) were quantified using enzymatic methods as a function of total cholesterol (TC), free cholesterol (FC), triglycerides (TG) and phospholipids (PL) as: TL = 1.677 (TC − FC) + FC + TG + PL, and were expressed in mg/dl [29
The primary outcome of interest was pregnancy loss. Early pregnancy loss was defined as a positive pregnancy test followed by a negative test or by clinical confirmation of pregnancy followed by clinical pregnancy loss, which was defined as loss following clinical confirmation of pregnancy. Clinical pregnancy losses were self- reported, though all women who had hCG confirmed pregnancies were followed for approximately eight weeks’ gestation. All clinical losses occurred prior to 20 weeks’ gestation as reported by women. In our analyses, total pregnancy loss included early and clinical loss.
Lifestyle covariates (caffeine, alcohol, and cigarettes smoked) were reported daily using the diary and were, subsequently, standardized to a 28-day cycle to account for the intra- and inter-woman variability in menses. Number of caffeinated (coffee, tea, or soda) and alcoholic drinks and number cigarettes smoked per day were recorded on the daily diary. Specifically, we summed the woman’s reported consumption over her menstrual cycle (which may be longer or shorter than 28 days), multiplied it by 28 and then divided by the number of days in her cycle. Failing to standardize lifestyle exposures could make it seem that women with shorter menstrual cycles consumed less, because they had less opportunity for consumption at each cycle; conversely, that women with longer cycles consumed more given their longer length of follow-up. Caffeine and alcohol consumption and cigarette smoking were selected because of their possible shared detoxification pathway with PCBs and DDE via cytochrome P450 (CYP450) [30
]. High levels of lifestyle covariates could lead to overloading this pathway; alternatively, individuals with high lifestyle covariate levels could have CYP450 pathways that are better suited to detoxification. Confounders were selected based on a priori
Descriptive population statistics were tabulated by study outcome variable, and by completion of the study protocol. Pearson’s Chi Square and ANOVA p-values were calculated to detect statistically significant differences (p < 0.05) between study outcome and completion of the study protocol.
Although all participants provided blood samples, some women did not have sufficient sample to quantify all congeners (≈10%) or individual lipid (30%) components necessitating the use of multiple imputation. Cigarette and alcohol consumption were imputed for two individuals with insufficient daily diary data, but who completed follow-up. Multiple imputation was used to estimate the missing values to avoid potential biases associated with mean or median imputation, or complete case analysis [31
] using Imputation by Chained Equations (ICE) in STATA version 9 (StataCorp, College Station, TX).
Cox proportional hazard regression was used to estimate hazard ratios (HRs) of pregnancy loss (early and clinical losses combined) when restricting to women achieving pregnancy while under observation. Censoring occurred after 20 weeks of pregnancy. We estimated the hazard of pregnancy loss for PCB and DDE concentrations when left as continuous or categorical exposures. Models for pregnancy loss were run separately for total PCBs, estrogenic PCBs, anti-estrogenic PCBs, other PCBs, and DDE. Additionally, estrogenic PCBs, anti- estrogenic PCBs, other PCBs, and DDE were run together as continuous variables in one model, to determine if the results from the individual models were altered, perhaps due to a competing mechanism of action. We presented all models with and without adjustment for age (categorized as 24 – 29 and 30 – 34) and average standardized alcohol and cigarette intake over a 28-day menstrual cycle (continuous). We did not include parity as a covariate because it may be on the causal pathway to pregnancy loss, and may induce over adjustment bias [34
]. Additionally, we emphasize that given our small sample size, our statistical models are not robust to stratifying by parity to further assess this issue. STATA version 9 was used for all analyses.