The BioCycle Study enrolled 259 healthy, regularly menstruating women (18–44 years of age) for up to two cycles to determine associations among biomarkers of oxidative stress, antioxidants, and hormonal levels during the menstrual cycle (Wactawski-Wende et al. 2009
). Women who self-reported a menstrual cycle length between 21 and 35 days, who were not trying to conceive, and who had not used hormonal contraception in the past 3 months were included in the study. Women were followed prospectively for one (n
= 9) or two (n
= 250) menstrual cycles. The present analysis includes 252 women. Metals were measured from a single whole-blood sample that was collected at enrollment, and hormones were measured in blood samples collected up to eight times per cycle, with 94% of women completing seven or eight clinic visits per cycle. Data collection occurred from 2005 to 2007 at the University at Buffalo in New York. The University at Buffalo Health Sciences Institutional Review Board (IRB) approved the study, and all participants provided written informed consent. Under a reliance agreement, the National Institutes of Health depends on the designated IRB of the University at Buffalo for review, approval, and continuing oversight of its human subject research for the BioCycle Study.
At the enrollment visit, which occurred an average of 16 days before the first clinic visit, whole blood was collected in purple-top Vacutainer tubes (Becton, Dickinson and Company, Franklin Lakes, NJ) that contained EDTA and that were prescreened for trace metals. After the samples were collected, they were refrigerated and later sent to the CDC’s Division of Laboratory Sciences, National Center for Environmental Health, for lead, cadmium, and mercury assessment by inductively coupled plasma mass spectrometry. Mercury levels represent the total concentration of mercury in blood from all relevant forms of mercury (e.g., inorganic, methyl). The limits of detection (LODs) for cadmium, lead, and mercury were 0.20 μg/dL (25% < LOD), 0.25 μg/dL (0% < LOD), and 0.30 μg/dL (12% < LOD). Lab-reported values < LOD were used without substitution to minimize potential bias (Craig et al. 2003
; Richardson and Ciampi 2003
; Schisterman et al. 2006
Fasting blood samples for hormone measurements were collected in the morning to minimize diurnal variation. Participants used Clearblue Easy fertility monitors (Inverness Medical, Waltham, MA) to assist in scheduling midcycle visits (Howards et al. 2009
), beginning on the sixth day after the start of menses and continuing daily testing until either the monitor indicated an LH surge or 20 test days had passed. If the monitor indicated peak fertility, the woman was instructed to come into the clinic that day and the next 2 days. Other visits were scheduled based on an algorithm that considered women’s reported cycle length. Visits were timed to correspond to early menstruation, mid- and late follicular phase, twice around expected ovulation, and early and late luteal phase.
Outcomes assessment. Reproductive hormones were measured in fasting serum samples collected in red-top tubes with no anticoagulant, spun, serum aliquoted, stored at –80°C, and shipped in batches that included a woman’s complete cycle samples. Estradiol was measured with radioimmunoassay. Progesterone, LH, and FSH were measured using solid-phase competitive chemiluminescent enzymatic immunoassays by Specialty Laboratory (Valencia, CA) on a DPC Immulite 2000 analyzer (Siemens Medical Solutions Diagnostics, Deerfield, IL). The interassay coefficients of variation reported by the laboratory for estradiol, LH, FSH, and progesterone were 9.7%, 4.8%, 4.8%, and 14.1%, respectively.
Cycles were classified as anovulatory if the progesterone level was ≤ 5 ng/mL throughout the entire cycle and if no serum LH peak was observed on the mid- or late-luteal-phase visit (Gaskins et al. 2009
). Based on this definition, 42 of the 509 (8%) cycles were considered anovulatory, including 24 (57%) that occurred during the first cycle and 18 (43%) during the second. We also conducted a sensitivity analysis using a less restrictive definition of anovulation based on progesterone across the cycle of < 5 ng/g (Abdulla et al. 1983
; Malcolm and Cumming 2003
) that resulted in 65 cycles (13%) being classified as anovulatory.
At enrollment, women provided written consent and then were asked to provide a health and reproductive history and lifestyle information (e.g., smoking, alcohol intake), and anthropometric measurements were taken by trained staff. In addition, usual physical activity was assessed using the International Physical Activity Questionnaire (IPAQ) (Yang et al. 2004
) and categorized according to standard IPAQ cut-points, and a food frequency questionnaire was used to estimate usual daily total energy, iron, shellfish, fish, and vegetable intakes during the previous 6 months (Fred Hutchinson Cancer Center, Seattle, WA).
Statistical methods. Descriptive statistics for continuous and categorical covariates were compared by tertiles of metal exposure and ovulatory status using analysis of variance, chi-square test, or Fisher’s exact test, as appropriate. Hormone levels were natural log transformed for normality. Metals were assessed as tertiles and continuous variables. Models were run separately for each metal. Potential confounders were identified (based on a review of the literature) and included age, body mass index (BMI), and race (white, black, Asian, other). We also evaluated smoking; income; education; physical activity; parity; dietary iron, fish, shellfish, and vegetables; and total calories as potential confounders; these factors were not included in the final models because they did not alter estimated associations by > 10% between metals and hormones or anovulation. Because of the small number of smokers, we restricted the sensitivity analyses to nonsmokers. One individual was excluded from continuous analyses of cadmium because her level was > 3 SDs from the mean.
We used generalized linear mixed models to estimate the association between metal exposures (modeled as continuous variables) and anovulation while accounting for dependence within women and cycles using SAS (version 9.2; SAS Institute Inc., Cary, NC). With these models, inferences are subject specific; the odds ratios correspond to a change in odds of anovulation associated with a one-unit change in blood metal level for an individual subject (Zeger et al. 1988
We used nonlinear mixed effects models with harmonic terms (version 2.10.1; R Foundation for Statistical Computing, Vienna, Austria) to evaluate associations between metals (grouped according to tertiles) and patterns of reproductive hormones over one to two menstrual cycles, accounting for cyclical hormonal changes and complex between-subject variation (Albert and Hunsberger 2005
). Outcomes for each hormone (natural-log–transformed estradiol, FSH, LH, and progesterone) included mean concentration, amplitude (the difference between nadir and peak hormone levels), and timing of shifts in the hormonal profile. Random effects for models of hormone means and amplitudes included a subject-specific term and, for each subject, a cycle-specific term. The number of harmonic terms used in each model was selected to minimize the Akaike information criteria (Akaike 1973
) and to reflect the expected shape of the hormonal profile. Time was scaled such that time was set to 0 on the first day of an individual’s cycle, 0.5 on the day of ovulation, and 1.0 on the last day of the cycle, regardless of the actual calendar days. Models to estimate associations with tertiles of metal exposure were run separately for each metal and were adjusted for age (continuous), BMI (continuous), and race (white, black, Asian, other).
We used generalized linear mixed models with stabilized weights to estimate effects of metals (continuous) on natural-log–transformed hormone levels while appropriately accounting for confounding by hormone levels at other times during the cycle (Cole and Hernan 2008
; Robins et al. 2000
). Inverse probability of treatment weights were conditioned upon estradiol, progesterone, LH, FSH, race (white, black, Asian, other), age, and BMI. Adujsting for smoking (current vs. nonsmoker) and average calorie intake (continuous) did not appreciably alter the estimates; thus, these factors were not included in the final models. An α-level ≤ 0.05 denoted statistical significance. We assessed interactions between each pair of metals separately with all three metals included in one model using a p
< 0.10 criterion.