Study population
CBCS study participants were women aged 20–74 years residing in 24 contiguous counties in central and eastern North Carolina [
26]. Patients were women with a first diagnosis of invasive breast cancer identified through a rapid ascertainment system implemented in cooperation with the North Carolina Central Cancer registry. Controls were selected from records of the North Carolina Division of Motor Vehicles and US Health Care Financing Administration, and were frequency-matched to patients on the basis of race and age. Information was collected through in-person interviews and included reproductive history, diet and lifestyle factors, a detailed family history of cancer, and occupational history. About 98% of participants who were interviewed agreed to give a 30 ml blood sample at the time of interview. Informed consent to obtain DNA and measurement of plasma concentration of PCBs was sought with the use of a form approved by the Institutional Review Board of the UNC School of Medicine in compliance with the Helsinki Declaration.
The CBCS was conducted in two phases. Phase I of the study was conducted between May 1993 and December 1996 [
26]. Phase II of the study was conducted between 1996 and 2001. The total numbers of participants enrolled in Phase I of the CBCS were 861 patients (335 African American, 526 white) and 790 controls (332 African American and 526 white). The response rates were 76% for patients and 55% for controls. Genotyping for
CYP1A1 was performed on the first 688 breast cancer patients and 702 controls enrolled in Phase I of the CBCS. Laboratory assays for organochlorines were conducted on blood samples from 748 patients and 659 controls in Phase I of the CBCS [
26]. The present analyses were based on Phase I participants for whom both
CYP1A1 genotyping and organochlorine measurements were available: 612 patients (242 African American, 370 white) and 599 controls (242 African American, 357 white).
Laboratory methods
Methods for identification of plasma PCBs on CBCS plasma samples using gas chromatography/electron capture detection have been described previously [
26]. Methods for genotyping of
CYP1A1 on CBCS samples using restriction fragment length polymorphism–polymerase chain reaction have been described previously [
27]. In a newly proposed nomenclature for
CYP1A1 alleles, the M1 allele corresponds to
CYP1A1*2A, M2 to
CYP1A1*2C, M3 to
CYP1A1*3, and M4 to
CYP1A1*4 [
28]. Data were missing on the following
CYP1A1 genotypes due to unreadable gels or failure to amplify: M1 (10 patients and 7 controls), M2 (none missing), M3 (10 patients and 7 controls), and M4 (none missing). To correct for differences in plasma concentrations of chlorinated hydrocarbons attributable to blood lipids, the quantity of different lipid components in the plasma of each subject was measured with an automated enzymic assay, as described previously [
26].
Statistical methods
Imputed total PCB values were used in the data analyses. Imputation was performed by setting zero values and values below the detection limit (0.0125 ng/ml) to the detection limit divided by the square root of 2 (26). Hornung and Reed [
29] recommended this method of imputation as an accurate approach for computing means and standard deviations for variables that include nondetectable values. In the present study, imputation did not affect the estimation of odds ratios (ORs) for breast cancer and total PCBs, because individuals in the lowest category (that is, below the median) remained in the lowest category after imputation. Lipid-adjusted total PCBs levels were calculated with Equation 2 of Phillips and colleagues [
30]. Univariate analyses were performed for total PCBs, lipid-adjusted total PCBs, low to moderate chlorinated PCBs, highly chlorinated PCBs, and individual PCB congeners 99, 105, 118, and 153. Among controls in the CBCS, levels of individual PCB congeners and congener groups were highly intercorrelated [
31]. ORs for the joint effects of
CYP1A1 genotypes and PCB exposure did not differ substantially according to specific PCB congeners or congener groups, and ORs were similar to results obtained with total PCBs. We did not observe differences in ORs for breast cancer and PCB exposure, or the joint effects of
CYP1A1 genotypes and PCBs, when we analyzed PCB congeners 99, 105, 118, or 153 alone, as suggested by Demers and colleagues [
32]. Therefore, only results for lipid-adjusted total PCBs are presented.
Adjusted ORs for breast cancer and 95% confidence intervals (CIs) were calculated from unconditional logistic regression models. Total PCBs were categorized as greater than or equal to versus below the median, on the basis of the distribution in African American or white controls separately. PROC GENMOD of software package SAS (version 8.1; SAS Institute, Cary, NC) was used to incorporate offsets derived from the sampling probabilities used to identify eligible participants. Covariates included age, race (whites/African Americans), parity (nulliparous, 1 or ≥ 2), use of hormone replacement therapy (never or ever), oral contraceptive use (never or ever), breast feeding (never or ever), smoking (never, current, or former smoker), alcohol consumption (yes/no), family history of breast cancer (yes/no), benign breast biopsy (yes/no), income (less than $30,000/year or at least $30,000/year), and education (lower than high school, or high school and above). Continuous covariates, which included height, waist/hip ratio, and body mass index, were categorized on the basis of the median values of each variable in all controls.
ORs did not differ after adjusting for additional covariates, so results are presented for African American and white women adjusting for sampling fractions and age. Menopausal status was defined as described previously [
27] and used to stratify analyses of
CYP1A1 genotypes and PCB levels. Stratified analyses were also conducted on the basis of smoking history (ever or never). In addition, we estimated the increase in odds of breast cancer per 0.10 ng/ml increase in total PCB levels by coding lipid-adjusted total PCBs as a continuous variable, and stratifying on
CYP1A1 genotypes. Linearity in the logit was tested using the Box–Tidwell transformation test as implemented in SAS (version 8.1). The assumption of linearity was not violated in premenopausal or postmenopausal African American or white women.
To assess the interaction on the additive scale between PCB levels and
CYP1A1 genotypes, indicator variables were created for each category of joint exposure of PCBs and
CYP1A1 genotype. Women with the homozygous common (non-M1, non-M2, non-M3, or non-M4) genotypes and the lowest level of exposure to PCBs were used as a common reference group. Interaction contrast ratios (ICRs) and 95% CIs were calculated for joint effects of the M1 genotype and PCBs [
33]. ICR values greater than zero indicate greater than additive effects (synergy), and 95% CIs for the ICR that exclude zero can be used as a test for statistical significance at an alpha level of 0.05. To test for interaction on a multiplicative scale, likelihood ratio tests (LRTs) were conducted comparing logistic regression models with main effect terms for lipid-adjusted total PCBs and
CYP1A1 genotypes and a product interaction term compared with models with main effects only.
P values for LRTs less than 0.20 were considered statistically significant [
34].
To test for independence of environmental exposure and genotype, means, medians, and distributions of total PCBs were determined in relation to CYP1A1 genotypes among African American and white controls. Medians for lipid-adjusted total PCBs were compared by using the Wilcoxon rank sum test. There were no statistically significant differences in medians for total PCBs comparing participants with CYP1A1 M1-containing versus non-M1-containing genotypes, M2 versus non-M2, M3 versus non-M3, or M4 versus non-M4 genotypes.