The NHS nested breast cancer case-control study (cases, n
= 1,270; controls, n
= 1,762) is derived from 32,826 women who were free of diagnosed breast cancer at blood collection in 1989 and 1990, since when they were followed for incident disease until 31 May 2000. Medical records were used to confirm the diagnoses in women who reported a diagnosis of breast cancer on the biennial questionnaires. Control subjects were matched to cases on the basis of age, menopausal status, recent hormone replacement therapy, and blood-draw specific variables (such as date and time of day). The NHS2 breast cancer case-control study (cases, n
= 317; controls, n
= 634) is nested in a study of 29,611 women who were free of diagnosed breast cancer at the time of blood draw (between 1996 and 1999) and were followed for incident disease until 1 June 2003. The nested breast cancer case-control study in the WHS began in 1993, when 28,263 women provided blood samples and were followed for incident disease until 7 March 2000, with 702 cases and 703 controls. Cases and controls were selected for the NHS2 and WHS with the use of the same criteria as in the NHS. All three cohorts use similar questionnaires to collect covariate information, administered biennially. Detailed descriptions of these three cohorts have been published previously [14
]. Informed consent was obtained from all participants, and the study was approved by the Institutional Review Board of the Brigham and Women's Hospital.
is a very small gene (less than 10 kilobase pairs from 5' UTR to 3' UTR), with only five known common polymorphisms; linkage disequilibrium between them is high (rs20417, rs5277, rs20432, rs5275, rs4648298 [17
]; see Table for more information). These five SNPs were genotyped by means of the 5' nuclease assay (TaqMan; Applied Biosystems, Foster City, CA, USA). The SNPs are located at base pairs 926, 3,050, 5,209, 8,473, and 9,850 on GenBank sequence D28235
. rs20417 is in the 5' region, rs5277 is a synonymous polymorphism in exon 3, rs20432 is in the intron between exons 5 and 6, and both rs5275 and rs4648298 are in the 3' UTR of PTGS2
. TaqMan primers, probes, and conditions for genotyping assays are available from the authors on request. All genotyping was done with laboratory personnel blinded to the case-control status of the samples and included quality-control samples (DNA from cohort participants, not part of the case-control study, repeated from 2 to 18 times across plates) for validation. Concordance for quality-control samples was 100%.
Pairwise linkage disequilibrium between SNPs selected for study in PTGS2
Statistical analysis was performed with SAS version 9.1 (SAS Institute, Cary, NC, USA). Haplotype estimation was performed with PROC HAPLOTYPE, tests for deviation from Hardy–Weinberg equilibrium were performed with PROC ALLELE, and haplotype-specific odds ratios and 95% confidence intervals were calculated with unconditional logistic regression with PROC LOGISTIC, controlling for age (at blood draw as a continuous variable), age at first birth (AFB), parity (nulliparous; one or two children and AFB 24 years or less; one or two children and AFB more than 24 years; more than two children and AFB 24 years or less; more than two children and AFB more than 24 years), menopausal status at diagnosis (premenopausal, postmenopausal, unknown), history of benign breast disease (yes/no), and family history of breast cancer (yes/no). Variables such as AFB, history of benign breast disease, or family history of breast cancer are updated in each questionnaire cycle, and the variable closest to the diagnosis (cases) or index date (controls) is used in these analyses. Cis-interactions, comparing associations with haplotypes to those observed with independent SNPs, were tested by comparing the model with the SNP genotype alone to the model with each haplotype carrying the risk allele by means of likelihood ratio testing. Interaction p values were calculated by likelihood ratio testing, comparing the model with main effects for each exposure (polymorphism and anti-inflammatory use) to the model including the cross-tabulation of the two exposures. Heterogeneity between cohorts was calculated with Cochran-Mantel-Haenszel statistics in PROC FREQ.