The PLCO Cancer Screening Trial is a multi-center, randomised controlled trial designed to evaluate screening methods for the early detection of prostate, lung, colorectal, and ovarian cancer (Prorok et al, 2000
). Briefly, 154
952 participants (78
217 women), aged 55–74 years, were recruited from 10 centres in the US between 1993 and 2001. On enrolment, the participants in the screened and non-screened arms of the trial completed a self-administered baseline questionnaire on demographics, personal/family cancer history, medical history, and lifestyle habits. Starting in 1998, diet was assessed using the Diet History Questionnaire (DHQ) (http://riskfactor.cancer.gov/DHQ/
), the National Cancer Institute's (NCI) self-administered validated food frequency questionnaire (FFQ) (Subar et al, 2001
). Each year, participants were sent annual study update questionnaires that asked whether they had been diagnosed with cancer by a health care provider. The study was approved by the institutional review boards at the NCI and PLCO study centres. All participants provided written informed consent.
Women were excluded from this analysis if they lacked the baseline questionnaire (n
=2095) or the DHQ (n
886); missed more than seven food items on the DHG (n
=1563); reported energy intake in the top or bottom 1% of women (n
=1226); had a history of cancer other than non-melanoma skin cancer before dietary assessment (n
=6819); or no follow-up time (n
=1110). After exclusions, with some subjects meeting multiple criteria, our analytic cohort consisted of 52
158 women. The vast majority of the cohort was postmenopausal based on self-reported age at last period and reason for last period. Menopausal status was ambiguous for 1.7%, but 89.1% of the women with ambiguous data were 57 or older, so the cohort was assumed to be postmenopausal.
Incident invasive breast cancer cases were identified through self-report from the annual study update questionnaire, physician reports, or through reports from the next of kin. This analysis includes only histologically confirmed invasive breast cancers based on pathology reports and medical records. Oestrogen receptor (ER) and progesterone receptor (PR) data collection is ongoing; we had receptor status for 388 of our cases. Entry date for the analytic cohort was the latest of the following: randomisation, completion of baseline questionnaire, or completion of the DHQ. Follow-up ended on 31 December 2006, with breast cancer cases exiting at the date of diagnosis and non-cases exiting at the date of the most recent annual study update questionnaire without a report of breast cancer.
The DHQ assessed usual intake (frequency and portion size) of 124 food items over the past year. Nutrient intake was estimated using the Diet*Calc Analysis Program (version 1.4.3, National Cancer Institute, Applied Research Program, 2005). Red meat (g per day) included bacon, beef, cheeseburgers, cold cuts, ham, hamburgers, hot dogs, liver, pork, sausage, veal, venison, and red meat from mixed dishes. White meat included chicken, fish, and turkey. Processed meat included bacon, cold cuts, ham, hot dogs, and sausage.
With the Computerised Heterocyclic Amines Resource for Research in Epidemiology of Disease (CHARRED) (http://www.charred.cancer.gov
) software application and data from a detailed meat-cooking module included in the DHQ, we generated intake estimates of three HCAs (ng per day): 2-amino-3,4,8-trimethylimidazo[4,5-f
]quinoxaline (DiMeIQx), 2-amino-3,8-dimethylimidazo[4,5-f
]quinoxaline (MeIQx), and 2-amino-1-methyl-6-phenyl-imidazo[4,5-b
]pyridine (PhIP), as well as benzo[a
]P), a marker of total PAH exposure, and mutagenic activity in meat (revertant colonies per day) (Sinha et al, 2005
). We estimated haem iron from meat using the NCI heme iron database based on the measured values of haem iron from meat samples cooked by a range of methods to varying doneness levels (Sinha et al, 2005
). The United States Department of Agriculture (USDA, 2007
) Survey Nutrient Database was used to estimate iron from meat (limited to meats in the haem iron database).
Hazard ratios (HRs) and 95% confidence intervals (CIs) were estimated using Cox proportional hazards regression with age at baseline as the underlying time metric; proportional hazards assumptions were not violated. Quintile cut points for the dietary exposures were based on intake in the analytic cohort, with the lowest quintile as referent. Dietary variables, except the meat mutagens, were energy adjusted using the multivariate nutrient density method; residual adjustment did not alter our findings (Willett, 1998
). Tests for linear trend were based on median values of each quintile. P
-values are two-sided and analyses were conducted using SAS (SAS Institute, Cary, NC, USA).
Multivariate models were adjusted for the following potential confounders, which were selected because inclusion in the age-adjusted model resulted in a 10% change in risk estimates, they were associated with breast cancer in this dataset, or are established breast cancer risk factors: age, race, education, study centre, randomisation group, family history of breast cancer, age at menarche, age at menopause (natural or surgical reasons), age at first birth and number of live births, history of benign breast disease, number of mammograms during past 3 years, menopausal hormone therapy, body mass index (BMI), and intakes of alcohol, total fat, and total energy. We included cross product terms in the multivariate models to assess effect modification by alcohol intake, parity, family history of breast cancer, BMI, menopausal hormone therapy, and number of mammograms.