Better Postdiagnosis Diet Quality Is Associated with Reduced Risk of Death among Postmenopausal Women with Invasive Breast Cancer in the Women's Health Initiative

Study population

The WHI has been previously described in depth (12–14). Briefly, between 1993 and 1998 through 40 clinical centers, postmenopausal women who were 50 to 79 years at study entry were recruited into either a Clinical Trials (CT) component (n = 68,132) or an Observational Study (OS) component (n = 93,676 women). The CT and OS were closed in 2004–2005 and the participants were invited to continue being followed in the WHI Extension Study, which currently has follow-up events through 2010. For the present analysis, we focused on women in the intervention and comparison arms of Dietary Modification Trial (WHI-DM) of the WHI CT and women in the OS who were diagnosed with invasive breast cancer during the course of follow-up and completed a food frequency questionnaire (FFQ) after being diagnosed with invasive breast cancer (overall = 2,412; OS = 1,156; WHI-DM = 1,256). In the WHI, only the WHI-DM and OS had multiple FFQs which enabled us to look at postdiagnosis diet quality, so we focused on these groups. Of these, 2,319 women [OS = 1,114 (96%); WHI-DM = 1,205 (96%)] were considered to have valid data for the FFQ, which we defined as reporting energy in the range of 600–5,000 kcals/day. Two women from the OS were missing data on postmenopausal hormone therapy use and were excluded from this analysis. Our final sample included 2,317 women.

Dietary assessment

A standardized written protocol, centralized training of staff, and quality assurance visits by the Clinical Coordinating Center (CCC) were used to ensure uniform administration of data collection. Diet was measured in WHI using a self-administered FFQ developed and validated for the study (19), adapted from the Health Habits and Lifestyle Questionnaire (20). The three sections of the WHI FFQ included 122 composite and single food line items asking about frequency of consumption and portion size, 19 adjustment questions related to type of fat intake, and 4 summary questions asking about the usual intake of fruits and vegetables and added fats for comparison with information gathered from the line items. In the WHI-DM, all participants completed an FFQ at baseline and year 1 of follow-up, and a one-third subset completed an FFQ each year on a rotating basis thereafter from years 2–9. In the OS, participants completed an FFQ at baseline and during year 3 of follow-up. For this analysis, we identified the FFQ closest to but after participants' diagnoses of invasive breast cancer. The postdiagnosis FFQ occurred, on average, 1.5 years after diagnosis for both WHI-DM and OS participants and the range was 0–6 years for WHI-DM participants and 0–4 years for OS participants.

The WHI-FFQ was designed to capture foods relevant for multiethnic and geographically diverse population groups, and has been shown to produce reliable (r_{all nutrients} = 0.76) and comparable estimates to 8 days of dietary intake from 4 24-hour dietary recalls and 4-day food records (r = 0.37, 0.62, 0.41, 0.36, with energy, percent energy from fat, carbohydrate, and protein; ref. 19).

The nutrient database used to analyze the WHI-FFQ is derived from the Nutrition Data Systems for Research (NDS-R, version 2005, University of Minnesota, Minneapolis, Minnesota; refs. 21, 22). NDS-R provides nutrient information for >140 nutrients and compounds, including energy, saturated fat, and sodium. We measured diet quality with the HEI-2005 (15–18), created by the U.S. Department of Agriculture and the National Cancer Institute. This index aligns with the U.S. Dietary Guidelines for Americans (23). HEI-2005 score is calculated using diet data in units of MyPyramid equivalents, so we established a customized link (8) between NDS-R and the MyPyramid Equivalents Database (24).

The HEI-2005 scores 12 components, using an energy-adjusted density approach to set standards (e.g., per 1,000 calories or as a percent of calories; refs. 16, 17, 25). Six components (total fruit; whole fruit; total vegetables; dark-green vegetables, orange vegetables, and legumes; total grains; whole grains) are worth 0–5 points; five components (milk; meats and beans; oils; saturated fat; and sodium) are worth 0–10 points; and one component (calories from solid fat, alcohol, and added sugar) is worth 0–20 points. The latter three components are reverse-scored. For each participant, we scored each component and calculated a total score (100 possible points). We classified HEI-2005 scores into quartiles to best separate those with “better quality” diets (Q4) and “poor quality” diets (Q1).

Other covariate assessments

At baseline, participants completed self-administered health history questionnaires. Weight and height were measured during clinic visits using standard methods at baseline and annually at years 1–9 in the WHI-DM and at baseline and year 3 in the OS. We calculated body mass index (BMI) as weight (kg)/height (m)². For height and weight, we chose the assessment closest to the postdiagnosis FFQ but no more than 30 days after the FFQ completion. Recreational moderate-vigorous intensity physical activity, including walking, was assessed by questionnaire at baseline in the WHI-DM and OS and we calculated metabolic equivalent task hours (MET-h)/week of physical activity for each participant, as described in detail in Irwin and colleagues (26) and defined physical activity level (0 MET-h/wk; 0.1–3 MET-h/wk; 3.1–8.9 MET-h/wk; 9 or more MET-h/wk; ref. 15).

Ascertainment of death

Vital status of participants was collected through clinical center follow-up of participants and surrogates. In addition, periodic searches of the National Death Index were conducted. Cause of death was determined by medical record and death certificate review at the CCC.

Statistical analysis

Participants were followed from postdiagnosis FFQ until death, loss to follow-up, or the end of the previously described WHI-Extension Study 1 on September 30, 2010. Participants who did not consent to the Extension Study and were alive at study closeout on September 12, 2005 were censored on that date.

Means, SDs, and frequencies of demographic, clinical, and lifestyle characteristics of the study sample were calculated by quartiles of HEI-2005 scores.

Cox proportional hazards models were fit to our data using person-years as the underlying time metric. We estimated multivariate HRs and 95% confidence intervals (CI) for death from any cause, death from breast cancer, and death from causes other than breast cancer associated with increasing quartiles of HEI-2005 scores. The test for linear trend across HEI-2005 quartiles was performed by assigning participants the median value of their categories and entering it as a continuous term in a regression model.

We considered covariates that have been shown to be important predictors of death among breast cancer survivors and death from any cause among general populations. These included age at study entry, race/ethnicity, education, income, breast cancer stage, estrogen receptor (ER) status, and progesterone receptor (PR) status, years from diagnosis to FFQ, number of alcohol servings per week, physical activity, and use of postmenopausal hormone therapy. We also adjusted all models for energy intake (27) and WHI study component/arm (WHI-DM-intervention, WHI-DM-comparison, OS). None of these covariates acted as significant confounders, changing the magnitude of HRs by at least 10%, but we retained them in models and presented unadjusted, age and energy-adjusted, and multivariate-adjusted models to illustrate this point. We did not have data on cancer treatment, but adjusted for stage and ER/PR status which influence types of treatment received, and treatment and stage have been shown to be correlated in other studies of early-stage breast cancer survivors (28). We also ran models with and without stage to examine any differences in results. We investigated confounding by BMI separately, given its potential role as a mediator of the diet-mortality and the disease specific-mortality relationships (29, 30). Then, we ran an additional sensitivity analysis excluding women whose postdiagnosis FFQs were completed within 6 months after their diagnoses (n = 430), during which time treatment may have affected dietary intake and reporting.

To test whether any one component was responsible for the overall association for diet quality and mortality, we ran models for each of the 12 HEI-2005 component scores, adjusting for the residual HEI-2005 score (HEI-2005 total score-HEI-2005 component score).

Given clinical trial evidence that the relationship with diet and breast cancer outcomes may vary by ER status (31), as an exploratory analysis, we planned a priori to conduct a stratified analysis by ER status. We investigated heterogeneity of the diet quality and all-cause mortality relationship by tumor ER status, by using likelihood ratio tests for both the interaction of diet quality with ER status (α = 0.1) and by evaluating the difference in model fit of full and reduced models.

All statistical tests were based on a priori hypotheses, and therefore there was no adjustment for multiple testing. All statistical analyses were conducted using SAS (version 9.2). All tests were two-sided with statistical significance set at P < 0.05.