Dietary Fat Reduction and Breast Cancer Outcome: Interim Efficacy Results From the Women's Intervention Nutrition Study

Abstract

The question of the influence of dietary fat on breast cancer has been controversial. Whereas preclinical and human ecologic studies have suggested an association of higher dietary fat intake with breast cancer risk ( 1 , 2 ) , cohort studies have reported less consistent effects ( 3 – 5 ) . Similarly, observational studies of dietary fat influence on breast cancer recurrence have had mixed results ( 6 , 7 ) , with some suggesting that higher fat intake is associated with higher risk of recurrence, especially in postmenopausal women ( 8 – 10 ) . The varying associations may be due to the modest range of fat intake seen and the difficulty in accurately measuring fat intake with current methods ( 11 , 12 ) .

Feasibility trials have demonstrated that dietary fat reduction can be achieved within the context of standard multimodality breast cancer management ( 13 , 14 ). The Women's Intervention Nutrition Study (WINS) was subsequently designed to test the hypothesis that a dietary intervention targeting fat intake reduction would prolong relapse-free survival in women with resected breast cancer.

P ATIENTS AND M ETHODS

Study Overview

In this phase III multicenter randomized trial, the effect on relapse of a dietary intervention designed to reduce fat intake was compared with that of a control condition with minimal dietary counseling in women with early-stage, resected breast cancer receiving conventional cancer management. Patients in this study were accrued in approximately 7 years, between February 1994 and January 2001. This report represents an interim analysis after a median of 60 months of follow-up. The WINS trial protocol is available online as supplementary data at http://jncicancerspectrum.oxfordjournals.org/jnci/content/vol98/issue24 .

Study Population

Women were recruited from 39 clinical centers. Eligibility criteria included histologically confirmed, resected, unilateral invasive breast carcinoma; lymph node evaluation; age between 48 and 79 years; life expectancy of at least 10 years excluding the cancer diagnosis; acceptable adjuvant systemic therapy (see below); at least 20% of calories obtained from fat; medically able to accept either randomization assignment; and trial entry within 365 days of surgery. Exclusion criteria included inflammatory carcinoma; chest wall or skin involved; tumor size less than 1 cm with negative nodes; tumor size greater than 5 cm with positive nodes; 10 or more nodes positive; preoperative chemotherapy; or any previous neoplasm other than carcinoma in situ of the cervix or basal cell skin carcinoma. Eligibility was confirmed by central review of medical records and pathology reports. The Institutional Review Board of each participating institution approved the study protocol. All patients gave written informed consent.

Standard Breast Cancer Management

Standard breast cancer management was protocol defined, including adequate surgery and radiotherapy. Estrogen receptor (ER) and progesterone receptor (PgR) status (positive or negative) was defined by local laboratory standards. Women with ER-positive tumors received tamoxifen (20 mg per day) for 5 years. Chemotherapy with a protocol-approved regimen (adriamycin plus cyclophosphamide; cyclophosphamide plus methotrexate plus 5-fluorouracil in two schedules; 5-fluorouracil plus adriamycin plus cyclophosphamide in two schedules [5-fluoruracil–adriamycin–cyclophosphamide or cyclophosphamide–adriamycin–fluorouracil]; or adriamycin–cyclophosphamide followed by paclitaxel) was required for women with ER-negative tumors and optional for those with ER-positive tumors. As an eligibility criterion, chemotherapy had to be initiated within 4 months after diagnosis for women given chemotherapy, and tamoxifen had to be initiated within 6 months after diagnosis for those given only tamoxifen. Recommended drug dosages and schedules paralleled those in use by cooperative group trials at the time ( 15 – 17 ).

Randomization

Adaptive randomization was carried out at the Statistical Coordinating Unit of the WINS study using a random stratified permuted block design. The trial was designed with an unbalanced randomization (60% control subjects, 40% dietary intervention subjects) to facilitate resource allocation to the dietary intervention. Women were initially stratified according to lymph node status (negative or positive) and systemic adjuvant therapy received (tamoxifen alone, tamoxifen plus chemotherapy, or chemotherapy alone). An additional stratum for sentinel node evaluation (yes or no) was introduced in 1999 to reflect clinical practice trends.

Dietary Intervention and Study Assessments

The goal of the dietary intervention was to reduce percentage of calories from fat to 15% while maintaining nutritional adequacy. Feasibility studies had indicated that this goal would result in a sustained reduction in fat intake to approximately 20% of calories ( 14 ) , which was the basis for the sample size calculation. Women in the dietary intervention group were given an individual fat gram goal and counseled by registered dieticians who implemented a previously developed low-fat eating plan ( 14 – 18 ) . Study dieticians were trained centrally on diet intervention and dietary and anthropometric data collection. Training continued with annual workshops incorporating training on motivational interviewing and with monthly conference calls.

The low-fat eating plan, which was based on nutritional and behavioral science principles ( 18 ) , incorporated social cognitive theory and included self-monitoring (fat gram counting and recording), goal setting, modeling, social support, and relapse prevention and management. Individual fat gram goals were based on energy intake needed to maintain weight, and no counseling on weight reduction was provided. The low-fat eating plan was initiated during eight biweekly individual, in-person counseling sessions, each lasting approximately 1 hour. Subsequent dietician contacts (visits or calls) occurred every 3 months, with available, optional monthly dietary group sessions. Women in the dietary intervention group were instructed to keep a written record of their fat gram intake daily throughout the trial using a previously developed “keeping score” book ( 18 ) . Control subjects had one baseline dietician visit and contacts with a dietician every 3 months subsequently. They received written information on general dietary guidelines and were counseled on nutritional adequacy for vitamin and mineral intake only.

Questionnaires administered at baseline were used to collect information on demographic characteristics; medical, reproductive, and family history; personal habits such as smoking and alcohol use; prior use of menopausal hormone therapy and oral contraceptives; and current use of medications and dietary supplements. Weight and height were measured at baseline and annually using standardized techniques on calibrated scales/stadiometer ( 19 ) . Body mass index (BMI) was calculated as weight (in kilograms) divided by height (in meters) squared.

Unannounced telephone calls were used to estimate dietary intakes. Trained interviewers who were blinded to randomization status collected dietary information; the information was entered into the Nutrition Data System for Research interactive software as previously described ( 18 ) . A multilayered quality control process was used to maintain nutrition data quality ( 20 ) . Data on dietary intakes from three follow-up calls (including one on a weekend day) were collected over 2 weeks for eligibility determination regarding dietary fat intake. Two additional follow-up calls (including one on a weekend day) were made annually to collect updated data on dietary intakes for all women.

Recurrence information was solicited during the every-3-month dietician contacts. Recurrences were confirmed initially by medical record and pathology report review by the clinical director (RTC) and subsequently by the WINS Pathology Committee, both of whom were blinded to random assignment.

Outcome Definitions

The primary study endpoint was relapse-free survival, defined as the time from random assignment to breast cancer recurrence at any site. Relapse events included local, regional, and distant recurrence; ipsilateral breast recurrence after lumpectomy; and contralateral breast cancer. Overall survival, defined as the time from randomization to death from any cause, was a secondary endpoint. For comparability to other adjuvant trials ( 15 , 21 – 24 ) , disease-free survival and recurrence-free survival were also examined. Disease-free survival events included any secondary invasive cancers, excluding basal and squamous skin cancers, and death without breast cancer recurrence. Recurrence-free survival events included local, regional, distant, or ipsilateral breast recurrence after lumpectomy but excluded contralateral breast cancer. Breast cancer size and nodal status were used to calculate tumor size–node–metastasis stage ( 25 ) at baseline.

Termination of the Dietary Intervention

WINS was supported by a multicenter RO1 grant from the National Cancer Institute, with continued funding dependent on ongoing peer review. Funding for the clinical centers and the active dietary intervention ended in May 2004, even though the protocol-defined follow-up period had not been completed and even though secular trends in therapy had reduced recurrence events compared with the number of events projected at the time the study was designed. The WINS External Advisory Committee reviewed results of the last (November 2003) interim analysis (which incorporated nutrient data through August 31, 2003, and all available efficacy data through October 31, 2003), after a median follow-up of 60 months. Based on the change in trial status, both the WINS External Advisory Committee and the WINS Executive Committee supported reporting the available results but recognized the need for additional follow-up. These results were presented at the American Society of Clinical Oncology annual meeting ( 26 ) on May 17, 2005, and were provided to study participants by mail. This interim efficacy report provides details of that dataset. The protocol design called for 3 years of follow-up after completion of recruitment through January 2004. Currently ongoing nonintervention follow-up will address the original design plans.

Statistical Analysis

Sample size was based on a model that generated power estimates via simulation. Using published information from cooperative group trials involving tamoxifen ( 27 , 28 ), we calculated a total sample size of 2502 under assumptions of 6 years of accrual, 3 years of follow-up after completion of accrual, and a 7.5% increase in relapse-free survival with a drop-in (defined as control group women with intake <20% of calories from fat at any interval while on study) rate of 10% and a drop-out rate of 30% for 84% power at a two-sided alpha level of .05.

The primary efficacy analysis was a stratified log-rank test. Exploratory Cox proportional hazards models for relapse-free survival were investigated, as were various Cox models for prognostic factors in addition to those used in the stratified randomization. The final Cox model included randomized group, stratification factors (ER status [positive versus negative] and tumor size [<2 versus ≥2 cm]), and surgery type (mastectomy versus lumpectomy) based on their established influences on breast cancer outcomes. The cumulative incidence method was applied for Cox model analyses. Model assessment suggested a reasonable data fit. The assumption of proportionality for Cox models was verified by graphical and numerical methods of Lin et al. ( 29 ) . The P value for the Kolmogorov-type supremum test (for group) based on 1000 simulations was 0.5170, supporting the proportional hazards assumptions.

Analyses included all randomly assigned patients, following the intent-to-treat paradigm. All P values were derived from a two-sided test for significance. The 95% confidence interval (CI) for the hazard ratio (HR) was obtained for each term in the final Cox model, with particular attention to the interaction between treatment and other factors in the model. The P values and confidence intervals are exploratory. Five protocol-planned interim analyses were carried out between February 2000 and November 2003. The Haybitle–Peto approach to the α-spending function was used to account for interim analyses ( 30 ) . The Kaplan–Meier method was used to calculate probability estimates for relapse-free and overall survival. In exploratory analyses, which were not defined prospectively, dietary effects on relapse-free survival in subgroups based on BMI, hormone receptor status, and nodal status were examined using the Cox model. Tests for interaction used Cox models and a likelihood ratio test. Differences in baseline variables between and within groups were analyzed using t tests or paired t tests or the appropriate tests with categorical variables. The t tests were performed to compare differences in the nutrient intakes and anthropometric variables, and corresponding 95% confidence intervals are reported.

R ESULTS

A total of 2437 apparently eligible women were randomly assigned, 975 to the dietary intervention group and 1462 to the control group ( Fig. 1 ). Thirty-four women (12 in the dietary group and 22 in the control group) were subsequently found to be ineligible, most commonly based on a longer interval from diagnosis and/or use of other chemotherapy (n = 10), pagetoid nipple involvement (n = 9), or size or margin issues (n = 5). However, all randomly assigned patients are included in the presented analyses.

Fig. 1.

Open in new tabDownload slide

CONSORT trial flow diagram. Detailed reasons for not completing study were as follows. Not receiving intervention, intervention group: lost interest in study (n = 4), personal–family problem (n = 2), did not like low-fat eating plan (n = 1). Not receiving intervention, control group: lost interest in study (n = 2), did not like control group allocation (n = 3). Lost to follow-up, intervention group: unable to contact participant (n = 24), not interested in study (n = 3), personal–family problem (n = 2), did not like low-fat eating plan (n = 3), medical problem (n = 3), time commitment (n = 2), 5 years was enough (n = 2), moved (n = 1), refused to be contacted (n = 4), unknown (n = 1). Lost to follow-up, control group: unable to contact participant (n = 39), not interested in study (n = 5), personal–family problem (n = 2), did not like control group (n = 2), medical problem (n = 2), time commitment (n = 2), 5 years was enough (n = 2), moved (n = 2), refused to be contacted (n = 6), unknown (n = 4). Discontinued study, intervention group: unable to contact the participant (n = 49), not interested in study (n = 15), personal–family problem (n = 26), did not like low-fat eating plan (n = 21), medical problem (n = 12), time commitment (n = 10), 5 years was enough (n = 23), moved (n = 7), refused to be contacted (n = 3), unknown (n = 4). Discontinued study, control group: unable to contact the participant (n = 51), not interested in study (n = 7), personal–family problem (n = 6), did not like control group (n = 1), medical problem (n = 3), time commitment (n = 5), 5 years was enough (n = 15), moved (n = 3), refused to be contacted (n = 8), unknown (n = 7).

The characteristics of the participants were well balanced across the two groups ( Table 1 ). The characteristics of the breast cancers of study participants and the therapy given for these cancers by treatment group are provided in Table 2 . Breast cancer characteristics were closely comparable in the two groups. Breast cancer therapy was also closely comparable in the two groups, except for the type of primary surgery. More women in the dietary intervention compared to control group had mastectomy rather than breast-conserving therapy ( P = .004).

As of October 31, 2003, 95% of the women in the dietary intervention group and 94% of those in the control group were being followed as part of the study or had experienced a study event, with 84% of women in the dietary intervention group and 89% of those in the control group contacted within 12 months of October 31, 2003 ( Fig. 1 ). The median interval between last contact and the analysis closeout date was 2.6 months for women in the dietary intervention group and 3.1 months for those in the control group.

Adherence to Dietary Intervention

The dietary intervention adherence results are based on telephone follow-up calls. The reported dietary intake differences by randomization group over time are outlined in Table 3 . In all, 80% of women provided dietary data for at least three time periods after baseline. After 1 year, mean daily fat gram intake was slightly reduced in the control group (from 56.3 g at baseline to 51.3 g at 12 months, mean difference = −5.09 g, 95% CI = −6.5 to −3.7, P <.0001) but was reduced to a statistically significantly greater extent in the dietary intervention group (from 57.3 g at baseline to 33.3 g at 12 months, mean difference = −24.4 g, 95% CI = −26.1 to −22.6, P <.0001; P <.001 comparing the mean difference between groups). The difference in fat gram intake was maintained through 5 years (difference in fat grams per day in dietary versus control groups of −18.0 g [95% CI = −19.9 to −16.1] at 12 months and −19.0 g [95% CI = −22.1 to −16.0] at 60 months [both P <.0001]). Similar differences were seen for all fat categories and in percentage of calories from fat ( Table 3 ). In addition, energy intake was somewhat lower and fiber intake slightly higher in intervention group participants ( Table 3 ). Nutrient adequacy was maintained in both groups, with the exception of calcium, vitamin D, and vitamin E, for which intakes were somewhat below recommended levels at baseline and throughout the trial in both groups, a situation that was addressed with counseling and supplement use. No adverse events were associated with the dietary intervention.

Body weight was not an intervention target. However, there was a modest but statistically significant ( P = .005) weight difference of about 6 pounds between groups, with dietary intervention women weighing less through 5 years of observation. Changes in BMI in both groups reflect the weight changes observed ( Table 3 ).

Efficacy

After a median of 60 months, 277 relapse-free survival events and 389 disease-free survival events were reported ( Table 4 ). Recurrence events were confirmed by central review in 99.5% of cases. For relapse-free survival—the primary endpoint—the hazard ratio of an event in the dietary intervention compared with the control group was 0.76 (95% CI = 0.60 to 0.98) ( Fig. 2 , P = .077 for stratified log rank and P = .034 for adjusted Cox model analysis). For recurrence-free survival (i.e., excluding contralateral breast cancers), the HR was 0.71 (95% CI = 0.53 to 0.94; stratified log rank P = .050). For disease-free survival, the HR was 0.81 (95% CI = 0.65 to 0.99; stratified log rank P = .078). There was no difference in overall survival comparing women receiving the dietary intervention with control group women (HR = 0.89; 95% CI = 0.65 to 1.21; stratified log rank P = .56). Based on the effects on the primary endpoint, 38 women would need to adopt a lifestyle intervention reducing dietary fat intake to prevent one additional breast cancer recurrence.

Fig. 2.

Open in new tabDownload slide

Kaplan–Meier estimates of relapse-free survival. Number of events/number of patients in the dietary intervention and control groups are indicated. Hazard ratio (HR) and 95% confidence interval (CI) were calculated from adjusted Cox proportional hazard model comparisons of control to dietary intervention groups through the 60-month median follow-up period. P value is two-sided. Numbers of patients at risk are indicated below the graph.

Dietary intervention effects were examined based on BMI, hormone receptor, and nodal status in subgroup analyses by using adjusted Cox model. A total of eight subgroup analyses were performed, and none of the interactions tested were statistically significant. However, the dietary intervention had a greater effect on relapse-free survival in women with ER-negative cancer (HR = 0.58; 95% CI = 0.37 to 0.91) than in women with ER-positive disease (HR = 0.85; 95% CI = 0.63 to 1.14); interaction test, P = .15 ( Fig. 3 ). Findings by PgR status were similar ( Table 5 ).

Fig. 3.

Open in new tabDownload slide

Kaplan–Meier estimates of relapse-free survival. ( A ) Estrogen receptor–positive subjects. ( B ) estrogen receptor–negative subjects. Number of events/number of patients in the dietary intervention and control groups are indicated. Hazard ratios (HRs) and 95% confidence intervals (CIs) were calculated from adjusted Cox proportional hazard model comparisons of control to dietary intervention groups through the 60-month median follow-up period. P values are two-sided. Numbers of patients at risk are indicated below the graph.

D ISCUSSION

WINS is, to our knowledge, the first large-scale randomized trial to test whether a dietary intervention can improve the clinical outcome of women with breast cancer. The WINS results indicate that a lifestyle intervention designed to reduce dietary fat intake can be successfully implemented in women with early-stage, resected breast cancer receiving conventional cancer management in a multicenter clinical trial setting. After approximately 5 years of follow-up, women in the dietary intervention group had a 24% lower risk of relapse than those in the control group (HR = 0.76; 95% CI = 0.60 to 0.98). Although these results are suggestive of benefit for the dietary intervention, given the level of statistical significance ( P = .034 for adjusted Cox model analysis), the higher relapse-free survival seen in the dietary group could also be a result of chance.

Secondary analyses suggested a stronger effect for dietary fat reduction on breast cancer recurrence in women with hormone receptor–negative cancers compared with women whose cancers were hormone receptor positive; however, the interaction between dietary intervention and hormone receptor status was not statistically significant. These findings are consistent with trends suggestive of a differential dietary influence by hormone receptor status on breast cancer incidence recently described in the Women's Health Initiative randomized primary prevention trial and in a Nurse's Health Study cohort ( 31 , 32 ) . A differential influence of dietary fat on breast cancer recurrence based on hormone receptor status will require confirmation. However, some of the most active adjuvant breast cancer interventions, including aromatase inhibitors ( 21 – 24 ) and trastuzumab ( 33 – 35 ) , are effective only in biologically defined subgroups, and interactions have been observed between ER status and adjuvant chemotherapy effect ( 36 ) . Thus, it would not be surprising to find that a lifestyle intervention has an effect on only some categories of breast cancer.

Because ER-positive cancers are under estrogen influence, a predominant influence of diet on ER-negative cancers would implicate factors other than estrogen ( 37 , 38 ) as mediators. Other potential mechanisms include reduced insulin levels ( 39 , 40 ) , reduced insulin resistance ( 41 ) , reduced insulin-like growth factor 1 ( 42 , 43 ) , or reduced inflammation markers—all factors that may be influenced by dietary fat decrease and/or weight loss ( 44 , 45 ) . Planned analyses of serial fasting blood samples in the two groups will address these issues.

The most appropriate endpoint for breast cancer adjuvant trials is controversial, and even the definitions of these endpoints are inconsistent ( 46 ) . Endpoints from recent selected adjuvant therapy trials are compared to the relapse-free survival endpoint of the WINS trial in Table 6 . Collectively, these reports define the endpoint of “disease-free survival” in four different ways ( 21 – 24 , 27 ) . We defined relapse-free survival, the prospectively identified primary study endpoint in WINS, in the same way that Goss et al. ( 22 ) recently defined a disease-free survival endpoint. To facilitate comparison of our study with others, we included death without breast cancer recurrence and second primary cancers as events in additional analyses; all results led to similar conclusions regarding the effects of the dietary intervention. Establishment of a common terminology for breast cancer adjuvant trial endpoints should be a future priority of the research community.

This study has several potential limitations. One is the imbalance of surgical treatments between the groups: 5.6% more women in the control group than in the intervention group had breast-conserving therapy. As a result, 55 more women in the control group with such surgery were at risk for an ipsilateral recurrence. However, considering all 1018 control group women with breast-conserving surgery, only 2.1% experienced ipsilateral recurrence. Consequently, only about one additional in-breast recurrence would be anticipated based on the surgical distribution. Thus, the modest imbalance in surgical management is unlikely to explain the difference in clinical outcome that we observed. In addition, in a large National Surgical Adjuvant Breast and Bowel Project Cancer randomized trial comparing mastectomy with lumpectomy plus radiation, women with mastectomy had a similar rate of local and regional recurrences after approximately 5 years of follow-up, even when ipsilateral breast recurrences were included (45 such recurrences were seen in the mastectomy group compared with 43 in the lumpectomy plus radiation group) ( 47 ) . Moreover, our adjustment for surgery type (mastectomy versus lumpectomy) in the Cox proportional hazards models should have controlled for the effect of the differences in surgical management.

A second limitation of the study is its reliance on self-report of dietary intake because no validated “gold standard” exists for assessing dietary fat intake ( 48 ) . However, body weight was statistically significantly lower in the dietary group, providing biologic plausibility that a dietary change did occur in women on the low-fat eating plan. Although the dietary intervention focused on reducing fat intake, intake of other nutrients changed, as did body weight. Thus, it is possible that weight change ( 49 , 50 ) and/or dietary factors other than fat intake influenced the breast cancer outcome. Planned future analyses will examine time trends and associations of changes in body weight and other dietary factors with breast cancer recurrence.

Study strengths include the randomized study design, closely comparable anticancer systemic therapies provided to women in the two randomization groups, and the degree of dietary adherence achieved by study participants. The reduction in dietary fat intake seen in the WINS trial is similar to or greater than that reported in other trials evaluating breast cancer and diet associations either for primary prevention ( 31 , 51 ) or recurrence ( 52 ) .

Replication in clinical practice of this dietary intervention will likely require on-going counseling with a dietician trained in these techniques. The WINS low-fat eating plan was intensive, individualized, and delivered using a standardized protocol by registered dieticians who had received centralized training which included motivational interviewing techniques.

In summary, an interim efficacy analysis suggests that a lifestyle intervention reducing dietary fat intake with modest body weight loss may improve the relapse-free survival of postmenopausal breast cancer patients. Ongoing follow-up will address original protocol design plans calling for 3 years of follow-up after completion of recruitment.

References