Breast cancer is one of the most common cancers among women in developed countries and significant advances in early detection and treatment have led to high 10-year survival rates and large prevalent survivor populations. Clinical outcome among women diagnosed with apparently similar cancers varies considerably, and does not appear to be explained by treatment differences. Laboratory studies have identified multiple bioactive constituents in foods that appear to either promote carcinogenesis or protect against it. As a result, there is considerable interest in the question of whether dietary pattern can influence a woman’s prognosis following breast cancer.
Prospective observational studies, or cohort studies, measure the dietary component or pattern and then monitor prognosis over time. These association studies do not provide sufficient evidence to conclude that changing dietary pattern would alter prognosis. The “gold standard” studies are randomized trials in which some participants are assigned to make a major dietary change. Several prospective observational studies have investigated the link between dietary pattern and breast cancer prognosis, and to date two randomized clinical trials have tested this association.
However, most dietary studies are compromised by measurement errors in the commonly used self-reported dietary patterns [1
] and there are few proven biomarkers of exposure to validate self-report. The extent of these measurement errors has been highlighted by two separate studies that have demonstrated that the self-report method used changes the study outcome [2
]. Another study showed an effect with a biomarker that was not seen with self-reported intake [4
Reviews of the observational studies of diet and breast cancer prognosis have identified 14 studies since 1990 [6
]. However, many of these studies measured pre-breast cancer dietary pattern shortly following diagnosis with breast cancer. As diets commonly change following diagnosis [22
], dietary recalls over extended periods that include diagnosis have additional measurement problems and are excluded from this review.
In 1994, the Canadian National Breast Screening Study reported 5-year survival data from the subsample of 678 breast cancer cases that had completed a diet history prior to diagnosis [23
]. In this study, lower saturated fat intake (but not intake of total fat or oleic acid) was associated with survival and women in the highest versus lowest quartiles of intake of beta carotene and vitamin C from food sources appeared protected.
The Nurses’ Health Study is a well known long-term investigation of a cohort of female registered nurses. Holmes et al [15
] reported on a subsample (n=1982) who were diagnosed with breast cancer from 1976–1990 and completed a dietary assessment in the post diagnosis period and followed through 1994. Consumption of fruit, red meat, or grain-based products was not associated with either all-cause mortality or breast carcinoma death. However, among women with metastatic cancer, consumption of vegetables, carotenoids and fiber had lower mortality rates. Kroenke et al [17
] reported on the 2619 nurses who were diagnosed from 1982–1998 and completed a dietary questionnaire at least 1 year post diagnosis. Over a median 9 years of follow-up, 9% of participants died from breast cancer. Using factor analysis, researchers identified two dietary patterns: one in which vegetables, fruit, whole grains, low fat dairy, poultry, fish and fiber consumption increased across quintiles (prudent dietary pattern) and the other in which consumption of refined grains, red meat, processed meat, high fat dairy, saturated fat and dessert increased across quintiles (western dietary pattern). Neither dietary pattern was related to breast cancer mortality. However, the prudent dietary pattern reduced and western dietary pattern increased deaths from other causes. Thus, the no-association result for dietary fat consumption was consistent across these two overlapping cohorts, but the conclusions for a mortality association with the consumption of vegetables, fruit and fiber were not.
The Women’s Healthy Eating and Living (WHEL) Study reported observational data from women randomized to the comparison group. Dietary assessment occurred an average of 2 years post diagnosis and dietary patterns were validated with a plasma carotenoid concentration, the accepted biomarker of vegetable and fruit intake. The first report [19
] focused on 205 women who had a breast cancer event prior to June 2004 (average of 5 years follow-up) and identified those in the lowest baseline tertile (compared to all others) of circulating carotenoid concentrations as having an approximate 40% increased risk. The second report [24**
] focused on overall mortality and included follow-up through December 2005. No association was found for energy from fat. Vegetable-fruit consumption and physical activity were weakly associated by themselves but there was an interaction between them with the combination of vegetable-fruit intake (at least 5 vegetables and fruits/day) and physical activity (equivalent to walking briskly for 30 minutes, 6 days a week) reduced the risk of dying from breast cancer by half, regardless of weight, although fewer obese women were physically active with a healthy dietary pattern (16% vs. 30%). The effect was stronger in women who had hormone receptor–positive cancers.
Comparing WHEL and WINS randomized trials
Both the Women’s Intervention Nutrition Study (WINS) and the WHEL Study randomized women who had been diagnosed with early stage breast cancer in the United States during the 1990s, before the widespread use of aromatase inhibitors. Both studies enrolled women with tumors ≥ 1 cm, but there were three important differences in the eligibility criteria. The first was age: WINS enrolled postmenopausal women aged 48–79 years at diagnosis whereas WHEL enrolled women aged 18–70 years. However, to compare studies it is possible to categorize WHEL participants by menopausal status (). Both the WHEL (n=2448) and the WINS (n=2437) studies had almost equivalently-sized populations of postmenopausal women.
Cancer characteristics of WINS and WHEL Study populations
The second difference was the time between diagnosis and enrollment. WINS enrolled women within 1 year of diagnosis, whereas the WHEL Study enrolled women within 4 years of diagnosis. () Thus, the WHEL Study undersampled the population who would recur within 4 years of diagnosis and is really a study of breast cancer events between 2 and 10 years from diagnosis [25
]. WINS on the other hand is focused much more on breast cancer events within the first 5 years from diagnosis. The third difference was that WINS excluded women with worse prognoses (based on tumor size and nodes), thus WHEL postmenopausal women had more advanced cancer characteristics at diagnosis. Over half of WINS participants were stage 1 compared to just over a third of the WHEL sample, and 25% of WHEL women were stage IIB or IIIA compared to only 10% of WINS. In both studies, about one fifth of postmenopausal women had ER negative tumors at diagnosis.
Despite the above differences in cancer characteristics, both studies recruited breast cancer survivors from similar populations. Socio-demographic characteristics of WINS and WHEL participants were markedly similar () in terms of age (55 vs 58 years), race (85% were Caucasian) and education (almost 50% college graduates). Approximately 27% of postmenopausal participants in both studies were obese and the mean energy intake from fat was 29%.
Comparison of WINS and WHEL Study populations
These WINS and WHEL comparisons include WHEL participants who were pre and peri menopausal. These women were younger, more highly educated and less likely to be obese. () They were also more likely to be recruited in the first year and more likely to be ER negative. ()
The effect of the WINS intervention
Both the WINS and the WHEL studies used sets of four 24 hour recalls to assess current dietary pattern at multiple points in time. The WINS intervention focused on reducing fat intake to 15% of energy. At 1 year, the intervention reported a 9% between-group difference in dietary fat intake [26**
] (). While it appeared that this between-group difference was maintained through 5 years, the low response rate for dietary assessments makes this result questionable (year 3 response ~70%; year 5 response ~40%). In reporting a completers-only analysis, the authors assumed that diets of the large proportion of nonrespondents were similar to the diets of the respondents – a very questionable assumption. Using a more conservative assumption (that nonrespondents didn’t lower their fat intake), an estimate of the between-group difference is ~6% at year 3 and ~3.5% at year 5. WINS also reported that the intervention group lost weight, resulting in a between group difference of 2.3 kg in year 1 (conservative assumption= ~1.6 kg) to 2.7 kg in year 5 (conservative assumption = ~1.1 kg).
Change achieved by WINS intervention
The effect of the WHEL Study intervention
The WHEL Study intervention encouraged women to adopt a daily dietary pattern including 5 vegetable servings, 16 oz of vegetable juice, 3 fruit servings, 30g of fiber and 20% energy from fat. Unlike WINS, there was only a small decline in completion rates for dietary assessment and 85% were assessed at year 6 (). The study has presented dietary change data in two ways: using a completers only analysis [24**
] as well as with the conservative assumption that nonresponders did not change their dietary pattern [27
Change achieved by WHEL Study intervention
To compare dietary change with WINS, we present the completers only data in [24**
]. The intervention was associated with a between-group difference of 4.7 vegetable-fruit servings/d at 1 year which decreased to 3 servings/d by 6 years. The between-group difference for fiber consumption was 8 g/d at 1 year, decreasing to 5g/d at year 6. The between-group difference in energy from fat was 5.7% at 1 year, decreasing to 3.5% at 6 years, similar to the conservative estimate of the WINS intervention effect. Body weight changed little over 6 years. The conservative analysis concluded that, at 4years, the relative between-group differences were 65% for vegetables (including juice), 25% for fruit, 30% for fiber, and 13% for energy from fat.
Study outcomes: WINS vs. WHEL
The WINS and the WHEL studies have reported conflicting results. WINS assessed the summary variable “any breast cancer event” for their analysis and the 2.6% between-group difference was borderline significant in the planned stratified log rank test (p=0.077) although statistically significant in the multivariate Cox model (p=0.034) [26**
]. Further, an exploratory analysis suggested that the between-group differences in breast cancer events might be confined to the 20% of the sample with initial tumors that were ER negative, however, the interaction between dietary intervention and hormone receptor status was not statistically significant.
The WHEL Study [24**
] reported no between-group differences in either any breast cancer event (adjusted Hazard Ratio= 0.63) or in overall mortality (HRadj
= 0.43). Additionally, for breast cancer events, the likelihood ratio tests for baseline dietary pattern by study group interaction were not significant for vegetable-fruit, fiber or energy from fat. However, for overall mortality, the likelihood ratio test was significant for quartiles of energy from fat (p=0.04), although the effect seemed to be in only the second quartile of baseline energy from fat and the intervention group difference was not in a protective direction. Further, hormone receptor status did not differ between groups (p=0.85).
Given the considerable similarities between these randomized trials, the marked difference in these findings require further investigation. presents details of the different study outcomes for postmenopausal women. As expected from the initial cancer characteristics () and the longer follow-up period, WHEL reported more study events than WINS () and particularly more distal recurrences that Tang [28
] has argued are the most important study outcome for patients with a pre-existing breast cancer diagnosis. The size of the between group difference in these distal recurrences was not that different for both of these studies (WINS = −1.1, WHEL = −0.9).
Comparison of cancer outcomes in WINS and WHEL Study populations
The difference in the studies occurred in the between-group differences in the proportion of women who had local recurrences and new primary breast cancer events (WINS = −1.3%, WHEL = +0.5%), and, in particular, the WINS group effect was among women who had a lumpectomy for their initial cancer and had an ipsilateral breast cancer recurrence during the follow-up period (C= 2.1%; I=1.1%). It is possible that the lack of a WHEL effect in this sub-population of women reflected the fact that WHEL underestimated events in the first few years from diagnosis. Another plausible explanation is that this finding is an artifact, particularly as the main hypothesis was only marginally statistically significant.
WHEL study secondary analysis
The initial WHEL Study protocol [25
] postulated that diet could affect prognosis by reducing circulating estrogen concentrations. The study’s nested case-control analysis showed that women with higher circulating estradiol concentrations at baseline were less likely to report hot flashes at that time and more likely to have a secondary cancer event during the follow-up period [29*
]. In two separate studies, early stage breast cancer survivors who did not report hot flashes shortly after treatment were ~30% more likely to have additional breast cancer events [30*
]. Given that the WHEL Study intervention has been shown to reduce circulating estrogen concentrations [32
], the study undertook a secondary analysis to see if an intervention effect may have been limited to this group of women who did not report hot flashes at baseline [33**
About one third of WHEL participants (n=900) were in the subgroup who did not report hot flashes at baseline (HF-) [24**
]. The intervention participants in this HF- subgroup had a similarly large change in dietary pattern (vegetables, fruit, fiber, and energy from fat) over the course of the study to that seen in the overall intervention group. Among the comparison HF- subgroup, 76.4% of the comparison group and 83.9% of the intervention group remained breast cancer free. (p=0.002) This effect was almost entirely restricted to the distal recurrence group (C=15.9%, I=9.4%), and the ~60% lower event rate in the intervention group was not explained by other variables in a series of sensitivity analyses [33**
]. This effect did not differ significantly by hormone receptor status (p=0.63).