Environmental Factors in Breast Cancer Supplement to Cancer
Diet and Breast Cancer A Review of the Prospective Observational Studies
Karin B. Michels, ScD, PhD1,2 Anshu P. Mohllajee, MPH1,2 Edith Roset-Bahmanyar, MD, MPH1 Gregory P. Beehler, MA3 Kirsten B. Moysich, PhD3
The role of diet for the risk of breast cancer is of great interest as a potentially modifiable risk factor. The evidence from prospective observational studies was reviewed and summarized on selected dietary factors, gene-diet interactions, and breast cancer incidence. Dietary factors were considered that, based on their nutritional constituents, are of particular interest in the context of breast cancer: fat intake, biomarkers of fat intake, fruit and vegetable consumption, antioxidant
Obstetrics and Gynecology Epidemiology Center, Brigham and Women’s Hospital, Harvard Medical School, Boston, Massachusetts.
vitamins (vitamins A, C, E, and beta-carotene), serum antioxidants, carbohydrate
2 Department of Epidemiology, Harvard School of Public Health, Boston, Massachusetts.
and adolescent diet. The PubMed database was searched for all prospective stu-
gene-diet interactions. Among the prospective epidemiologic studies conducted
Department of Epidemiology, Roswell Park Cancer Institute, Buffalo, New York.
intake, glycemic index and glycemic load, dairy consumption (including vitamin D), consumption of soy products and isoflavones, green tea, heterocyclic amines, dies that relate these dietary items to the incidence of breast cancer or consider on diet and breast cancer incidence and gene-diet interactions and breast cancer incidence, to date there is no association that is consistent, strong, and statistically significant, with the exception of alcohol intake, overweight, and weight gain. The apparent lack of association between diet and breast cancer may reflect a true absence of association between diet and breast cancer incidence or may be due to measurement error exceeding the variation in the diet studied, lack of sufficient follow-up, and focus on an age range of low susceptibility. The risk of breast cancer can be reduced by avoidance of weight gain in adulthood and limiting the consumption of alcohol. Cancer 2007;109(12 Suppl):2712–49. 2007 American Cancer Society.
KEYWORDS: epidemiology, clinical trials, breast cancer, diet, nutrition, fat, fruits and vegetables, carbohydrates, soy, green tea.
Rationale role for diet in cancer etiology has been suggested in part because of the large international variation in cancer rates and may be ascribed to the antioxidant properties of selected nutrients, influence on DNA repair, DNA mutations, DNA adducts, metabolic detoxification, stimulation of growth factors, and potential antiestrogenic influence of some nutrients.1 Conversely, some foods and nutrients have been suggested to increase the risk of breast cancer through an increase in circulating levels of endogenous estrogen, insulin-like growth factor 1, or other growth factors. Energy balance, the interplay of caloric intake, physical activity, and metabolic rate is another important factor impacting breast risk through mechanisms not entirely understood.
Supported by Susan G. Komen for the Cure as part of the Environmental Factors and Breast Cancer Science Review project led by Silent Spring Institute with collaborating investigators at Harvard Medical School, Roswell Park Cancer Institute, and the University of Southern California. Address for reprints: Karin B. Michels, ScD, PhD, Obstetrics and Gynecology Epidemiology Center, Brigham and Women’s Hospital, Harvard Medical School, 221 Longwood Ave., Boston, MA 02115; Fax: (617) 732-4899; E-mail: [email protected]
harvard.edu Received July 18, 2006; revision received January 26, 2007; accepted January 30, 2007.
ª 2007 American Cancer Society
Type of Studies The majority of epidemiologic studies of diet and breast cancer are case-control studies. Some ecologic studies have been conducted
DOI 10.1002/cncr.22654 Published online 14 May 2007 in Wiley InterScience (www.interscience.wiley.com).
Diet and Breast CA/Michels et al.
and in more recent years data from cohort studies have become increasingly available. Few randomized clinical trials on diet have been conducted. Whereas some important leads may arise from case-control studies, they have the potential for recall bias because individuals with breast cancer may well associate their malignancy with a previous ‘‘unhealthy diet’’ or ‘‘bad diet’’ and thus overreport foods considered less healthy, whereas healthy control subjects may not have selective recall. By using prospectively and retrospectively collected diet data, Giovannucci et al.2 found that fat intake was associated with the risk of breast cancer only if the dietary data were retrospectively assessed, but not if they were collected prospectively. A similar study nested within the Canadian National Breast Screening Study did not evidence recall bias,3 suggesting that the degree of bias is likely to differ between studies. Furthermore, selection bias is a problem in case-control studies of diet and cancer. Selecting a comparable control group for a cancer case series is difficult and participation rates among individuals approached to be controls are often lower than participation rates among cases, introducing substantial bias. Ecologic studies are subject to confounding because correlations are made on a population level and cannot provide estimates of causal associations. Migrant studies can provide important information on the role of environmental factors but are again not able to provide specific information on the role of diet and cancer. Randomized control trials of diet are problematic (unless randomizing dietary supplements) because the randomized dietary scheme has to be adhered to for many years; it is difficult for healthy participants to change their diet substantially and maintain this altered diet over a long period of time. The recently released results from the Women’s Health Initiative (WHI) Randomized Controlled Dietary Modification Trial support this concern: As evidenced by the lack of difference in biomarkers of dietary intake between the intervention and control groups, women in the intervention group were largely unable to follow the low-fat diet (20% of calories from fat) they were assigned to, having consumed a high-fat diet (38% of calories) most of their lives.4 This lack of adequate contrast in diet between the intervention and control groups leaves the WHI results difficult to interpret. The number of randomized trials on diet and breast cancer is limited. It is possible that randomizing women with breast cancer to different dietary regimens to study survival or recurrence may be more successful because individuals with a severe illness may be more compliant with an assigned food regimen than healthy individuals. The gold standard
would be a trial providing participants with all meals; however, such an approach is not feasible for cancer outcomes due to the required duration of intervention. Given the wealth of studies available on diet and breast cancer, we deemed it appropriate to restrict the present review to cohort studies and nested casecontrol studies in which diet intake is assessed before diagnosis of disease.
Diet Assessment Methods The valid and reproducible assessment of dietary intake in free-living populations is difficult because of the variety of foods and complexity of dishes consumed. With more meals prepared away from home, reporting of ingredients becomes increasingly difficult for participants in an observational study. All dietary instruments are subject to misclassification of food consumption, nutrient intake, and total caloric intake. The most commonly used dietary assessment instrument in observational studies is the food frequency questionnaire (FFQ).5 The FFQ requests information on the frequency of consumption of a prespecified list of between 50 and 200 food items. The semiquantitative FFQ provides simple measures of portion size such as a glass or a cup. The strengths of this dietary assessment instrument are that dietary preferences and average frequencies of consumption of individual food items are generally captured reasonably well.5 On the other hand, because of the restricted list of food items, other food items, which were consumed but not listed, may be missed and nutrient calculations may be misclassified, especially if the contribution to a particular nutrient is high.5 The structured format of the FFQ with a specified list of selected foods allows the FFQ to be easily scanned and transformed into a food data file. Nutrient intake can be calculated using a standard nutrient database.5 The 7-day diet diary (7DD) collects more detailed information on food consumption during a limited period of time.5 The 7DD requires participants to keep records of every food item and beverage they consume during 1 week. The free format of the 7DD makes it much more laborious to computerize as the list of recorded foods is indefinite and dieticians have to determine foods that are part of prepared dishes and ready-to-eat meals and calculate their nutrient contents. Hence, the 7DD is rarely used in large-scale epidemiologic studies, with the notable exception of Epic Norfolk.6 The 24-hour recall provides a 1-day snippet of a person’s diet, which can be compromised by day-today and seasonal variation of diet.5 Whereas the 24hour recall can provide adequate population means, it can be substantially misclassified on the individual
June 15, 2007 / Volume 109 / Number 12
level. For this reason, often several recalls are administered on different weekdays and during different seasons of the year.5 For the reasons delineated above, and because it is easiest and most cost-efficient to administer, the FFQ has emerged as the most popular dietary assessment instrument in large-scale observational studies.
Analysis of Dietary Data With a large number of foods and nutrients ascertained with a diet questionnaire the optimal analytic model is not obvious. Statistical models including a large number of foods suffer from collinearity, whereas models including only 1 food or 1 nutrient at a time are subject to confounding by foods and nutrients not included in the model. Several analytic approaches have attempted to overcome these limitations including the use of dietary patterns (using factor or cluster analysis),7 dietary indices, food groups (eg, fruits and vegetables), or the use of more complex statistical models such as hierarchical models that account for correlation of foods and nutrients using a 2-stage model.8
FINDINGS FROM COHORT STUDIES ON DIET AND BREAST CANCER Outline A large number of studies have addressed the association between diet and breast cancer. The results from these studies have dampened previous optimistic expectations that adult life diet may play an important role in breast cancer etiology. In this review we summarize the most important findings from prospective cohort studies on diet and breast cancer. We will restrict our review to the following foods and nutrients of particular interest in the context of breast cancer: fat, fruits and vegetables, antioxidants, carbohydrates, glycemic index and glycemic load, dairy and vitamin D, soy and isoflavones, green tea, and heterocyclic amines. Because the relations between weight, body mass, and the incidence of breast cancer and regular alcohol consumption and the incidence of breast cancer are fairly well established9 we will only provide brief summaries of the available evidence. On the basis of our review we will give recommendations for future directions of research on diet and breast cancer. Methods Search strategy We searched PubMed database for all articles in English published in peer-reviewed journals from January 1950 through May 2005 for evidence relevant to
diet and breast cancer. We only included studies that were cohorts, nested case-control studies, meta-analyses, or pooled analyses and that reported a point estimate with an appropriate confidence interval (CI). Among nested case-control studies we only included those that had collected information on diet prospectively before disease occurrence. We specifically examined 11 dietary exposures: fat intake, biomarkers of fat intake, fruit and vegetable consumption, antioxidant vitamins (vitamins A, C, E, and beta-carotene), serum antioxidants, carbohydrate intake, glycemic index and glycemic load, dairy consumption (including vitamin D), consumption of soy products and isoflavones, green tea, heterocyclic amines, and adolescent diet. For this review we were interested in intake of antioxidants from diet and not from supplements. Studies that included only information from supplements were excluded. We were also interested in studies that examined gene-environment interactions and searched the Medline database from January 1966 to May 2005, using the same criteria stated above. These molecular epidemiologic investigations represent efforts to examine the associations between dietary exposures and breast cancer risk among subgroups of women that are considered to be susceptible or nonsusceptible to the potential protective or harmful effects of these exposures on risk. For instance, a potential protective effect of high antioxidant intake might be more pronounced among women with genetic traits associated with high DNA repair capacity. On the other hand, women with reduced capacity to detoxify and excrete dietary carcinogen may be at greater risk of breast cancer than women with genetic traits associated with enhanced detoxification capacity. All diet and breast cancer search strategies were limited to humans and included a phrase for breast cancer, study design, and the specific dietary exposure (Table A). Gene-environment interaction strategies were limited to humans and included a phrase for breast cancer, gene-environment interaction, and the specific dietary exposure.
Study selection The search strategy identified a total of 1477 articles. We reviewed the abstracts of these articles to determine whether they met our criteria for review. Once we included a specific article in our review, we examined reference lists for additional articles. We did not attempt to identify unpublished articles or abstracts from scientific conferences. If there were 2 or more reports from 1 cohort, we included the report with the most up-to-date analysis. For fat intake we included 19 cohort studies,10–28 1 meta-analysis,29 and 1 pooled
Diet and Breast CA/Michels et al.
TABLE A Search Strategies for Systematic Review
TABLE B Abbreviations of Studies in Tables
Breast Cancer Detection Demonstration Project Follow Up Cohort Study Canadian National Breast Screening Study Cancer Prevention Study II Nutrition Cohort California Teacher’s Study European Prospective Investigation Into Cancer and Nutrition Finnish Social Institution’s Mobile Clinic Health Survey Japan Public Health Center-based Prospective Study on Cancer and Cardiovascular Disease Iowa Women’s Health Study Malmo Diet Cancer Cohort Monitoring of Trends in Cardiovascular diseases Monitoring Project on Cardiovascular Disease Risk Factors Mammary Screening Project Netherlands Cohort Study National Health and Nutrition Examination Survey I National Health Epidemiologic Follow-up Study Nurses’ Health Study Nurses’ Health Study II Norwegian National Health Screening Services Norwegian Women and Cancer Study New York University Women’s Health Study Hormone and Diet Etiology of Breast Cancer Swedish Mammography Screening Cohort Vasterbotten Intervention Project Women’s Health Study
Diet And Breast Cancer Search Strategy In PubMed Overall search (‘‘Breast Neoplasms’’[MeSH] OR ‘‘breast cancer’’[All strategy Fields]) AND (‘‘Cohort Studies’’[MeSH] OR ‘‘meta analysis’’[Publication Type] OR cohort* OR ‘‘Retrospective Studies’’[MeSH]) Fat (‘‘Dietary Fats/adverse effects’’[MeSH] OR ‘‘Fatty Acids/ adverse effects’’[MeSH] OR ‘‘fat’’[All Fields]) Biomarkers of (‘‘Fatty Acids/blood’’[MeSH]); fruits and vegetables fat intake (‘‘Fruit’’[MeSH] OR ‘‘Vegetables’’[MeSH] OR ‘‘fruit and vegetable’’[All Fields] OR fruit [text word] or fruit* OR vegetable [text word] OR vegetable*) Antioxidants (‘‘Antioxidants’’[MeSH] OR ‘‘Vitamin A’’[MeSH] OR ‘‘Ascorbic Acid’’[MeSH] OR ‘‘Vitamin E’’[MeSH] OR ‘‘Carotenoids’’[MeSH] OR antioxidants OR vitamin A OR ascorbic acid OR vitamin E or carotenoids) Carbohydrates (‘‘Glycemic Index’’[MeSH] OR ‘‘Carbohydrates’’[MeSH] OR glycemic index or carbohydrates) Dairy (‘‘Milk’’ [MeSH] OR ‘‘Dairy Products’’[MeSH] OR (‘‘milk’’[MeSH Terms] OR milk[Text Word]) OR dairy[All Fields]) Vitamin D (‘‘Vitamin D’’[MeSH] OR (‘‘vitamin d’’[MeSH Terms] OR vitamin D[Text Word])) Soy (‘‘Soy Foods’’[MeSH] OR ‘‘Isoflavones’’[MeSH] OR ‘‘Phytoestrogens’’[MeSH] OR soy* [All Fields] OR (‘‘isoflavones’’[MeSH Terms] OR isoflavones[Text Word]) OR (‘‘phytoestrogens’’[MeSH Terms] OR ‘‘phytoestrogens’’[Pharmacological Action] OR phytoestrogens[Text Word])) Heterocyclic (‘‘Heterocyclic Compounds’’[MeSH] OR heterocyclic[All amines Fields] AND (‘‘amines’’[MeSH Terms] OR amines[Text Word])) OR fried[All Fields] AND (‘‘meat’’[MeSH Terms] OR meat[Text Word]) OR (well[All Fields] AND (‘‘meat’’[MeSH Terms] OR meat[Text Word])) Adolescent diet ((‘‘Adolescent’’[MeSH] or adolescence or young adult or childhood AND (‘‘diet’’[MeSH Terms] OR diet[Text Word] or nutrition)) Gene-Environment Interaction Search Strategy in Medline Breast cancer Subject heading: breast neoplasms Keywords: breast cancer Gene-Environment Subject heading: genotype; polymorphism (genetics); interaction epidemiology, molecular Keywords: gene-environment interactions; polymorphism; molecular epidemiology; CYP; COMT; MnSOD; GST; SULT; EGFR; XRCC, MTHFR, MPO, PROGINS, MAT Diet Subject heading: diet, Mediterranean; diet, surveys; diet, cariogenic; diet, protein-restricted; diet, reducing; diet, therapy; diet, atherogenic; diet, macrobiotic; diet, records; diet, fads; diet, diabetic; diet, vegetarian; diet, sodium-restricted; diet, fat-restricted Keywords: Diet; adolescent diet Diet specific Subject heading: fats, unsaturated; fats, saturated; dietary nutrients, fats; fats; fruit; vegetables; antioxidants; dietary exposures, carbohydrates; carbohydrates; glycemic index; dietary or measures fiber; food handling; meat; meat products; heterocyclic compounds; milk; dairy products; soy; phytoestrogens; vitamin D Keywords: monounsaturated fat; saturated fat; polyunsaturated fat; fruit and vegetable; food processing; heterocyclic amines; dairy; soy milk; soy foods
CNBSS CPS II CTS EPIC FSIIMCHS JPHC IWHS MDC MONICA MPCDRF MSP NCS NHANES I NHEFS NHS NHS II NHSS NOWAC NYUWHS ORDET SMSC VIP WHS
analysis.30 The meta-analysis and the pooled analysis included several of the 19 cohorts identified in our review. For fruits and vegetables we included 6 cohorts,31–36 1 meta-analysis37 and 1 pooled analysis38; 11 cohorts for carbohydrates11–13,18,21,22,39–43; 11 cohorts for antioxidants16,18,32,33,35,36,44–48; and 11 cohorts for milk10,13,14,24,25,27,31,49–52; 2 cohorts for vitamin D51,53; 5 cohorts for soy18,31,54–56; 5 cohorts31,57–59 and 1 pooled analysis for green tea58; 1 cohort for heterocyclic amines60; 2 retrospective cohorts for adolescent diet49,51; and 4 nested case-control studies for biomarkers of fat composition.61–64 We also included 8 studies analyzing biomarkers of antioxidants.65–72 We also included 1 metaanalysis for milk, biomarkers of fat composition, and green tea.73–75 We also identified 7 nested case-control studies that examined gene-environment interactions.59,76–81
Data extraction and synthesis For each of the dietary exposures we have summarized the evidence in tables and presented evidence separately for pre- and postmenopausal status, if available. Information for each table was taken directly from the published manuscript of each individual study. When determining the size of the popu-
June 15, 2007 / Volume 109 / Number 12
lation for each study, we extrapolated information from the abstract or the methods section of the article. If the numbers differ, the number was identified from the methods section that best described the diet cohort for the study that was used in the analysis. We provide covariate-adjusted measures of association. In all studies, age and total caloric intake was adjusted unless noted. Although we have listed all variables that were either considered confounders or adjusted for in the analysis, we systematically record adjustment for body mass index (BMI) and family predisposition to breast cancer, which are factors likely associated with dietary patterns. We have not provided information about the participation rates because most of the studies are part of larger cohorts, ie, the Nurses’ Health Study (NHS) or the Iowa Women’s Study, in which no information is given about the original participant rate. The average follow-up time in the cohort studies was 8 to 9 years. Table B contains all the abbreviations for the studies included in the review. In addition, a critical review for each study was entered in a database (accessible at www.silentspring.org/sciencereview and www. komen.org/environment).
Studies on Diet and Breast Cancer Weight, body mass, and weight gain Epidemiologic evidence consistently demonstrates that a high BMI increases the incidence of breast cancer after menopause.82 Pooling data from 8 prospective cohort studies including 337,819 women and 3208 incident cases of invasive breast cancer after menopause resulted in a relative risk (RR) ¼ 1.26 (95% CI, 1.09–1.46) for postmenopausal breast cancer comparing women with a BMI above 28 kg/m2 to women with a BMI of less than 21.83 Adult weight gain has been associated with postmenopausal breast cancer incidence in several studies.84–87 Alcohol Regular alcohol consumption has been consistently linked to a modest increase in the incidence of breast cancer. In a recent pooled analyses of 6 prospective cohort studies including 322,647 women and 4335 cases of incident invasive breast cancer, consumption of each additional 10 g of alcohol per day was associated with a 9% (95% CI, 4%–13%) increase in the risk of breast cancer.9 Fat intake Fat intake may be related to the risk of breast cancer because it may raise endogenous estrogen levels. A total of 19 cohort or nested case-control studies have been conducted on the association between fat
intake and breast cancer incidence (Table 1).10–28 Two of these studies considered premenopausal women separately17,24 and 6 studies considered postmenopausal women separately.16,17,22,23,26,27 Overall fat intake was not related to the incidence of breast cancer. The most notable exception is NHANES I, which found a marked decrease in the risk of breast cancer associated with total fat intake (RR ¼ 0.34; 95% CI, 0.16–0.73), but diet was assessed using a 24hour recall.20 An increase in postmenopausal breast cancer incidence associated with total fat intake was found in the Rancho Bernardo study (RR ¼ 2.01 per 28 g increase in fat intake; 95% CI, 1.19–3.41),12 and for the highest category of total fat intake in the Italian ORDET study (RR ¼ 3.47; 95% CI, 1.43–8.44).11 Associations of borderline statistical significance were reported from the Nurses’ Health Study II (NHS II)13 (RR ¼ 1.25; 95% CI, 0.98–1.59) and among postemenopausal women of the Malmo Diet Cancer Cohort (MDC)23 (RR ¼ 1.36; 95% CI, 0.96–1.94), comparing the highest versus the lowest quintile of intake. NHS II participants in the highest quintile of animal fat intake had a relative risk of breast cancer of 1.33 (95% CI, 1.02–1.73) compared with women in the lowest quintile.13 Among other studies, Gaard et al.14 observed an increase in the association between monounsaturated fat intake and breast cancer (RR ¼ 1.72; 95% CI, 1.19–2.49) in a Norwegian study. Two studies pooled the prospective evidence (Table 2): Boyd et al.29 in 1993 conducted a metaanalysis including 7 prospective studies with a total of about 3000 cases of breast cancer and found an overall RR ¼ 1.03 (95% CI, 0.92–1.96) for the highest category of intake compared to the lowest. Smith-Warner et al.30 in 2001 pooled 8 prospective studies on fat intake and breast cancer and derived an overall RR ¼ 1.00 (95% CI, 0.98–1.03) per 5% increase in fat. In this pooled analysis the RR associated with intake of saturated fat was 1.09 (95% CI, 1.00–1.99) and of animal fat 1.01 (95% CI, 0.96–1.06) per 5% increase in fat.
Biomarkers of fat intake The use of biochemical markers that reflect dietary intake has potential advantages compared with the assessment of dietary intake through self-reports, as reporting errors and limitations of food composition tables are avoided. However, levels of biomarkers can be affected by several factors other than diet, such as smoking and metabolic factors. Furthermore, biomarkers that reflect dietary intake with sufficient accuracy are available for only a limited number of foods and nutrients. In particular, few recovery biomarkers have been identified. Recovery
Size of cohort/no. of cases; years followed
56,837/519 (1182 controls) Nested case-control study 1982–1987
14,291/180 (829 controls) Nested case-control 1985–1991
Howe19 1991 Canada CNBSS
Toniolo25 1994 USA NYUWHS
Gaard14 1995 Norway Norwegian National Health Screening Services Byrne10 1996 USA NHEFS/NHANES 1 6156/53 1982/1984–1986/1987
Mills24 1989 USA Seventh Day Adventist Study Knekt21 1990
Premenopausal and postmenopausal women combined 5485/99 Jones20 1987 USA 1971/1975 (10 y mean follow-up) NHANES 1
Author, year, country, parent cohort
TABLE 1 Studies on Fat intake and Breast Cancer Incidence
1.36 (0.50–3.73) 2.70 (0.99–7.37) 1.23 (0.55–2.75) 1.30 (0.90–1.88)
H vs 1 tertile (97.3 vs