Search tips
Search criteria 


Logo of nihpaAbout Author manuscriptsSubmit a manuscriptHHS Public Access; Author Manuscript; Accepted for publication in peer reviewed journal;
Breast Cancer Res Treat. Author manuscript; available in PMC 2013 April 1.
Published in final edited form as:
PMCID: PMC3574573

A dietary pattern derived to correlate with estrogens and risk of postmenopausal breast cancer


Circulating estrogens are an established risk factor for breast cancer and some data suggested that diet may influence estrogen levels. Therefore, using a subsample (n=550) of women from a large cohort, we applied reduced rank regression to identify a dietary pattern that is correlated with estradiol and estrone sulfate. We then adapted the pattern to be used with the full cohort (n=67,802) and prospectively assessed its association with postmenopausal breast cancer. The estrogen food pattern, characterized by higher intakes of red meat, legumes, and pizza, but lower intakes of coffee and whole grains, was modestly but significantly correlated with estradiol (r=0.14) and estrone sulfate (r=0.20). During 22 years of follow-up, we ascertained 4,596 incident breast cancer, with 2,938 estrogen receptor positive tumors and 689 estrogen receptor negative tumors. However, after adjusting for potential confounders, we did not observe any association with overall, estrogen receptor positive, or estrogen receptor negative breast cancer. In conclusion, diet pattern appeared to only have modest association with estrogens, and was not associated to postmenopausal breast cancer risk. Although these results were null, it should be repeated in other populations as differences in food intake may yield a dietary pattern with stronger association with estrogens.

Keywords: diet, reduced rank regression, estrogen, breast cancer


Serum estrogen is an established risk factor for post-menopausal breast cancer [1]. However, data on the potential influence of food on estrogen levels in post-menopausal women is scarce. There are suggestions that a higher fiber [2] intake and adherence to the Mediterranean diet [3] may be associated with lower estradiol levels, while higher milk [4] intake may be associated with higher levels. In the Nurses’ Health Study, we previously found that the Alternate Healthy Eating Index (AHEI) was associated with lower levels of estradiol among overweight post-menopausal women [5]. Because the AHEI was not designed to correlate with estrogen, there might be other food combinations that could have stronger associations. Reduced rank regression (RRR) is a statistical procedure that can identify specific foods that would most strongly correlate with biomarkers. This process uses “response variables” (biomarkers), and derives food patterns that are associated with them [6]. The patterns can in turn be used to predict cancer risk. The advantage of this method is that it generates patterns linked to biological pathways of cancer development. This approach has been used to derive dietary patterns that is simultaneously associated to multiple biomarkers for predicting diabetes [7,8] and cardiovascular [9,10] disease. However, it has not been widely used in cancer research. Two studies have used the RRR approach to study breast cancer. One identified a food pattern associated glycemic load but this pattern was not associated with breast cancer [11], and the other identified a pattern associated with fatty acids [12] intake and noted a direct association. However, the response variables used in these two studies were dietary factors and not biomarkers, thus the ability to elucidate the mechanistic link between food and disease remained limited.

In this analysis, we used RRR to identify a dietary pattern that is significantly associated with estradiol and estrone sulfate, and applied this pattern to a large cohort of women and assessed its association with postmenopausal breast cancer. These two estrogens were selected for their established association with breast cancer. We also stratified our analysis by estrogen receptor status hypothesizing that the association may be stronger with estrogen receptor positive tumors.


Study population

The Nurses’ Health Study (NHS) is a cohort study established in 1976 and consists of 121,700 female nurses aged 30–55 years living in 11 U.S. states at the time [13]. Questionnaires are sent biennially to collect medical, lifestyle, and other health-related information. In 1980, participants completed a 61-item food frequency questionnaire (FFQ). This was expanded to 116 items in 1984 and similar FFQs were sent in 1986, 1990, 1994, 1998, and 2002. For this analysis, we used 1986 as baseline for the cancer analyses because this FFQ was used used to compute the RRR pattern in the subsample of women with estrogen data. After excluding those with a history of cancer (except non-melanoma skin cancer), we included 67,802 women with follow-up from 1986 through 2008. This study was approved by the Institutional Review Board of the Brigham and Women's Hospital, Boston, MA.

Biomarker subsample and essay

Blood was collected in 1989–1990 in a subsample of women from the Nurses’ Health Study [14]. Each willing participant was sent a blood collection kit containing instructions and needed supplies (e.g., blood tubes and needles). Participants made arrangements for blood to be drawn, packaged in an enclosed cool pack, and sent to the laboratory by overnight courier. Almost all the samples arrived within 26 hr of the blood draw. Upon their arrival at the laboratory, the whole blood samples were centrifuged and aliquotted and stored at −130°C or colder. The lifestyles and dietary intakes of women who returned a blood sample were in general similar to those who did not provide a blood sample. The women in this analysis were controls for previous nested case-control studies in breast cancer [15] and wrist factures (n=550). Estradiol and estrone sulfate levels were measured using radioimmunoassay and intra-assay coefficients of variation were less than 15%.

Dietary assessment

Self-administered semi-quantitative FFQs were designed to assess average food intake over the preceding year. A standard portion size and nine possible consumption frequency categories, from “never, or <1/month” to “6+ times per day” were given for each food. Total energy and nutrient intake was calculated by using the sum from all foods. Previous validation studies revealed reasonably good correlations between food intake and energy-adjusted nutrients assessed by the FFQ and multiple food records completed over the preceding year [16,17].

Case ascertainment

Incident breast cancer was ascertained between 1986 and 2006. In each biennial questionnaire, participants self-report any diagnosis of breast cancer in the previous 2 years. We then sought permission to obtain medical records to confirm the diagnosis. 99% of self-reported cases were confirmed by medical records. We also included 1% of cases confirmed by the participants. Estrogen and progesterone receptor status was obtained from pathology reports and each receptor was classified as positive, negative, or uncertain. Deaths were reported by the postal service, family members, or by searching the National Death Index. In this study, we included only postmenopausal breast cancer cases to reduce potential etiologic heterogeneity.

Covariate ascertainment

Body mass index (BMI) was calculated from weight reported on each biennial questionnaire and height reported on the first questionnaire. Smoking, history of hypertension, aspirin use, multivitamin intake, menopausal status and use of postmenopausal hormone therapy, history of benign breast disease, parity, and age at first birth were assessed every 2 years. Family history of breast cancer was assessed six times during follow-up. Leisure-time physical activity was measured with validated questions on 10 common activities every 2 years [18].

Derivation of dietary pattern

To account for laboratory drift because biomarkers were not assayed all at the same time, the biomarkers were adjusted by dividing the original value with a ratio that is calculated by dividing the geometric mean of the batch by the mean of all batches [19]. To derive a pattern using RRR, we used average diet intake collected by a validated FFQ in 1986 and 1990 to reduce within subject variation. Then we grouped foods into 37 groups and applied RRR to 550 women who had both estradiol and estrone sulfate (ES) data [6]. RRR produces a linear function of food groups that explains variations in the response variables, which in this case was estradiol and estrone sulfate. The first factor (i.e. pattern) derived from RRR was retained as it had the strongest correlation with the two estrogens. Because the RRR factor score cannot be directly computed for women without estradiol and ES values, we generated a simplified pattern using stepwise linear regressions with the RRR factor score as outcome and food groups as predictors. Food groups were retained to form a simplified pattern if the p-value of the coefficient in the stepwise regression model was less than 0.1. Simplified pattern scores were calculated for all postmenopausal women in the cohort by summing the retained standardized food group items for each FFQ collected in 1986, 1990, 1994, 1998, 2002, and 2006.

Statistical analysis

We used Cox proportional hazard models to assess the association between the simplified pattern score and risk of breast cancer between 1986 and 2008 (n=67,802 at baseline). To reduce random within-person variation and to best represent long-term dietary intake, we calculated cumulative averages of the scores from our repeated FFQs [20]. We adjusted for age, energy intake (quintiles), multivitamin use (yes/no), alcohol intake (4 categories), weight change since age 18 (7 categories), BMI (5 categories), BMI at age 18 (4 categories), family history (yes/no), history of benign breast disease (yes/no), physical activity in METs (quintiles), and age at menopause and post-menopausal hormone use (11 categories). We also examined associations separately by estrogen receptor status.


Estradiol and estrone sulfate were moderately correlated (r=0.57, p<0.0001). The first RRR pattern was modestly but significantly correlated with estradiol (r=0.22, p<0.0001) and ES (r=0.24, p<0.0001). This pattern explained 6.5 % of variation in estradiol and 5.7% variation in estrone sulfate. Stepwise linear regression retained a simplified pattern (i.e. “estrogen food pattern”) characterized by higher intake of red meat, legumes, and pizza, but lower intakes of coffee and whole grains, with a correlation coefficient of 0.69 (p<0.0001) with the original RRR pattern score. The estrogen food pattern was weakly but significantly correlated with estradiol (r=0.14, p<0.0001) and ES (r=0.20, p<0.0001) (table 1). Partial correlation adjusting for BMI only slightly attenuated the correlation coefficients, and they remained statistically significant. In a sensitivity analysis, we included estrone into the RRR procedure but that did not improve the correlation between the resulting pattern and the three estrogens, therefore we did not include estrone sulfate in the RRR procedure in this analysis. Among women who provided blood samples, changes in food group intakes were similar across quintiles of estradiol and estrone sulfate (table 2).

Table 1
Pearson correlation coefficients (p value) for the estrogen food pattern and its components, and RRR pattern score, estradiol, and estrone sulfate (n=550).
Table 2
Mean dietary characteristics (average of 1986 and 1990 FFQ) (standard error) according to quintiles of estrogen levels

In the main cohort, we ascertained 4596 incident breast cancer over 22 years of follow-up, with clear classification of 2938 estrogen receptor positive tumors and 689 estrogen receptor negative tumors. Women with high estrogen food pattern score had higher BMI, consumed more calories, but less likely to smoke, consumed less folate, and physically less active (table 3).

Table 3
Age standardized baseline (1986) characteristics according to quintiles of estrogen food pattern score*

Multivariate analysis showed no association between the estrogen food pattern score and total breast cancer, ER+ or ER− tumors (table 4). Results stratified by progesterone receptor status did not differ. There was no effect modification by BMI. Adding estrone to RRR did not improve the correlation between the pattern and estrogens and the derived pattern remained null for breast cancer (total, ER+ and ER−).

Table 4
Relative risk (95% CI) for quintiles of estrogen pattern score and risk of post-menopausal breast cancer


This is the first study that derived a dietary pattern specifically correlated with estrogen levels. The dietary pattern derived with this method had modest association with estradiol and estrone sulfate but was not associated with post-menopausal breast cancer. Data on the association between dietary patterns and estrogen levels are scarce. In this cohort, a higher Alternate Healthy Eating Index (AHEI) was associated with lower concentrations of estradiol [5] and in a small 6-month intervention study, the Mediterranean diet resulted in lower levels of estradiol without any appreciable difference in weight change compared with the typical diet group [3]. Although the AHEI was associated with a lower risk of ER- breast cancer [21], this subtype is unlikely to be strongly influenced by estrogen levels. Therefore, although lower estrogen was observed with higher AHEI score, the components of AHEI likely influenced breast cancer risk through other mechanisms.

RRR overcomes the weak association between single foods and biomarker levels by being able to derive a dietary pattern that is correlated with multiple biomarkers. Although the RRR procedure has been applied to breast cancer in two studies, results were mixed and the response variables were dietary factors [11,12]. A dietary pattern associated with fatty acid intake, characterized by low intake of bread, fruit juices, but high intake of processed meat, fish, and fats was associated with fatty acids intake, was associated with higher risk of breast cancer [12]. On the other hand, a diet correlated with glycemic load, characterized by higher intakes of sweets, refined grains, and salty snacks, had no association with breast cancer [11]. The strength of association between RRR patterns and breast cancer depends on the correlation between the RRR patterns and biomarkers, as well as the strength of association between biomarkers and breast cancer. Although we observed a significant association between the dietary pattern and estrogens, the correlation may be was insufficiently strong to have represented a change in a magnitude that would influence breast cancer risk.

This study included a substantial sample to identify food groups correlated with estrogen, and a large number of cases to separately examine ER+ and ER− tumors. We had detailed information on potential confounders but since lifestyle factors were self-reported, some level of measurement error was inevitable. The sample size for the estrogens was substantial, but that does not guarantee that the distribution would sufficiently reflect the distribution of these biomarkers in the main cohort.

In conclusion, although a diet pattern characterized by higher intakes of red meat, legumes, and pizza, but lower intakes of coffee and whole grains was modestly associated with estradiol and ES, it was not associated with postmenopausal breast cancer risk. Although we did not identify a dietary pattern with strong association with estrogens, other populations may consume different foods in different amounts that may find otherwise. Therefore, the association between food groups and estrogen levels and their relationship with breast cancer should be explored in other populations.


We would like to thank the participants and staff of the Nurses' Health Study, for their valuable contributions as well as the following state cancer registries for their help: AL, AZ, AR, CA, CO, CT, DE, FL, GA, ID, IL, IN, IA, KY, LA, ME, MD, MA, MI, NE, NH, NJ, NY, NC, ND, OH, OK, OR, PA, RI, SC, TN, TX, VA, WA, WY.

This work is funded by National Institutes of Health grants CA87969, HL60712, CA95589, and 1U54CA155626-01.


Financial disclosure: The authors declare that they have no conflict of interest.


1. Key TJ, Appleby PN, Reeves GK, Roddam A, Dorgan JF, Longcope C, Stanczyk FZ, Stephenson HEJ, Falk RT, Miller R, Schatzkin A, Allen DS, Fentiman IS, Key TJ, Wang DY, Dowsett M, Thomas HV, Hankinson SE, Toniolo P, Akhmedkhanov a, Koenig K, Shore RE, Zeleniuch-Jacquotte A, Berrino F, Muti P, Micheli a, Krogh V, Sieri s, Pala V, Venturelli E, Secreto G, Barrett-Connor E, Laughlin GA, Kabuto M, Akiba S, Stevens RG, Neriishi K, Land CE, Cauley JA, Kuller LH, Cummings SR, Helzlsouer KJ, Alberg AJ, Bush TL, Comstock GW, Gordon GB, Miller SR, Longcope C. Group. EHBCC. Body mass index, serum sex hormones, and breast cancer risk in postmenopausal women. Journal of the National Cancer Institute. 2003;95:1218–1226. [PubMed]
2. Monroe KR, Murphy SP, Henderson BE, Kolonel LN, Stanczyk FZ, Adlercreutz H, Pike MC. Dietary fiber intake and endogenous serum hormone levels in naturally postmenopausal Mexican American women: the Multiethnic Cohort Study. Nutrition and Cancer. 2007;58:127–135. [PubMed]
3. Carruba G, Granata OM, Pala V, Campisi I, Agostara B, Cusimano R, Ravazzolo B, Traina A. A traditional Mediterranean diet decreases endogenous estrogens in healthy postmenopausal women. Nutrition and Cancer. 2006;56:253–259. [PubMed]
4. Brinkman MT, Baglietto L, Krishnan K, English DR, Severi G, Morris Ha, Hopper JL, Giles GG. Consumption of animal products, their nutrient components and postmenopausal circulating steroid hormone concentrations. European Journal of Clinical Nutrition. 2010;64:176–183. [PubMed]
5. Fung TT, Hu FB, Barbieri RL, Willett WC, Hankinson SE. Dietary patterns, the Alternate Healthy Eating Index and plasma sex hormone concentrations in postmenopausal women. International Journal of Cancer. 2007;121:803–809. [PubMed]
6. Hoffmann K, Schulze MB, Schlenkiewitz A, Nothlings U, Boeing H. Application of a new statistical method to derive dietary patterns in nutritional epidemiology. American Journal of Epidemiology. 2004;159:935–944. [PubMed]
7. Heidemann C, Hoffmann K, Spranger J, Klipstein-Grobusch K, Mohlig M, Pfeiffer AF, Boeing H. cohort EPIiCaNE-PS. A dietary pattern protective against type 2 diabetes in the European Prospective Investigation into Cancer and Nutrition (EPIC)--Potsdam Study cohort. Diabetologia. 2005;48:1126–1134. [PubMed]
8. Schulze MB, Hoffmann K, Manson JE, Willett WC, Meigs JB, Weikert C, Heidemann C, Colditz GA, Hu FB. Dietary pattern, inflammation, and incidence of type 2 diabetes in women. American Journal of Clinical Nutrition. 2005;82:675–684. [PMC free article] [PubMed]
9. Hoffmann K, Zyriax BC, Boeing H, Windler E. A dietary pattern derived to explain biomarker variation is strongly associated with the risk of coronary artery disease. American Journal of Clinical Nutrition. 2004;80:633–640. [PubMed]
10. Liu L, Nettleton JA, Bertoni AG, Bluemke DA, Lima Ja, Szklo M. Dietary pattern, the metabolic syndrome, and left ventricular mass and systolic function: the Multi-Ethnic Study of Atherosclerosis. American Journal of Clinical Nutrition. 2009;90:362–368. [PubMed]
11. McCann SE, McCann WE, Hong CC, Marshall JR, Edge SB, Trevisan M, Muti P, Freudenheim JL. Dietary patterns related to glycemic index and load and risk of premenopausal and postmenopausal breast cancer in the Western New York Exposure and Breast Cancer Study. American Journal of Clinical Nutrition. 2007;86:465–471. [PMC free article] [PubMed]
12. Schulze M, Hoffmann K, Weikert C, Nothlings U, Schulze MB, Boeing H. Identification of a dietary pattern characterized by high-fat food choices associated with increased risk of breast cancer: the European Prospective Investigation into Cancer and Nutrition (EPIC)-Potsdam Study. British Journal of Nutrition. 2008;100:942–946. [PubMed]
13. Colditz GA, Martin P, Stampfer MJ, Willett WC, Sampson L, Rosner BA, Hennekens CH, Speizer FE. Validation of questionaire information on risk factors and disease outcomes in a prospetive cohort of women. American Journal of Epidemiology. 1986;123(5):894–900. [PubMed]
14. Hankinson SE, Willett WC, Manson JE, Hunter DJ, Colditz GA, Stampfer MJ, Longscope C, Speizer FE. Alcohol, height, and adiposity in relation to estrogen and prolactin levels in postmenopausal women. Journal of the National Cancer Institute. 1995;87:1297–1302. [PubMed]
15. Missmer SM, Eliassen AH, Barbieri RL, Hankinson SE. Endogenous estrogen, androgen, and progesterone concentrations and breast cancer risk among postmenopausal women. Journal of the National Cancer Institute. 2004;96:1856–1865. [PubMed]
16. Feskanich D, Rimm E, Giovannucci EL, Colditz GA, Stampfer MJ, Litin LB, Willett WC. Reproducibility and validity of food intake measurements from a semiquantitative food frequency questionnaire. Journal of the American Dietetic Association. 1993;93:790–796. [PubMed]
17. Willett WC. Nutritional Epidemiology. New York: Oxford Univeristy Press; 1998.
18. Ainsworth BE, Haskell WL, Whitt MC, Irwin ML, Swartz AM, Strath SJ, O'Brien WL, Bassett DRJ, Schmitz KH, Emplaincourt PO, Jacobs DRJ, Leon AS. Compendium of physical activities: an update of activity codes and MET intensities. Medicine and Science in Sports and Exercise. 2000;32(9 Suppl):S498–S504. [PubMed]
19. Rosner BA, Cook N, Portman r, Daniels S, Falkner B. Determination of blood pressure percentiles in normal-weight children: some methodological issues. American Journal of Epidemiology. 2008;167:653–666. [PubMed]
20. Hu FB, Stampfer MJ, Rimm E, Ascherio A, Rosner BA, Spiegelman D, Willett WC. Dietary fat and coronary heart disease: a comparison of approaches for adjusting for total energy intake and modeling repeated dietary measurements. American Journal of Epidemiology. 1999;149:531–540. [PubMed]
21. Fung TT, Hu FB, McCullough ML, Newby PK, Willett WC, Holmes MD. Diet quality is associated with the risk of estrogen receptor-negative breast cancer in postmenopausal women. Journal of Nutrition. 2006;136:466–472. [PubMed]