Abstract

Since evidence relating diet to breast cancer risk is not sufficiently consistent to elaborate preventive proposals, the authors examined the association between dietary patterns and breast cancer risk in a large French cohort study. The analyses included 2,381 postmenopausal invasive breast cancer cases diagnosed during a median 9.7-year follow-up period (1993–2005) among 65,374 women from the E3N-EPIC cohort. Scores for dietary patterns were obtained by factor analysis, and breast cancer hazard ratios were estimated by Cox proportional hazards regression for the highest quartile of dietary pattern score versus the lowest. Two dietary patterns were identified: “alcohol/Western” (essentially meat products, French fries, appetizers, rice/pasta, potatoes, pulses, pizza/pies, canned fish, eggs, alcoholic beverages, cakes, mayonnaise, and butter/cream) and “healthy/Mediterranean” (essentially vegetables, fruits, seafood, olive oil, and sunflower oil). The first pattern was positively associated with breast cancer risk (hazard ratio = 1.20, 95% confidence interval (CI): 1.03, 1.38; P = 0.007 for linear trend), especially when tumors were estrogen receptor-positive/progesterone receptor-positive. The “healthy/Mediterranean” pattern was negatively associated with breast cancer risk (hazard ratio = 0.85, 95% CI: 0.75, 0.95; P = 0.003 for linear trend), especially when tumors were estrogen receptor-positive/progesterone receptor-negative. Adherence to a diet comprising mostly fruits, vegetables, fish, and olive/sunflower oil, along with avoidance of Western-type foods, may contribute to a substantial reduction in postmenopausal breast cancer risk.

Breast cancer incidence varies widely between countries, suggesting the influence of environmental factors. The Japanese have traditionally been at low risk of breast cancer (1), but breast cancer incidence in Japan has recently increased (2) concomitantly with major changes in traditional habits, especially diet (3). Indeed, during the past 50 years in Japan, the proportion of energy obtained from fat increased until it represented up to 25% of total energy, and the consumption of dairy products increased 10-fold (4). Thus, the increasing incidence of breast cancer in Japan can be attributed at least partly to the adoption of a Western diet, which is notably characterized by higher intakes of meat, dairy products, and saturated fat, and decreased consumption of traditional Japanese foods such as seafood products (3). However, to date, evidence for associations between breast cancer risk and specific foods or nutrients has been limited, except for alcohol (5).

The recent approach to dietary patterns (6) classifies subjects according to dietary behavior and facilitates public health recommendations; in contrast, the nutrient/food approach permits better assessment of biologic mechanisms involved (7). Several epidemiologic studies have investigated the association between dietary pattern and breast cancer risk (8). A significant inverse relation has been described between breast cancer risk and a “prudent” (9, 10) or “healthy” (11) dietary pattern and specific regional patterns (12–14), in contrast to positive associations with a “drinker” (15) or “Western” pattern (11, 12, 16). However, the World Cancer Research Fund concluded that the evidence was not sufficient to draw firm conclusions (5). Thus, further large-scale prospective studies are needed to strengthen the observed associations.

We investigated the association between dietary pattern and risk of postmenopausal invasive breast cancer, considering potential interactions with known risk factors for breast cancer.

MATERIALS AND METHODS

The E3N cohort

The E3N [Etude Epidémiologique auprès de Femmes de la Mutuelle Générale de l'Education Nationale] Study is a prospective cohort study that was initiated in 1990 to investigate factors associated with the most common types of cancer (17). It involves 98,995 women living in France who were born between 1925 and 1950 and are covered by a national health insurance plan for teachers and coworkers. Participants complete biennial self-administered follow-up questionnaires on health status, medical history, and lifestyle. All subjects signed an informed consent form at study entry, and the study protocol was approved by the French National Commission for Computed Data and Individual Freedom.

Dietary assessment

Dietary data were collected via a self-administered diet history questionnaire assessing consumption of 208 foods and beverages. It included quantitative questions, using a booklet of photographs with portion sizes and frequency of food group consumption, and qualitative questions on food groups. Both questionnaire and booklet were validated and tested for reproducibility after 1 year (18, 19).

The dietary questionnaire was sent between June 1993 and July 1995 to the responders to the previous questionnaire. Women who completed a valid dietary questionnaire represent the French component of the European Prospective Investigation into Cancer and Nutrition (EPIC) (20).

Analytic cohort

In this study of postmenopausal breast cancer, follow-up began on the return date of the dietary questionnaire for women who were already postmenopausal at that time, or the date of menopause if this occurred later. Women contributed person-time until the date of cancer diagnosis, the date of the last completed questionnaire, or the date on which the last available follow-up questionnaire was mailed (July 2005), whichever occurred first.

Among the 74,524 women with available dietary data, we excluded 1,490 women with extreme values (1% on both sides) for the ratio between energy intake and energy requirement (21, 22). We also excluded women with cancer other than basal cell carcinoma or in situ breast lobular carcinoma before the start of follow-up (n = 5,361); those who did not answer the second questionnaire (n = 795); those who had never menstruated (n = 21); and those who had not reached menopause at the time of the last follow-up questionnaire (n = 1,483). Thus, 65,374 women contributed to the analyses, accruing 568,084 person-years of follow-up.

Determining cases

Cases were defined as cases of first primary invasive postmenopausal breast cancer (International Classification of Diseases for Oncology codes C50.0–C50.6 and C50.8–C50.9). We classified breast cancer by estrogen receptor (ER) and progesterone receptor (PR) status and by histologic type (ductal, including mixed ductal-lobular; lobular; or other, including tubular carcinomas). Women with in situ breast cancer were censored as noncases at the date of diagnosis. Each questionnaire asked about cancer occurrence, and we subsequently collected pathology reports. Pathology reports were obtained for 94% of declared breast cancer cases; 98% of them confirmed the diagnosis; and 88% of confirmed breast cancers were invasive. Among the 65,374 women studied, 2,381 developed postmenopausal invasive breast cancer during a median follow-up period of 9.7 years. Cases with missing data for hormone receptor status (n = 532) or histologic type (n = 173) were excluded from corresponding analyses.

Dietary patterns

Dietary patterns were produced from principal-components analysis based on 57 predefined food groups (Appendix Table), using the SAS “Proc Factor” procedure (SAS Institute Inc., Cary, North Carolina). This factor analysis forms linear combinations of the original food groups, thereby grouping together correlated variables. Coefficients defining these linear combinations are called factor loadings. A positive factor loading means that the food group is positively associated with the factor, whereas a negative loading reflects an inverse association with the factor. For interpreting the data, we considered foods with a loading coefficient under −0.25 or over 0.25. We rotated factors by orthogonal transformation using the SAS “Varimax” option to maximize the independence (orthogonality) of retained factors and obtain a simpler structure for easier interpretation (15, 23). In determining the number of factors to retain, we considered eigenvalues greater than 1.25 (as in the article by Slattery et al. (24)), the scree test (25, 26) (with values being retained at the break point between components with large eigenvalues and those with small eigenvalues on the scree plot), and the interpretability of the factors. For each subject, we calculated the factor score for each pattern by summing observed consumption from all food groups, weighted by the food group factor loadings. The factor score measures the conformity of a woman's diet to the given pattern. Labeling was descriptive, based on foods most strongly associated with the dietary patterns.

Statistical analysis

We used Cox proportional hazards regression with age as the underlying time metric for estimating hazard ratios and 95% confidence intervals. Control for potential confounders was ensured by adjustment for a number of factors (see tables for definitions): age, educational level, geographic area at baseline, body mass index (weight (kg)/height (m)2), height, family history of breast cancer in first- or second-degree relatives, age at menarche, age at first full-term pregnancy combined with number of livebirths, menopausal hormone therapy, personal history of benign breast disease or lobular carcinoma in situ at baseline, use of oral contraceptives at baseline, lifetime duration of breastfeeding, frequency of Papanicolaou (Pap) testing at baseline (as an indicator of compliance with gynecologic screening), physical activity at baseline, smoking status at baseline, energy intake (excluding alcohol), current use of phytoestrogen supplements, and current use of vitamin/mineral supplements. P for linear trend was estimated in models into which quartiles of factor scores were entered as an ordinal variable. Potential interactions either suggested in the literature or related to plausible underlying mechanisms were tested. We thus considered energy intake, use of menopausal hormone therapy, physical activity, body mass index, and smoking status in models including an interaction term (potential effect modifier × factor score with ordinal values corresponding to quartiles). Specific types of breast cancer were studied in separate Cox models, and tests of homogeneity for the association between each dietary pattern and risk of different types of breast cancer were based on Wald chi-square statistics (27): For each dietary pattern, the coefficients and their standard errors estimated from the Cox models were used to compute test statistics with degrees of freedom equal to the number of subtypes of breast cancer minus 1. All covariates had fewer than 5% of values missing; therefore, missing values were replaced by the modal value in subjects with complete data (all were qualitative variables). All statistical tests were 2-sided, and significance was set at the 0.05 level. Analyses were performed using SAS software, version 9.1.

RESULTS

Description of dietary patterns

Factor analysis identified 2 main dietary patterns, which accounted for 10% of the variance in consumption of the 57 food and beverage items (Table 1). Pattern 1 was positively correlated with consumption of processed meat and meat products (ham, offal), French fries, appetizers, sandwiches, rice/pasta, potatoes, pulses, pizza/pies, canned fish, eggs, crustaceans, alcoholic beverages, cakes, mayonnaise, and butter/cream and was thus termed “alcohol/Western.” Pattern 2 was characterized by a high intake of vegetables and fruits, fish and crustaceans, olives, and sunflower oil. It was labeled “healthy/Mediterranean” because of its Mediterranean traits (fish, fruits, vegetables, olive oil), as well as its healthy traits (use of sunflower oil was highly advocated in the 1990s).

Table 1.

Factor Loadingsa for Dietary Patterns Identified by Factor Analysis (n = 65,374) in the E3N-EPIC Cohort, France, 1993–2005

Food Group “Alcohol/Western” Pattern “Healthy/Mediterranean” Pattern 
Fruits  0.34 
Raw vegetables  0.70 
Cooked vegetables  0.66 
Potatoes 0.33  
Pulses 0.29  
Rice, pasta, semolina 0.39  
French fries 0.48  
Appetizers 0.45  
Pizza, pies 0.39  
Sandwiches 0.32  
Cakes 0.36  
Processed meatb 0.59  
Ham 0.31  
Offal 0.29  
Eggs 0.36  
Canned fish 0.37  
Crustaceans 0.32 0.30 
Fish  0.51 
Mayonnaise 0.39  
Butter, cream 0.31  
Olive oil  0.46 
Sunflower oil  0.26 
High-alcohol beverages 0.37  
Wine 0.26  
Food Group “Alcohol/Western” Pattern “Healthy/Mediterranean” Pattern 
Fruits  0.34 
Raw vegetables  0.70 
Cooked vegetables  0.66 
Potatoes 0.33  
Pulses 0.29  
Rice, pasta, semolina 0.39  
French fries 0.48  
Appetizers 0.45  
Pizza, pies 0.39  
Sandwiches 0.32  
Cakes 0.36  
Processed meatb 0.59  
Ham 0.31  
Offal 0.29  
Eggs 0.36  
Canned fish 0.37  
Crustaceans 0.32 0.30 
Fish  0.51 
Mayonnaise 0.39  
Butter, cream 0.31  
Olive oil  0.46 
Sunflower oil  0.26 
High-alcohol beverages 0.37  
Wine 0.26  

Abbreviations: E3N, Etude Epidémiologique auprès de Femmes de la Mutuelle Générale de l'Education Nationale; EPIC, European Prospective Investigation into Cancer and Nutrition.

a

For both patterns, food groups with factor loadings of less than ±0.25 (soup, tea, coffee, chicory, chocolate beverages, juices, soft drinks—regular, soft drinks—light, mineral water, tap water, low-alcohol beverages, white bread, whole-grain bread, crisp bread, breakfast cereals, biscuits, sweets, croissant-like/Danish pastries, dairy-based sweet puddings, high-fat dairy products, low-fat dairy products, canned fruits, meat, poultry and rabbit, cheese, margarine, vegetable oil (except olive oil and sunflower oil), other fats, manufactured salad dressing, salad dressing—light, chocolate, added sugar and artificial sweeteners) were omitted for simplicity.

b

Except cooked ham.

Increasing scores of the “alcohol/Western” pattern were associated with younger age, decreasing prevalence of nulliparity, decreasing duration of breastfeeding, increasing prevalence of overweight, greater height, a higher proportion of relatives with a history of breast cancer, a higher proportion of oral contraceptive use, biennial Pap smears, and a higher prevalence of current smoking. Increasing scores of the “healthy/Mediterranean” pattern were associated with older age, higher education, a higher prevalence of overweight, a higher proportion of personal history of benign breast disease, increasing use of menopausal hormone therapy, increasing duration of breastfeeding, an increasing proportion of annual Pap smears, higher levels of physical activity, and an increasing proportion of former smokers. Energy intake and alcohol use also increased with increasing score, but to a lesser extent than for the alcohol/Western pattern (Table 2).

Table 2.

Characteristics of Postmenopausal Women at Baseline, by Quartile of Dietary Pattern Score (n = 65,374), in the E3N-EPIC Cohort, France, 1993–2005a

 Quartile of “Alcohol/Western” Pattern Score
 
Quartile of “Healthy/Mediterranean” Pattern Score
 
 
Age, years 55.1 (6.8) 53.5 (6.6) 52.2 (6.3) 51.2 (6.0) 52.4 (6.8) 53.0 (6.7) 53.3 (6.6) 53.3 (6.4) 
<12 years of education, % 11.8 10.9 10.5 11.8 12.2 11.3 10.7 10.8 
Age at menarche, years 12.8 (1.4) 12.8 (1.4) 12.8 (1.4) 12.8 (1.4) 12.9 (1.4) 12.8 (1.4) 12.8 (1.4) 12.6 (1.4) 
Age at first full-term pregnancy and no. of livebirths, %         
    Nulliparous 14.2 12.0 10.7 9.5 12.3 11.1 11.4 11.7 
    <30 years and 1–2 births 48.4 50.2 50.2 51.4 48.3 50.0 50.5 51.4 
    <30 years and ≥3 births 27.4 27.5 28.1 28.8 27.5 28.3 28.2 27.8 
    ≥30 years and ≥1 birth 10.0 10.3 11.0 10.3 11.9 10.7 10.0 9.1 
Body mass indexb, %         
    <18.5 5.0 3.3 3.1 2.5 5.5 3.5 2.7 2.3 
    ≥18.5 and <25 76.5 76.9 76.3 71.0 78.1 77.9 75.6 69.2 
    ≥25 18.5 19.8 20.6 26.5 16.4 18.6 21.7 28.5 
Height, cm 160.6 (5.6) 161.1 (5.6) 161.7 (5.6) 162.2 (5.6) 161.2 (5.7) 161.3 (5.6) 161.4 (5.6) 161.7 (5.6) 
Family history of breast cancer in first- or second-degree relative, % 22.6 23.3 23.6 24.4 23.3 23.1 23.8 23.8 
Menopausal hormone therapy, % 49.7 50.9 51.7 49.8 46.6 51.3 51.5 52 
Personal history of benign breast diseasec or lobular carcinoma in situ, % 28.2 29.6 29.9 29.1 28.4 28.7 29.6 30 
Ever use of oral contraceptives, % 51.2 57.7 63.2 68.1 59.9 59.5 59.8 60.6 
Lifetime duration of breastfeeding, monthsd 3.7 (5.8) 3.5 (5.2) 3.3 (5.1) 3.1 (4.9) 3.2 (5.0) 3.4 (5.1) 3.5 (5.3) 3.6 (5.6) 
Frequency of Papanicolaou testing, %         
    Never 3.4 2.7 2.5 2.7 3.8 2.8 2.6 2.2 
    Irregularly 12.1 11.6 10.6 11.5 13.2 11.8 10.9 10.1 
    Every 4–5 years 3.7 3.0 3.1 3.4 3.4 3.2 3.2 3.4 
    Every 2–3 years 24.7 26.2 26.9 27.6 26.8 26.8 26.3 25.4 
    Every year 56.1 56.4 56.9 54.8 52.7 55.4 57.1 59.0 
Physical activity (metabolic equivalents/week), %         
    <29.0 32.4 33.1 33.2 32.5 36.0 33.4 31.2 30.7 
    29.0–46.7 34.2 35.3 35.5 34.9 35.4 35.6 34.8 34.2 
    >46.7 33.3 31.6 31.3 32.6 28.7 31.0 34.0 35.1 
Alcohol consumption, g/day 5.6 (8.3) 9.1 (11.1) 12.0 (13.3) 18.2 (18.8) 10.5 (13.8) 10.8 (13.6) 11.3 (13.8) 12.1 (15.3) 
Smoking status, %         
    Never smoker 61.4 57.8 54.7 51.7 58.9 58.1 55.9 52.8 
    Former smoker 28.3 29.7 30.7 31.0 26.9 29.2 30.7 32.9 
    Current smoker 10.3 12.5 14.6 17.3 14.2 12.7 13.3 14.4 
Energy intake (excluding alcohol), kcal/day 1,662.2 (410.5) 1,933.3 (407.6) 2,175.2 (424.9) 2,594.4 (513.5) 1,974.9 (555.1) 2,024.6 (527.1) 2,097.4 (534.2) 2,255.3 (574.4) 
 Quartile of “Alcohol/Western” Pattern Score
 
Quartile of “Healthy/Mediterranean” Pattern Score
 
 
Age, years 55.1 (6.8) 53.5 (6.6) 52.2 (6.3) 51.2 (6.0) 52.4 (6.8) 53.0 (6.7) 53.3 (6.6) 53.3 (6.4) 
<12 years of education, % 11.8 10.9 10.5 11.8 12.2 11.3 10.7 10.8 
Age at menarche, years 12.8 (1.4) 12.8 (1.4) 12.8 (1.4) 12.8 (1.4) 12.9 (1.4) 12.8 (1.4) 12.8 (1.4) 12.6 (1.4) 
Age at first full-term pregnancy and no. of livebirths, %         
    Nulliparous 14.2 12.0 10.7 9.5 12.3 11.1 11.4 11.7 
    <30 years and 1–2 births 48.4 50.2 50.2 51.4 48.3 50.0 50.5 51.4 
    <30 years and ≥3 births 27.4 27.5 28.1 28.8 27.5 28.3 28.2 27.8 
    ≥30 years and ≥1 birth 10.0 10.3 11.0 10.3 11.9 10.7 10.0 9.1 
Body mass indexb, %         
    <18.5 5.0 3.3 3.1 2.5 5.5 3.5 2.7 2.3 
    ≥18.5 and <25 76.5 76.9 76.3 71.0 78.1 77.9 75.6 69.2 
    ≥25 18.5 19.8 20.6 26.5 16.4 18.6 21.7 28.5 
Height, cm 160.6 (5.6) 161.1 (5.6) 161.7 (5.6) 162.2 (5.6) 161.2 (5.7) 161.3 (5.6) 161.4 (5.6) 161.7 (5.6) 
Family history of breast cancer in first- or second-degree relative, % 22.6 23.3 23.6 24.4 23.3 23.1 23.8 23.8 
Menopausal hormone therapy, % 49.7 50.9 51.7 49.8 46.6 51.3 51.5 52 
Personal history of benign breast diseasec or lobular carcinoma in situ, % 28.2 29.6 29.9 29.1 28.4 28.7 29.6 30 
Ever use of oral contraceptives, % 51.2 57.7 63.2 68.1 59.9 59.5 59.8 60.6 
Lifetime duration of breastfeeding, monthsd 3.7 (5.8) 3.5 (5.2) 3.3 (5.1) 3.1 (4.9) 3.2 (5.0) 3.4 (5.1) 3.5 (5.3) 3.6 (5.6) 
Frequency of Papanicolaou testing, %         
    Never 3.4 2.7 2.5 2.7 3.8 2.8 2.6 2.2 
    Irregularly 12.1 11.6 10.6 11.5 13.2 11.8 10.9 10.1 
    Every 4–5 years 3.7 3.0 3.1 3.4 3.4 3.2 3.2 3.4 
    Every 2–3 years 24.7 26.2 26.9 27.6 26.8 26.8 26.3 25.4 
    Every year 56.1 56.4 56.9 54.8 52.7 55.4 57.1 59.0 
Physical activity (metabolic equivalents/week), %         
    <29.0 32.4 33.1 33.2 32.5 36.0 33.4 31.2 30.7 
    29.0–46.7 34.2 35.3 35.5 34.9 35.4 35.6 34.8 34.2 
    >46.7 33.3 31.6 31.3 32.6 28.7 31.0 34.0 35.1 
Alcohol consumption, g/day 5.6 (8.3) 9.1 (11.1) 12.0 (13.3) 18.2 (18.8) 10.5 (13.8) 10.8 (13.6) 11.3 (13.8) 12.1 (15.3) 
Smoking status, %         
    Never smoker 61.4 57.8 54.7 51.7 58.9 58.1 55.9 52.8 
    Former smoker 28.3 29.7 30.7 31.0 26.9 29.2 30.7 32.9 
    Current smoker 10.3 12.5 14.6 17.3 14.2 12.7 13.3 14.4 
Energy intake (excluding alcohol), kcal/day 1,662.2 (410.5) 1,933.3 (407.6) 2,175.2 (424.9) 2,594.4 (513.5) 1,974.9 (555.1) 2,024.6 (527.1) 2,097.4 (534.2) 2,255.3 (574.4) 

Abbreviations: E3N, Etude Epidémiologique auprès de Femmes de la Mutuelle Générale de l'Education Nationale; EPIC, European Prospective Investigation into Cancer and Nutrition.

a

Data are presented as mean values with standard deviations in parentheses, unless otherwise indicated.

b

Weight (kg)/height (m)2.

c

Including fibrocystic breast disease, mastosis, and fibroadenoma.

d

Among parous women (n = 57,776).

Alcohol/Western pattern and breast cancer risk

The alcohol/Western dietary pattern was positively associated with breast cancer risk (Table 3). The multivariable hazard ratio for the highest quartile versus the lowest was 1.20 (95% confidence interval (CI): 1.03, 1.38; P = 0.007 for linear trend). After adjustment for alcohol intake, the association was not substantially modified, but it was weakened (for quartile 4 vs. quartile 1, multivariable hazard ratio (HR) = 1.14, 95% CI: 0.98, 1.33; P = 0.054 for linear trend).

Table 3.

Hazard Ratios for Invasive Postmenopausal Breast Cancer, by Quartile of Score for the Alcohol/Western Dietary Pattern (n = 65,374), Among Women in the E3N-EPIC Cohort, France, 1993–2005

 Quartile of “Alcohol/Western” Pattern Score
 
P for Linear Trend 
 1 (HR = 1)a 2
 
3
 
4
 
 HR 95% CI HR 95% CI HR 95% CI 
Total         
    No. of cases 574 583 620 604  
    Age-adjusted model 1.00 1.08 0.96, 1.21 1.22 1.08, 1.36 1.26 1.12, 1.41 <0.0001 
    Multivariable modelb 1.00 1.05 0.93, 1.19 1.17 1.03, 1.33 1.20 1.03, 1.38 0.007 
ER+/PR+         
    No. of cases 259 257 276 292  
    Age-adjusted model 1.00 1.07 0.90, 1.27 1.22 1.03, 1.45 1.38 1.17, 1.64 <0.001 
    Multivariable modelb 1.00 1.03 0.86, 1.24 1.18 0.97, 1.43 1.33 1.07, 1.65 0.005 
ER−/PR−         
    No. of cases 79 72 80 68  
    Age-adjusted model 1.00 0.94 0.68, 1.30 1.09 0.79, 1.49 0.96 0.69, 1.33 0.96 
    Multivariable modelb 1.00 0.95 0.68, 1.32 1.05 0.74, 1.49 0.84 0.56, 1.27 0.56 
ER−/PR+         
    No. of cases 18 16 13 15  
    Age-adjusted model 1.00 0.90 0.46, 1.77 0.75 0.37, 1.54 0.90 0.45, 1.80 0.65 
    Multivariable modelb 1.00 0.83 0.41, 1.68 0.64 0.29, 1.42 0.75 0.32, 1.79 0.42 
ER+/PR−         
    No. of cases 92 106 101 105  
    Age-adjusted model 1.00 1.22 0.92, 1.61 1.22 0.92, 1.62 1.33 1.00, 1.77 0.06 
    Multivariable modelb 1.00 1.23 0.92, 1.65 1.24 0.91, 1.71 1.38 0.97, 1.97 0.09 
 Quartile of “Alcohol/Western” Pattern Score
 
P for Linear Trend 
 1 (HR = 1)a 2
 
3
 
4
 
 HR 95% CI HR 95% CI HR 95% CI 
Total         
    No. of cases 574 583 620 604  
    Age-adjusted model 1.00 1.08 0.96, 1.21 1.22 1.08, 1.36 1.26 1.12, 1.41 <0.0001 
    Multivariable modelb 1.00 1.05 0.93, 1.19 1.17 1.03, 1.33 1.20 1.03, 1.38 0.007 
ER+/PR+         
    No. of cases 259 257 276 292  
    Age-adjusted model 1.00 1.07 0.90, 1.27 1.22 1.03, 1.45 1.38 1.17, 1.64 <0.001 
    Multivariable modelb 1.00 1.03 0.86, 1.24 1.18 0.97, 1.43 1.33 1.07, 1.65 0.005 
ER−/PR−         
    No. of cases 79 72 80 68  
    Age-adjusted model 1.00 0.94 0.68, 1.30 1.09 0.79, 1.49 0.96 0.69, 1.33 0.96 
    Multivariable modelb 1.00 0.95 0.68, 1.32 1.05 0.74, 1.49 0.84 0.56, 1.27 0.56 
ER−/PR+         
    No. of cases 18 16 13 15  
    Age-adjusted model 1.00 0.90 0.46, 1.77 0.75 0.37, 1.54 0.90 0.45, 1.80 0.65 
    Multivariable modelb 1.00 0.83 0.41, 1.68 0.64 0.29, 1.42 0.75 0.32, 1.79 0.42 
ER+/PR−         
    No. of cases 92 106 101 105  
    Age-adjusted model 1.00 1.22 0.92, 1.61 1.22 0.92, 1.62 1.33 1.00, 1.77 0.06 
    Multivariable modelb 1.00 1.23 0.92, 1.65 1.24 0.91, 1.71 1.38 0.97, 1.97 0.09 

Abbreviations: CI, confidence interval; E3N, Etude Epidémiologique auprès de Femmes de la Mutuelle Générale de l'Education Nationale; EPIC, European Prospective Investigation into Cancer and Nutrition; ER, estrogen receptor; HR, hazard ratio; PR, progesterone receptor.

a

Reference category.

b

Adjusted HR from Cox proportional hazards regression. HRs were adjusted for age (years; time scale), educational level (<12 or ≥12 years of schooling), region at baseline (7 categories), body mass index (weight (kg)/height (m)2; <18.5, ≥18.5 and <25, or ≥25, as a time-dependent variable), height (cm; continuous), family history of breast cancer in a first- or second-degree relative (yes or no), age at menarche (years; continuous), age at first full-term pregnancy combined with number of livebirths (nulliparous, <30 years and 1–2 births, <30 years and ≥3 births, or ≥30 years and ≥1 birth), menopausal hormone therapy initiated before the previous year (yes or no; time-dependent variable), personal history of benign breast disease (fibrocystic breast disease, mastosis, or adenoma) or lobular carcinoma in situ (yes or no) at baseline, use of oral contraceptives at baseline (yes or no), lifetime duration of breastfeeding (0, <7, ≥7 and <12, or ≥12 months), frequency of Papanicolaou testing at baseline as an indicator of adherence to gynecologic screening (never, irregularly, every 4–5 years, every 2–3 years, or every year), physical activity (metabolic equivalents/week, in tertiles), smoking status at baseline (never, former, or current smoker), energy intake excluding alcohol (kcal/day, in quartiles), current use of phytoestrogen supplements (yes or no; time-dependent variable), and current use of vitamin/mineral supplements (yes or no; time-dependent variable).

The association between this pattern and breast cancer risk was statistically significant only for ER-positive (ER+)/PR-positive (PR+) tumors (for quartile 4 vs. quartile 1, multivariable HR = 1.33, 95% CI: 1.07, 1.65; P = 0.005 for linear trend). The test for homogeneity between ER+/PR+ tumors and ER+/PR-negative (PR−) tumors was not statistically significant, while the P value for all ER+ tumors versus ER-negative (ER−) tumors was 0.06.

Associations did not vary significantly across ductal (n = 1,619) and lobular (n = 396) tumors (P for homogeneity = 0.50). The multivariable hazard ratios (quartile 4 vs. quartile 1) were 1.17 (95% CI: 0.98, 1.40; P = 0.06 for linear trend) and 1.36 (95% CI: 0.96, 1.95; P = 0.09 for linear trend) for ductal and lobular tumors, respectively.

A statistically significant interaction was observed between pattern score and body mass index (P = 0.02). In women with a body mass index less than 25, a positive association was observed (multivariable HR = 1.34, 95% CI: 1.13, 1.60; P = 0.001 for linear trend). In heavier women, the multivariable hazard ratio was 0.97 (95% CI: 0.76, 1.25; P = 0.99 for linear trend).

Healthy/Mediterranean pattern and breast cancer risk

The healthy/Mediterranean dietary pattern was inversely associated with breast cancer risk (for quartile 4 vs. quartile 1, HR = 0.85, 95% CI: 0.75, 0.95; P = 0.003 for linear trend) (Table 4). The association was not modified by further adjustment for alcohol intake (for quartile 4 vs. quartile 1, multivariable HR  = 0.85, 95% CI: 0.75, 0.95; P = 0.003 for linear trend).

Table 4.

Hazard Ratios for Invasive Postmenopausal Breast Cancer, by Quartile of Score for the Healthy/Mediterranean Dietary Pattern (n = 65,374), Among Women in the E3N-EPIC Cohort, France, 1993–2005

 Quartile of “Healthy/Mediterranean” Pattern Score
 
P for Linear Trend 
 1 (HR = 1)a 2
 
3
 
4
 
 HR 95% CI HR 95% CI HR 95% CI 
Total         
    No. of cases 593 606 594 588  
    Age-adjusted model 1.00 0.97 0.87, 1.09 0.94 0.84, 1.05 0.92 0.82, 1.03 0.12 
    Multivariable modelb 1.00 0.95 0.85, 1.07 0.90 0.80, 1.01 0.85 0.75, 0.95 0.003 
ER+/PR+         
    No. of cases 262 273 273 276  
    Age-adjusted model 1.00 0.99 0.83, 1.17 0.97 0.82, 1.15 0.97 0.82, 1.15 0.71 
    Multivariable modelb 1.00 0.96 0.81, 1.14 0.92 0.78, 1.09 0.88 0.74, 1.05 0.13 
ER−/PR−         
    No. of cases 76 74 77 72  
    Age-adjusted model 1.00 0.93 0.68, 1.28 0.96 0.70, 1.31 0.88 0.64, 1.22 0.50 
    Multivariable modelb 1.00 0.90 0.65, 1.24 0.90 0.65, 1.24 0.78 0.56, 1.10 0.17 
ER−/PR+         
    No. of cases 14 15 14 19  
    Age-adjusted model 1.00 1.02 0.49, 2.11 0.94 0.45, 1.96 1.23 0.62, 2.46 0.60 
    Multivariable modelb 1.00 1.02 0.49, 2.12 0.92 0.43, 1.94 1.18 0.58, 2.42 0.71 
ER+/PR−         
    No. of cases 117 117 87 83  
    Age-adjusted model 1.00 0.96 0.74, 1.24 0.70 0.53, 0.93 0.67 0.50, 0.88 0.001 
    Multivariable modelb 1.00 0.95 0.74, 1.23 0.70 0.53, 0.92 0.65 0.49, 0.87 0.001 
 Quartile of “Healthy/Mediterranean” Pattern Score
 
P for Linear Trend 
 1 (HR = 1)a 2
 
3
 
4
 
 HR 95% CI HR 95% CI HR 95% CI 
Total         
    No. of cases 593 606 594 588  
    Age-adjusted model 1.00 0.97 0.87, 1.09 0.94 0.84, 1.05 0.92 0.82, 1.03 0.12 
    Multivariable modelb 1.00 0.95 0.85, 1.07 0.90 0.80, 1.01 0.85 0.75, 0.95 0.003 
ER+/PR+         
    No. of cases 262 273 273 276  
    Age-adjusted model 1.00 0.99 0.83, 1.17 0.97 0.82, 1.15 0.97 0.82, 1.15 0.71 
    Multivariable modelb 1.00 0.96 0.81, 1.14 0.92 0.78, 1.09 0.88 0.74, 1.05 0.13 
ER−/PR−         
    No. of cases 76 74 77 72  
    Age-adjusted model 1.00 0.93 0.68, 1.28 0.96 0.70, 1.31 0.88 0.64, 1.22 0.50 
    Multivariable modelb 1.00 0.90 0.65, 1.24 0.90 0.65, 1.24 0.78 0.56, 1.10 0.17 
ER−/PR+         
    No. of cases 14 15 14 19  
    Age-adjusted model 1.00 1.02 0.49, 2.11 0.94 0.45, 1.96 1.23 0.62, 2.46 0.60 
    Multivariable modelb 1.00 1.02 0.49, 2.12 0.92 0.43, 1.94 1.18 0.58, 2.42 0.71 
ER+/PR−         
    No. of cases 117 117 87 83  
    Age-adjusted model 1.00 0.96 0.74, 1.24 0.70 0.53, 0.93 0.67 0.50, 0.88 0.001 
    Multivariable modelb 1.00 0.95 0.74, 1.23 0.70 0.53, 0.92 0.65 0.49, 0.87 0.001 

Abbreviations: CI, confidence interval; E3N, Etude Epidémiologique auprès de Femmes de la Mutuelle Générale de l'Education Nationale; EPIC, European Prospective Investigation into Cancer and Nutrition; ER, estrogen receptor; HR, hazard ratio; PR, progesterone receptor.

a

Reference category.

b

HRs were adjusted as described in Table 3.

The association was statistically significant only for ER+/PR− tumors (for quartile 4 vs. quartile 1, multivariable HR = 0.65, 95% CI: 0.49, 0.87; P = 0.001 for linear trend), although nonsignificant results for other case groups pointed in the same direction (except for ER−/PR+, for which the number of cases was very small). The P value for homogeneity between ER+/PR− and ER+/PR+ tumors was 0.03, and the P value for homogeneity between ER+/PR− tumors and tumors of all other receptor statuses was 0.14.

Associations between this pattern and breast cancer risk did not vary significantly across ductal and lobular tumors (P for homogeneity = 0.74). Multivariable hazard ratios (quartile 4 vs. quartile 1) were 0.83 (95% CI: 0.72, 0.96; P = 0.007 for linear trend) and 0.86 (95% CI: 0.65, 1.15; P = 0.30 for linear trend) for ductal and lobular tumors, respectively.

There was a significant interaction between healthy/Mediterranean pattern scores and energy intake (P = 0.03) (Table 5). In women with energy intake below the median (2,037 kcal/day), the pattern was inversely associated with breast cancer risk (for quartile 4 vs. quartile 1, multivariable HR = 0.75, 95% CI: 0.63, 0.90; P = 0.002 for linear trend), while no association was observed among women with higher energy intake (for quartile 4 vs. quartile 1, multivariable HR = 0.93, 95% CI: 0.79, 1.10; P = 0.29 for linear trend).

Table 5.

Hazard Ratios for Invasive Postmenopausal Breast Cancer by Quartile of Dietary Pattern Score, According to Body Mass Index or Daily Energy Intake (n = 65,374), Among Women in the E3N-EPIC Cohort, France, 1993–2005

 Quartile of Dietary Pattern Score
 
P for Linear Trend 
 1 (HR = 1)a 2
 
3
 
4
 
 HR 95% CI HR 95% CI HR 95% CI 
Alcohol/Western dietary pattern         
       Body mass indexb <25         
        No. of cases 387 410 431 386  
        Age-adjusted model 1.00 1.17 1.02, 1.34 1.33 1.15, 1.52 1.42 1.23, 1.64 <0.001 
        Multivariable modelc 1.00 1.14 0.99, 1.32 1.27 1.09, 1.49 1.34 1.13, 1.60 0.001 
    Body mass index ≥25         
        No. of cases 187 173 189 218  
        Age-adjusted model 1.00 0.91 0.74, 1.11 1.01 0.82, 1.24 1.01 0.82, 1.23 0.70 
        Multivariable modelc 1.00 0.88 0.71, 1.10 0.97 0.77, 1.21 0.97 0.76, 1.25 0.99 
Healthy/Mediterranean dietary pattern         
    Energy intake ≤ mediand         
        No. of cases 344 314 275 204  
        Age-adjusted model 1.00 0.92 0.79, 1.08 0.89 0.76, 1.05 0.83 0.70, 0.99 0.03 
        Multivariable modele 1.00 0.89 0.77, 1.04 0.84 0.72, 0.99 0.75 0.63, 0.90 0.002 
    Energy intake > mediand         
        No. of cases 249 292 319 384  
        Age-adjusted model 1.00 1.03 0.87, 1.22 0.97 0.82, 1.14 0.95 0.81, 1.11 0.37 
        Multivariable modele 1.00 1.03 0.87, 1.22 0.96 0.81, 1.14 0.93 0.79, 1.10 0.29 
 Quartile of Dietary Pattern Score
 
P for Linear Trend 
 1 (HR = 1)a 2
 
3
 
4
 
 HR 95% CI HR 95% CI HR 95% CI 
Alcohol/Western dietary pattern         
       Body mass indexb <25         
        No. of cases 387 410 431 386  
        Age-adjusted model 1.00 1.17 1.02, 1.34 1.33 1.15, 1.52 1.42 1.23, 1.64 <0.001 
        Multivariable modelc 1.00 1.14 0.99, 1.32 1.27 1.09, 1.49 1.34 1.13, 1.60 0.001 
    Body mass index ≥25         
        No. of cases 187 173 189 218  
        Age-adjusted model 1.00 0.91 0.74, 1.11 1.01 0.82, 1.24 1.01 0.82, 1.23 0.70 
        Multivariable modelc 1.00 0.88 0.71, 1.10 0.97 0.77, 1.21 0.97 0.76, 1.25 0.99 
Healthy/Mediterranean dietary pattern         
    Energy intake ≤ mediand         
        No. of cases 344 314 275 204  
        Age-adjusted model 1.00 0.92 0.79, 1.08 0.89 0.76, 1.05 0.83 0.70, 0.99 0.03 
        Multivariable modele 1.00 0.89 0.77, 1.04 0.84 0.72, 0.99 0.75 0.63, 0.90 0.002 
    Energy intake > mediand         
        No. of cases 249 292 319 384  
        Age-adjusted model 1.00 1.03 0.87, 1.22 0.97 0.82, 1.14 0.95 0.81, 1.11 0.37 
        Multivariable modele 1.00 1.03 0.87, 1.22 0.96 0.81, 1.14 0.93 0.79, 1.10 0.29 

Abbreviations: CI, confidence interval; E3N, Etude Epidémiologique auprès de Femmes de la Mutuelle Générale de l'Education Nationale; EPIC, European Prospective Investigation into Cancer and Nutrition; HR, hazard ratio.

a

Reference category.

b

Weight (kg)/height (m)2.

c

HRs were adjusted as described in Table 3, with the exception of body mass index.

d

Median energy intake was 2,037 kcal/day.

e

HRs were adjusted as described in Table 3, with the exception of energy intake.

Among women classified in the fourth quartile of the healthy/Mediterranean pattern, we compared mean intakes of foods between women with energy intakes above and below the median. Differences between these 2 subgroups were mostly due to “unhealthy” food items (i.e., foods associated with the alcohol/Western pattern, such as sandwiches, French fries, cakes, or processed meats), while intakes of “healthy” foods were quite similar (fruit, raw vegetables, cooked vegetables, olive oil, sunflower oil, or fish).

DISCUSSION

In the present study, 2 independent dietary patterns were found to be associated with postmenopausal breast cancer risk. The first, characterized by Western-type foods and alcohol, was associated with a high risk of ER+ tumors; the second, a Mediterranean pattern high in fruits, vegetables, fish, and olive and sunflower oils, was associated with reduced risk of breast cancer, especially for ER+/PR− tumors. However, associations were restricted to slim or normal-weight women for the alcohol/Western pattern and to women with low energy intake for the healthy/Mediterranean pattern.

The recent global approach to dietary patterns is particularly advantageous when planning dietary preventive strategies, as it considers diet in all its complex relations and potential interactions. Subjective definition of food groups included in factor analysis, along with labeling of identified patterns, has been described as a major limitation of this approach to standardizing patterns (6, 28, 29). However, major dietary patterns derived from factor analysis have been found to be stable over time and reproducible across populations (23, 30, 31). A Western pattern has been consistently described in several epidemiologic studies (5, 10, 23, 32, 33). Patterns associated with healthy behavior are less homogeneous, although most of them are described as diets high in fruits, vegetables, whole grains, fish, poultry, and salad dressing (10, 15, 32–34) or a Mediterranean diet rich in fruits, vegetables, and olive oil (35).

An important finding of the present study was the interaction with energy intake. There was no association between breast cancer risk and a high score in the healthy/Mediterranean pattern in cases of high energy intake due to concomitant intake of energy-dense Western foods. This may explain previously published negative findings regarding healthy or prudent diets (13, 14); a beneficial effect of such diets could be offset by simultaneous intake of “unhealthy” foods, that is, foods characteristic of a Western diet. This hypothesis was supported by our results regarding the higher increased intakes of “unhealthy” foods than of “healthy” foods among women with an above-median energy intake and classified in the fourth quartile of the healthy/Mediterranean pattern. Our findings indicate that a healthy/Mediterranean pattern is associated with a reduced risk of breast cancer only if energy intake remains within recommendations and if “unhealthy” foods are not consumed in large quantities.

It is difficult to determine which components of the healthy/Mediterranean pattern explain the inverse association with breast cancer risk. Although some components of fruits and vegetables, such as folates (36) and lignans (17), have been inversely associated with postmenopausal breast cancer risk in our cohort, a protective effect of fruits and vegetables has not been solidly established (5, 37, 38). Other candidates, although not established either, include n-3 fatty acids from fish and the overall balance between fatty acids (5).

The association between alcohol intake and breast cancer risk is convincing (5). However, the above-described alcohol/Western pattern was only moderately associated with alcohol intake (correlation coefficient = 0.36). Other components of the pattern, such as processed foods rich in trans-fatty acids, may participate in the association (39). The observed interaction with body mass index suggests that being overweight has an impact on postmenopausal breast cancer risk (5) that outweighs any dietary effect. Therefore, avoiding Western-type foods might reduce breast cancer risk only in normal-weight women.

In agreement with our findings, in a prospective study investigating the association between food patterns and breast cancer risk in Uruguay, Ronco et al. (11) found no heterogeneity according to histologic type. In contrast, it can be hypothesized that a hormonal pathway effect could be involved in the etiology of the association between dietary patterns and breast cancer, as investigators in several studies (40–46) have described heterogeneity of this association according to hormone receptor status. Indeed, results regarding prudent (40) or fruit- and vegetable-rich (41, 44) diets are inconsistent. Regarding components of the Western diet, a stronger association with alcohol intake was observed with ER+ tumors than with ER− tumors (46); a high fat intake has been associated with increased risk of ER+/PR+ tumors (42), although not consistently (43). In 1 intervention study, decreased fat intake was associated with risk reduction, mainly for ER+/PR− tumors (45).

Nevertheless, our study had some limitations. First, despite the use of a validated detailed questionnaire, some degree of misclassification of dietary intake is to be expected, as in similar dietary studies. In addition, we estimated the usual diet through a single dietary assessment; thus, we cannot rule out the possibility that some changes occurred in the diet during follow-up. Second, associations derived from an observational study may partly result from residual confounding, although we carefully adjusted all results for known breast cancer risk factors. Third, findings from cohorts of volunteers demand cautious extrapolation to the general population. Indeed, our population involved mostly teachers or their families—persons with a high level of education and health-consciousness, especially regarding dietary practices, but also a higher rate of breast cancer than the general French population (47). Finally, the reduced amount of total variance explained by our 2 dietary patterns may be a subject of concern, although it was comparable with that in other dietary studies (9, 48). The proportion of variance explained by the factors strongly depends on the number of food groups included in the principal-components analysis (48). The fewer food items (i.e., the broader the food groups), the higher the proportion of variance explained. We chose to consider a relatively high number of food groups in order to better comprehend the diversity and complexity of French food.

Strengths of our study include the prospective design, the study size, the availability of data on histologic type and hormone receptor type, careful adjustment for breast cancer risk factors, and validated dietary data. All cases of prevalent tumors at baseline were excluded so as to produce dietary patterns from cancer-free subjects. The median 9.7-year follow-up period provided a large latency period for potential disease occurrence.

In conclusion, our findings suggest that postmenopausal breast cancer risk in women may be influenced by dietary habits. Public health advice should emphasize the importance of increasing intake of foods associated with a healthy/Mediterranean pattern while maintaining energy intake within recommendations, in view of reducing the breast cancer burden. Avoidance of Western-type foods may reduce breast cancer risk in normal-weight women.

Abbreviations

    Abbreviations
  • CI

    confidence interval

  • ER

    estrogen receptor

  • HR

    hazard ratio

  • PR

    progesterone receptor

Author affiliations: Institut National de la Santé et de la Recherche Médicale (INSERM), Equipe Région INSERM 20, Institut de Cancérologie Gustave Roussy, Villejuif, France (Vanessa Cottet, Mathilde Touvier, Agnès Fournier, Marina S. Touillaud, Françoise Clavel-Chapelon, Marie-Christine Boutron-Ruault); and Direction of Risk Assessment for Nutrition and Food Safety (DERNS)/Office of Scientific Support for Risk Assessment (PASER), French Food Safety Agency (AFSSA), Maisons-Alfort, France (Mathilde Touvier, Lionel Lafay, Marie-Christine Boutron-Ruault).

This work was supported by the French League Against Cancer, the European Community, the 3M Company, Mutuelle Générale de l'Education Nationale, Institut de Cancérologie Gustave Roussy, INSERM, AFSSA, and several general councils in France.

The authors are indebted to the E3N Study physicians for their active collaboration. They are grateful to all members of the E3N study group; to Maryvonne Niravong, Rafika Chaït, Lyan Hoang, and Marie Fangon for technical assistance; to Dr. Emmanuelle Kesse for advice on dietary pattern assessment; and to Dr. Dimitrios Trichopoulos and J. Bram for assistance with manuscript writing.

Conflict of interest: none declared.

Appendix Table.

Food Groups and Food Items Introduced Into Factor Analysis, E3N-EPIC Cohort, France, 1993–2005

Food Group Food Item(s) Included 
Appetizers Savory biscuits, olives, and nuts 
Artificial sweeteners Artificial sweeteners (mostly aspartame) added to hot drinks, yogurt, etc. 
Biscuits Sweet biscuits such as cookies and chocolate-coated biscuits 
Breakfast cereals Sweetened and unsweetened cereals 
Butter, creama Salted and regular butter, clotted cream 
Cakes Cakes, sweet pies 
Canned fish Canned fish: anchovies, sardines, tuna 
Canned fruit Canned fruits in light syrup 
Cheese All cheeses except cottage cheese and yogurt 
Chicory Chicory as a hot drink (substitute for coffee) 
Chocolate Chocolate, chocolate bars 
Chocolate beverages Beverages consisting of mostly milk plus sweetened cocoa powder 
Coffee Espresso, instant coffee, coffee from a machine, etc. 
Cooked vegetables Cooked vegetables 
Crisp bread Manufactured rusks 
Crustaceans Crustaceans and mollusks 
Dairy-based sweet puddings Cream or milk desserts, rice or semolina puddings, ice cream 
Eggs Hard-boiled eggs, omelettes, etc. 
Fish Fresh or deep-frozen fish 
French fries Homemade, frozen, or fast-food deep-fried potatoes 
Fruits All fresh and preserved fruits except nuts, olives, and juices 
Hamb Cured and cooked ham 
High-alcohol beverages Spirits, vodka, gin, whisky, aniseed beverages, and cocktails 
High-fat dairy products Full-fat milk, full-fat yogurt and cottage cheese 
Juices Homemade or commercial pure fruit juice 
Low-alcohol beverages Beer and cider 
Low-fat dairy products Half-fat and semi-skimmed milk, low-fat yogurt 
Margarinea Margarine used as a spread and for home cooking 
Sugar, marmalade, honey Added sugar, honey; homemade and commercial jam and marmalade 
Mayonnaise Homemade or manufactured mayonnaise 
Meat Pork, beef, veal, mutton, lamb 
Mineral water Bottled mineral water, spring water (plain or sparkling) 
Offal Liver, kidney, tongue, etc. 
Olive oila Olive oil used for cooking and dressings 
Other fatsa Goose, duck fat 
Croissant-like/Danish pastries Breakfast pastries such as croissants 
Pizza, pies Pizza, savory tarts and pies 
Potatoes Potatoes, except French fries 
Poultry and rabbit Chicken, turkey, duck, goose, and rabbit 
Processed meatb All processed meats (sausages, pâté, etc.) except ham 
Pulses Dried peas, lentils 
Raw vegetables Raw vegetables 
Rice, pasta, semolina Rice, pasta, wheat or corn semolina 
Salad dressing Manufactured salad dressing 
Salad dressing—light Low-fat manufactured salad dressing 
Sandwiches Sandwiches, including hamburgers 
Soft drinks—diet Soda and fruit beverages (except pure fruit juice) with artificial sweeteners 
Soft drinks—regular Soda and fruit beverages (except pure fruit juice) 
Soup Soups and broths (homemade or commercial) 
Sunflower oil Sunflower oil used for cooking and dressings 
Sweets Sweets, assorted candy, caramels, toffee, gum, liquorice 
Tap water Tap water 
Tea Hot tea 
Other vegetable oila Oils used for cooking and dressings, except olive oil and sunflower oil 
White bread White bread, toast 
Whole-grain bread Whole-grain bread 
Wine Wine (red and white), champagne 
Food Group Food Item(s) Included 
Appetizers Savory biscuits, olives, and nuts 
Artificial sweeteners Artificial sweeteners (mostly aspartame) added to hot drinks, yogurt, etc. 
Biscuits Sweet biscuits such as cookies and chocolate-coated biscuits 
Breakfast cereals Sweetened and unsweetened cereals 
Butter, creama Salted and regular butter, clotted cream 
Cakes Cakes, sweet pies 
Canned fish Canned fish: anchovies, sardines, tuna 
Canned fruit Canned fruits in light syrup 
Cheese All cheeses except cottage cheese and yogurt 
Chicory Chicory as a hot drink (substitute for coffee) 
Chocolate Chocolate, chocolate bars 
Chocolate beverages Beverages consisting of mostly milk plus sweetened cocoa powder 
Coffee Espresso, instant coffee, coffee from a machine, etc. 
Cooked vegetables Cooked vegetables 
Crisp bread Manufactured rusks 
Crustaceans Crustaceans and mollusks 
Dairy-based sweet puddings Cream or milk desserts, rice or semolina puddings, ice cream 
Eggs Hard-boiled eggs, omelettes, etc. 
Fish Fresh or deep-frozen fish 
French fries Homemade, frozen, or fast-food deep-fried potatoes 
Fruits All fresh and preserved fruits except nuts, olives, and juices 
Hamb Cured and cooked ham 
High-alcohol beverages Spirits, vodka, gin, whisky, aniseed beverages, and cocktails 
High-fat dairy products Full-fat milk, full-fat yogurt and cottage cheese 
Juices Homemade or commercial pure fruit juice 
Low-alcohol beverages Beer and cider 
Low-fat dairy products Half-fat and semi-skimmed milk, low-fat yogurt 
Margarinea Margarine used as a spread and for home cooking 
Sugar, marmalade, honey Added sugar, honey; homemade and commercial jam and marmalade 
Mayonnaise Homemade or manufactured mayonnaise 
Meat Pork, beef, veal, mutton, lamb 
Mineral water Bottled mineral water, spring water (plain or sparkling) 
Offal Liver, kidney, tongue, etc. 
Olive oila Olive oil used for cooking and dressings 
Other fatsa Goose, duck fat 
Croissant-like/Danish pastries Breakfast pastries such as croissants 
Pizza, pies Pizza, savory tarts and pies 
Potatoes Potatoes, except French fries 
Poultry and rabbit Chicken, turkey, duck, goose, and rabbit 
Processed meatb All processed meats (sausages, pâté, etc.) except ham 
Pulses Dried peas, lentils 
Raw vegetables Raw vegetables 
Rice, pasta, semolina Rice, pasta, wheat or corn semolina 
Salad dressing Manufactured salad dressing 
Salad dressing—light Low-fat manufactured salad dressing 
Sandwiches Sandwiches, including hamburgers 
Soft drinks—diet Soda and fruit beverages (except pure fruit juice) with artificial sweeteners 
Soft drinks—regular Soda and fruit beverages (except pure fruit juice) 
Soup Soups and broths (homemade or commercial) 
Sunflower oil Sunflower oil used for cooking and dressings 
Sweets Sweets, assorted candy, caramels, toffee, gum, liquorice 
Tap water Tap water 
Tea Hot tea 
Other vegetable oila Oils used for cooking and dressings, except olive oil and sunflower oil 
White bread White bread, toast 
Whole-grain bread Whole-grain bread 
Wine Wine (red and white), champagne 

Abbreviations: E3N, Etude Epidémiologique auprès de Femmes de la Mutuelle Générale de l'Education Nationale; EPIC, European Prospective Investigation into Cancer and Nutrition.

a

Various types of seasoning and cooking fats were studied separately because of large regional differences, as well as differences in perception as healthy or unhealthy.

b

Ham was studied separately from other processed meats, as it is often part of a low-energy diet, unlike other processed meats such as sausage or pâté.

References

1.
Pisani
P
Bray
F
Parkin
DM
Estimates of the world-wide prevalence of cancer for 25 sites in the adult population
Int J Cancer
 , 
2002
, vol. 
97
 
1
(pg. 
72
-
81
)
2.
Minami
Y
Tsubono
Y
Nishino
Y
, et al.  . 
The increase of female breast cancer incidence in Japan: emergence of birth cohort effect
Int J Cancer
 , 
2004
, vol. 
108
 
6
(pg. 
901
-
906
)
3.
Katanoda
K
Matsumura
Y
National Nutrition Survey in Japan—its methodological transition and current findings
J Nutr Sci Vitaminol (Tokyo)
 , 
2002
, vol. 
48
 
5
(pg. 
423
-
432
)
4.
Yoshiike
N
Matsumura
Y
Iwaya
M
, et al.  . 
National Nutrition Survey in Japan
J Epidemiol
 , 
1996
, vol. 
6
 
3 suppl
(pg. 
S189
-
S200
)
5.
World Cancer Research Fund/American Institute for Cancer Research. Food, Nutrition, Physical Activity and the Prevention of Cancer: A Global Perspective
 , 
2007
Washington, DC
American Institute for Cancer Research
6.
Hu
FB
Dietary pattern analysis: a new direction in nutritional epidemiology
Curr Opin Lipidol
 , 
2002
, vol. 
13
 
1
(pg. 
3
-
9
)
7.
Jacobs
DR
Jr
Steffen
LM
Nutrients, foods, and dietary patterns as exposures in research: a framework for food synergy
Am J Clin Nutr
 , 
2003
, vol. 
78
 
3 suppl
(pg. 
508S
-
513S
)
8.
Edefonti
V
Randi
G
La Vecchia
C
, et al.  . 
Dietary patterns and breast cancer: a review with focus on methodological issues
Nutr Rev.
 , 
2009
, vol. 
67
 
6
(pg. 
297
-
314
)
9.
Hirose
K
Matsuo
K
Iwata
H
, et al.  . 
Dietary patterns and the risk of breast cancer in Japanese women
Cancer Sci.
 , 
2007
, vol. 
98
 
9
(pg. 
1431
-
1438
)
10.
Adebamowo
CA
Hu
FB
Cho
E
, et al.  . 
Dietary patterns and the risk of breast cancer
Ann Epidemiol
 , 
2005
, vol. 
15
 
10
(pg. 
789
-
795
)
11.
Ronco
AL
De Stefani
E
Boffetta
P
, et al.  . 
Food patterns and risk of breast cancer: a factor analysis study in Uruguay
Int J Cancer
 , 
2006
, vol. 
119
 
7
(pg. 
1672
-
1678
)
12.
Murtaugh
MA
Sweeney
C
Giuliano
AR
, et al.  . 
Diet patterns and breast cancer risk in Hispanic and non-Hispanic white women: the Four-Corners Breast Cancer Study
Am J Clin Nutr
 , 
2008
, vol. 
87
 
4
(pg. 
978
-
984
)
13.
Sieri
S
Krogh
V
Pala
V
, et al.  . 
Dietary patterns and risk of breast cancer in the ORDET cohort
Cancer Epidemiol Biomarkers Prev
 , 
2004
, vol. 
13
 
4
(pg. 
567
-
572
)
14.
Velie
EM
Schairer
C
Flood
A
, et al.  . 
Empirically derived dietary patterns and risk of postmenopausal breast cancer in a large prospective cohort study
Am J Clin Nutr
 , 
2005
, vol. 
82
 
6
(pg. 
1308
-
1319
)
15.
Terry
P
Suzuki
R
Hu
FB
, et al.  . 
A prospective study of major dietary patterns and the risk of breast cancer
Cancer Epidemiol Biomarkers Prev
 , 
2001
, vol. 
10
 
12
(pg. 
1281
-
1285
)
16.
Cui
X
Dai
Q
Tseng
M
, et al.  . 
Dietary patterns and breast cancer risk in the Shanghai Breast Cancer Study
Cancer Epidemiol Biomarkers Prev
 , 
2007
, vol. 
16
 
7
(pg. 
1443
-
1448
)
17.
Touillaud
MS
Thiébaut
AC
Fournier
A
, et al.  . 
Dietary lignan intake and postmenopausal breast cancer risk by estrogen and progesterone receptor status
J Natl Cancer Inst
 , 
2007
, vol. 
99
 
6
(pg. 
475
-
486
)
18.
Lucas
F
Niravong
M
Villeminot
S
, et al.  . 
Estimation of food portion size using photographs: validity, strengths, weaknesses and recommendations
J Hum Nutr Diet
 , 
1995
, vol. 
8
 (pg. 
65
-
74
)
19.
van Liere
MJ
Lucas
F
Clavel
F
, et al.  . 
Relative validity and reproducibility of a French dietary history questionnaire
Int J Epidemiol
 , 
1997
, vol. 
26
 
suppl 1
(pg. 
S128
-
S136
)
20.
Riboli
E
Nutrition and cancer: background and rationale of the European Prospective Investigation into Cancer and Nutrition (EPIC)
Ann Oncol
 , 
1992
, vol. 
3
 
10
(pg. 
783
-
791
)
21.
Food and Agriculture Organization of the United Nations; World Health Organization; United Nations University
Energy and Protein Requirements: Report of a Joint FAO/WHO/UNU Expert Consultation. (WHO Technical Report Series no. 724)
 , 
1985
Geneva, Switzerland
World Health Organization
22.
Ferrari
P
Slimani
N
Ciampi
A
, et al.  . 
Evaluation of under- and overreporting of energy intake in the 24-hour diet recalls in the European Prospective Investigation into Cancer and Nutrition (EPIC)
Public Health Nutr
 , 
2002
, vol. 
5
 
6B
(pg. 
1329
-
1345
)
23.
Newby
PK
Weismayer
C
Akesson
A
, et al.  . 
Long-term stability of food patterns identified by use of factor analysis among Swedish women
J Nutr
 , 
2006
, vol. 
136
 
3
(pg. 
626
-
633
)
24.
Slattery
ML
Boucher
KM
Caan
BJ
, et al.  . 
Eating patterns and risk of colon cancer
Am J Epidemiol
 , 
1998
, vol. 
148
 
1
(pg. 
4
-
16
)
25.
Cattell
RB
The scree test for the number of factors
Multivariate Behav Res.
 , 
1966
, vol. 
1
 
2
(pg. 
245
-
276
)
26.
Kim
J-O
Mueller
CW
Factor Analysis: Statistical Methods and Practical Issues
 , 
1978
Thousand Oaks, CA
Sage Publications
27.
Lagakos
SW
A covariate model for partially censored data subject to competing causes of failure
Appl Stat
 , 
1978
, vol. 
27
 
3
(pg. 
235
-
241
)
28.
Jacques
PF
Tucker
KL
Are dietary patterns useful for understanding the role of diet in chronic disease?
Am J Clin Nutr
 , 
2001
, vol. 
73
 
1
(pg. 
1
-
2
)
29.
Martínez
ME
Marshall
JR
Sechrest
L
Invited commentary: factor analysis and the search for objectivity
Am J Epidemiol
 , 
1998
, vol. 
148
 
1
(pg. 
17
-
19
)
30.
Borland
SE
Robinson
SM
Crozier
SR
, et al.  . 
Stability of dietary patterns in young women over a 2-year period
Eur J Clin Nutr
 , 
2008
, vol. 
62
 
1
(pg. 
119
-
126
)
31.
Hu
FB
Rimm
E
Smith-Warner
SA
, et al.  . 
Reproducibility and validity of dietary patterns assessed with a food-frequency questionnaire
Am J Clin Nutr
 , 
1999
, vol. 
69
 
2
(pg. 
243
-
249
)
32.
Bamia
C
Orfanos
P
Ferrari
P
, et al.  . 
Dietary patterns among older Europeans: the EPIC-Elderly study
Br J Nutr
 , 
2005
, vol. 
94
 
1
(pg. 
100
-
113
)
33.
McNaughton
SA
Mishra
GD
Stephen
AM
, et al.  . 
Dietary patterns throughout adult life are associated with body mass index, waist circumference, blood pressure, and red cell folate
J Nutr
 , 
2007
, vol. 
137
 
1
(pg. 
99
-
105
)
34.
Schulze
MB
Fung
TT
Manson
JE
, et al.  . 
Dietary patterns and changes in body weight in women
Obesity (Silver Spring)
 , 
2006
, vol. 
14
 
8
(pg. 
1444
-
1453
)
35.
Trichopoulou
A
Costacou
T
Bamia
C
, et al.  . 
Adherence to a Mediterranean diet and survival in a Greek population
N Engl J Med
 , 
2003
, vol. 
348
 
26
(pg. 
2599
-
2608
)
36.
Lajous
M
Romieu
I
Sabia
S
, et al.  . 
Folate, vitamin B12 and postmenopausal breast cancer in a prospective study of French women
Cancer Causes Control
 , 
2006
, vol. 
17
 
9
(pg. 
1209
-
1213
)
37.
Smith-Warner
SA
Spiegelman
D
Yaun
SS
, et al.  . 
Intake of fruits and vegetables and risk of breast cancer: a pooled analysis of cohort studies
JAMA
 , 
2001
, vol. 
285
 
6
(pg. 
769
-
776
)
38.
van Gils
CH
Peeters
PH
Bueno-de-Mesquita
HB
, et al.  . 
Consumption of vegetables and fruits and risk of breast cancer
JAMA
 , 
2005
, vol. 
293
 
2
(pg. 
183
-
193
)
39.
Chajès
V
Thiébaut
AC
Rotival
M
, et al.  . 
Association between serum trans-monounsaturated fatty acids and breast cancer risk in the E3N-EPIC Study
Am J Epidemiol
 , 
2008
, vol. 
167
 
11
(pg. 
1312
-
1320
)
40.
Fung
TT
Hu
FB
Holmes
MD
, et al.  . 
Dietary patterns and the risk of postmenopausal breast cancer
Int J Cancer
 , 
2005
, vol. 
116
 
1
(pg. 
116
-
121
)
41.
Gaudet
MM
Britton
JA
Kabat
GC
, et al.  . 
Fruits, vegetables, and micronutrients in relation to breast cancer modified by menopause and hormone receptor status
Cancer Epidemiol Biomarkers Prev
 , 
2004
, vol. 
13
 
9
(pg. 
1485
-
1494
)
42.
Kushi
LH
Potter
JD
Bostick
RM
, et al.  . 
Dietary fat and risk of breast cancer according to hormone receptor status
Cancer Epidemiol Biomarkers Prev
 , 
1995
, vol. 
4
 
1
(pg. 
11
-
19
)
43.
Löf
M
Sandin
S
Lagiou
P
, et al.  . 
Dietary fat and breast cancer risk in the Swedish women's lifestyle and health cohort
Br J Cancer
 , 
2007
, vol. 
97
 
11
(pg. 
1570
-
1576
)
44.
Olsen
A
Tjønneland
A
Thomsen
BL
, et al.  . 
Fruits and vegetables intake differentially affects estrogen receptor negative and positive breast cancer incidence rates
J Nutr
 , 
2003
, vol. 
133
 
7
(pg. 
2342
-
2347
)
45.
Prentice
RL
Caan
B
Chlebowski
RT
, et al.  . 
Low-fat dietary pattern and risk of invasive breast cancer: the Women's Health Initiative Randomized Controlled Dietary Modification Trial
JAMA
 , 
2006
, vol. 
295
 
6
(pg. 
629
-
642
)
46.
Suzuki
R
Ye
W
Rylander-Rudqvist
T
, et al.  . 
Alcohol and postmenopausal breast cancer risk defined by estrogen and progesterone receptor status: a prospective cohort study
J Natl Cancer Inst
 , 
2005
, vol. 
97
 
21
(pg. 
1601
-
1608
)
47.
Remontet
L
Estève
J
Bouvier
AM
, et al.  . 
Cancer incidence and mortality in France over the period 1978–2000
Rev Epidemiol Sante Publique
 , 
2003
, vol. 
51
 
1
(pg. 
3
-
30
)
48.
McCann
SE
Marshall
JR
Brasure
JR
, et al.  . 
Analysis of patterns of food intake in nutritional epidemiology: food classification in principal components analysis and the subsequent impact on estimates for endometrial cancer
Public Health Nutr
 , 
2001
, vol. 
4
 
5
(pg. 
989
-
997
)