Association of Mediterranean diet, dietary supplements and alcohol consumption with breast density among women in South Germany: a cross-sectional study

Study population

Between April 2007 and September 2008 494 women aged 21 to 84 years, who underwent mammography at the breast center in University Hospital Ulm, were recruited in this cross-sectional study. After informed consent they completed a self-administered questionnaire. From the statistical analysis 65 participants were excluded due to history of breast ablation and breast cancer and 5 participants filled in the questionnaire twice. Thus 424 women remained for the analysis.

A study protocol was approved by the ethics committee of the medical faculty of Ulm University.

Data assessment

The questionnaire assessed socio-demographic characteristics, including date of birth, nationality, body weight, height and education. Furthermore, participants were asked to comment on their consumption of food, beverages, physical activity and their smoking status. The food questionnaire of this study was partially adapted from the food questionnaire of the Million Women Study [12]. The choice of the components was based on the hypothesis that a low consumption of “unhealthy” food, such as meat (contains saturated fatty acids) [13] and a high consumption of “healthy” food, such as fruits (contain antioxidants) [14] protect against a high mammographic density as indicator for higher breast cancer risk. Participants provided information on reproductive factors, such as age at menarche, menopausal status and parity, and family history of breast cancer. Data on hormone replacement therapy, biopsy of breast and past medical history was also included.

In this study, we collected information of the supplementary intake of vitamins and minerals both individually and combined. Due to the small numbers on single vitamin and mineral use, we investigated the association between intake of multivitamin-multimineral supplements and mammographic density. The corresponding question in our questionnaire was: “Do you take vitamins, minerals or dietary supplements regularly?”

Data on the consumption of wine, beer and spirits beverages in glasses per week has been collected to assess a drinking of alcoholic beverages. We calculated how much alcohol in g (gram) each woman consumed according to conversions described in other German studies [15, 16]. The following conversions were made:

wine: 1 glass = 0.25 liter [15].

beer: 1 glass = 0.3 liter.

spirits beverage: 1 glass = 0.02 liter.

1 liter wine = 100 g alcohol, and

1 liter beer = 40 g alcohol [16].

There were 25 g alcohol in 1 glass wine; 12 g alcohol in 1 glass bier and 6.2 g alcohol in 1 glass spirits beverage [16]. On the basis of this data, we calculated the total alcohol consumption (wine, beer and spirits beverage) per day (d) in gram for each participant. We created quantitative (increase per 10 g/d) and qualitative variables to analyze alcohol consumption. Categorized variable included four classes: > 10.0 g/d total alcohol; 5.0 - 10.0 g/d total alcohol; 0.1 - 5.0 g/d total alcohol and 0 g/d total alcohol.

We selected the dietary components contributing to a score based on other studies of adherence to Mediterranean diet. Based on other studies on the adherence to Mediterranean dietary pattern [11, 17] we selected the dietary components in our study. The following foods were included: boiled and raw vegetables, fruit, fish, nuts and olive oil (pro-Mediterranean diet); butter, beef, pork, sausages, ham, hamburger and lemonade/soft drinks (contra-Mediterranean diet). Points were assigned for pro-Mediterranean food: low consumption = 1 point, regular consumption = 2 points, frequent consumption = 3 points. Points for contra-Mediterranean food were reversed: low consumption = 3 points, regular consumption = 2 points, frequent consumption = 1 point. In cases of no consumption or where no information was provided, 0 points were given. In total, a maximum of 29 points could be achieved. High score indicates adherence to Mediterranean diet.

As potential confounders were considered: physical activity at recruitment (yes/no), Body-Mass-Index (BMI) at recruitment (kg/m²), ever use of hormone replacement therapy (HRT) (yes/no), menopausal status at recruitment (proxy variable with premenopausal (<50) and postmenopausal (≥50) women), age at recruitment (years), mother with history of breast cancer (yes/no), school education (9 years of school/13 years of school), ever use of oral contraceptives (yes/no), age at menarche (years), smoker at recruitment (yes/no), ever breast feeding (yes/no), number of live births (quantitative variable; 0,1,2,3,4 or 5 live births), age at the first birth (quantitative variable; years), alcohol consumption at recruitment (yes/no) if not exposure variable.

Mammographic density assessment

Breast density was assessed by two radiologists of the University Hospital Ulm applying the classification of the American College of Radiology (ACR) including four categories: ACR 1 = almost entirely fat, glandular tissue <25%; ACR 2 = scattered fibroglandular densities (ca. 25–50% of the breast); ACR 3 = heterogeneously breast dense (ca. 51–75% of the breast); ACR 4 = extremely dense (>75% of the breast) [7]. For the analyses the ACR categories were combined: category ACR 1/2 (including ACR 1 and ACR 2) means lower density and category ACR 3/4 (including ACR 3 und ACR 4) means higher density of breast. The measurement of the left breast was incorporated into the statistical analysis. Breast densities were assessed by two radiologists separately to determine inter-rater reliability. The differences in assessment of mammographic density between these two raters were calculated with the kappa-value. According to Altman’s interpretation of kappa-values [18], the inter-rater variability assessed by kappa statistics showed moderate agreement (0.48) for ACR classes, but for the categories ACR1/2 and ACR 3/4 good agreement (0.61) was found.

Statistical analysis

The X ²-test, the t-test and the Wilcoxon-test were used to analyze the significance of different frequencies and distribution, respectively. We defined the level of statistical significance as 0.05. We used these tests to investigate the significance of relationships between possible confounders (such as BMI, education, etc.) and either the target variable (mammographic density) or influencing variables (such as current intake of multivitamin-multimineral supplements, current alcohol consumption and Mediterranean dietary pattern). Tests for trend across quartiles were performed by assigning the mean level within specific quartiles of alcohol intake to all individuals in that quartile, and using this as a continuous variable in a linear regression. Logistic regression models were fitted to obtain Odds Ratios (ORs) with 95% confidence intervals (95%CIs) for high mammographic density. Additionally interactions between influencing variables and the target variable were tested. Variables, which were identified as potential confounders because they had a biological plausibility and statistically significant relationships with the target variable and/or influencing variables, were tested with manual step-by-step regression. A variable showed an effect and was identified as a confounder, if it had changed estimator more than 10%. The parsimonious models were adjusted for age and BMI based on results of the manual step-by-step regression and/or the biological plausibility. These models were computed stratified by smoking status in order to account for differential association according to hormonal and oxidative stress, respectively. Also the fully adjusted models were calculated, adjusted for physical activity, BMI, HRT use, menopausal status, age, mother with history of breast cancer, school education, ever use of oral contraceptives, age at menarche, smoker at recruitment, number of live births, alcohol consumption. Statistical analysis was assessed with the Analyst Interface of the SAS (Statistical Analysis System) Version 9.2.

Mediterranean diet

In line with former research on diet and mammographic density, we found an inverse association between the adherence to Mediterranean diet score and mammographic density [11, 19, 20]. Our results did not change by additional adjustment for alcohol consumption and multivitamin-multimineral supplements.

Previous analyses of single food groups and mammographic density showed that components of the Mediterranean diet such as vegetables and olive oil could be inversely associated [20]. Consistent with our finding of an inverse association among postmenopausal women, prospective studies on breast cancer have shown that conformity to Mediterranean diet is associated with reduced breast cancer risk among postmenopausal women [17, 21]. However, among postmenopausal women in Japan intakes of protein, total fat, and saturated fat were significantly positively associated with percentage of mammographic density [22], while carbohydrates were inversely associated. For components of the Mediterranean diet biological mechanisms have been described related to antioxidants, phytoestrogens and fatty acids [23]. In an observational study inverse associations between phytoestrogens and mammographic density have been reported [24].

In contrast to Tseng et al. (2008) we observed a statistically significant inverse association of Mediterranean diet score with mammographic density among non-smoking women.

Multivitamin-multimineral supplements

Our findings of an inverse association between the current use of multivitamin-multimineral supplements and mammographic density in postmenopausal women are in contrast to results from a case–control study showing that multivitamin-multimineral supplement use is associated with higher breast density among premenopausal women [25]. Our results did not substantially change by additional adjustment for educational levels or other adjustment variables. Due to the antioxidant property of vitamins and minerals an inverse association seems to be biologically plausible [26].

Epidemiological studies on diet have shown inverse associations between vitamin D intake and mammographic density [27]. For the intake of vitamin C and E positive associations with mammographic density [19] and inverse associations for vitamin C intake have been reported [20].

Women who take multivitamin-multimineral supplement may have a healthier lifestyle than others. However, differences in the composition of the multivitamin-multimineral supplement pills and the inclusion of potential confounders could have influenced the associations. Data on use of vitamin and mineral supplements and its combination are scarce [25].

Alcohol consumption

In our study, current alcohol intake of more than 10g/d was associated with high mammographic density in pre- and postmenopausal women. Results from prospective [20, 28], case–control [29] and cross-sectional [30] studies have shown that increased alcohol intake is associated with mammographic more dense breast parenchyma. However, also no association between alcohol intake and mammographic density was found in a cross-sectional study [31]. Flom et al. investigated alcohol beverages and found that particularly beer and white wine consumption may increase mammographic density [28].

There is evidence linking alcohol by Insulin-like growth factor (IGF) and Insulin-like growth factor binding protein (IGFBP)-3 to mammographic density [32, 33]. In addition, alcohol increases estrogen levels in blood [34] which were found to increase mammographic density [35].

Strengths and limitations

One potential limitation of this study is that measurement error is inherent in dietary assessment. However, our data for alcohol were consistent with other data from Germany [16], and for dietary variables long-term reproducibility was fairly [36] indicating that dietary habits are stable over longer time intervals.

In the Mediterranean diet score in cases with no information on food items (e.g. for fish), 0 points were given. However, exclusion of individuals without information revealed similar results in the fully adjusted model without elimination of these women.

Since a cross-sectional study was performed, we cannot appraise the time sequence of the associations. The data on exposure variables were collected retrospectively, which may have introduced recall bias. However, it is likely to be non differential.

The assessment of the mammograms by experienced radiologists is one of the strengths of our study. In addition, we calculated the fully adjusted and the parsimonious models which revealed similar results but only statistical significance was reached in the parsimonious models.

The inter-rater reliability for the analyzed categories of ACR classification as indicator for mammographic density was good. ACR classification of mammographic density is widely applied and it has been shown that it is correlated with more refined measurement methods [8]. Information on reproductive and use of exogenous hormones has been collected by a standardized questionnaire. However, in our data these variables did not substantially affect the estimates.