Association of Mediterranean diet with survival after breast cancer diagnosis in women from nine European countries: results from the EPIC cohort study | BMC Medicine

Study population

We used data from the European Prospective Investigation into Cancer and Nutrition (EPIC) study, a prospective, multicenter European cohort with more than half a million women and men recruited between 1992 and 2000. Full details of the study design and data collection have been described elsewhere [13, 14]. Participants completed questionnaires on diet, lifestyle, and medical history at the time of recruitment and anthropometric measurements were also obtained. All participants provided written informed consent and the study was approved by the ethical review committees of the International Agency for Research on Cancer (IARC-Lyon, France) and all local centers.

Dietary assessment

The dietary assessment was conducted using a combination of methods, including detailed dietary questionnaires, and food frequency questionnaires (FFQs). At recruitment, participants completed a validated country- or center-specific dietary questionnaire that included questions on the frequency and portion sizes of foods and drinks consumed in the previous year and was designed to capture the geographical specificity of the diet [13].

Derivation of the arMED score

To measure adherence to the MD, we used the arMED score [8], a variant of the original MD scale defined by Trichopoulou et al. [15]. The arMED score was based on tertiles of energy-adjusted intake of eight foods/food groups to reflect consumption in relation to the individual’s total daily energy intake. Unlike the original score [15], the arMED includes olive oil instead of monounsaturated fats, and alcohol was excluded from the list of components due to its causal association with BC carcinogenesis. For five items presumed to fit the MD, a score of 0 to 2 was assigned to tertiles of intake: fruits (including nuts and seeds), vegetables (excluding potatoes), legumes, fish, and cereals. The scoring was inverted for the components presumed to not fit MD: meat (red meat and processed meat) and dairy products. The score was slightly modified for olive oil due to the relatively high proportion of non-consumers in some countries; a score of 0 was assigned to non-consumers, 1 to participants below the median (calculated among consumers), and 2 to participants at or above the median. Thus, the arMED score ranged from 0 to 16, with higher scores indicating greater adherence to MD.

Ascertainment of breast cancer cases

The International Classification of Diseases for Oncology (ICD-O-2) codes C50.0–50.9 were used to define BC cases. Women with prevalent tumors at recruitment, no follow-up data, no information on lifestyle and diet, or implausible diets were excluded; furthermore, BC cases (N = 50) with unknown vital status, inconsistent follow-up data, or with non-epithelial morphology were also excluded. Out of 318,686 women from nine countries (Denmark, France, Germany, Italy, the Netherlands, Norway, Spain, Sweden, and the UK), a total of 13,270 incident primary malignant breast cancers (including 14 in situ) were diagnosed during the follow-up and were included in the present analysis.

Statistical analyses

The baseline characteristics of the participants were described as mean (SD) for continuous variables and frequencies for categorical variables. Cox proportional hazard models were used to prospectively analyze associations between the arMED score and overall mortality. Fine-Gray competing risks models were performed to evaluate the association with BC-specific mortality, with other causes of death considered competing events. Entry time was defined as the date of diagnosis of primary breast cancer, and exit time was defined as the date of death or end of follow-up. The arMED score was assessed as a categorical variable according to low (score 0–5), medium (score 6–8), and high (score 9–16) adherence, using the medium category as the reference, as well as per 3-unit increase in the score. Restricted cubic spline models with five knots were fitted, and non-linearity was tested using the likelihood (LR) ratio test.

All models were stratified by country, menopausal status at diagnosis (women aged ≥ 55 years at diagnosis were considered postmenopausal regardless of the baseline information) and stage of the tumor (non-metastatic, metastatic, unknown) and adjusted for: age at diagnosis (5-years categories), education level (no formal education, primary school, secondary school, technical or professional training, university, and not specified), body mass index (BMI) (kg/m2, continuous), physical activity (inactive, moderately inactive, moderately active, active, unknown), alcohol consumption (non-drinker, 0 to < 3 g/day, 3 to < 12 g/day, 12 to < 24 g/day, ≥ 24 g/day, unknown), smoking status and intensity (never smokers, current smokers 1–15, 16–25, and > 25 cigarettes/day, former quit ≤ 10, 11–20, and > 20 years before recruitment, current smoker of cigars, pipes and occasional current smokers, current smokers with unknown intensity, and not specified), ever use of hormone replacement therapy for menopause at diagnosis (yes, no, unknown), grade of tumor (well differentiated, moderately differentiated, poorly differentiated or undifferentiated, not determined), and tumor receptor status (positive, negative, unknown) for estrogen receptor (ER), progesterone receptor (PR), and human epidermal growth factor receptor 2 (HER2). BMI was modeled as a restricted cubic spline to account for its non-linear association with mortality [5, 16].

Separate models for pre- and postmenopausal cases were performed and heterogeneity was tested by comparing models with and without the cross-product terms using the Likelihood ratio (LR) test. The proportional hazards assumptions were tested by using the Schoenfeld goodness-of-fit test.

Stratified analyses were performed for overall mortality according to potential modifiers of the association with arMED: BMI, physical activity, smoking status, tumor stage, and hormone receptor status (ER, PR, HER2, and triple negative), and adherence to dietary patterns related to underlying biological mechanisms of BC previously associated with BC survival [17]: low/high adherence to the Diabetes Risk Reduction Diet [DRRD] [18] and Inflammatory Score of Diet [ISD] [19].

In sensitivity analyses, we examined whether further adjustment for the time interval between recruitment (time at which dietary information was collected) and BC diagnosis modified our main results, as well as for the period of diagnosis, to account for the potential influence of improvements in treatment and diagnosis over time. Comorbidities, including cardiovascular disease and presence of diabetes at baseline, and a combined variable with mechanistic dietary patterns (DRRD-ISD) were also used to further adjust separate models and test the robustness of the results.

Direct adjusted cumulative incidence function (CIF) curves for the three levels of adherence to the arMED score and overall mortality were derived from the multivariable Cox proportional hazards model [20].

All analyses were performed using R version 4.2.2. We used a significance level of 0.05, but also considered the confidence intervals and point estimate magnitudes. Data analysis was conducted from October 1, 2022, to January 13, 2023.

Source link