Mediterranean diet adherence and risk of postmenopausal breast cancer: results of a cohort study and meta-analysis

International Journal of CancerVolume 140, Issue 10 p. 2220-2231 Cancer EpidemiologyFree Access Mediterranean diet adherence and risk of postmenopausal breast cancer: results of a cohort study and meta-analysis Piet A. van den Brandt, Corresponding Author Piet A. van den Brandt [email protected] orcid.org/0000-0001-8781-8099 Department of Epidemiology, GROW – School for Oncology and Developmental Biology, Maastricht University Medical Centre, Maastricht, the Netherlands Department of Epidemiology, CAPHRI – School for Public Health and Primary Care, Maastricht University Medical Centre, Maastricht, the NetherlandsCorrespondence to: Piet A. van den Brandt, Department of Epidemiology, Maastricht University Medical Centre, PO Box 616; 6200 MD Maastricht, The Netherlands, E-mail: [email protected]; Tel: +31 043 3882361Search for more papers by this authorMaya Schulpen, Maya Schulpen Department of Epidemiology, GROW – School for Oncology and Developmental Biology, Maastricht University Medical Centre, Maastricht, the NetherlandsSearch for more papers by this author Piet A. van den Brandt, Corresponding Author Piet A. van den Brandt [email protected] orcid.org/0000-0001-8781-8099 Department of Epidemiology, GROW – School for Oncology and Developmental Biology, Maastricht University Medical Centre, Maastricht, the Netherlands Department of Epidemiology, CAPHRI – School for Public Health and Primary Care, Maastricht University Medical Centre, Maastricht, the NetherlandsCorrespondence to: Piet A. van den Brandt, Department of Epidemiology, Maastricht University Medical Centre, PO Box 616; 6200 MD Maastricht, The Netherlands, E-mail: [email protected]; Tel: +31 043 3882361Search for more papers by this authorMaya Schulpen, Maya Schulpen Department of Epidemiology, GROW – School for Oncology and Developmental Biology, Maastricht University Medical Centre, Maastricht, the NetherlandsSearch for more papers by this author First published: 05 March 2017 https://doi.org/10.1002/ijc.30654Citations: 114AboutSectionsPDF ToolsRequest permissionExport citationAdd to favoritesTrack citation ShareShare Give accessShare full text accessShare full-text accessPlease review our Terms and Conditions of Use and check box below to share full-text version of article.I have read and accept the Wiley Online Library Terms and Conditions of UseShareable LinkUse the link below to share a full-text version of this article with your friends and colleagues. Learn more.Copy URL Abstract The Mediterranean Diet (MD) has been associated with reduced mortality and risk of cardiovascular diseases, but there is only limited evidence on cancer. We investigated the relationship between adherence to MD and risk of postmenopausal breast cancer (and estrogen/progesterone receptor subtypes, ER/PR). In the Netherlands Cohort Study, 62,573 women aged 55–69 years provided information on dietary and lifestyle habits in 1986. Follow-up for cancer incidence until 2007 (20.3 years) consisted of record linkages with the Netherlands Cancer Registry and the Dutch Pathology Registry PALGA. Adherence to MD was estimated through the alternate Mediterranean Diet Score excluding alcohol. Multivariate case–cohort analyses were based on 2,321 incident breast cancer cases and 1,665 subcohort members with complete data on diet and potential confounders. We also conducted meta-analyses of our results with those of other published cohort studies. We found a statistically significant inverse association between MD adherence and risk of ER negative (ER−) breast cancer, with a hazard ratio of 0.60 (95% Confidence Interval, 0.39–0.93) for high versus low MD adherence (ptrend = 0.032). MD adherence showed only nonsignificant weak inverse associations with ER positive (ER+) or total breast cancer risk. In meta-analyses, summary HRs for high versus low MD adherence were 0.94 for total postmenopausal breast cancer, 0.98 for ER+, 0.73 for ER− and 0.77 for ER − PR− breast cancer. Our findings support an inverse association between MD adherence and, particularly, receptor negative breast cancer. This may have important implications for prevention because of the poorer prognosis of these breast cancer subtypes. Abstract What's new? When it comes to diet and breast cancer risk, dietary patterns may be of greater importance than individual foods or nutrients. It remains uncertain, however, whether specific dietary patterns impact breast cancer risk. The Mediterranean Diet (MD), which is linked to reduced cardiovascular disease risk, is of particular interest. Here, MD adherence was investigated for potential associations with risk of postmenopausal breast cancer. Analyses of data on 62,573 women ages 55–69 enrolled in the Netherlands Cohort Study show that increased MD adherence is associated with reduced risk of estrogen receptor-negative breast cancer. A meta-analysis of cohort studies confirmed the finding. Abbreviations AIC Akaike Information Criterium AICR American Institute for Cancer Research aMED alternate Mediterranean Diet Score aMEDr alternate Mediterranean Diet Score excluding alcohol BMI body mass index EPIC European Prospective Investigation into Cancer ER+PR+ estrogen receptor positive and progesterone receptor positive ER−PR− estrogen receptor negative and progesterone receptor negative HR hazard ratio HRT hormone replacement therapy MD Mediterranean Diet mMED modified Mediterranean Diet Score mMEDr modified Mediterranean Diet Score excluding alcohol NHS Nurses' Health Study NLCS Netherlands Cohort Study SD standard deviation WCRF World Cancer Research Fund 95% CI 95% confidence interval. Introduction Breast cancer is the most commonly diagnosed cancer in Western countries, and prevention is of paramount importance to reduce the burden of this disease. Thus far, very few modifiable (lifestyle) risk factors, such as overweight and alcohol consumption, have been identified. Intake of individual dietary factors has been extensively studied in relation to breast cancer risk, but only for alcohol there is convincing evidence for an increased risk.1 Because individuals do not consume isolated foods or nutrients, studying dietary patterns in relation to breast cancer seems more fruitful, thereby acknowledging interactions between individual components as well as existing collinearity between components. Dietary patterns might also yield more actionable information on dietary change needed for prevention. In contrast with dietary patterns that are a posteriori derived from factor or cluster analyses of a dataset, the Mediterranean diet (MD) score is a dietary quality index, a priori constructed on the basis of dietary recommendations.2 The traditional MD is characterized by a high intake of plant proteins, whole grains, fish and monounsaturated fat, moderate alcohol intake and low intake of refined grains, red meat and sweets.3, 4 MD adherence is associated with decreased risk of mortality and cardiovascular diseases; however, for cancer risk, results are still rather limited. A recent meta-analysis5 reported a lower incidence of overall breast cancer for women adhering to the highest category of MD-scores in case–control studies, but not in cohort studies. It is important to distinguish between pre- and postmenopausal breast cancer, as well as hormone receptor subtypes, because of differences in etiology. The meta-analysis suggested that evidence for an inverse association with MD was more convincing for postmenopausal breast cancer. Furthermore, differences were noted between different estrogen/progesterone receptor (ER/PR) subtypes of breast cancer in the associations with MD, but this observation was based on very few cohort studies. Recent evidence from a randomized controlled trial on primary prevention of cardiovascular diseases indicated a strong protective effect of MD on the risk of postmenopausal breast cancer in Spain.6 We investigated the association between adherence to MD and postmenopausal breast cancer risk, overall and stratified by hormone receptor status, in the Netherlands Cohort Study (NLCS). Based on earlier findings,7 we hypothesized that MD-adherence would show a stronger inverse association with ER− breast cancer than ER+ breast cancer, which may have important implications for prevention because of the poorer prognosis of ER− breast cancer. Because alcohol is a risk factor for breast cancer,8 we excluded it from the MD-score that normally includes moderate alcohol consumption, and tested the effect of this exclusion. We also conducted meta-analyses on MD-adherence and breast cancer risk by subtype. Material and Methods Study design and cancer follow-up The NLCS started in September 1986 and the female part included 62,573 women aged 55–69 years.9 At baseline, participants completed a mailed, self-administered questionnaire on cancer risk factors. The NLCS study was approved by institutional review boards from Maastricht University and the Netherlands Organization for Applied Scientific Research. All cohort members consented to participation by completing the questionnaire. For data processing and analysis the case–cohort method was used.10 Accumulated person-years in the cohort were estimated from a subcohort (n = 2,589 women), randomly sampled from the cohort immediately after baseline. These subcohort members were actively followed up biennially for vital status information. The follow-up of the subcohort was 100% complete at 20.3 years of follow-up. Follow-up for cancer incidence was established by annual record linkage with the Netherlands Cancer Registry and PALGA, the nationwide Dutch Pathology Registry.11 Completeness of follow-up through record linkage with cancer registries and PALGA was estimated to be >95%.12 After 20.3 years of follow-up (September 17, 1986 until January 1, 2007), a total of 3,354 incident breast cancer cases were detected among women. Cases and subcohort members were excluded if they reported a history of cancer (except skin cancer) at baseline and if their dietary data were incomplete or inconsistent. Figure S1 (Supplementary data) shows the selection and exclusion steps that resulted in the number of cases and female subcohort members that were included in the analysis. There were 1,665 subcohort members and 2,321 breast cancer cases available for analysis. Exposure assessment The 11-page baseline questionnaire measured dietary intake, detailed smoking habits, anthropometry, physical activity and other risk factors related to cancer.9 Habitual consumption of food and beverages during the year preceding baseline was assessed using a 150-item semi-quantitative food-frequency questionnaire, which was validated against a 9-day diet record.13 Nutrient intakes were calculated using the computerized Dutch food composition table.14 Non-occupational physical activity was calculated by adding the minutes spent per day on cycling or walking, shopping, walking the dog, gardening and sports or exercise as reported previously.15 Mediterranean diet score Conformity with the MD was assessed using the alternate Mediterranean Diet Score (aMED),16, 17 which is an adapted version of the traditional Mediterranean Diet Score created by Trichopoulou et al.18, 19 The aMED contains 9 dietary components that are typical of the Mediterranean diet. To control for energy intake, the intake of each component was first adjusted to a daily intake of 2,000 kcal.16, 17, 19 For each of the presumed beneficial food items (vegetables (without potatoes), legumes, fruits, nuts, whole grains, fish and the ratio of monounsaturated to saturated fatty acid intake (MUFA:SFA)), one point was given when the intake was at least the sex-specific median intake, and zero otherwise. For red and processed meat, 1 point was given (and 0 otherwise) when the intake was below the sex-specific median intake. In the full aMED, 1 additional point is normally given when alcohol intake is between 5 and 25 g/day, and 0 otherwise.17 However, since alcohol is a risk factor for breast cancer,8 we excluded alcohol from the score in the present analysis. The reduced 9-point sum score (aMEDr) ranged from zero to eight points (minimal to maximal conformity). Mediterranean diet adherence was also assessed using the modified MD score by Trichopoulou et al.,20 abbreviated as mMED. Apart from alcohol, this score differs from aMED as follows: fruits and nuts are combined in one component; dairy is considered as component; cereals are considered as component instead of whole grains, total meat is used instead of red and processed meat, and for fatty acids the ratio of (MUFA + PUFA):SFA is used. Statistical analysis The reduced scores (aMEDr and mMEDr) were categorized in three categories: 0–3, 4–5 and 6–8 points. The distribution of the subcohort members by aMEDr-score, mMEDr-score and various characteristics was examined by cross-tabulations and summary statistics. The relationship between Mediterranean diet adherence and breast cancer risk was evaluated using Cox proportional hazards models. It was verified that the proportional hazards assumption was not violated using scaled Schoenfeld residuals21 and -ln(-ln) survival plots. Standard errors were estimated using the robust Hubert–White sandwich estimator to account for additional variance introduced by the subcohort sampling.22 We conducted age- and multivariable-adjusted survival analyses in which aMEDr and mMEDr were tested on categorical and continuous scales. In the multivariable analyses, hazard ratios (HRs) were corrected for potential confounding by age at baseline (55–59, 60–64, 65–69 years), cigarette smoking (status (never, former, current), frequency (number of cigarettes per day; continuous, centered), duration (number of years; continuous, centered)), body height (continuous, cm), BMI (<18.5, 18.5–<25, 25–<30, ≥30 kg/m2), non-occupational physical activity (≤30, >30–60, >60–90, >90 min/day), highest level of education (primary school or lower vocational, secondary or medium vocational and higher vocational or university), family history of breast cancer in mother or sisters (no, yes), history of benign breast disease (no, yes), age at menarche (≤12, 13–14, 15–16, ≥17 years), parity (nulliparous, 1–2, ≥3 children), age at first birth (<25, ≥25 years), age at menopause (<45, 45–49, 50–54, ≥55 years), oral contraceptive use (never, ever), postmenopausal hormone replacement therapy (never, ever), energy intake (continuous, kcal/day) and, depending on the analysis, alcohol intake (0, 0.1–<5, 5–<15, 15–<30, ≥30 g/day). Tests for trends were assessed by fitting ordinal exposure variables as continuous terms. The Akaike Information Criterium (AIC)23 was used to compare performance of models with aMEDr and mMEDr scores. We also analyzed associations with the full aMED and mMED-scores (including alcohol) to compare these with models using aMEDr and mMEDr, using the AIC. Besides overall postmenopausal breast cancer, we conducted these analyses for subtypes defined by hormone receptor status: ER+, ER−, PR+, PR−, ER + PR+ and ER − PR−. Differences in associations with MD-scores between breast cancer subtypes were tested using a heterogeneity test,24 in which the standard error for the observed difference in rate ratios was estimated using a bootstrapping method developed for the case–cohort design.25 To evaluate potential residual confounding by breast cancer risk factors, and effect modification, analyses of MD-scores and breast cancer were also conducted within strata of age at baseline, smoking status, alcohol intake, BMI and physical activity and family history of breast cancer. Interactions with these factors were tested using Wald tests and cross-product terms. In sensitivity analyses, we repeated analyses after excluding cancers (and person-years) occurring in the first 2 years of follow-up, and we also split the follow-up period in 3 periods. Population attributable fractions were calculated26 to estimate the potentially avoidable proportion of cancer if all participants would shift towards the highest MD-score category. As a more realistic scenario, preventable proportions were also calculated to estimate the preventable proportion of cancer if all participants in the lowest 2 categories of MD-scores would shift their pattern 1 category upward.27, 28 The STATA-command “punafcc” was used to calculate the population attributable fractions and 95% CIs.29 To investigate possible dominance of certain components of the MD-scores,30 we ran analyses in which all components were entered as dichotomous variables simultaneously in Cox regression models. We then subtracted alternately one component at a time from the original 9-point sum score (thus reducing it to an 8-point score), and estimated HRs per 2-point increment in the reduced score (corrected by 8/9 before exponentiating them to preserve comparability), as in Trichopoulou et al.30 The MD-scores are relative measures, using cohort-specific medians as cut-offs. We compared the MD-score findings with a score that uses absolute cut-offs, based on the dietary part of the cancer prevention recommendations issued by World Cancer Research Fund/American Institute for Cancer Research (WCRF/AICR).1 We operationalized their dietary recommendations by using the same absolute cut-offs per recommendation (sometimes subrecommendations) as in EPIC,31, 32 using scores of 1 if the recommendation was met, 0.5 if half met, and 0 if not met. This concerned intake of energy-dense foods (≤125, >125–<175, ≥175 kcal/100 g per day) and sugary drinks (0, ≤250, >250 g/day); vegetables and fruit (≥400, 200–<400, <200 g/day), and dietary fiber (≥25, 12.5–<25, <12.5 g/day); red and processed meat (red& processed meat <500 g/week and processed meat <3 g/day; red& processed meat <500 g/wk and processed meat 3–<50 g/day; red& processed meat ≥500 g/wk and processed meat ≥50 g/day); and alcohol (<10, 10–20, >20 g/day). We additionally operationalized the WCRF/AICR-recommendation on salt intake by categorizing the calculated33 total salt intake (from food and salt added during cooking or consumption) into <6, 6–<9, 9+ g/day with scores 1, 0.5 and 0, respectively (based on Dutch dietary guidelines 201534). The resulting sum score (ranging from 0 to 5) was used in survival analyses; an additional sum score without alcohol was also made. The AIC was used to compare the fit of models with these sum scores to models with MD-scores. Meta-analyses Using PubMed with search terms Mediterranean diet, and breast cancer/neoplasm/tumor, or mammary carcinoma/tumors, cohort studies of the association between MD-adherence (a priori defined) and breast cancer were identified up to August 2016. Six articles on breast cancer (ER/PR subtypes) were identified.7, 35-39 Because Buckland et al.37 presented EPIC-wide results, the results from specific EPIC countries35, 39 were not included in the meta-analysis to avoid overlap. In addition to EPIC-results, the publication by Pot et al.39 also included results of Cade et al.36 Data on total postmenopausal breast cancer and subtypes of the remaining four cohorts (Nurses' Health Study (NHS), UK Women's Cohort Study (UKWCS), European Prospective Investigation into Cancer (EPIC), Women's Lifestyle and Health (WLH)) were combined with NLCS-data in the meta-analysis. HRs for the contrast between highest versus lowest category of MD-adherence from each study were pooled using random-effects models. In these analyses, the HR estimate for each study was weighted by the inverse of the variance of the log HR to calculate the summary HR and its 95% confidence interval (CI). Heterogeneity between studies was estimated using the Cochran's Q test and I2 (the proportion of variation in HRs attributable to heterogeneity40). Publication bias was assessed by the Begg test.41 Analyses were performed using Stata version 12; presented p-values are two-sided, with p < 0.05 considered as statistically significant. Results The mean (SD) score of aMEDr among subcohort members was 4.0 (1.6), and for mMEDr 4.0 (1.5). Table 1 summarizes several baseline characteristics according to adherence to aMEDr and mMEDr. Conformity with the Mediterranean diet was lower in older women, in nulliparous women, current smokers, in those with a positive family history of breast cancer (for aMEDr), and was higher in physically active women, higher educated women and ever oral contraceptive user. Alcohol intake was somewhat higher in those scoring higher on aMEDr, but this was reversed for mMEDr. Women with a high score on mMEDr more often reported a history of benign breast disease. Table 1. Baseline characteristics (mean (SD), or percentage) according to category of Mediterranean diet adherence (excluding alcohol) in subcohort women (in those with complete data on aMEDr and mMEDr), the Netherlands Cohort Study aMEDra mMEDrb Characteristic 0–3 pts 4–5 pts 6–8 pts 0–3 pts 4–5 pts 6–8 pts Number of subjects 769 901 357 730 981 316 Age (years) 61.7 (4.4) 61.3 (4.2) 60.8 (4.0) 61.7 (4.3) 61.3 (4.2) 61.1 (4.2) Height (cm) 165.0 (6.4) 165.6 (6.0) 165.4 (5.8) 165.5 (6.2) 165.1 (6.1) 165.4 (6.1) BMI (kg/mb) 25.1 (3.6) 25.1 (3.6) 24.7 (3.1) 24.9 (3.6) 25.1 (3.5) 24.9 (3.5) Physical activity (min/day) 62.0 (55.1) 63.8 (45.8) 72.6 (52.6) 59.7 (48.1) 66.5 (50.0) 70.3 (57.0) Age at menarche (year) 13.7 (1.8) 13.7 (1.7) 13.5 (1.7) 13.7 (1.7) 13.6 (1.8) 13.6 (1.8) Age at menopause (year) 48.6 (4.5) 48.7 (4.4) 49 (4.5) 48.6 (4.4) 48.9 (4.5) 48.7 (4.5) Alcohol intake (g/day) 5.1 (9.1) 6.3 (9.7) 6.5 (9.5) 6.2 (9.4) 5.9 (9.7) 5.4 (9.0) Energy intake (kcal/day) 1,696 (392) 1,676 (391) 1,698 (397) 1,709 (384) 1,682 (398) 1,654 (392) Nulliparous (%) 20.2 19.2 13.9 21.3 17.3 16.6 Age at first birth ≥ 30 years (% of parous) 22.7 22.3 22.6 22.2 22.3 23.8 Ever used OC (%) 23.2 23.9 33.5 24.7 25.2 27.3 Ever used HRT (%) 12.6 14.0 13.7 14.2 12.8 13.4 Family history breast ca (%) 10.0 8.3 7.3 8.5 9.0 8.9 History benign breast disease (%) 7.2 8.7 7.6 7.7 7.2 10.4 Current cigarette smoker (%) 26.1 19.8 14.6 25.1 20.1 16.1 University or higher vocational education (%) 6.9 10.7 12.1 9.3 8.8 12.3 Alcohol in range 5–25 g/day (%) 22.1 28.3 31.7 28.6 25.4 25.3 a aMEDr: alternate Mediterranean Diet Score excluding alcohol. b mMEDr: modified Mediterranean Diet Score excluding alcohol. Table 2 shows results of the age-adjusted and multivariable-adjusted analyses of the associations of MD-scores with total breast cancer risk. While the aMEDr-score was significantly inversely associated with breast cancer risk in age-adjusted analyses, in multivariable-adjusted continuous analyses, the HR per 2-point increment was 0.92 (95% CI, 0.84, 1.01). In multivariable-adjusted categorical analyses, only the medium category showed a significantly decreased risk with a HR of 0.82 (95% CI, 0.70, 0.96), compared to low adherence score, and there was no clear decreasing trend across categories (p-trend = 0.066). The AIC of the model using aMEDr was smaller compared to the mMEDr-model, indicating a better fit using aMED-scoring. For comparison, the table also shows analyses when using the full aMED and mMED including alcohol. The AIC-values indicated a worse fit for both aMED and mMED when alcohol was included in the scores (Table 2). Based on this, ensuing analyses were conducted primarily with aMEDr; at several places we also present results for mMEDr for reasons of comparison. Table 2. Hazard ratio of breast cancer, according to adherence to aMED and mMED, without and with alcohol, in multivariable-adjusteda analyses, the Netherlands Cohort Study MD-score without alcohol in score (aMEDr and mMEDr) MD-score with alcohol in score (aMED and mMED)b 0–3 pts 4–5 pts 6–8 pts p Trend Cont, per 2 pts 0–3 pts 4–5 pts 6–9 pts p Trend Cont, per 2 pts aMED No. of cases 928 987 406 789 996 536 Person-years in subcohort 10,438 13,016 5,478 9,163 12,599 7,170 Age-adjusted HR 1 0.85 0.83 0.023 0.91 1 0.92 0.87 0.094 0.94 (95% CI) (0.74–0.98) (0.69–1.00) (0.84–0.99) (0.79–1.06) (0.73–1.03) (0.87–1.01) Multivariable-adjusted HR 1 0.82 0.87 0.066 0.92 1 0.88 0.88 0.142 0.94 (95% CI) (0.70–0.96) (0.72–1.06) (0.84–1.01) (0.75–1.03) (0.73–1.06) (0.87–1.03) AIC 33,363 33,384 mMED No. of cases 877 1074 370 731 1077 513 Person-years in subcohort 9,871 14,224 4,837 8,198 14,242 6,491 Age-adjusted HR 1 0.85 0.86 0.056 0.94 1 0.85 0.89 0.143 0.96 (95% CI) (0.74–0.98) (0.71–1.05) (0.86–1.03) (0.73–0.98) (0.74–1.06) (0.88–1.04) Multivariable-adjusted HR 1 0.84 0.85 0.052 0.92 1 0.83 0.88 0.135 0.94 (95% CI) (0.72–0.98) (0.69–1.05) (0.84–1.01) (0.71–0.97) (0.73–1.07) (0.86–1.03) AIC 33,366 33,377 a Multivariable analyses were adjusted for: age at baseline (55–59, 60–64, 65–69 years), cigarette smoking (status (never, former, current), frequency (number of cigarettes per day; continuous, centered), duration (number of years; continuous, centered)), body height (continuous, cm), BMI (<18.5, 18.5–<25, 25–<30, ≥30 kg/m2), non-occupational physical activity (≤30, >30–60, >60–90, >90 min/day), highest level of education (primary school or lower vocational, secondary or medium vocational, and higher vocational or university), family history of breast cancer in mother or sisters (no, yes), history of benign breast disease (no, yes), age at menarche (< 12, 13–14, 15–16, >17 years), parity (nulliparous, 1–2, >3 children), age at first birth (< 25, >25 years), age at menopause (<45, 4,549, 50–54, >55 years), oral contraceptive use (never, ever), postmenopausal HRT (never, ever), energy intake (continuous, kcal/day) and alcohol intake (0, 0.1–<5, 5–<15, 15–30, >30 g/day). b Without additional adjustment for alcohol intake. Table 3 shows age- and multivariable-adjusted associations between aMEDr and risk of estrogen and progesterone receptor subtypes of breast cancer. There was a stronger inverse association with aMEDr for ER− breast cancer than for ER+ breast cancer, with HRs when comparing high versus low adherence of 0.60 (95% CI, 0.39, 0.93), p-trend = 0.032 for ER−, and 0.87 (95% CI, 0.69, 1.10), p-trend = 0.101 for ER+, respectively. The same pattern was seen for PR subtypes, albeit somewhat less strongly inverse in PR− than ER− subtype. Similarly, ER − PR− breast cancer was significantly inversely related to MD-adherence (p-trend= 0.047) with a HR per 2-point increment of 0.75 (95% CI, 0.60, 0.94), while the ER + PR+ subtype showed no significant association. Heterogeneity tests across subtypes using bootstrapping were not significant. The analyses in Table 3 were also conducted with mMEDr. When mMEDr was used, the HRs per 2-point increment were 0.95 (95% CI, 0.85–1.07) in ER+ breast cancer, 0.85 (0.71–1.03) for ER− breast cancer, 0.95 (0.83–1.08) in PR+, 0.90 (0.76–1.07) in PR−, 0.94 (0.83–1.08) in ER + PR+ and 0.79 (0.63–0.99) in ER−PR− breast cancer, i.e. all somewhat weaker associated than with aMEDr. Table 3. Hazard Ratio of breast cancer subtypes, according to adherence to Mediterranean diet (aMEDr) in multivariable-adjusteda analyses, the Netherlands Cohort Study aMEDr 0–3 pts 4–5 pts 6–8 pts p Trend AIC Cont, per 2 pts Total breast cancer No. of cases 928 987 406 Person-years in subcohort 10,438 13,016 5,478 Age-adjusted HR 1 0.85 0.83 0.023 0.91 (95% CI) (0.74–0.98) (0.69–1.00) (0.84–0.99) Multivariable-adjusted HR 1 0.82 0.87 0.066 33,363 0.92 (95% CI) (0.70–0.96) (0.72–1.06) (0.84–1.01) ER+ breast cancer No. of cases 460 466 195 Age-adjusted HR 1 0.82 0.80 0.022 0.89 (95% CI) (0.69–0.97) (0.64–1.00) (0.81–0.99) Multivariable-adjusted HR 1 0.81 0.87 0.101 16,119 0.91 (95% CI) (0.68–0.97) (0.69–1.10) (0.82–1.02) ER− breast cancer No. of cases 100 116 32 Age-adjusted HR 1 0.93 0.59 0.024 0.81 (95% CI) (0.69–1.24) (0.39–0.91) (0.69–0.97) Multivariable-adjusted HR 1 0.92 0.60 0.032 3,623 0.81 (95% CI) (0.67–1.25) (0.39–0.93) (0.67–0.96) PR+ breast cancer No. of cases 276 305 122 Age-adjusted HR 1 0.89 0.83 0.139 0.93 (95% CI) (0.73–1.09) (0.64–1.08) (0.82–1.04) Multivariable-adjusted HR 1 0.90 0.90 0.378 10,101 0.94 (95% CI) (0.73–1.11) (0.69–1.19) (0.83–1.07) PR− breast cancer No. of cases 158 157 60 Age-adjusted HR 1 0.79 0.69 0.017 0.80 (95% CI) (0.61–1.01) (0.48–0.96) (0.69–0.93) Multivariable-adjusted HR 1 0.76 0.72 0.047 5,422 0.81 (95% CI) (0.59–1.00) (0.52–1.05) (0.69–0.96) ER+PR+ breast cancer No. of cases 270 295 120 Age-adjusted HR 1 0.88 0.84 0.146 0.93 (95% CI) (0.72–1.08) (0.65–1.09) (0.83–1.04) Multivariable-adjusted HR 1 0.89 0.91 0.400 9,838 0.95 (95% CI) (0.71–1.10) (0.69–1.21) (0.83–1.08) ER−PR− breast cancer No. of cases 71 75 24 Age-adjusted HR 1 0.83 0.61 0.042 0.77 (95% CI) (0.59–1.18) (0.37–0.99) (0.63–0.95) Multivariable-adjusted HR 1 0.79 0.61 0.047 2,483 0.75 (95% CI) (0.55–1.14) (0.36–1.01) (0.60–0.94) a Multivariable analyses were adjusted for: age at baseline (55–59, 60–64, 65–69 years), cigarette smoking (status (never, former, current), frequency (number of cigarettes per day; continuous, centered), duration (number of years; continuous, centered)), body height (continuous, cm), BMI (<18.5, 18.5–<25, 25–<30, ≥30 kg/m2), non-occupational physical activity (≤30, >30–60, >60–90, >90 min/day), highest level of education (primary school or lower vocational, secondary or medium vocational and higher vocational or university), family history of