Alcohol consumption and breast cancer risk by estrogen receptor status: in a pooled analysis of 20 studies

Abstract

Introduction

Estrogen receptor negative (ER-) breast cancer is a less common but more aggressive breast cancer subtype than ER positive (ER+) breast cancer. ¹ Little is known about the aetiology of ER− breast cancer because most studies are limited by the relatively small number of ER− breast cancer cases. Although alcohol consumption is a convincing cause of breast cancer, ² the evidence from cohort studies for an association between alcohol consumption and ER− breast cancer risk is limited. A recent meta-analysis of 19 studies (3 cohort studies and 16 case-control studies) reported that the summary relative risk (RR) for ER− tumours, comparing the highest with lowest alcohol intake category, was 1.14 [95% confidence interval (CI): 1.03-1.26] for all studies and 1.21 (95% CI: 0.84-1.74) for three cohort studies. ³ Of six subsequently published cohort studies, ^4–9 only the Multiethnic Cohort Study showed a statistically significant positive association between alcohol consumption and ER− breast cancer risk. ⁸ Furthermore, alcohol-disrupted folate metabolism increases DNA instability and aberrant DNA methylation, which alters epigenetic regulation of gene expression; ^10,11 previously, two prospective cohort studies ^12,13 found that the increased risk of ER− breast cancer for drinkers was attenuated among those with high folate intake.

To address the limited number of ER− cases in most individual studies, we expanded our initial analyses ¹⁴ of alcohol consumption and overall breast cancer risk and evaluated the association between alcohol intake and risk of breast cancer defined by ER status. These analyses include 14 additional cohort studies and longer follow-up for studies in the original analyses, resulting in an additional 33 590 cases. We also examined whether associations with alcohol consumption varied by folate intake. In secondary analyses, we examined associations with tumours classified by progesterone receptor (PR) status.

Methods

Study population

This analysis includes 20 prospective cohort studies ( Table 1 ) within the Pooling Project of Prospective Studies of Diet and Cancer. ¹⁵ Each study met the following inclusion criteria: (i) at least one publication on a diet and cancer association; (ii) ascertainment of at least 25 incident ER− or PR− breast cancer cases; (iii) long-term diet assessment; (iv) validation of diet assessment; and (v) assessment of alcohol or dietary folate intake.

Case ascertainment

Incident invasive breast cancer cases were ascertained by self-report and medical record reviews, ^16–19 linkage to cancer registries, ^8,20–30 or both. ^4,31–33 Some studies additionally identified cases using linkage to mortality registries. ^{16,18,22,23,27,28,31} Hormone receptor status was obtained through cancer registries, ^{8,20,23,25,27–31,33,34} pathology reports, ^{16,18,24,26,32,34} medical records ^4,17,19 or laboratory determinations. ²² Cases with borderline hormone receptor status were considered positive for that hormone receptor.

Dietary and non-dietary factors assessment

Dietary intake at baseline was assessed using a food frequency questionnaire (FFQ) or diet history. Each study enquired as to the frequency of consumption of specific alcoholic beverages and the usual number and/or usual consumption of alcoholic drinks consumed on each occasion ( Table 1 ). We computed total alcohol consumption by summing alcohol intake as grams/day from each alcoholic beverage. All studies assessed folate intake from food (dietary folate). Total folate intake was calculated by summing the folate intake from diet, individual supplements and multivitamins among the 12 studies that assessed use of folate supplements and/or multivitamins. Total folate intake was not available in the Netherlands Cohort Study because multivitamins used in The Netherlands did not include folate when the study was initiated. Folate intake was adjusted for energy intake. ³⁵ The correlation coefficients between intakes estimated from the FFQ or a closely related instrument, and the reference method, generally exceeded 0.7 for alcohol consumption ¹⁵ and 0.4 for dietary folate intake across studies. ¹⁵

Studies collected non-dietary information including medical history, demographics, lifestyles and menopausal status at baseline. All studies had information on age, height and weight. Most studies had information on reproductive factors, education, physical activity, smoking status and family history of breast cancer.

Statistical analysis

Participants were excluded if they met the exclusion criteria specified within each study, were diagnosed with cancer (except non-melanoma skin cancer) prior to baseline, or reported energy intakes beyond three standard deviations from their study-specific log _e -transformed mean energy intake.

The Nurses’ Health Study was analysed as two separate cohorts [1980–86, NHS (a); 1986–2006, NHS (b)] to utilize the comprehensive dietary assessment administered in 1986. The Netherlands Cohort Study was analysed as a case-cohort study. ³⁶

Alcohol, dietary folate and total folate intakes were categorized using common intake cut-points and/or study-specific quintiles. We tested for trend from a Wald test using the median intake category as a continuous variable.

Study-specific rate ratios were calculated using Cox proportional hazards regression models separately for overall, ER−, ER+, PR− and PR+ breast cancer and tumours defined jointly by ER and PR status; ³⁷ cases without information on hormone receptor status or with a subtype different from the one being evaluated were censored at their diagnosis date. We calculated follow-up time from the baseline questionnaire return date to the date of incident breast cancer diagnosis, death, loss to follow-up or end of the follow-up, whichever came first. Age at baseline and calendar year of questionnaire return were used as stratification variables. In multivariable analyses, we included potential confounders assessed at baseline (see Table 2 ) in the model for studies with ≥ 200 cases of the outcome being evaluated. Otherwise, we used propensity scores to adjust for confounding. ^38–40 Missing indicators were created for missing responses for each covariate, if applicable.

The study-specific rate ratios were pooled using a random-effects model ⁴¹ weighted by the sum of the inverse of the variance and the estimated between-studies variance components. We examined between-studies heterogeneity using the Q statistic. ^41,42 Sources of heterogeneity by region and median follow-up time also were examined. The statistical significance of the differences in the associations by cancer subtype was assessed using the contrast test. ^{43, 44}

We tested whether the association with alcohol intake varied by folate intake, multivitamin use, menopausal status during follow-up (using a previously described algorithm ¹⁴ ), hormone-replacement therapy use, family history of breast cancer and smoking status by using a mixed-effects meta-regression model. ^15,45 All statistical tests were two sided.

Results

During 6–18 years of follow-up across 20 prospective cohort studies including 1 089 273 women, 37 191 women were diagnosed with breast cancer including 21 624 ER+ and 5113 ER− tumours ( Table 1 ). Across studies, 11–86% women were drinkers, among whom median alcohol consumption varied 8-fold. Median dietary folate intake varied more than 2-fold across studies. The prevalence of multivitamin use ranged from 3–69% across studies, with higher prevalence observed in the North American studies.

We present only the pooled multivariable RRs because the age-adjusted results were similar. Alcohol intake was significantly associated with higher risk of overall, ER+, ER−, PR+ and PR− breast cancer ( Table 2 ; Supplementary Figure 1 , available as Supplementary data at IJE online). The risk estimates for ER+ and ER− breast cancer were not statistically significantly different ( P_{common-effects} by ER status for ≥ 30 g/day category: 0.57). Comparing more extreme alcohol intakes (≥ 45 g/day vs. 0 g/day) yielded similar results [pooled multivariable RRs (95% CI): 1.57 (1.39-1.77) for ER+ and 1.35 (1.05-1.73) for ER− breast cancer; P_{common-effects} by ER status for ≥ 45 g/day category: 0.29]. Similarly, alcohol consumption was significantly positively associated with risk of PR+ and PR− breast cancer [ P_{common-effects} by PR status: 0.57 for ≥ 30 g/day category ( Table 2 ) and 0.40 for ≥ 45 g/day category; pooled multivariable RRs (95% CI): 1.48 (1.29-1.70) for PR+ and 1.36 (1.11-1.67) for PR−breast cancer]. Statistically significant heterogeneity was observed for the study-specific risk estimates for PR+ breast cancer ( P_{between-studies heterogeneity} for ≥ 30 g/day category: 0.01). A potential source of heterogeneity was geographical region; we observed a stronger association in North American studies [pooled multivariable RR (95% CI) comparing ≥ 30 g/day with 0 g/day: 1.43 (1.28-1.59); N of cases: 785] compared with studies in other continents [pooled multivariable RR (95% CI): 1.07 (0.64-1.79); N of cases: 54; P_difference by region: 0.02). Excluding cases diagnosed within the first 5 years of enrolment to reduce potential lifestyle changes due to prediagnostic symptoms did not alter the results materially for any of the outcomes evaluated (data not shown).

When we categorized alcohol consumption by 5 g/day increments in an aggregated dataset of all studies, a positive linear trend was observed for alcohol intakes up to 55 g/day, beyond which further increases in risk were not observed but < 1% of women reported alcohol intakes > 55 g/day; Figure 1 ). Therefore, when modeling alcohol consumption as a continuous variable, we restricted analyses to women consuming alcohol less than 55 g/day. The pooled multivariable RRs (95% CI) for alcohol consumption per 10 g/day (approximately 0.75–1 drink/day) were 1.08 (1.07-1.09) for overall breast cancer [ P_{between-studies heterogeneity} : 0.47]; 1.10 (1.08-1.12) for ER+ breast cancer; and 1.06 (1.02-1.10) for ER− breast cancer ( P_{common-effects} by ER status: 0.08); 1.10 (1.08-1.12) for PR+ breast cancer; and 1.07 (1.04-1.10) for PR− breast cancer ( P_{common-effects} by PR status: 0.07).

Figure 1.

Multivariable relative risks (95% CIs) for 5 g/d categories of alcohol consumption and total breast cancer, compared with non-drinkers; multivariable model included the covariates listed in footnote ‘a’ in Table 2 . The proportion of the study population in each category was 38% for non-drinkers, 35% for > 0–< 5 g/d, 10% for 5–< 10 g/d, 7% for 10–< 15 g/d, 4% for 15–< 20 g/d, 2% for 20–< 25 g/d, 1% for 25–< 30 g/d, 1% for 30–< 35 g/d, 1% for 35–< 40 g/d, 0.4% for 40–< 45 g/d, 0.4% for 45–< 50 g/d, 0.2% for 50–< 55 g/d, 0.1% for 55–< 60 g/d, 0.1% for 60–< 65 g /d, 0.1% for 65–< 70 g/d and 0.4% for ≥ 70 g/d. ^* RR and 95% confidence interval for the ≥ 70 g/d alcohol intake category vs non-drinkers.

Open in new tabDownload slide

Alcohol consumption was positively associated with risk of breast cancer defined jointly by ER and PR status. However, only associations for ER+PR+, ER+PR-, and ER−PR- were statistically significant; risk estimates were similar across subtypes ( Supplementary Table 1 , available as Supplementary data at IJE online). Analysing alcohol intake as a continuous variable yielded similar results ( P_{common-effects} by receptor status: > 0.26; data not shown).

Alcohol intakes from beer, wine and liquor were each significantly associated with 23–33% higher breast cancer risk comparing ≥ 15 g/day alcohol consumption from a specific alcohol beverage vs non-drinkers of that beverage. Results did not differ significantly by ER or PR status ( P_{common-effects} by ER or PR status for ≥ 15 g/day category: ≥ 0.51) ( Supplementary Table 2 , available as Supplementary data at IJE online). The results were similar when alcohol intakes from beer, wine and liquor, modelled as continuous variables, were simultaneously included in the multivariable analyses.

The positive association between alcohol consumption and overall breast cancer risk was not modified by total folate intake, multivitamin use, family history of breast cancer, postmenopausal hormone therapy use or smoking status ( Table 3 ). However, there was a modestly stronger positive association for postmenopausal women compared with premenopausal women ( P_interaction : 0.04). Associations of alcohol consumption with risk of ER+ and ER− breast cancer were not significantly modified by total folate intake, multivitamin use, family history of breast cancer, smoking status or menopausal status ( P_interaction : ≥ 0.05; Supplementary Table 3 , available as Supplementary data at IJE online). Although there was a stronger positive association for alcohol intake and risk of ER− breast cancer among current postmenopausal hormone therapy users than among never and past users ( P_interaction : 0.01), postmenopausal hormone therapy use did not modify the association between alcohol consumption and risk of ER+ breast cancer ( P_interaction : 0.76). Similar positive associations between alcohol consumption and risk of PR+ and PR− breast cancer were observed across strata of the breast cancer risk factors mentioned above ( P_interaction : ≥ 0.17; data not shown), with one exception; the association between alcohol intake and risk of PR+ breast cancer was stronger among women with a family history of breast cancer compared with those without a family history of breast cancer ( P_interaction : 0.01).

Dietary and total folate intakes were not associated with risk of total, ER+, ER−, PR+, PR− ( Table 4 ), ER+PR+, ER+PR-, ER−PR- or ER−PR+ breast cancer ( Supplementary Table 4 , available as Supplementary data at IJE online). Risk estimates comparing the extreme quintiles of dietary or total folate intake ranged from 0.98 to 1.03 for overall breast cancer and breast cancer subtypes defined by ER and/or PR status ( P_{common-effects} for highest quintile: ≥ 0.32) ( Table 4 ). Restricting analyses of dietary folate intake to participants who had not used supplements containing folic acid or to studies that assessed supplemental folate intake did not change results materially. Separate analyses for studies in North America or for North American studies containing cases diagnosed before mandatory folate fortification of grain products in the USA were almost identical to the main results (data not shown). Analysing dietary and total folate intake as categorical variables using study-specific deciles or common intake cut-points did not substantially change the results for all breast cancer outcomes (data not shown). Excluding cases diagnosed during the first 5 years of follow-up did not alter the results for dietary or total folate intake (data not shown).

Discussion

In this large pooled analysis, total alcohol consumption and alcohol intake from beer, wine and liquor were each positively associated with risk of overall, ER+, ER−, PR+ and PR− breast cancer. Associations for breast tumours were not significantly different regardless of ER or PR status. These associations were not modified by total folate intake, multivitamin use, family history of breast cancer, postmenopausal hormone therapy use or smoking status except that a stronger positive association was observed for postmenopausal women compared with premenopausal women. Dietary and total folate intakes were not associated with total or hormone-receptor-defined breast cancer risk.

Our result of a positive association between alcohol consumption and total breast cancer risk is consistent with our previous pooled analyses from six cohort studies ¹⁴ and a meta-analysis of 39 cohort studies and 74 case-control studies. ⁴⁶ Our analyses included four ^7,48–50 of the six prospective cohort studies ^{7,6,47,48–50} that had previously examined associations for specific alcoholic beverages with breast cancer risk. Similar to our results, the two studies ^6,47 not included in our analyses reported positive associations between wine, beer and liquor and breast cancer risk.

For breast tumours defined by hormone receptor status, we found associations of similar magnitude for ER+ and ER− breast tumours. However, a meta-analysis of three cohort studies (all are included in our analyses) and 16 case-control studies found a slightly stronger association for ER+ [summary RR: 27% (95% CI 17–38%)] compared with ER− [summary RR: 14% (95% CI 3–26%)] breast cancer, comparing the highest with lowest alcohol consumption categories. ³ Six subsequent prospective cohort studies have examined associations between alcohol consumption and breast cancer risk defined by hormone receptor status. ^4–9 Of these, only two ^5,6 were not included in our analyses; both of these studies showed statistically significant positive associations with ER+ breast cancer risk ^5,6 but non-significant inverse associations for ER− ⁵ or ER−PR− ⁶ breast tumours. Our study included all four prospective cohort studies ^5,7,9,12 that had previously examined associations between alcohol consumption and breast cancer risk by PR status.

Positive associations between alcohol consumption and risk of both ER− and ER+ breast cancer are biologically plausible. Alcohol may operate through both hormone-independent and hormone-dependent aetiological pathways. Alcohol increases mammary susceptibility to DNA damage by producing acetaldehyde, which increases oxidative stress and causes epigenetic alterations inhibiting the absorption of methyl donors. ⁵¹ In addition, alcohol stimulates cell proliferation by elevating endogenous estrogen levels ^52,53 and promotes breast carcinogenesis.

We observed no change in the alcohol-breast cancer association with high folate consumption. Some, ^54–61 but not all, ^{4,7,50,62–67} studies have found that the positive association between alcohol consumption and breast cancer risk was attenuated among women with high folate intake. Only three studies examined whether folate intake modified associations of alcohol consumption with ER− and/or ER+ breast cancer risk, ^12,13,68 all of which were included in our analyses. Similarly, two studies that examined the interaction between folate and alcohol intake and breast cancer risk defined by PR status were included in our analyses. ^12,13 Our observation of no interaction between alcohol and folate intake and breast cancer risk requires cautious interpretation. We could not take into account diet changes over time with baseline dietary intake data; given that the US studies in our analyses started before the mandatory folate fortification of grain products, potential misclassification of folate intake during follow-up might have attenuated the association. ⁶⁹ Future study with time-integrated measures of endogenous folate level is warranted.

We observed a stronger positive association between alcohol consumption and total breast cancer risk for postmenopausal women than for premenopausal women. Our result is qualitatively similar to our previous analysis ¹⁴ and the most recent Nurses’ Health Study analysis, with cumulatively updated alcohol intake data over 28 years. ⁴⁹ The underlying mechanism of this finding is unclear, although associations of breast cancer risk with body mass index ² and polymorphisms in genes encoding the estrogen receptor and some oxidative enzymes have been found to differ by menopausal status. ^70,71 Alcohol consumption may have a more pronounced effect on breast tissue after menopause. Alternatively, our result may be due to chance.

Our study has several strengths. The large number of breast cancer cases allowed us a detailed investigation of associations between alcohol and specific alcoholic beverages and folate intakes and risk of overall and hormone-receptor-defined breast cancer subtypes, and of the interaction between alcohol and folate intake. Median alcohol and folate intake varied more than 2-fold across the studies, minimizing the likelihood of missing an association due to homogeneous dietary habits. Standardizing categorizations for nutrient and covariate data and using similar analytical models across studies reduced heterogeneity among studies. Further, the prospective design of individual studies minimized recall bias.

Our study had limitations. Baseline alcohol or folate intake data may not incorporate possible diet changes during follow-up attenuating associations. However, a US survey found a correlation of 0.63 for alcohol intake collected 10 years apart, suggesting that alcohol consumption at baseline may reasonably represent long-term intake. ⁷² Baseline folate intake does not take into account mandatory folate fortification of grain products beginning in 1997 in North America; ⁶⁹ nonetheless, the pooled results from the North American studies for follow-up occurring before folate fortification were similar to those for follow-up occurring after fortification. We could not examine associations at young ages or with the frequency, pattern and duration of alcohol consumption or very high alcohol intakes. Hormone receptor status information was missing for 3–56% of cases across studies, but associations for overall breast cancer were similar for cases with and without hormone receptor data. We could not investigate other tumour subtypes, including HER2 and Ki67, ⁷³ because of a lack of information on these tumor characteristics. Finally, we could not distinguish breast cancers detected by symptoms from those diagnosed by mammography. As tumors detected by mammography, particularly ER+ tumors, ⁷⁴ are more likely to be indolent, ⁷⁵ our results for ER+ breast cancer may have been attenuated.

In summary, in this large pooled analysis, alcohol consumption was positively associated with risk of breast cancer overall and of subtypes defined by ER and/or PR status. Dietary and total folate intakes were not associated with risk of overall breast cancer or breast cancer subtypes defined by ER and/or PR status. Higher total folate intake did not modify the positive associations between alcohol consumption and risk of overall or hormone-receptor-defined breast cancer.

Funding

This work was supported by a National Institute of Health grant CA55075 and a Breast Cancer Research Foundation grant.

Acknowledgements

We thank Shiaw-Shyuan Yaun, Tao Hou and Ruifeng Li for their invaluable contributions to data management and statistical support. We are grateful to the participants and personnel who contributed to each study that was included in the Pooling Project.

All authors contributed to these pooled analyses and read and approved the final manuscript.

Conflict of interest: None declared.

References