Alcohol Use and Breast Cancer Survival among Participants in the Women's Health Initiative

Alcohol intake, even at low levels, is an established risk factor for breast cancer (1, 2), possibly due to its effects on pathways related to estrogen metabolism and/or DNA damage and repair (3–6). It is not clear, however, whether alcohol intake influences breast cancer mortality, and results from prior studies are mixed. Existing data have been summarized in a recent report that included both a meta-analysis and three pooled analyses (7). The meta-analysis of 11 published studies (n = 29,239 cases) reported 20% better overall survival associated with prediagnosis alcohol consumption [95% confidence interval (CI), 0.73–0.88] and no association with postdiagnosis alcohol intake (HR, 0.95; 95% CI, 0.85–1.05); however, it did not examine breast cancer–specific mortality or if risk varied according to estrogen receptor (ER) status. The pooled analyses (n = between 8,000 and 11,000 cases each) were stratified by ER status, and among women with ER⁺ disease, alcohol intake (before or after diagnosis) was not associated with breast cancer–specific survival, but there was some evidence for an association between prediagnosis intake and improved overall survival. For ER⁻ disease, there was no association with prediagnosis intake, but some evidence of improved survival (overall and breast cancer–specific) associated with postdiagnosis intake (7). However, several potential limitations related to variations in the compositions of the study populations included, and in some cases the length of follow-up, may have influenced these findings (see Discussion section). Only one prior study investigated the possible relationship between a change in alcohol consumption before and after breast cancer diagnosis, and observed that an increase of one or more drinks per week was associated with a 24% lower risk of death due to any cause (95% CI, 0.60–0.97) compared with women who never drank; decreasing alcohol consumption did not affect risk (8).

Given that alcohol consumption is a common exposure and that the population of breast cancer survivors continues to grow as survival rates have continued to improve, further characterization of alcohol's potential associations with breast cancer mortality is needed. Understanding this relationship may have implications for improving survival in women diagnosed with breast cancer, as alcohol intake is a potentially modifiable factor. In this study, we investigated whether various aspects of alcohol consumption are associated with mortality (overall and breast cancer–specific) in a prospective cohort of breast cancer patients who participated in the Women's Health Initiative (WHI).

Details of the design and execution of the WHI have been described previously (9–11). Briefly, 161,808 postmenopausal women ages 50 to 79 years were enrolled from 40 clinical centers around the United States, including 93,676 women in the observational study (OS) and 68,132 women in the clinical trial (CT). CT enrollees were included in one, two, or all three of the following randomized trials: hormone therapy, dietary modification, or supplementation with both calcium and vitamin D. Women were recruited by a population-based mail campaign plus media awareness programs. They were eligible if, at baseline screening, they were between the ages of 50 and 79 years, postmenopausal, and likely to remain in the same geographic region for at least 3 years. They were excluded for conditions likely to present challenges in terms of study retention, adherence, or survival (for example, cancer, alcoholism, dementia, or transportation challenges).

The WHI collected data on a range of exposures, including alcohol use and other dietary factors, hormone therapy and other medications, supplement use, and other demographic and reproductive factors. Data used in this study were collected similarly for OS and CT participants. At an initial clinic visit, all participants completed a self-administered questionnaire, including questions on alcohol use, and underwent a physical exam. During follow-up, participants completed health questionnaires every 6 months and also attended annual clinic visits, which included standardized interviews to capture data on health conditions and symptoms, annual mammograms, and clinical breast exams. Follow-up also included self-reported alcohol use at years 1, 3, 6, and 9 for CT participants, and year 3 for OS participants. Prediagnosis alcohol intake was defined based on self-report of the number of servings of alcohol consumed per week using the closest available data prior to diagnosis, and was available for all women in these analyses. To define postdiagnosis alcohol intake, the closest data available following diagnosis were used (such data were available for a subset of the sample). Change in alcohol intake was calculated based on pre- and postdiagnosis intake, and was defined as no change (same number of drinks before and after diagnosis), increase (higher alcohol intake after diagnosis, by at least 1 drink/week), or decrease (lower intake after diagnosis, by at least 1 drink/week). The average time elapsed between assessment of pre- and postdiagnosis intake was 1,270 days (SD, 528 days; range, 365–3,287 days).

The average follow-up time of study participants was 7.9 years. Women were censored at their date of last follow-up. The present analyses (n = 7,835) include all subjects who were diagnosed with breast cancer after enrollment (and after randomization, for CT subjects; n = 8,819), except that women with incomplete alcohol and/or covariate data were excluded (n = 1,034).

Cox proportional hazards regression was used to estimate HR and 95% CIs for mortality (all-cause and breast cancer–specific) associated with alcohol intake, among women diagnosed with breast cancer. Analyses were adjusted for confounders which were selected a priori, including age, income, CT enrollment, family history of breast cancer, smoking status, menopausal hormone therapy at baseline, and body mass index (BMI). Analyses were conducted for prediagnosis intake, postdiagnosis intake, and changes in intake after diagnosis. We explored whether associations differed by ER status (positive or negative) of the disease at diagnosis.

We assessed (and ruled out) potential effect modification by CT/OS membership (P for interaction >0.05, data not shown). We also conducted a sensitivity analysis where models were refit with BMI as a time-varying confounder, in consideration of estrogen-related mechanisms between alcohol and obesity. However, results did not change appreciably and so the time-varying version of this variable was not included in the final model. We conducted secondary analyses limited to just those women for whom data on all 3 exposures of interest (before diagnosis, after diagnosis, and change in alcohol consumption) were available (n = 20,633 person-years, 352 deaths). The results, including strength and direction of associations, were all similar, though were no longer statistically significant for the association between prediagnosis alcohol intake and all-cause mortality (likely due to the smaller sample size; not shown). We also conducted analyses using an alternate cutoff of ≥9 drinks/week, instead of ≥7, for the highest level of alcohol intake, in order to more directly compare our results with those reported in with the meta-analysis by Ali and colleagues (7); results and conclusions were the same for all comparisons regardless of cutoff level (data not shown). We additionally conducted an analysis excluding women with stage IV cancers (n = 81, or 1% of study population); results were essentially identical and did not affect our conclusions (data not shown). The statistical programs SAS 9.3 and R 3.0.1 were used for all analyses.

Similarities in distributions of age, first-degree family history of breast cancer, and parity were observed across prediagnosis alcohol use categories (Table 1). Women who were alcohol consumers before diagnosis had somewhat higher household incomes and were somewhat more likely to be past or current smokers, current users of estrogen+progestin menopausal hormones, non-Hispanic whites, and college graduates compared with never and past alcohol consumers (Table 1). Women from the OS study were overall similar to those in the CT study (not shown). OS participants were somewhat more likely to be ≥60 years old (70% vs. 66%), to be white (89% vs. 86%), to have an income ≥$75,000 (22% vs. 18%), to have ever smoked (52% vs. 50%), to have a lower BMI (median 26 vs. 28 kg/m²) to be nulliparous (15% vs. 12%), and to have ever used estrogen only (36% vs. 33%) or estrogen+progestin (36% vs. 29%) compared with CT participants. OS participants were also more likely to have data on postdiagnosis alcohol intake (26% vs. 22%), and they had slightly shorter average times to diagnosis (5.4 vs. 5.9 years), times between prediagnosis alcohol assessment and diagnosis (3.4 vs. 4.3 years), and times between diagnosis and postdiagnosis alcohol assessment (1.4 vs. 2.0 years). Among cases who died, OS participants had a slightly longer time from postdiagnosis alcohol assessment to death (average 9.4 vs. 8.5 years). Stage at diagnosis and disease histology were very similar for OS and CT participants (not shown).

Adjusting for age, income, study (CT vs. OS), family history of breast cancer, smoking history, BMI, and history of menopausal hormone therapy, there were similar 24% (95% CI, 0.56–1.04) and 26% (95% CI, 0.61–0.89) reduced risks of breast cancer–specific and all-cause mortality, respectively, among women who consumed alcohol prior to diagnosis, though the former risk estimate was within the limits of chance (Table 2). However, neither alcohol consumption after diagnosis nor changes in alcohol consumption before versus after diagnosis were related to either breast cancer–specific or all-cause mortality. In analyses stratified by breast cancer subtype, reduced risks of both breast cancer–specific and all-cause mortality were observed for ER⁻ (HR = 0.51; 95% CI, 0.31–0.83 and HR = 0.49; 95% CI, 0.34–0.72, respectively), but not for ER⁺ disease (HR = 0.97; 95% CI, 0.31–1.54 and HR = 0.81; 95% CI, 0.63–1.04, respectively; Table 3). However, neither of the interaction terms assessing heterogeneity by ER status were statistically significant (P = 0.30 and 0.09 for breast cancer–specific and all-cause mortality, respectively).

We investigated three aspects of alcohol use (prediagnosis and postdiagnosis alcohol intake, and change in drinking habits after diagnosis) in relation to mortality after breast cancer diagnosis. Overall, there was some suggestion that women who were consuming alcohol prior to their breast cancer diagnosis had a reduced risk of mortality that appeared to be primarily limited to patients with ER⁻ disease. However, there were no associations between alcohol intake after diagnosis or changes in alcohol consumption patterns and mortality. These findings are relatively consistent with most previous studies, which have generally reported either null or slightly protective associations between alcohol intake and mortality among women with breast cancer. However, past studies have investigated varying aspects of alcohol intake (before vs. after diagnosis; categories of alcohol quantity consumed), on different outcomes (breast cancer–specific vs. overall), and in different populations (varying ages and levels of alcohol consumption), which complicates comparison of findings. A recent meta-analysis of 11 studies investigating only “moderate” intake (defined as ≥14 units, or approximately 9 drinks, per week) reported 20% better overall survival associated with prediagnosis intake (95% CI, 0.7–3–0.88; based on 6 studies), and no association with postdiagnosis intake (HR, 0.95; 95% CI, 0.85–1.05; based on 5 studies). These analyses were not stratified by ER status, and authors did not perform a meta-analysis for breast cancer–specific survival, citing heterogeneity across studies.

This publication also included pooled analyses from three separate cohorts, stratifying by ER status, and assessing both overall and breast cancer–specific survival. Of the two analyses of prediagnosis alcohol intake, one found no association with either all-cause or breast cancer–specific survival, overall or when stratified by ER status, though the follow-up period was as brief as 2 years for some cases (7). In the other, prediagnosis intake was associated with 36% to 44% better overall survival, in women with ER⁺ disease only (95% CIs all within 0.47–0.88; ref. 7), which differs from our findings. However, the original goal of this larger international pooled cohort was to investigate genetic causes of breast cancer, and several of its component studies enrolled only younger cases, making it less comparable with our study (overall average age, 54 years, vs. 63 years in our study). It also had higher proportions of ER⁻ cases (20% vs. 13%) and of non-drinkers (34% vs. 9%). There was also greater potential for uncontrolled confounding based on disproportionate representation of different countries in the referent (nondrinker) group compared with the exposure groups (e.g., 66% of study participants from Japan were nondrinkers, vs. <1% of those from Finland). This could lead to confounding by country-level differences (e.g., in breast cancer treatment and prognosis) and/or genetic differences (in disease subtype and course; in alcohol metabolism). Such differences could also potentially explain discrepancies in findings between our study and those analyses.

The third pooled analysis investigated postdiagnosis alcohol intake and observed 46% and 49% lower all-cause mortality in ER⁻ cases who consumed moderate (>7–≤14 units/week) and high levels of alcohol (>14 units/week) after diagnosis (95% CIs, 0.33–0.78 and 0.30–0.97, respectively), and 42% better breast cancer–specific survival in ER⁻ moderate drinkers (95% CI, 0.37–0.92) compared with nondrinkers (7). However, that cohort was also from a study designed to assess genetic risk factors for breast cancer, and the study population had a greater proportion of younger cases (45% were ≤50 years at diagnosis), as well as cases with longer survival (median follow-up time: 7 years; 5-year survival: 87%), limiting its comparability.

Another more recent study, not included in the meta-analysis, reported a 15% to 20% lower all-cause mortality associated with moderate prediagnosis intake (HR, 0.80; 95% CI, 0.74–0.86 for 3 to 6 drinks per week; HR, 0.85; 95% CI, 0.77–0.93 for 7 to 9 drinks per week), but no association with higher levels of intake (8). That study also reported 15% lower breast cancer–specific mortality in women who drank 3 to 6 drinks per week before diagnosis (95% CI, 0.75–0.95), and weaker, borderline associations at higher and lower levels of intake (i.e., only 5% to 10% better survival, with 95% CIs just barely including the null) compared with nondrinkers; ER⁻status-specific estimates were not provided. That study also assessed postdiagnosis alcohol consumption, and unlike ours, observed 23% to 37% lower all-cause mortality for consumption of ≥3 drinks/week compared with nondrinkers (95% CIs all within 0.45–0.98), though study subjects had much longer survival time (postdiagnosis alcohol intake was assessed a median of 6 years after diagnosis, vs. an average of 1.7 years in our study; ref. 8). Perhaps any protective effect of moderate postdiagnosis alcohol on overall survival is limited to women with less advanced or less aggressive disease. Alcohol has been shown to have a protective effect on cardiovascular disease (CVD) mortality (12), but presumably this effect is not immediate. Thus, we might expect that an effect of postdiagnosis alcohol intake on all-cause mortality would only be apparent among women with longer survival.

To our knowledge, only one other study has reported on the association with a change in drinking habits after diagnosis: A 24% lower overall survival was observed in women who reported drinking ≥1 additional drink per week after diagnosis than before diagnosis (95% CI, 0.60–0.97), compared with women who never drank. This is somewhat similar in magnitude to our observation of a 38% lower mortality, though our finding was not statistically significant (95% CI, 0.33–1.16; ref. 8). No association was reported for a decrease in alcohol, or for any change in alcohol and breast cancer–specific mortality. It is unclear why an increase in drinking might be associated with longer survival, though moderate intake has a protective effect on CVD mortality (12, 13), and the same study noted that an increase, but not a decrease, in alcohol intake was associated with a lower risk of CVD mortality, compared with never-drinkers (8). Alternatively, there could be confounding by disease severity and/or overall health and well-being.

It is also not clear how alcohol might influence mortality, among women diagnosed with ER⁻ but not ER⁺ breast cancer. The protective effect of alcohol on CVD mortality would predict better survival among ER⁺ and ER⁻ cases alike. However, among women diagnosed with ER⁺ disease, the adverse effects of alcohol on estrogen levels (and thus, perhaps, on disease progression and recurrence) may offset this benefit. Alcohol is thought to act on estrogen pathways, increasing serum levels of estrogen, disrupting estrogen metabolism (4), and enhancing expression of ERs in breast cells (14). These factors could conceivably affect breast-cancer progression and/or recurrence of ER⁺, but not ER⁻, disease.

This study has several strengths. The WHI study sample is a broad population from 40 clinics across the United States, making it more generalizable than many previous studies of this question. This study is one of only two which were able to measure alcohol intake both before and after diagnosis in same population, allowing for assessment of the potential association of a change in drinking habits and breast cancer survival. Other strengths include prospective study design, use of incident cases, postdiagnosis follow-up, and high levels of retention. Highly detailed data were available on alcohol use, allowing for categorization by level of intake, and data on prediagnosis alcohol use were collected before diagnosis, unlike many studies, leading to greater accuracy. We were able to consider time-varying variables in our analyses due to the availability of repeated follow-up measures in BMI.

A key limitation of this study is that data on postdiagnosis alcohol use were only available for 25% of the population. However, given the size of the study, these analyses still included 1,928 women. This subset is less likely to include women with the shortest survival; a mean time of 1.7 years elapsed between breast cancer diagnosis and postdiagnosis alcohol ascertainment, and so our findings on postdiagnosis alcohol consumption may be less generalizable to women with particularly advanced or aggressive disease. However, as noted, this was much more inclusive than some other recent studies, and a study which addressed this question by excluding early events in some analyses (deaths within 2 years of diagnosis) observed no difference in results (15). The potential for confounding by unmeasured factors is another concern; for example, overall fitness could influence both pre- and postdiagnosis alcohol intake, as well as survival, but no fitness data were available. Disease severity could also influence both postdiagnosis alcohol intake and survival. Treatment could affect alcohol consumption, and the association of alcohol with survival could be dependent on its timing in relation to treatment.

Our findings, together with those of earlier studies, suggest that alcohol intake among women diagnosed with breast cancer has little to no impact on either breast cancer–specific or all-cause mortality. From a public health perspective, the evidence does not suggest that modification of alcohol intake after a breast cancer diagnosis is warranted, nor does it warrant concerns about prediagnosis intake in relation to breast cancer survival. If future studies provide stronger evidence for the observed suggestion of a small protective association of moderate alcohol intake with mortality in women with ER⁻ breast cancer, then future etiologic studies into the biologic role of alcohol on mortality, survival, recurrence, and disease progression could potentially shed more light on the reason for this. However, research into exposures other than alcohol that are hypothesized to affect breast cancer survival may be of greater relevance for understanding and ultimately improving breast cancer survival.

R. Chlebowski has received honoraria from the speakers bureau of Pfizer and Novartis. He is also a consultant/advisory board member for Genentech, Genomic Health, Novartis, and Pfizer. No potential conflicts of interest were disclosed by the other authors.

Conception and design: C.I. Li

Development of methodology: C.I. Li

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): R. Chlebowski

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S.J. Lowry, K. Kapphahn, R. Chlebowski, C.I. Li

Writing, review, and/or revision of the manuscript: S.J. Lowry, K. Kapphahn, R. Chlebowski, C.I. Li

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): K. Kapphahn, C.I. Li

Study supervision: C.I. Li

The authors acknowledge the WHI investigators who were responsible for the design, oversight, and execution of WHI including: Program Office: (National Heart, Lung, and Blood Institute, Bethesda, Maryland) Jacques Rossouw, Shari Ludlam, Dale Burwen, Joan McGowan, Leslie Ford, and Nancy Geller. Clinical Coordinating Center: (Fred Hutchinson Cancer Research Center, Seattle, WA) Garnet Anderson, Ross Prentice, Andrea LaCroix, and Charles Kooperberg. Investigators and Academic Centers: (Brigham and Women's Hospital, Harvard Medical School, Boston, MA) JoAnn E. Manson; (MedStar Health Research Institute/Howard University, Washington, DC) Barbara V. Howard; (Stanford Prevention Research Center, Stanford, CA) Marcia L. Stefanick; (The Ohio State University, Columbus, OH) Rebecca Jackson; (University of Arizona, Tucson/Phoenix, AZ) Cynthia A. Thomson; (University at Buffalo, Buffalo, NY) Jean Wactawski-Wende; (University of Florida, Gainesville/Jacksonville, FL) Marian Limacher; (University of Iowa, Iowa City/Davenport, IA) Robert Wallace; (University of Pittsburgh, Pittsburgh, PA) Lewis Kuller; (Wake Forest University School of Medicine, Winston-Salem, NC) Sally Shumaker. Women's Health Initiative Memory Study: (Wake Forest University School of Medicine, Winston-Salem, NC) Sally Shumaker. For a list of all the investigators who have contributed to WHI science, please visit: https://www.whi.org/researchers/Documents%20%20Write%20a%20Paper/WHI%20Investigator%20Long%20List.pdf.

The WHI program is funded by the National Heart, Lung, and Blood Institute, NIH, U.S. Department of Health and Human Services through contracts HHSN268201100046C, HHSN268201100001C, HHSN268201100002C, HHSN268201100003C, HHSN268201100004C, and HHSN271201100004C (to R. Chlebowski, coinvestigator). S.J. Lowry and this publication were also supported by grant number T32CA09168 from the NCI.