Postdiagnosis Alcohol Consumption and Breast Cancer Prognosis in the After Breast Cancer Pooling Project

This article is featured in Highlights of This Issue, p. 1

In 2011, the estimated number of breast cancer survivors in the US was over 2.6 million, and the population continues to grow (1, 2). Many women living with breast cancer are keenly interested in how they can improve their prognosis and survival by making lifestyle changes after their diagnosis. One factor of particular interest is alcohol consumption, which is generally recognized to increase the risk of breast cancer (3–5). However, inconclusive results have emerged from the limited number of studies that have examined alcohol intake (primarily before diagnosis) and breast cancer-related outcomes.

Previous studies have reported increased (6–9) and decreased risks of death (10–13) with pre- and postdiagnosis alcohol consumption, as well as no association (14–24). Fewer studies have examined the association of alcohol on breast cancer recurrence, with mixed results (6, 13, 25, 26).

Overall, studies of alcohol and breast cancer prognosis have suffered from methodologic challenges, namely, varied endpoints, narrow exposure (drinking) range, prediagnosis rather than postdiagnosis assessment of intake, and failure to adjust for prognostic factors (27, 28). Alcohol might influence breast cancer risk and prognosis by increasing estrogen metabolism and endogenous estrogen levels, particularly in postmenopausal women (29–33). However, epidemiologic studies to date have not had adequate sample size to conduct subgroup analyses by estrogenic factors.

To address these limitations, we conducted one of the largest pooled investigations of postdiagnosis alcohol consumption and breast cancer recurrence and total mortality using data from 9,329 US breast cancer survivors in the After Breast Cancer Pooling Project (ABCPP). We also examined potential effect modification by a priori selected estrogenic factors including menopausal status, tumor estrogen receptor (ER) status, and obesity.

The ABCPP is an international collaboration of pooled data from 4 prospective studies of breast cancer survivors to examine the roles of physical activity, adiposity, dietary factors, supplement use, and quality of life in breast cancer prognosis. Details about cohort creation and characteristics have been published previously (34). Briefly, the ABCPP includes 18,333 breast cancer survivors diagnosed with stage I-IV invasive breast cancer from 4 population-based prospective cohort studies recruited from the US and Shanghai, China: the Shanghai Breast Cancer Survival Study (SBCSS; ref. 35), the Life after Cancer Epidemiology (LACE) Study (36), the Women's Healthy Eating and Living (WHEL) Study (37), and breast cancer patients from the Nurses' Health Study (NHS; ref. 38). Each cohort collected data on clinical and reproductive factors, family history of breast cancer, quality of life, medical history, anthropometry, smoking history, alcohol intake, supplement use, physical activity, and diet. In the ABCPP, these data have been harmonized into a common dataset. Individual studies received Institutional Review Board approval from their respective institution(s) to participate in this collaboration.

Data from WHEL, LACE, and NHS were included in this analysis, as almost no women in the SBCSS reported alcohol consumption. Postdiagnosis alcohol intake was collected from participants at baseline (entry into cohort for LACE and WHEL) or regular follow-up after diagnosis (NHS) using validated food frequency questionnaires (FFQ).

Alcohol was assessed at baseline (on average 2 years postdiagnosis) in WHEL using the Arizona Food Frequency Questionnaire (39, 40) and in LACE using the Fred Hutchinson Cancer Research Center Food Frequency Questionnaire (FHRCC-FQ; ref. 41). For wine, beer, and liquor, women reported their average frequency of consumption over the past 12 months, and a medium serving size was defined as 1 6-oz glass, 1 12-oz can or bottle, and 1 1.5-oz shot, respectively. For WHEL, other minor sources of alcohol (e.g., vanilla extract and bread) were included in calculations of alcohol intake. The NHS assessed dietary habits using a validated semi-quantitative FFQ approximately every 4 years since diagnosis (38). For wine, beer, and liquor, women reported their average frequency of consumption over the past 12 months, and single serving sizes were defined as 1 4-oz glass, 1 12-oz can or bottle, and 1 standard shot, respectively.

Servings per day in oz was converted to grams (g) per day of alcohol [one standard drink in the United States is commonly 12.0 to 14.0 g of pure alcohol depending on the alcohol by volume of the drink (42)] and categorized as none (<0.36 g/day, considered nondrinkers), 0.36 to less than 6.0 g/day (occasional drinkers), and 6.0 g/day or more (regular drinkers). Half a drink per day (6.0 g) was selected as the cutoff point, given that the median level of intake among drinkers in the combined cohorts was 5.3 g/day and the current guidelines for cancer prevention is 1 drink per day for women with no cancer history (3, 43). Higher categories of consumption were also created to examine dose response relationships among regular drinkers (6.0 to less than 12.0 g/day, 12.0 to less than 24.0 g/day, and ≥24.0 g/day).

Data included race/ethnicity (non-Hispanic white, non-Hispanic black, Asian, Hispanic, and other), education (less than high school, high school, some college, and college graduate), smoking history at diagnosis (never, past, and current), and menopausal status at diagnosis (premenopausal, postmenopausal, and unknown). Postdiagnosis recreational physical activity in metabolic equivalents (MET-hours/week) was determined on average 2.4 years postdiagnosis (range: 1.4–4.0 years) from semiquantitative questionnaires and classified as meeting or not meeting the 2008 Physical Activity Guidelines of 10 MET-hours/week or more (44). Prediagnosis body mass index (BMI) was calculated from self-reported prediagnosis weight and height at baseline (entry into cohort for WHEL and LACE) or regular follow-up before diagnosis (NHS). The BMI measures from all 3 studies were between 1 and 2 years before breast cancer diagnosis (45).

Data included age at diagnosis (years), AJCC stage (I, II, III, and IV), ER/progesterone receptor (PR) status (ER+/PR+, ER+/PR−, ER−/PR+, ER−/PR−), surgery (none, lumpectomy, and mastectomy), chemotherapy (no and yes), radiation therapy (no and yes), and hormonal therapy (no and yes). Any comorbidity history consisted of reporting at least one of the following conditions: diabetes, hypertension, myocardial infarction, and/or stroke.

All studies ascertained outcome events by self-report and regular linkage to electronic medical records and vital statistics registries. Reported events were verified by medical record review except self-reported recurrences in the NHS. Cause of death, including breast cancer and cardiovascular disease (CVD), was determined from death certificates and supplemented with medical records if necessary. Details of outcome ascertainment have been published (34).

Primary analytic outcomes were a new breast cancer event, defined as a first recurrence/metastasis or new primary breast cancer (hereafter referred to as recurrence), all-cause mortality, breast cancer mortality, and CVD mortality. New primary breast cancers were longitudinally recorded in all cohorts except the NHS. For NHS participants who did not report a recurrence but died from breast cancer (7.2%), the recurrence date was set 2 years before the date of death (46).

Women were excluded from the analysis if they were diagnosed before 1990 to ensure comparable diagnosis dates and treatment information across the cohorts (n = 2,965 NHS cases), had stage IV breast cancer (n = 126), or had no postdiagnosis alcohol data (n = 1,021) or follow-up time (n = 6). A total of 9,329 breast cancer survivors comprised the final cohort. For the analyses depending on the outcome of interest, we further excluded women with missing outcome data (n = 119 with no recurrence information; n = 19 with no cause of death information), missing covariate data (n = 4), and/or date of recurrence was before start of study follow-up (n = 55). A total of 9,151 observations were available for the recurrence analyses, 9,325 observations for the total mortality analyses, and 9,306 observations for the breast cancer and CVD mortality analyses.

Sociodemographic, lifestyle, and clinical characteristics of the pooled cohort and by category of postdiagnosis alcohol intake were summarized by frequency distributions for categorical variables and means with SD for continuous variables.

The multivariable analysis involved 3 steps. First, delayed entry Cox proportional hazards regression models were used to estimate study-specific adjusted HRs and 95% confidence intervals (CI) with time since diagnosis as the time scale (47). The entry date was the date of the first survey after breast cancer diagnosis. For models with breast cancer recurrence as the outcome, the exit date was date of recurrence or date of death (whichever was first), or date of last contact (i.e., date of last follow-up survey or date of last registry linkage, whichever occurred first) for women without an event. For models with mortality as the outcome, the exit date was the date of death or date of last contact. Second, a meta-analysis was conducted with study-specific HRs using inverse-variance weights in random-effects models (48). The Q test statistic was used to assess heterogeneity in risk estimates across studies (49). Evidence for heterogeneity was not observed (P > 0.05); hence, pooled analyses were conducted and presented herein for the alcohol-outcome associations of interest using delayed entry Cox proportional hazards regression models stratified by study (50). The proportional hazards assumption for all Cox models was assessed by Kolmogorov-type supremum tests. Tests for linear trend were conducted by modeling categorical variables on an ordinal scale. Statistical significance was considered as P < 0.05 or 95% CI not overlapping with 1.0.

Covariates for the final Cox models were chosen based on a priori determination from literature review or if a covariate produced a 10% change in the main effect estimate when the covariate was added individually to the Cox model (51). Age at diagnosis, AJCC stage, race/ethnicity, education, menopausal status at diagnosis, hormone receptor status, surgery, treatment, smoking, physical activity, prediagnosis BMI, and comorbidity were retained. All missing values for each covariate were included as dummy variables. To test for the impact of missing covariate information, we ran a complete case analysis, and results were similar to treating missing data as an indicator category. We also conducted a subgroup analysis excluding women who had an early recurrence within the first 2 years of their breast cancer diagnosis (n = 152) to rule out potential effects of reverse causality attributable to underlying disease.

Possible a priori effect modification was evaluated in the associations of alcohol consumption with recurrence and mortality by menopausal status (premenopausal vs. postmenopausal), ER status (ER+ vs. ER−), and obesity (obese vs. nonobese) in stratified analyses. Statistical significance of multiplicative interaction terms was assessed by a Wald test of the cross-product terms between the main exposure (3-level alcohol variable) and the potential effect modifier (dichotomous variable as described above) in the Cox models.

Over a mean follow-up of 10.3 years in the entire cohort, a total of 1,646 recurrences and 1,543 deaths (156 related to CVD) were confirmed. Mean times (range) from index diagnosis to recurrence and death were 5.6 (0.1–16.6) years and 7.8 (1.0, 18.6) years among the women who had the events, respectively. Alcohol consumption was assessed on average 2.1 years postdiagnosis (range: 0.1–14.2 years).

Characteristics of the pooled cohort by nondrinkers, occasional drinkers, and regular drinkers (g/day of alcohol consumption) are given in Table 1. The mean age at breast cancer diagnosis was 58.8 years, and 70.9% were postmenopausal at diagnosis. The majority was non-Hispanic white (88.6%) and had at least some college education (89.7%). Most women (88.4%) were diagnosed with stage I or II tumors, and 83.1% of the tumors were ER+ and/or PR+. Compared with nondrinkers (41.9%), drinkers were more likely to be younger and premenopausal, more educated, diagnosed with earlier stage, hormone receptor positive breast cancer, and received radiation therapy or hormonal therapy. Drinkers also tended to be current or past smokers, more physically active, leaner, and with fewer comorbidities.

In the cohort, the mean (SD) g/day of alcohol consumption was 5.79 (11.15), whereas the median (range) was 1.00 (0–149.32; Table 2). Thus, the distribution of alcohol consumption was skewed by a small proportion of heavy drinkers. Among the drinkers (≥0.36 g/day), the mean (SD) consumption was 9.94 (13.14) g/day (median 5.30; range: 0.39–149.32).

Table 3 gives the associations between postdiagnosis alcohol intake (nondrinker, occasional, regular low, regular medium, and regular high) and breast cancer recurrence and mortality in the overall pooled cohort. Compared with no consumption, regular consumption of alcohol (≥6.0 g/day) was not associated with risk of recurrence at any level (HR for 6 to less than 12 g/day, 1.03; 95% CI, 0.86–1.24; HR for 12 to less than 24 g/day, 1.12; 95% CI, 0.93–1.34; HR for ≥24 g/day, 1.04; 95% CI, 0.84–1.31). Occasional drinking (0.36 to less than 0.6 g/day) was also not associated with recurrence risk (HR, 0.99; 95% CI, 0.87–1.12). No dose–response effect for increasing alcohol consumption was found. When excluding the n = 152 women who had an early recurrence within the first 2 years of their breast cancer diagnosis to examine the potential impact of reverse causality, all the associations remained nonsignificant with no dose–response effect.

Examining risk of overall mortality, breast cancer mortality, and CVD mortality, no significant associations were found for any amount of alcohol intake. However, a borderline statistically significant trend of increasing consumption with decreasing risk of CVD death was observed for regular consumption of 24 g/day or more (HR, 0.48; 95% CI, 0.21–1.12; P for trend = 0.07). A corresponding decreased risk of overall death was also found for regular consumption of 24 g/day or more (HR, 0.79; 95% CI, 0.63–1.00; P for trend = 0.06).

Table 4 gives stratified analyses of postdiagnosis alcohol consumption (nondrinker, occasional, and regular) and recurrence by menopausal status at diagnosis, ER status, and prediagnosis obesity. Significant effect modification was present for risk of recurrence by menopausal status (P for interaction = 0.027) and ER status (P for interaction = 0.012). Among postmenopausal women at diagnosis, increasing consumption of alcohol postdiagnosis was associated with an increased risk of recurrence (P for trend = 0.04), and the increased risk associated with regular consumption (≥6.0 g/day) was statistically significant (HR, 1.19; 95% CI, 1.01–1.40). In contrast, among premenopausal women at diagnosis, occasional drinking (0.36 to less than 0.6 g/day) postdiagnosis was associated with a decreased risk of recurrence (HR, 0.75; 95% CI, 0.59–0.94) with no apparent dose–response effect. While no associations were observed among ER+ women, occasional alcohol consumption (0.36 to less than 6.0 g/day) in ER− women was associated with decreased risk of recurrence (HR, 0.70; 95% CI, 0.53–0.92), whereas regular consumption (≥6.0 g/day) had no association with recurrence (HR, 1.03; 95% CI, 0.78–1.36). Associations were similar across the obese and nonobese groups with no evidence of effect modification (Table 4).

Stratified analyses of alcohol intake and mortality outcomes (total and breast cancer specific) by menopausal status, hormone receptor status, and obesity were not significant (data not shown).

In this large pooled analysis of 9,329 US breast cancer survivors, alcohol consumption after a breast cancer diagnosis was not associated with increased risk of recurrence or total mortality in general, yet some varying risk associations were observed by estrogenic factors. Specifically, compared with no drinking, regular consumption of more than 3 drinks per week (approximately half a drink or more per day) was associated with an approximate 20% increased relative risk of recurrence among postmenopausal women. While not statistically significant yet consistent with previous studies, there was a suggestive inverse trend of alcohol intake with decreased cardiovascular-related mortality despite the limited number of CVD deaths in the cohort. These results suggest that alcohol intake is generally not associated with poor outcomes after a breast cancer diagnosis in the largest analytic sample size to date. However, alcohol might be associated with a modestly increased risk of recurrence in postmenopausal women, but this finding requires independent corroboration in other large studies.

A possible explanation as to why postdiagnosis alcohol intake was not associated with risk of breast cancer recurrence among all women yet associated with increased risk among postmenopausal women only is that the estrogenic effects of alcohol on circulating estrogens might vary depending on the endogenous estrogen environment (29, 33). Among postmenopausal women with breast cancer, sex hormone concentrations have been shown to be stable over time (52, 53), yet those in premenopausal women with breast cancer can fluctuate greatly because of treatment effects and age (53–55). Furthermore, women who experience menopause resulting from treatment likely have higher estrogen levels than women who are already postmenopausal at diagnosis. Overall, more research is needed to elucidate the potentially differing biologic mechanisms of alcohol intake among premenopausal and postmenopausal women with breast cancer.

We report an increased risk of recurrence among women who were postmenopausal at breast cancer diagnosis and a corresponding decreased risk of recurrence among those who were premenopausal at diagnosis. Previous large studies on alcohol and incident breast cancer have reported an elevated risk among postmenopausal women (56–58), and when we excluded the second primary breast cancers (n = 135) from our menopausal analysis, the recurrence findings were essentially unchanged. Furthermore, a linear increase in estrone sulfate levels (the most abundant, long-lasting estrogen in postmenopausal women) was previously observed across alcohol intakes of 0 to 30 g/day or more in the NHS (59).

While our results suggest a possible interaction of postdiagnosis alcohol intake by ER tumor status, because of the limited sample size of women with ER− tumors who had an event of interest, we cannot rule out chance and residual confounding. Nonetheless, alcohol has been found to increase the expression and proliferation of ERs in cultured human breast cancer cells (32, 60) and has been found to be associated with the development of incident ER+ breast cancer in postmenopausal women (61, 62). When we conducted a subgroup analysis among the 5,082 postmenopausal, ER+ women in our cohort, any level of alcohol intake was associated with a borderline significant, increased risk of recurrence (HR, 1.21; 95% CI, 0.99–1.45 for <6.0 g/day and HR, 1.19; 95% CI, CI: 0.99–1.45 for ≥6.0 g/day).

The 3 studies included in this pooled analysis have previously published individual analyses on alcohol and breast cancer prognosis, yet they either did not examine recurrence specifically (NHS) or were limited by sample size to adequately examine potential interactions by estrogen-related factors (all studies). This pooled analysis includes new analyses of recurrence, longer follow-up, and an expanded number of NHS women.

In the earlier WHEL study of 3,088 early-stage breast cancer survivors followed for 7.3 years, moderate (≥1 drink/day) alcohol intake postdiagnosis was associated with reduced overall mortality, particularly among nonobese women (13), yet was not associated with recurrence. In the NHS, moderate alcohol intake postdiagnosis was not associated with total mortality among 1,982 women diagnosed with invasive breast cancer and followed for 13.1 years (19). Similarly, in the LACE study of 1,897 women diagnosed with early-stage breast cancer and followed for 7.4 years, regular consumption of alcohol (≥0.5 drink/day) after diagnosis was not associated with overall death (25). In agreement with this pooled analysis, an elevated risk of recurrence with alcohol intake was observed among postmenopausal women in LACE, whereas WHEL and NHS did not examine alcohol associations by menopausal status.

Strengths of this pooled study include being the largest to date of breast cancer survivors with data on postdiagnosis alcohol consumption, and having the ability to adjust for important prognostic and treatment-related factors, including BMI, smoking, menopausal status, cancer stage, hormone receptor status, and treatment. Our study population had a wide range of alcohol intake, allowing adequate exploration of higher levels of drinking and outcomes. The mean level in the pooled cohort among drinkers was 9.9 g/day (median 5.3), which is in agreement with mean levels reported in studies of healthy female drinkers (3.2–12.6 g/day; ref. 5). Pooling data also improved the statistical power to conduct stratified analyses by estrogen-related factors.

Of note, our analysis did not examine patterns of change in postdiagnosis alcohol consumption, as follow-up assessments across cohorts were not temporally uniform relative to baseline. However, most women (98%) remained in the same or adjacent category of intake at their subsequent follow-ups. Furthermore, change in alcohol consumption from prediagnosis to postdiagnosis may also be a contributing factor in associations of alcohol and mortality. Specifically, women who more regularly consumed alcohol before their breast cancer diagnosis and then substantially reduced or stopped their intake after diagnosis may have lower overall mortality compared with lifelong nondrinkers or low drinkers. While we do not have information on prediagnosis alcohol intake on the entire pooled cohort, perhaps this hypothesis explains, in part, the attenuated CVD mortality association observed in our analyses. We also relied on self-reported alcohol use, yet in NHS and WHEL, levels have been shown to be highly reproducible using food records and 24-hour recalls, respectively (13, 63).

In summary, we observed that alcohol consumption after a breast cancer diagnosis was generally not associated with risk of recurrence, or total or breast cancer-related mortality. A suggestive trend of alcohol intake and decreased risk of cardiovascular-related deaths was noted. However, regular drinking of more than 3 drinks per week (half a drink or more per day) was associated with higher risk of recurrence in postmenopausal women. Given that consuming alcohol is a potentially modifiable lifestyle factor following a breast cancer diagnosis, further confirmation is warranted in other large prospective studies of breast cancer survivors with long-term follow-up and detailed information on alcohol intake and estrogen-related covariates.

No potential conflicts of interest were disclosed.

Conception and design: M.L. Kwan, W.Y. Chen, X.-O. Shu, J.P. Pierce, B.J. Caan

Development of methodology: M.L. Kwan, W.Y. Chen, J.P. Pierce, B.J. Caan

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): W.Y. Chen, S. Nechuta, M. Holmes, J.P. Pierce, B.J. Caan

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M.L. Kwan, W.Y. Chen, S.W. Flatt, E.K. Weltzien, S. Nechuta, R.E. Patterson, X.-O. Shu, B.J. Caan

Writing, review, and/or revision of the manuscript: M.L. Kwan, W.Y. Chen, S.W. Flatt, E.K. Weltzien, S. Nechuta, E.M. Poole, M. Holmes, R.E. Patterson, X.-O. Shu, J.P. Pierce, B.J. Caan

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M.L. Kwan, S.W. Flatt, E.K. Weltzien, S. Nechuta, E.M. Poole

Study supervision: B.J. Caan

This work was supported by the National Cancer Institute (3R01CA118229-03S1). Funding for each individual cohort participating in the After Breast Cancer Pooling Project is as follows: The SBCSS was supported by the Department of Defense (DAMD 17-02-1-0607); and the National Cancer Institute (R01 CA118229). The WHEL Study was supported by the Susan G. Komen Foundation (#KG100988). The LACE Study was supported by the National Cancer Institute (R01 CA129059). The NHS was supported by the National Cancer Institute (P01 CA87969, T32 CA009001).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.