Alcohol Consumption and Survival after a Breast Cancer Diagnosis: A Literature-Based Meta-analysis and Collaborative Analysis of Data for 29,239 Cases

This article is featured in Highlights of This Issue, p. 893

Many studies have investigated the association of alcohol consumption and prognosis in women diagnosed with breast cancer. However, the results of these studies have been inconsistent. Many studies have reported no significant association between pre- or postdiagnosis alcohol consumption and overall survival (OS; refs. 1–10), whereas other studies showed a protective effect (11–16) or an adverse effect (17). Fewer studies have examined the association of alcohol consumption with breast cancer-specific survival (BCSS; refs. 1, 2, 7, 12, 14, 18–22) and even fewer studies investigated disease-free survival (DFS; refs. 1, 2, 7, 9, 13, 19, 20). Two reviews have been published and concluded that alcohol intake was not associated with survival in patients with breast cancer (23, 24). More recently, Kwan and colleagues reported the results of a joint analysis of data from three large cohorts with information on postdiagnosis alcohol consumption (1). Overall they found no association between regular alcohol consumption and breast cancer mortality, but regular alcohol consumption was associated with an increased risk of recurrence in postmenopausal women.

There are several possible reasons for the heterogeneity of the published evidence: different studies used different endpoints, many studies had a small number of events and limited statistical power, the timing of the exposure varied and included both pre- and postdiagnosis alcohol intake, the range of alcohol consumption was limited in some studies, the classification of the exposure variable varied widely, and finally different studies adjusted for different covariates.

It is important to clarify the impact of alcohol intake on prognosis in women with breast cancer because alcohol is a well-established risk factor for breast cancer (25), and public health advice to women is to limit alcohol consumption. It is not clear, however, whether it is safe to continue with moderate alcohol consumption after breast cancer diagnosis. Thus, the key question relates to the influence of postdiagnosis alcohol consumption on outcome.

The aim of this study was to carry out a systematic review and meta-analysis of published data to provide more precise estimates of mortality risk after breast cancer diagnosis and, if possible, to identify the causes of heterogeneity across published studies. In addition, we evaluated the association between alcohol consumption and prognosis in large case cohorts: the Studies of Epidemiology and Risk Factors in Cancer Heredity breast cancer cohort (SEARCH), the European Prospective Investigation into Cancer and Nutrition (EPIC), and nine studies from the Breast Cancer Association Consortium (BCAC). The association between self-reported postdiagnosis alcohol consumption and all-cause mortality has been previously reported for SEARCH (11), but the sample size is now considerably larger, data on breast cancer-specific mortality are now available, and data on other variables such as tumor estrogen receptor (ER) status are more complete.

A MEDLINE search of the literature up to and including January 2013 was performed using the following search terms: [“survival” (Mesh) or “mortality” (Mesh) or “survival rate” (Mesh) or “DFS” (Mesh) or “recurrence” (Mesh)] or “prognosis” (Mesh) or “death” (Mesh) or survival or mortality or relapse or recurrence or outcome or prognosis or death) and [“ethanol” (Mesh) or “alcohols” (Mesh) or “alcohol drinking” (Mesh) or “alcoholic beverages” (Mesh) or alcohol or wine or spirits or beer] and [“breast neoplasms” (MeSH) or breast cancer or breast neoplasm]. Overall 1,096 hits were retrieved; of these 50 articles were relevant based on skimming the titles and the abstracts. Each article was reviewed and included in the analysis if the following criteria were met: (i) case cohort published as an original article and (ii) findings expressed as HRs. In addition, the bibliographies of all retrieved articles were reviewed for any relevant publications missed by the search. Overall 22 original studies were relevant and were systematically reviewed. Only the most recent and complete article of studies published more than once was included in the meta-analysis. From each article, we abstracted the HR and 95% confidence limits for different exposure categories of alcohol consumption associated with any of the following endpoints: BCSS, OS, and DFS. If adjusted HR estimates were reported, we used the maximally adjusted estimates.

Different studies used different units to measure alcohol consumption, including grams, milliliters, ounces, or drinks consumed per day, week, or month. We converted these to units per week as a standard measure of ethanol intake according to the following equivalencies: 1 mL = 0.8 g, 1 oz = 28 g, 1 drink = 12.5 g, and 1 U = 8 g. We defined patients who consumed not more than 2 U of alcohol per day (14 U per week) as moderate drinkers, and compared them with nondrinkers in the meta-analysis. However, some studies used different cutoff points; we excluded reported categories that included women with alcohol consumption of more than 2 U per day (Supplementary Fig. S1). Where a single study included more than one moderate drinker category, these estimates were pooled.

We performed fixed-effect meta-analysis using the inverse-variance weighting method (26). Statistical heterogeneity between studies was assessed using the among-study variance (τ2) and the I² statistic (27). When two estimates were used from the same study, they were combined in the same way.

SEARCH is an ongoing, population-based study of breast cancer in the region covered by the Eastern Cancer Registration and Information Centre (ECRIC). A detailed description of the study has previously been published (11). The study was set up to investigate genetic susceptibility to breast cancer. All patients diagnosed with invasive breast cancer before the age of 55 years since 1991 and still alive at the start of the study in 1996 (prevalent cases; median age, 48 years) together with all those diagnosed under 70 years of age between 1996 and the present (incident cases; median age, 54 years), are eligible to take part. The study was approved by the Cambridgeshire Research Ethics Committee. The present analysis is based on data from 8,446 participants (98% of whom were White British) with a diagnosis of invasive breast cancer.

A self-administered questionnaire was used to collect information on lifestyle factors, including height, weight, smoking history, and current (postdiagnosis) alcohol intake. Reported weekly alcohol intake was converted into standard units. The local area Index of Multiple Deprivation was used as a proxy for socioeconomic status (SES; ref. 28). Age at diagnosis, vital status, and data on tumor characteristics were obtained through the ECRIC. The registry actively follows up individuals at 3 and 5 years after diagnosis and every 5 years afterwards with continuous passive follow-up through notifications of death received from the Office for National Statistics.

Data from incident breast cancer cases from EPIC were included in the analysis. EPIC is an ongoing multicenter prospective cohort study designed to investigate the associations between diet, lifestyle, genetic and environmental factors, and various types of cancer. A detailed description of the methods has previously been published (29). In summary, 521,448 participants from 10 European countries (∼70% women) mostly ages 35 years or above were recruited between 1992 and 2000. Written informed consent was provided by all study participants. Ethical approval for the EPIC study was provided from the Review Boards of the International Agency for Research on Cancer and local participating centers. Self-administered lifestyle questionnaires were used to obtain information on alcohol consumption, smoking status, and education, which was used as a proxy for SES. Height and weight were measured by trained research staff.

Incident cancer cases were identified through record linkages with regional cancer registries in all EPIC study centers except those in France, Germany, Greece, and Naples (Italy) where follow-up is conducted by review of health insurance records, contacts with cancer and pathology registries, and/or direct contact with cohort members. Vital status was ascertained through linkages with regional and national mortality registries and data collected by active follow-up (Germany and Greece). For the present study, the latest dates of complete follow-up for cancer incidence and vital status in the EPIC centers ranged from 2002 to 2006.

BCAC comprises multiple studies investigating inherited susceptibility to breast cancer susceptibility (30). Many of these have detailed pathologic data on the breast cancer cases linked to follow-up data. All BCAC studies that had collected data on alcohol intake, and had data available on survival time were eligible for inclusion in this analysis. In all the studies, the prediagnosis alcohol intake was estimated using a self-reported questionnaire that was filled in after diagnosis. A multistep data, harmonization procedure was used to reconcile differences in individual study questionnaires (31). In total, nine studies from Europe, North America, Japan, and Australia contributed unpublished data on 10,232 cases. A full description of all studies included is given in Supplementary Table S1.

Cox regression was used to assess the association of alcohol consumption and survival for the SEARCH, EPIC, and BCAC breast cancer case cohorts. Because cases were enrolled at variable times after diagnosis, analyses were conducted allowing for left-truncated data. Time to failure was considered from the date of diagnosis. Time at risk began on the date of receipt of the completed questionnaire and ended at the date of death from any cause or, if death did not occur, at date of last follow-up. Follow-up of all breast cancer cases was censored at 15 years. EPIC data were stratified by country and BCAC data were stratified by study. We modeled alcohol consumption both as a categorical and as an ordinal variable in four categories: nondrinkers, up to 7, >7 to 14, and >14 U per week. We also carried out multivariable analyses adjusting for body mass index (BMI; in quartiles), smoking status (never, former, and current), menopausal status at diagnosis (<45 years, premenopausal; 45–55 years, perimenopausal, ≥55 years, menopausal), and SES. SES was categorized into five groups (1 = least deprived in SEARCH and the most educated in EPIC). Tumor characteristics considered included clinical stage (I–IV), histopathological grade (well, moderately and poorly differentiated), and ER status (negative/positive). Stage and grade were modeled as ordinal variables in the multivariable analysis. ER-positive and ER-negative disease were analyzed separately. HRs with 95% confidence intervals (CI) were estimated. For categorical variables, a likelihood-ratio test for heterogeneity of risk between groups was carried out by comparing the fit of the full model with the intercept-only model. A similar procedure was used for a trend test for ordinal and continuous variables. All tests were two sided. The assumption of proportional hazards was assessed using standard log–log plots and tested using Schoenfeld residuals. ER status and stage were time dependent (in addition to grade in SEARCH) and were therefore treated as time-dependent variables in an extended Cox model. Intercooled Stata version 12 (STATA statistical software, release 12; Stata Corporation) was used for all analyses.

We identified 22 studies that investigated the relationship between alcohol consumption and survival (Table 1). Half of these studies had been conducted in the United States. Sample size varied considerably (range: 125–9,325 cases), with 12 studies including more than 1,000 cases. Median follow-up ranged from 3 to 13 years. Ten studies reported findings based on alcohol intake before breast cancer diagnosis and 12 studies measured alcohol intake after diagnosis.

Seventeen studies reported the relationship between alcohol consumption and OS. Most studies reported no significant association between pre- or postdiagnosis alcohol consumption and OS (1–10). However, some studies found a protective effect of alcohol (11–16) and one showed an adverse effect (28). We excluded six of these studies from the meta-analysis as follows. One study reported a protective effect of alcohol but provided HRs without 95% CIs and an overall P value across multiple levels of exposure (15). Another study reported no association for a comparison of drinkers with nondrinkers (3). Two studies (9, 11) modeled alcohol intake as a continuous variable, one of which used data from SEARCH that we have updated in the new analysis reported in this paper (11). The other study reported no association between alcohol intake and prognosis (9). One study was excluded because there were large overlaps with the data reported in other publications (7). Finally one study was excluded because of lack of details about levels of exposure, although a trend toward lower risk of death from any cause with higher alcohol consumption was reported (P_trend = 0.01; ref. 12).

Thus, 11 studies were included in the meta-analysis. For the purposes of this meta-analysis, we pooled the results for the reported comparison that we considered most closely approximating moderate alcohol consumption (up to 14 U per week) compared with nondrinkers. Six studies reported the association between prediagnosis alcohol intake and OS. Moderate alcohol consumption was associated with better survival (Fig. 1; HR, 0.80; 95% CI, 0.73–0.88). Five studies reported the association with postdiagnosis alcohol intake which was not associated with all-cause mortality (Fig. 2; HR, 0.95; 95% CI, 0.85–1.05).

Of the 11 studies that reported on BCSS, six studies reported no association (1, 12, 14, 18, 20, 22) and four studies reported that increased alcohol intake was associated with increased breast cancer-specific mortality (2, 7, 19, 21). The reporting of alcohol consumption in these studies was very heterogeneous. Therefore, it was not possible to combine estimates in a formal meta-analysis. One study did not report details about levels of exposure (12), one study did not use non- or minimal drinkers as a reference group (20), and two studies modeled alcohol intake as a continuous variable (19, 21). Both of these reported that alcohol consumption was associated with a poorer prognosis, but the sample sizes were small and in one, the effect was limited to beer drinkers (19). Kwan and colleagues (7) showed that drinking ≤6 g/day of alcohol compared with no drinking was associated with an increased risk of death due to breast cancer (HR, 1.51; 95% CI, 1.00–2.29). However, in a more recent analysis with a large sample size, this association was not observed (1). One study (2) showed increased risk of death from breast cancer in drinkers especially among those who consumed 12 or more grams of alcohol per day in comparison with nondrinkers (HR, 1.74; 95% CI, 1.13–2.67). Holm and colleagues (20) showed no statistically significant association (HR, 1.06; 95% CI, 0.66–1.72), but their reference group was women who consumed one unit or less per day. Similarly, Harris and colleagues (14) showed no significant association between alcohol intake and BCSS. In that study, women who consumed 10 g per day or more of alcohol had an adjusted HR of breast cancer-specific death of 1.36 (95% CI, 0.82–2.26; P_trend: 0.47) compared with nondrinkers. Similar conclusions of null association were made by others (12, 18, 22).

Three studies examined the relationship between prediagnosis alcohol consumption and breast cancer recurrence and four examined postdiagnosis alcohol consumption; four showed no association (1, 2, 9, 13) and three showed increased risk of recurrence (7, 19, 20). Again, heterogeneity prevented the combining of these studies. Two studies modeled alcohol intake as a continuous variable (9, 19), one study did not use non- or minimal drinkers as a reference group (20) and there was an overlap between two studies (1, 7). In one of the studies (19), a statistically significant increased risk of recurrence with beer drinking (drinks/day) was reported (HR, 1.41; 95% CI, 1.02–1.97). Similarly, Kwan and colleagues (7) showed that drinking ≥ 6 g/day of alcohol compared with no drinking was associated with an increased risk of breast cancer recurrence (HR, 1.35; 95% CI, 1.00–1.83). This association has been confirmed only in postmenopausal women in their recent analysis (1); in which postmenopausal women who regularly consumed alcohol had an increased risk of recurrence (HR, 1.19; 95% CI, 1.01–1.40). Like other studies (9, 13), Holm and colleagues (20) showed no statistically significant association with moderate drinking (HR, 1.31; 95% CI, 0.81–2.11). However, they found a modest but significant association among heavy drinkers between prediagnostic alcohol consumption and recurrence, both when using baseline measures of alcohol intake (HR, 1.65; 95% CI, 1.02–2.67) and cumulated alcohol intake (HR, 2.02; 95% CI, 1.06–3.85).

Postdiagnosis alcohol intake data were available for 8,446 patients with breast cancer in SEARCH with 55,684 person-years of follow-up (median 6 years). Age at diagnosis varied from 23 to 73 years (mean 54 years). There were 1,506 deaths of which 1,213 were due to breast cancer. The annual mortality rate was 2 per 100 and the 5-, 10-, and 15-year breast cancer survival rates were 88% (95% CI, 87%–89%), 75% (95% CI, 74%–77%), and 43% (95% CI, 36%–50%), respectively. Supplementary Table S2 shows the characteristics of the SEARCH study by postdiagnosis alcohol consumption. Alcohol intake tended to be higher for women with low-grade and early-stage tumors and varied by ER status, BMI, SES, smoking, and menopausal status.

Prediagnosis alcohol intake data were available for 10,561 patients in EPIC with 69,383 person-years of follow-up (median 7 years). Age at diagnosis varied from 25 to 93 years (mean 60 years). There were 1,422 deaths of which 749 were due to breast cancer. The 5-, 10-, and 15- year breast cancer survival rates were 90% (95% CI, 90%–91%), 81% (95% CI, 80%–82%), and 71% (95% CI, 69%–75%), respectively. Supplementary Table S3 shows the characteristics of the EPIC study by prediagnosis alcohol consumption. Alcohol consumption varied by menopausal status, BMI, and SES. There was a trend for alcohol consumption to decrease with increasing BMI. There were missing data in most of the variables studied, especially ER status (35%). Supplementary Table S4 shows the distribution of alcohol consumption by country in the EPIC cohort.

Prediagnosis alcohol intake data were available for 10,232 patients in BCAC with 69,710 person-years of follow-up (median 6 years). Age at diagnosis varied from 18 to 95 years (mean 54 years). There were 1,911 deaths of which 860 were due to breast cancer. The 5-, 10-, and 15-year breast cancer survival rates were 88% (95% CI, 87%–88%), 77% (95% CI, 76%–78%), and 66% (95% CI, 64%–67%), respectively. Supplementary Table S5 shows the characteristics of the BCAC data by study.

The associations between self-reported alcohol intake and prognosis in women with ER-positive breast cancer are shown in Table 2. There was a weak inverse association between postdiagnosis alcohol intake and all-cause mortality for women with ER-positive tumors (SEARCH data) with an apparent dose-response effect in univariate analysis. However, the association was substantially attenuated after adjustment for stage, grade, BMI, smoking status, SES, and menopausal status. There was also an inverse association between postdiagnosis alcohol intake and BCSS in ER-positive disease, but the effect was weaker than for OS and no association remained in the multivariate analysis. There was a very similar pattern for the inverse association of prediagnosis alcohol intake and both OS and BCSS in women with ER-positive breast cancer (EPIC and BCAC datasets).

The associations between self-reported alcohol intake and prognosis in women with ER-negative breast cancer are shown in Table 3. In unadjusted analyses, there was a strong inverse relationship in SEARCH between postdiagnosis alcohol intake and all-cause mortality (P = 0.0003) with an apparent dose-response effect (per-category HR, 0.74; 95% CI, 0.65–0.87; P_trend = 0.0001). A slightly weaker inverse association was observed for breast cancer-specific mortality (P = 0.006) with a similar dose-response effect (per-category HR, 0.78; 95% CI, 0.67–0.91; P_trend = 0.001). This association was attenuated slightly in a multivariable adjusted model (per-category HR, 0.81; 95% CI, 0.69–0.96; P_trend = 0.013). However, no association between prediagnosis alcohol intake and mortality was observed in EPIC or the BCAC dataset. Supplementary Table S6 shows the association of alcohol consumption and mortality without stratification by ER status. We also carried out analyses stratified by BMI and menopausal status but no significant differences were found (data not shown). We also restricted the analysis stage I and II disease for SEARCH and again there was no significant difference in the findings (Supplementary Table S7).