The design and procedures followed in the WCHS have been described elsewhere (Ambrosone et al, 2009). In summary, the WCHS was a case–control study conducted in NY and NJ. Eligibility criteria for cases included AA or Caucasian females, 20–75 years of age, no previous history of cancer except for non-melanoma skin cancer, able to understand and speak English and newly diagnosed with histologically confirmed invasive breast cancer or ductal carcinoma in situ (DCIS). Controls had the same eligibility criteria as cases except that they had no breast cancer. Study recruitment in NY ended in December 2008 while data collection in NJ pertaining to this investigation was completed in March 2012. Cases in NY were identified through hospitals with the largest referral patterns for AA women while in NJ, rapid case ascertainment methodology was employed by the NJ State Cancer Registry to identify eligible cases. African-American controls in NY were recruited through random digit dialling (RDD) and in NJ through RDD supplemented by recruitment through community sources (Bandera et al, 2013). Although WCHS included both AA and Caucasian women, because of the plethora of evidence relating alcohol and breast cancer risk in Caucasian women and the scarce literature in AA women, these analyses focus on AA women. Overall, the participation rate for those who were contacted and eligible was 78.7% and 48.2% in AA cases and controls, respectively, and 79% and 49% in Caucasian cases and controls, respectively.

There were no meaningful differences between cases and controls with regard to frequencies of education, age at menarche, parity, breastfeeding history or BMI (Table 1). A higher proportion of cases reported a family history of breast cancer, personal history of benign breast disease and HRT use. Proportions of OC use and menopausal status were very similar in cases and controls. The distributions of recent and lifetime alcohol intakes for different alcoholic beverages are presented in Table 2. Overall, reported recent alcohol consumption in this study of AA women was very low, with median intake of around 16 g of ethanol in a week in cases and controls who reported being alcohol consumers. Among drinkers, mean intakes of total ethanol from recent drinking and beer were higher among controls than in cases, whereas the opposite was true for ethanol consumed from wine and liquor.

Associations between recent alcohol consumption and breast cancer risk are presented in Table 3. There was no clear relationship between recent alcohol drinking and breast cancer risk, both when evaluated compared with non-drinkers (OR=0.95; 95% CI: 0.77–1.17) or for increasing levels of consumption (OR=0.95; 95% CI: 0.65–1.39, comparing >28 g per week vs <14 g per week). There was also no evidence of an association between drinking any specific alcoholic beverage and disease risk, although the risk estimate for consuming >2 mixed drinks or liquor a week compared with <1 drink was strong, albeit statistically nonsignificant (OR=1.53; 95% CI: 0.84–2.79).

There was a suggestion of decreased breast cancer risk for women in the highest category of lifetime drinking compared with non-drinkers (OR=0.77; 95% CI: 0.58–1.03), although the association for ever drinkers compared with never drinkers was null (Table 4). Inverse statistically significant associations were further observed among women who drank alcohol when they were <20 years of age with an OR of 0.65 (95% CI: 0.47–0.89) compared with women who were non-drinkers in that age interval. This association was significant (P=0.01) even after adjusting for multiple testing (Bonferroni correction alpha=0.01). Similarly, a reduction in risk was also suggested among women who drank after age 60 (OR=0.42; 95% CI: 0.17–1.01); however, the sample size was small for drinkers in this age interval, and results could have occurred by chance (P-value=0.05). There was no association between drinking during any other age interval and breast cancer risk.

When stratified by menopausal status at the time of breast cancer diagnosis/reference date (Table 5), the ORs for premenopausal (OR=1.21; 95% CI: 0.90–1.62) and postmenopausal women (OR=0.78; 95% CI: 0.57–1.07) were in the opposite direction, albeit not statistically significant. There was no evidence of an association for increasing levels of consumption in either group. However, the inverse association observed between drinking under 20 years of age and breast cancer risk was observed in both premenopausal and postmenopausal women, although the estimate was statistically significant in premenopausal women (OR=0.60; 95% CI: 0.38–0.94). Similarly, the inverse association with drinking under 20 years of age was also found for both ER+/PR+ and ER−/PR− tumours although borderline significance was found among women with hormone receptor-positive cancers (OR=0.64; 95% CI: 0.41–0.99).

There was no clear evidence of effect modification by BMI or smoking status for both recent and lifetime alcohol consumption, as magnitude of ORs were generally comparable across BMI and smoking categories, with most CIs including the null value (data not shown). However, the inverse association between drinking under 20 years of age and breast cancer risk was stronger among women who were overweight (OR=0.49; 95% CI: 0.27–0.87) and among never smokers (OR=0.39; 95% CI: 0.21–0.73).

There was no change in the main findings on lifetime alcohol consumption (overall or drinking under 20 years of age) and breast cancer risk when multivariable models were further adjusted for recent alcohol intake or when both recent and lifetime alcohol risk estimates were also adjusted for physical activity (data not shown). Excluding HRT users, community controls or DCIS cases also did not alter the main conclusions, although the OR for recent drinking compared with non-drinkers among postmenopausal women changed from 0.78 to 0.64 when excluding community controls (data not shown). Results from the main analyses did not change in parsimonious models (data not shown).

In this case–control study involving a large sample of AA women, neither recent, light to moderate alcohol intake nor lifetime drinking significantly increased breast cancer risk, even after stratifying by menopausal or hormone receptor status. Drinking under 20 years of age was associated with a reduced disease risk when compared with non-drinkers during that age interval, which remained regardless of menopausal status and tumour receptor status. The inverse relationship observed for drinking during adolescence was also significant among overweight women and non-smokers, although magnitude of associations were generally similar across BMI and smoking levels.

Despite a larger sample size compared with previous studies, the overall null findings for recent alcohol intake and breast cancer risk were in general agreement with the current evidence in AA women (reviewed in Chandran et al, 2012). There were also no meaningful differences in risk by beverage type, except for a stronger OR observed with increasing levels of liquor consumption. In a small case–control study of AA women, an inverse association was observed between alcohol intake (mostly consisting of wine) and breast cancer risk (Llanos et al, 2012). In our study, we observed nonsignificant opposite relationships in premenopausal and postmenopausal women for drinkers compared with non-drinkers, which also could have contributed to the null findings when combining all women, although there was limited power to detect a significant association. The likelihood of spurious associations because of relatively small numbers especially among alcohol consumers cannot be discounted. As this is the first study to present risk estimates stratified by menopausal status in AA women, findings need replication.

In this study, we also found a borderline inverse association between lifetime alcohol consumption and breast cancer risk. The only other study on lifetime consumption in AA women reported null results albeit with an OR below 1 (OR=0.7; 95% CI: 0.3–1.6) for average lifetime consumption of over 182 g per week (Kinney et al, 2000), but drinking in our study was much lower. It is possible that the inverse association in our study was driven by drinking during a particular lifetime period, during which time alcohol consumption may not have a detrimental effect or perhaps alcohol intake was only a marker for other factors that may reduce risk. The marginal inverse relationship observed for lifetime drinking in AA women appeared to be reflected mostly through alcohol consumption under age 20 including the adolescent period. This association remained statistically significant even when adjusted for multiple testing. Excluding HRT users also did not alter the inverse relationship. It is unlikely that this finding was affected by the participant's current age because not only was this age category applicable to all participants, but the association was found in both premenopausal and postmenopausal women. However, findings related to lifetime consumption could be spurious in view of potential recall bias described later; hence should be interpreted with caution. Some past studies mostly involving Caucasian populations have reported no risk increase for drinking at 10–15 years (Marcus et al, 2000), 15–20 years (Berstad et al, 2008), 16 years (Freudenheim et al, 1995), 18 years (Smith et al, 1994) and below 22 years (Garland et al, 1999), while other studies have shown increased risks associated with drinking below 30 years of age (Harvey et al, 1987) and during ages 18–40 years (Chen et al, 2011). Terry et al (2006) observed an inverse relationship between breast cancer risk and drinking >15 g of alcohol per day (compared with non-drinkers) under age 20 (OR=0.66; 95% CI: 0.41–1.06) and between 20 and 30 years (OR=0.57; 95% CI: 0.36–0.89) in women who were overweight or obese, and the sample included Caucasian women. Similarly, although the interaction between smoking and alcohol intake on breast cancer risk was not significant, perhaps the stronger inverse finding in non-smokers could be because of reduced exposure to toxic compounds in the absence of tobacco. However, mechanisms to explain both an increased and decreased risk of breast cancer associated with smoking have been postulated (Terry and Rohan, 2002) and, the evidence remains inconclusive. Hence, the inverse finding in non-smokers cannot be emphasised.

Racial differences in oestrogen (Taioli et al, 1996) and ethanol (Yu et al, 1995) metabolism have been noted, such that even when AA have reported heavier drinking habits, on average, they had lower levels of urinary ethanol as compared with Caucasian drinkers, which could be due to identified racial differences in distributions of genetic polymorphisms related to ethanol metabolism (Dumitrescu and Shields, 2005). A variant in the regulatory region of CYP2E1 has been associated with increased metabolic activity of this gene (McCarver et al, 1998), which could increase concentrations of acetaldehyde. The incidence of this variant has been shown to be higher in AA than in Caucasians with greater levels of metabolic activity observed mostly among obese individuals or among those who had recently consumed ethanol (McCarver et al, 1998). However, animal studies have also shown that CYP2E1 could metabolise both ethanol and acetaldehyde (Terelius et al, 1991), which could potentially result in lowered levels of acetaldehyde in the body, thus reducing toxicity. Research on the role of alcohol in promoter methylation of tumour-suppressor genes has indicated an inverse association between p16 promoter methylation and lifetime drinking as well as drinking intensity (Tao et al, 2011). This inverse relationship was more prominent for intakes at earlier age periods. Hence, race-specific investigations on biological mechanisms involved in the impact of ethanol intake and oestrogen levels on breast tissue are warranted.

Certain limitations of the study should be noted. The main issue was the limited consumption of alcohol in our study because of light drinking (<1 drink per day). Alcohol consumption in AA women has been generally lower than reported intakes in Caucasian women (Liu et al, 1996; US Department of Health and Human Services; Centers for Disease Control and Prevention; National Center for Health Statistics, 2012), which was also found in our study. Although the prevalence of recent alcohol consumption in our study (39%) is similar to national estimates reporting 32% of AA women as current regular drinkers (US Department of Health and Human Services; Centers for Disease Control and Prevention; National Center for Health Statistics, 2012), only 10% of women reported having more than half a drink per day, which is lower than levels reported among AA women in the Carolina Breast Cancer Study (Kinney et al, 2000). Therefore, we could not assess breast cancer risk associated with drinking patterns, heavy drinking (3 or more drinks per day) or binge drinking. There was also limited power to detect associations in stratified analyses. Low alcohol consumption in our study was not restricted to AA women. In our study, associations between recent drinking, lifetime consumption and breast cancer risk were mostly null in Caucasian women, possibly because of limited range of intake (median among drinkers was <3 drinks per week) even in this group (data not shown). Although majority of studies in Caucasians have shown increased breast cancer risk with increasing levels of drinking, few studies have also shown inverse or nonsignificant associations (Rosenberg et al, 1990; Friedenreich et al, 1993; Freudenheim et al, 1995).

Recall bias is also often an issue in case–control studies especially for exposures such as lifetime alcohol consumption where there could be possible errors in recalling alcohol intake from the distant past. There is some evidence showing that alcohol consumption recall (Giovannucci et al, 1993) is similar for breast cancer cases and controls. A validation study reported that recall of alcohol intake for 20 years in the past was good but not reports at 18 years of age (Dwyer et al, 1989). Another study reported that recall of alcohol intake at different age periods: 16–19 years, 20–29 years, 30–39 years, and recent consumption was very good (Longnecker et al, 1992). Specifically, a reliability study (Liu et al, 1996) using NHANES data reported that correlations for alcohol recalled from 10 years in the past for AA women were moderate and were only slightly lower than correlations observed in Caucasian women. Nevertheless, since adolescent drinking has been illegal in the United States for several years, incorrect reporting of drinking when under 20 years of age is likely. Hence, the observed inverse association between drinking under 20 years and the marginal decreased risk for overall lifetime drinking and breast cancer risk should be viewed with caution.

Mode of data collection (interviewer-administered vs self-report) could also affect reporting for certain exposures such as alcohol. In our study, >300 women reported to be non-drinkers on the interviewer-administered questionnaire, but disclosed alcohol intake in the self-administered FFQ, whereas only 17 women seem to have the opposite discrepancy. However, results were similar when we used the FFQ and interviewer-administered responses. Moreover, assessment of alcohol intake using self-reported methods has been shown to be both valid and reliable (Feunekes et al, 1999; Del Boca and Darkes, 2003). Nevertheless, analyses were repeated after excluding women with inconsistent data, and while there was a loss in precision as expected, study conclusions did not change.

Cases may also have changed their alcohol intake after their diagnosis and incorrectly report recent consumption. In our study, as compared with controls, greater proportion of AA cases had quit drinking since diagnosis (33.9% vs 19.4%), and a lesser proportion of cases had started drinking more (5.4% vs 18%). Drinking levels could be systematically different among controls than cases, especially with higher socioeconomic status generally observed among RDD controls. Finally, to evaluate potential selection bias arising from community control recruitment, analyses were repeated excluding them. But, main results did not substantially change. Moreover, a combination of community and RDD controls appeared to be more representative of the underlying population with regard to characteristics such as education, income, and marital status (Bandera et al, 2013). Response rates as low as 50% are not uncommon in population studies (Hartge, 2006), and case participation for studies of cancer have shown to be lower than for studies of other diseases (Olson, 2001).

Strengths of the study include the large sample of AA women allowing further evaluation by important subgroups as well as collection of both recent and lifetime drinking data. To our knowledge, this is the first and largest study of AA women to evaluate the association between both recent and lifetime alcohol intakes and breast cancer risk along with assessment of risk associated with drinking during different life periods, and stratification by several important factors.

To conclude, recent moderate alcohol intake was not associated with breast cancer risk in AA women. Lifetime drinking, particularly early age drinking, appeared to decrease risk. However, as this is the first study to report these findings in AA women, and in light of potential biases that have been discussed, no definitive conclusions can be drawn at this time. Hence, until future research is conducted to explore the role of alcohol as well as potential gene–environment interactions to evaluate racial differences in response to intake, expert guidelines of limiting alcohol consumption for cancer prevention should continue to be recommended.