Alcohol Consumption in Relation to Breast Cancer Risk in a Cohort of United States Women 25–42 Years of Age¹

Alcohol consumption may be one of the few modifiable breast cancer risk factors (1). Positive associations between alcohol and breast cancer have been observed in most case-control and cohort studies of this relationship (2). In a meta-analysis of 28 case-control and 10 cohort studies, a modest positive association between alcohol and breast cancer was found; the RR³ (95% CI) was 1.24 (1.15–1.34) for consumption of 2 drinks/day relative to no drinking (3). In a recent pooled analysis of six prospective studies of mainly postmenopausal women, an increment of 10 g of alcohol/day was associated with a RR (95% CI) of 1.09 (1.04–1.13; Ref. 4).

Despite the consistency of the epidemiological findings, several important issues pertaining to the relationship between alcohol and breast cancer remain unresolved (1, 2). The measure of alcohol exposure in most studies has been limited to current or usual alcohol drinking; few studies have addressed temporal aspects of drinking. Drinking in early adult life has been hypothesized to be particularly harmful for breast cancer risk (5). If the association between current or recent alcohol consumption and breast cancer is attributable to a carcinogenic effect of alcohol intake at younger ages, then a substantial association may exist between drinking at young ages and breast cancer. Although drinking early in adult life has been found to be more strongly related to breast cancer risk than later drinking in several studies (6, 7), in other studies, recent drinking appeared to be more important (8, 9). Data on the relationship between alcohol consumption and breast cancer are limited in young women. We therefore examined alcohol consumption in various time periods in relation to breast cancer risk in a large cohort of mostly premenopausal women.

The Nurses’ Health Study II is a prospective cohort study of female registered nurses who were 25–42 years old and living in one of 14 states in the United States when they responded to a baseline questionnaire in 1989. Women who reported cancer at enrollment (with the exception of nonmelanoma skin cancer, hydatidiform mole, or cervical cancer/dysplasia) were excluded, resulting in a cohort of 116,671 women. Questionnaires are mailed to participants every 2 years to obtain information on the occurrence of breast cancer and other diseases and to update information on risk factors for these outcomes. For each of the first three follow-up questionnaires (1991, 1993, and 1995), the response rate among living participants was greater than 90%.

On the baseline questionnaire, participants were asked about their alcohol consumption during five different periods: from 15 to 17 years of age, from 18 to 22 years of age, from 23 to 30 years of age, from 31 to 40 years of age, and during the previous year. For all periods except for the previous year, participants were asked to report the frequency of consumption of alcoholic drinks (the total of beer, wine, and liquor), with nine frequency response categories ranging from “none or less than 1 per month” to “40 or more per week.” Average lifetime alcohol consumption (drinks/week) was calculated by multiplying the frequency of alcohol consumption in each individual time period by the length of the time period (truncated according to the woman’s age for the last two periods), summing the contributions from each time period, and dividing by the sum of time from all periods.

To assess alcohol consumption over the previous year, three separate questions were asked on the frequency of consumption of beer (one glass, bottle, or can), wine (one 4-oz. glass), and liquor (one drink or shot). The frequency responses were as described above. Total alcohol consumption over the previous year was calculated by adding the contributions from beer, wine, and liquor. The estimated weights of alcohol (ethanol) contributed by the above serving sizes were as follows: 12.8 g for beer, 11.0 g for wine, and 14.0 g for liquor.⁴ In two similar prospective studies, alcohol intake over the previous year as assessed in this manner was shown to be highly correlated with alcohol intake as assessed by diet records (r = 0.90 in women; Ref. 10). On the 1991 questionnaire, alcohol consumption was assessed in a similar manner, except that there were two separate questions on consumption of red and white wine, as well as two separate questions on regular and light beer. The estimated weight of alcohol contributed by a serving of light beer was 11.3 g;⁴ the estimated alcohol contents of the other individual beverages were the same as above.

The baseline questionnaire also contained questions on history of benign breast disease, family history of breast cancer, height, weight, pregnancy history, and menopausal status. Information on benign breast disease, weight, pregnancy, and menopausal status was updated on the 1991 and 1993 questionnaires.

Participants were asked on the 1991, 1993, and 1995 questionnaires whether they had been diagnosed with breast cancer in the previous 2 years. Deaths in the cohort are reported by family members and the postal service and are detected by a search of the National Death Index for participants whose vital status is unknown. The National Death Index has been shown to have a sensitivity of at least 96.5% in a similar cohort of nurses (11). When a case of breast cancer was identified, we asked the participant (or next of kin for those who had died) for confirmation of the diagnosis and for permission to seek relevant hospital records and pathology reports. There were 40 reports of breast cancer that were not reconfirmed; for 32 such reports, we were unable to contact the nurse, and for 8 reports, the nurse did not reconfirm the diagnosis upon being contacted. Pathology reports were obtained for 92% of the reconfirmed diagnoses, and of these, 99% confirmed the self-reported diagnosis of breast cancer. After exclusion of six cases whose diagnosis was rejected based on the pathology reports and 87 cases of carcinoma in situ, there were 445 cases of invasive breast cancer available for analysis. We included 43 cases whose diagnosis was based on the reconfirmed self-report only.

For each participant, follow-up time was allocated to the stratum defined by her exposure status at baseline. Participants who were missing follow-up questionnaires and for whom there was no report of a breast cancer diagnosis between return of the 1989 questionnaire and June 1, 1995, were considered lost to follow-up at the date of their last questionnaire return. Participants contributed person-time in each 2-year interval until the earliest of the following: a diagnosis of breast cancer, death, loss to follow-up, or June 1, 1995. We calculated incidence rates for each exposure category by dividing the number of cases in that exposure category by the amount of person-time accrued in that exposure category. We calculated RRs, measured as incidence rate ratios, by dividing the incidence rate in the exposed category by the incidence rate in the unexposed category. RRs were adjusted for age in 5-year categories.

To adjust for age and other potential confounders simultaneously, we used proportional hazard models (12, 13). The time period between successive questionnaires constituted a risk set, and thus, we considered three such risk sets in the analysis, each consisting of women who were still alive, had not previously been lost to follow-up, and had not been diagnosed with breast cancer. Covariate values were assigned at baseline and updated based on the information from subsequent questionnaires. Alcohol intake was defined at baseline and was generally not updated, except for in a subset of analyses in which it was updated in 1991 (the start of the second interval).

We calculated two-sided 95% CIs for all RRs. To test for trend, the exposure variable to be investigated was coded as 1, 2, 3, and so forth, and this ordinal variable entered into the model.

A total of 445 cases of invasive breast cancer were diagnosed during 619,691 person-years of follow-up between 1989 and 1995. At baseline, few women reported drinking more than moderately in the previous year: for example, for alcohol consumption above 20 g/day (approximately 1.5 drinks/day), there were 11,628 person-years of follow-up, representing 2% of the total person-years of follow-up for those with nonmissing data on alcohol consumption (Table 1). When we examined alcohol consumption in the previous year in relation to breast cancer risk in an age-adjusted model, we observed slightly elevated risks associated with the higher categories of alcohol consumption (Table 1); however, the CIs overlapped the null, and the test for trend was not significant. The multivariate results were similar to the age-adjusted results. When alcohol consumption in the previous year was treated as a continuous variable, the multivariate RR (95% CI) associated with an increment of 10 g/day was 1.03 (0.89–1.20). In a secondary analysis in which we updated alcohol information based on the information on the 1991 questionnaire, results were similar to those presented in Table 1. For example, the multivariate RR (95% CI) associated with alcohol consumption above 20 g/day was 1.28 (0.71–2.32), and the P for trend was 0.29. Because of the possibility that alcohol consumption may be influenced by preclinical disease, we examined the association between alcohol consumption in the previous year, as assessed on the 1989 questionnaire, and breast cancer after exclusion of person-time and cases in the first 2-year cycle of follow-up. In this analysis conducted among 278 cases and 388,031 person-years of follow-up, results were similar to those presented in Table 1 (data not shown). Results were also similar to those in Table 1 upon exclusion of 851 women reporting a diagnosis of cervical cancer/dysplasia at baseline (data not shown). When we examined the consumption of specific beverages (wine, beer, and liquor) in the previous year in relation to breast cancer risk, no notable associations were observed (data not shown).

The relationships of drinking in each of four time periods to breast cancer risk are shown in Table 2. In each period, the proportion of women drinking more than 6 drinks/week was small. For example, even at ages 18–22, the time period during which this level of drinking was most frequent, only 9.5% of follow-up time was allocated to this level of consumption. Consumption of more than 6 drinks/week at ages 15–17 was associated with an elevated risk, but this was not statistically significant, nor was there a significant trend. No notable association was observed between drinking at ages 18–22 and breast cancer risk nor between drinking at ages 31–40 and breast cancer risk. Drinking at ages 23–30 was positively associated with breast cancer risk. In a multivariate model adjusting for alcohol consumption in each of the other time periods, the RR associated with consumption of more than 6 drinks/day, relative to no alcohol consumption, was 1.72 (95% CI, 1.01–2.91), and the P for trend was 0.05. We examined average lifetime alcohol consumption in relation to breast cancer risk (Table 3). In the multivariate analysis, we observed a slightly elevated RR associated with consumption of 10 or more drinks/week, but the CI overlapped the null, and the test for trend was not significant. When average lifetime alcohol consumption was treated as a continuous variable, the multivariate RR (95% CI) associated with a 5 drink/week increment was 1.12 (0.97–1.29).

At baseline, 97% of the cohort was premenopausal. Of the breast cancer cases, 92% were among premenopausal women. Results of analyses were similar when limited to premenopausal women (data not shown).

Because of concern that results may have been affected by ascertainment bias, we examined the age-adjusted proportions of women with various screening practices (self-examination, examination by a health care provider, and mammography) by level of alcohol consumption. Screening practices, as assessed on the baseline questionnaire, were similar across categories of alcohol consumption in the previous year (as assessed on the baseline questionnaire). For example, the age-adjusted proportion of women who reported ever having had a mammogram was 36% among nondrinkers, as well as among those women drinking more than 20 g/day of alcohol.

In this study, few women drank more than moderately. In 1991, 4.8% of US women of reproductive age drank ≥7 drinks/week (14). Results from the current study are consistent with this finding: 4.6% of follow-up time was allocated to recent alcohol consumption of greater than 13 g/day (approximately 1 drink/day). In the current study of mainly premenopausal women, neither alcohol consumption in the previous year nor average lifetime alcohol consumption was found to be significantly positively associated with an increased risk of breast cancer. When we examined drinking in four different time periods in relation to breast cancer risk, a significant positive association was observed only for drinking at ages 23–30.

Because this study is prospective, it is unlikely that our results are affected by recall bias. A bias due to the influence of preclinical disease on alcohol consumption is also unlikely because our results were essentially unchanged upon exclusion of cases diagnosed early in follow-up. It is also improbable that substantial nondifferential misclassification of alcohol consumption occurred, as available data suggest that both recent alcohol consumption and alcohol consumption in the more remote past are recalled well (10, 15). Ascertainment bias is unlikely, because screening practices were similar across categories of alcohol consumption. Because we controlled for known breast cancer risk factors in our analyses, our results are not likely to be explained by confounding.

Temporal aspects of drinking have been addressed in several case-control studies. Results from some studies are consistent with the hypothesis that drinking early in life is more strongly related to breast cancer risk than recent drinking (6, 7). However, results from other studies indicate that recent drinking may be more strongly related to breast cancer risk (8, 9); one of these studies in particular (8) was specifically designed to examine temporal aspects of drinking and was quite large, with 6662 cases. In other studies, neither recent nor early adult drinking was related to breast cancer risk (16, 17). The study population was limited to young women in only two of these studies (9, 16). The association between drinking in early life and breast cancer may differ by age of the study population, due to a latency effect. We observed a positive association between drinking at ages 23–30 and breast cancer risk, whereas no association was observed between drinking in any of the other time periods and breast cancer risk. Although drinking in adolescence has been hypothesized to be particularly harmful for breast cancer risk (based on animal data and by analogy to the effects of other early life exposures such as menarche and radiation; Ref. 5), why drinking at ages 23–30 would confer the greatest risk is unclear. This finding may be due to chance. However, in the large case-control study mentioned above (8), alcohol consumption before age 30 was more strongly related to risk than recent drinking among premenopausal women. The authors of this report state that “the importance of this finding is unclear.” Thus, although the plausibility of a particularly detrimental effect of drinking at ages 23–30 in young women is uncertain, this should be examined in further studies.

In a recent case-control study of 1645 cases of breast cancer among women under the age of 45, an elevated risk of breast cancer was observed only among those women consuming 14 or more drinks/week (9). In our study, there were only five cases who reported alcohol consumption above 25 g/day (approximately 2 drinks/day); thus, we did not have sufficient information to investigate the effects of heavier drinking.

Results from previous studies conducted mainly among postmenopausal women suggest that low to moderate alcohol consumption is associated with an increased risk of breast cancer but that the effect is modest. For example, in a pooled analysis of six large prospective studies of alcohol in relation to breast cancer risk, an increment of 10 g of alcohol/day was associated with a RR (95% CI) of 1.09 (1.04–1.13; Ref. 4). Our results for both current and average lifetime alcohol consumption are consistent with this finding, although they are also compatible with a lack of effect.

In summary, in this cohort of young women, neither current nor average lifetime alcohol consumption was significantly associated with breast cancer risk. We examined drinking in several time periods of life, and only drinking at ages 23–30 was significantly positively associated with risk. This may represent a chance finding but merits further study. Because drinking levels in this population were low, we had limited data with which to investigate the effects of heavy drinking. Overall, these results suggest that there is unlikely to be a large effect of moderate alcohol consumption on breast cancer risk in young women. However, given the width of the CIs in this study, we cannot exclude a modest effect. Our findings suggest that the association between alcohol consumption and breast cancer is unlikely to be substantially stronger among premenopausal women than among postmenopausal women.

We thank Lori Ward for expert assistance.