Alcohol consumption-associated breast cancer incidence and potential effect modifiers: the Japan Public Health Center-based Prospective Study

Abstract

Epidemiological studies have evaluated whether the impact of alcohol intake on breast cancer risk is modified by use of exogenous estrogens, folate intake, body weight and smoking status, but results have been inconsistent. Further, effect modification by intake of isoflavones and alcohol-induced facial flushing, which are prevalent in Asian populations, have not been investigated. We investigated the association between alcohol intake and breast cancer risk and whether the association is modified by these factors among 50,757 premenopausal and postmenopausal women (aged 40–69 years) in the population-based Japan Public Health Center-based Prospective Study. Alcohol consumption and other related factors were assessed using self-reported questionnaires. Through to the end of 2006, 572 patients were identified. Relative risks (RRs) and 95% confidence intervals (CIs) were estimated by hazard ratios derived from Cox proportional hazards regression models. Compared with never-drinkers, regular alcohol drinkers (>150 g of ethanol/week) had a higher risk of the development of breast cancer; the multivariable-adjusted RRs were 1.75 (95% CI = 1.16–2.65; p_trend = 0.035) for overall, 1.78 (95% CI = 1.09–2.90) for premenopausal and 1.21 (95% CI = 0.53–2.75) for postmenopausal women. There was no statistical evidence for effect modification by menopausal status, use of exogenous estrogens, intakes of isoflavone and folate, body weight, alcohol-induced facial flushing or smoking (All p_interactions ≥ 0.15). Excessive alcohol intake was associated with an increase in the risk of breast cancer in this population. There was no statistical evidence for effect modification.

Alcohol-related carcinogenesis of the breast has been identified in a number of animal and epidemiological studies, and confirmed by the International Agency for Research on Cancer (IARC) working group,1, 2 the Second Expert Report from the World Cancer Research Fund,3 and meta-analyses.4, 5 Apparently consistent with this, the steep increase in the incidence of breast cancer in Japan over the last 3 decades6 has coincided with an upward trend in alcohol consumption,7 although other explanations may include increased height, early menarche, low birthrate or late first birth.8–11 Nevertheless, a recent qualitative review12 of 11 epidemiological studies conducted in Japanese women suggested that the association remains inconclusive, albeit that one prospective study in 35,844 women, including 151 cases, reported a positive association.13

Proposed mechanisms of ethanol-associated breast carcinogenesis include both hormone-dependent14–19 as well as hormone-independent carcinogenic pathways, such as the induction of carcinogenesis, mutagenesis and DNA damage by acetaldehyde, a toxic ethanol metabolite; reactive oxygen species and an effect on one carbon metabolism, which is involved with folate level and vitamin B intake.20–24

Although the results of a meta-analysis5 supported both pathways by showing a positive association for estrogen receptor-positive (ER+) progesterone receptor-positive (PR+) and ER+PR− tumors, the summarized results appeared to differ across Western and Asian populations, likely due to the availability of only 2 hospital-based case-control studies in Asia.25, 26

Epidemiological studies have evaluated whether the impact of alcohol intake on breast cancer risk may vary by folate intake27–30 or use of exogenous hormone replacement therapy.31–33 To our knowledge, however, no large prospective study has investigated whether the association between alcohol intake and breast cancer risk is modified by the intake of dietary isoflavones, which are plant-derived estrogen-like compounds with estrogenic and/or anti-estrogenic activity,34 although these agents have attracted attention as substitutes for exogenous hormones.

In addition, alcohol-induced facial flushing due to acetaldehyde accumulation is more prevalent among oriental than Western populations,35 and is considered a predictor of aldehyde dehydrogenase 2 (ALDH2)-deficiency,36, 37 since ALDH2 activity is responsible for detoxifying acetaldehyde in ethanol metabolism. It has been hypothesized that the effect of alcohol on carcinogenesis could be modified by genetic variation in ALDH2 in esophageal cancer,38 but the presence of such an effect in breast cancer remains unclear.39

Here, we prospectively evaluated the association between alcohol consumption and breast cancer risk in consideration of the ER/PR status of tumors, and investigated whether the association is modified by menopausal status, use of exogenous estrogens, the intake of isoflavones and folate, body weight, alcohol-induced facial flushing and smoking status among 50,757 Japanese women in the Japan Public Health Center-based Prospective Study (JPHC Study).

ALDH2: aldehyde dehydrogenase 2; BMI: body mass index; CIs: confidence intervals; ER: estrogen receptor; FFQ: food frequency questionnaire; OC: oral contraceptives; PMH: postmenopausal hormones; PR: progesterone receptor; RE: risk estimate; SD: standard deviation

Materials and methods

Study population

The JPHC Study was launched in 1990–1994, to evaluate the association between lifestyle factors and cancer and cardiovascular disease in a Japanese population, as described previously.40 Conduct of the study was approved by the Institutional Review Board of the National Cancer Center, Tokyo, Japan. In brief, the target population was all Japanese residents aged 40–69 years enrolled with the residential registries of areas served by 11 public health centers. Subjects were categorized by area into 2 cohorts, namely Cohort I (Iwate-Ninohe, Akita-Yokote, Nagano-Saku, Okinawa-Chubu, Tokyo-Katsushikaku) and Cohort II (Ibaraki-Mito, Nigata-Nagaoka, Kochi-Chuohigashi, Nagasaki-Kamigoto, Okinawa-Miyako, Osaka-Suita) at the baseline survey. Initially, 140,420 subjects were invited to join the JPHC cohort, including 71,698 female subjects. Data from one public health center (Tokyo-Katsushika) were not included in the present analysis (n = 4,178) because information on cancer incidence was not available.

Of invitees, 55,907 women completed a baseline questionnaire, giving a response rate of 82.8%. In 1995–1998, we sent a new questionnaire for the 5-year follow-up survey to all eligible cohort members (response rate 79.4%). Further follow-up questionnaire for the 10-year follow-up survey gave a response rate of 77.4%.

We excluded ineligible subjects, including women who moved before the start of follow-up or women with missing information on the date of moving out of the study area or date of death (n = 69); those with a self-reported history of cancer before the start of follow-up (n = 1,509) and those without information on alcohol intake at baseline (n = 1,856). Further, we excluded women without information on weight, height, smoking, leisure-time physical activity or who reported unreasonably high or low estimates of total energy intake (±3 SD) (n = 1,716). Finally, 50,757 were included in the analyses.

Exposure measurement

Alcohol consumption was assessed using a self-administered food frequency questionnaire (FFQ) and calculated as previously described.41, 42

At baseline survey, 2 validated FFQs were used in Cohort I43 and II.41 In Cohort I, subjects were asked the average frequency of intake in 6 frequency categories (almost never, 1–3 days per month, 1–2 days per week, 3–4 days per week, 5–6 days per week and every day). In Cohort II, we first classified alcohol drinking status (never-drinkers, ex-drinkers and current drinkers) and then asked about frequency among ex-drinkers or current drinkers (1–3 days per month, 1–2 days per week, 3–4 days per week and almost daily). For the calculation of alcohol consumption, we assigned a score to each category of frequency, namely 1.5 for 1–2 days per week, 3.5 for 3–4 days per week, 5.5 for 5–6 days per week and 7 for every day in Cohort I at baseline survey. In Cohort II, scores were assigned as 1.5 for 1–2 days per week, 3.5 for 3–4 days per week and 6 for almost daily. Types of alcoholic beverages in the questionnaires and the amount of ethanol used in the calculations were as follows: sake (23 g of ethanol/180 ml), shochu or awamori (36 g of ethanol/180 ml), beer (23 g of ethanol/633 ml), whiskey or brandy (10 g of ethanol/30 ml) and wine (6 g of ethanol/60 ml). We also asked about average daily consumption for each type of alcoholic beverage.

At the 5-year and 10-year follow-up surveys, alcohol drinking was evaluated by a similar validated FFQ, which included 6 frequency categories (i.e., almost never, 1–3 days per month, 1–2 days per week, 3–4 days per week, 5–6 days per week and daily) as well as questions about the average daily consumption of each of 5 types of alcohol, namely sake, shochu or awamori, beer, whiskey and wine. Among women who drank at least 1 day per week, weekly ethanol consumption was quantitatively assessed by multiplying the amount of ethanol by the frequency of alcohol drinking.41 In the calculations of alcohol consumption, the score for each category of frequency and the amount of ethanol in each type of alcohol beverage was applied in the same way as in baseline survey.

Validity of this FFQ-based estimated alcohol consumption was evaluated in a subsample of the JPHC cohort who completed dietary records (DRs). Spearman rank correlation coefficients between the FFQs and DRs were 0.44 (107 women) in Cohort I43 and 0.40 (178 women) in Cohort II for the baseline survey,41 and 0.51 (113 women) in Cohort I44 and 0.48 (176 women) in Cohort II45 for the 5-year follow-up survey.

Dietary assessment was also done using validated FFQs, which included the 44 or 52 food items in the baseline survey46 and 138 food items in the 5-year follow-up survey.44 Subjects were asked to report the average frequency of consumption of each food item during the preceding 1 year. For seasonal fruits and vegetables products, we asked about the frequency of consumption when they were in season and calculated total consumption by taking into account seasonal length. The estimated nutrient intakes were calculated based on the fourth47 and the fifth revised editions of the Standard Tables of Food Composition in Japan.48 Nutritional covariates, except alcohol intake, were adjusted for total caloric intake using the residual method.49 The questionnaires were also used to obtain information on anthropometric and reproductive characteristics, and other lifestyle-related factors, such as physical activity, smoking status and menopausal status.

Ascertainment of breast cancer cases and follow-up of the cohort

Incident cases of breast cancer were identified through linkage with the JPHC Cancer Registry, which is administered by the National Cancer Center (NCC) in collaboration with major local hospitals and the Regional Cancer Registry in each study area. Cases of breast cancer were defined as codes C500–509 according to the Third Edition of the International Classification of Diseases for Oncology.50 Ten breast cancer cases (1.7% of total cases) were identified through information on death certificates (i.e., Death Certificate Notification, DCN), of which 6 (1.0% of total cases) had no information on diagnosis (i.e., Death Certificate Only; DCO). Diagnosis was microscopically verified in 97% of all case patients. ER and PR status were evaluated either by enzyme-linked immunoassay or immunohistochemical assay. Receptor status (i.e., positive or negative) was determined using either the cut-off point for the assay method at the clinical laboratory performing the assay or by clinical estimation at the hospital treating the case patient.

Date of death during follow-up was verified through linkage with the Death Registries at the PHCS, which are required by the Ministry of Health, Labor and Welfare. Date of migration from the study area was verified through linkage with the Residence Registries at the regional PHCs in Japan. Follow-up started on the date of administration of the baseline questionnaire, and subjects were censored on the date of death, date of migration out of the study area or end of the follow-up (December 31, 2006), whichever occurred first.

Statistical analysis

To reduce misclassification of exposure and improve statistical efficiency, we used time-dependent multivariable Cox proportional hazards regression models to estimate relative risks (RRs) and 95% confidence intervals (CIs) using age as the time scale.51 The proportional hazards assumptions were satisfied as evaluated by Kaplan–Meier curves.52 Participants were subdivided into 6 categories by frequency of alcohol drinking [non-drinkers (past drinkers); non-drinkers (never-drinkers); occasional drinkers (i.e., 1–3 days/month); regular drinkers (1–2 times/week); regular drinkers (3–4 times/week) and regular drinkers (≥5 times/week)]. For the amount of alcohol consumption, subjects were classified into 5 groups [i.e., non-drinkers (past drinkers); non-drinkers (never-drinkers); occasional drinkers (non-quantitative estimation); regular drinkers (≤150 g ethanol/week) and regular drinkers (>150 g ethanol/week)]. The multivariable adjusted model included height (continuous, cm), BMI (continuous, kg/m²), smoking status (never, ever) and leisure-time physical activity (no, 1–3 days/month, 1–2 days/week, 3–4 days/week, every day). We also adjusted for age at menarche (≤13, 14, 15, ≥16 years, missing), age at first birth (nulliparous, <26, ≥26 years, missing), parity (nulliparous, 1–2, 3, ≥4 children, missing), menopausal status or age (premenopausal, <48, 48–53, ≥54 years, missing), use of exogenous hormones (never, ever, missing) and energy-adjusted isoflavone intake (quintiles). Trend tests were conducted by creating a continuous variable in order of the rank of each alcohol drinking category and by including it in the regression model among all subjects except past drinkers.

We also calculated the population attributable fraction (PAF) of overall breast cancer for regular drinkers (>150 g ethanol/week) in this population. PAF was calculated as pd×{RR-1}/RR; where pd = proportion of exposed cases,53 95% CI of adjusted PAF was estimated according to the formula of Greenland.54

Among 37,681 women with quantitative information on alcohol consumption, we performed analyses stratified by menopausal status (pre or post), use of exogenous estrogens (never, ever), intakes of dietary isoflavones and folate [upper or lower intake; approximate median (30.5 mg/day for isoflavones; 351 mg/day for folate)], BMI (<25 or ≥25 kg/m²), alcohol-induced facial flushing (yes or no), and smoking status (never, ever) and assessed possible interactions of alcohol intake with these factors in relation to breast cancer.

The cross-product terms of these factors and alcohol consumption (continuous) were introduced into the Cox proportional hazards regression model. The p-value for interaction was calculated by a likelihood ratio test which compared models with and without the interaction terms.

All analyses were performed using the PROC PHREG procedure of the SAS statistical package version 9.1 (SAS Institute, Cary, NC). All statistical tests were 2-sided, and statistical significance was defined as p < 0.05.

Results

After an average 13.8 years of follow-up, corresponding to 698,081 person-years, 572 cases of invasive breast cancer were diagnosed among 50,757 women. Mean age at recruitment was 52 years, while mean age at diagnosis was 50 years. Among the 572 cases, information on ER status was available for 275 (48% of total cases) and on PR for 262 (46% of total cases). Of these, 176 cases were ER+ (64% of all known ER cases) and 99 were ER−, whereas 129 cases were PR+ and 133 were PR−.

At baseline, ∼79% of women were non-drinkers and 12.5% were regular drinkers (Table 1). During a long follow-up period (an average 13.8 years of follow-up), alcohol drinking habit in this population was slightly changed over time.

Table 1. Distribution of alcohol consumption and age-standardized¹ characteristics according to category of alcohol consumption among 50,757 women in the Japan Public Health Center-based Prospective Study, Cohort I (1990-) and Cohort II (1993-)

In present study, 85.2% of the cohort had a repeat measure at the 5-year follow-up survey. Of these women, 6.4% had stopped drinking, 7.9% had started drinking and 79.7% stayed in the same category as at baseline (Spearman rank correlation coefficient among women with quantitative information; r = 0.89); 82.9% of the cohort had information at the 10-year follow-up survey. Of these, 5.9% had stopped drinking, 7.9% had started drinking and 77% stayed in the same category as at baseline survey (Spearman rank correlation coefficient; r = 0.74).

Although height and relative body weight did not appear to differ by level of alcohol consumption, women with high alcohol consumption were more likely to be younger and ever-smokers and less likely to have a facial flushing response on alcohol drinking compared to non-drinkers. At baseline, alcohol consumption was inversely correlated with the intake of isoflavones (r = −0.11) and folate (r = −0.07) among women with quantitative information on alcohol.

In the present cohort, distribution of alcoholic beverages by type was sake 36%, shochu or awamori 30%, beer 26%, whiskey 8% and others 0.2% at baseline survey; sake 33%, shochu 36%, beer 25%, whiskey 5% and wine 1% at the 5-year follow-up survey and sake 29%, shouchu 39%, beer 26%, whiskey 4% and wine 1% at the 10-year follow-up survey.

Increased frequency of alcohol drinking was marginally positively associated with the development of breast cancer. Compared with never-drinkers, women who drank with a frequency of ≥5 times/week had an ∼56% higher risk of breast cancer, with a multivariable-adjusted RR of 1.56 (95% CI = 1.09–2.23; Table 2).

Table 2. Relative risks (RRs) and 95% confidence intervals (Cls) for the association between alcohol consumption and breast cancer risk over 698,081 person-years among 50,757 women in the Japan Public Health Center-based Prospective Study, 1990–2006

The amount of alcohol consumption was also associated with an elevated risk of overall breast cancer. Regular drinkers (>150 g ethanol/week; = ∼2 drinks/day) had an ∼75% higher risk of overall breast cancer than never-drinkers (95% CI = 16–165%; p_trend = 0.035; Table 2). The corresponding risk estimate was 1.78 (95% CI = 1.09–2.90) among premenopausal and 1.21 (95% CI = 0.53–2.75; p_trend <0.05) among postmenopausal women (text only). The PAF of overall breast cancer incidence that was attributable to regular drinkers (>150 g ethanol/week) was 1.93% (95% CI = 1.76–2.12).

Among the 37,681 women with quantitative information on alcohol consumption, an increase in consumption of 10 g of ethanol/day (continuous) was associated with a 6% (95% CI = 1–13; p_trend = 0.047) increase in risk of overall breast cancer after adjustment for all covariates. Because we did not have complete information on past drinkers in Cohort I at baseline, a further analysis was conducted among regular drinkers only. Results showed a non-statistically significant 8% increase in risk (95% CI = −4 to 18; p_trend = 0.24: text only).

In consideration of the ER/PR status of breast tumors, the corresponding results were 1.08 (95% CI = 0.99–1.18; p_trend = 0.098) for all ER+; 0.88 (95% CI = 0.57–1.37; p_trend = 0.58) for all ER−; 1.06 (95% CI = 0.95–1.19; p_trend = 0.28) for ER+PR+; 1.17 (95% CI = 0.93–1.48; p_trend = 0.17) for ER+PR−; 0.89 (95% CI = 0.53–1.49; p_trend = 0.66) for ER−PR− and 1.07 (95% CI = 0.99–1.15; p_trend = 0.11) for unknown ER/PR tumors (text only). Although we did not observe an increased risk for the development of all ER− tumors and ER−PR− tumors among regular drinkers compared with non-drinkers, past drinkers had a statistically significant increased risk over never-drinkers; multivariable-adjusted RRs = 2.39 (95% CI = 1.42–4.05) for all ER− tumors, and 2.08 (95% CI = 1.12–3.87) for ER−PR− tumors (Table 2).

In the analyses stratified by menopausal status, the observed risk estimates were not statistically heterogeneous across menopausal status (p_interaction = 0.46 for overall; Table 3). Therefore, further analyses stratified by other factors were performed without stratification by menopausal status and were conducted for overall and all ER+ tumors.

Table 3. Multivariable relative risks (RRs)¹ and 95% confidence intervals (CIs) for the association between an increase in consumption of 10 g of ethanol/day and breast cancer risk with stratification by several epidemiological factors among 37,681 women² in the Japan Public Health Center-based Prospective Study, 1990–2006

No evidence was seen for statistical interaction between alcohol intake and the use of exogenous estrogens in relation to breast cancer risk among postmenopausal women (p_interaction = 0.24 for overall; 0.87 for ER+; Table 3).

There was no statistical significant interaction between isoflavone intake (< or ≥30.5 mg/day) and alcohol drinking in relation to breast cancer risk (p_interaction = 0.15 for overall; 0.23 for ER+; Table 3).

There was no evidence for effect modification by folate intake for overall (p_interactions = 0.25) or for all ER+ tumors (p_interactions = 0.60), although there was a statistically significant positive association among women with low folate intake (<351 mg/day); the multivariable-adjusted RR for an increase in consumption of 10 g of ethanol/day was 1.08 (95% CI = 1.02–1.16; p_trend 0.012) for overall.

The observed association between alcohol intake and breast cancer risk was not modified with regard to BMI (<25, ≥25) (p_interaction = 0.68 for overall; 0.96 for ER+), alcohol-induced facial flushing (p_interaction = 0.29 for overall; 0.45 for ER+) or smoking status (p_interaction = 0.39 for overall; 0.16 for ER+). Additional stratified analyses with changes to the cut-off point of BMI for Asian population (e.g., 22, 23 or 24)55 produced similar results to those in Table 3 (data not shown).

Discussion

In this large population-based prospective cohort study among Japanese women, we found a statistically significantly positive association between alcohol consumption and the development of breast cancer, as has also been reported in Western populations. The observed positive association was not modified by menopausal status, use of exogenous estrogens, intake of dietary isoflavones and folate, BMI level, alcohol-induced facial flushing or smoking status.

Similar to Western populations, alcohol consumption was positively associated with the development of breast cancer among a Japanese population, although the types of alcoholic beverages consumed by this cohort were largely different from those in Western populations. We found a 6% increase in the risk of breast cancer per 10 g increment of ethanol/day among this Japanese population after adjustment for all covariates. The corresponding results in Western populations reported a statistically significant 9%56 and 12%57 increase in the risk in a pooled analysis of 7 cohort studies56 and the Million Women Study,57 respectively. In the category of regular drinkers >150 g/week, the PAF of overall breast cancer incidence in this population (1.9%) was relatively lower than that in Western population as (∼2.6%).56 This may be partly due to the lower prevalence of regular drinkers in this population.

The biological plausibility of our main result has been supported by a number of proposed mechanisms of ethanol-associated breast carcinogenesis at the molecular, hormonal and other carcinogenic pathway levels.23 Although these mechanisms have not been precisely determined, the involvement of ER-mediated14, 15, 58 estrogen-dependent16–19 mechanisms, and hormone-independent mechanisms20–23 has been suggested.

For ER+ tumors, our results suggested no statistically significant association between alcohol intake and the development of breast cancer, albeit that a weak positive trend was indicated. In the current study, 54% of breast cancer cases did not have information on the ER/PR status of breast tumors and the results of unknown tumors showed a statistically significant positive association in the highest category. The observed null association for ER+ tumors should be carefully evaluated by further follow-up.

Our result for ER− tumors among past drinkers did not eliminate the possible involvement of hormone-independent mechanisms, including carcinogenic effects by acetaldehyde, the most toxic metabolite of ethanol20; and chromosomal21 or DNA damage22 due to alcohol-induced reactive oxygen species and/or lipid peroxidation,23 as previously suggested.28

Consistent with the statement of the Second Expert Report from the World Cancer Research Fund,3 our results suggest that the association between alcohol intake and breast cancer risk is not heterogeneous across menopausal status.

In line with the results of the European Prospective Investigation cohort,59 we observed no evidence of effect modification by ever-use of exogenous estrogens among postmenopausal women. However, effect modification has been suggested by several previous studies,31–33 which had a relatively high prevalence of exogenous estrogen ever-users (∼20–45%),31–33versus our 11% at baseline. Further investigation of this effect is warranted.

To our knowledge, this is the first large prospective study to investigate the interaction between alcohol intake and intake of dietary isoflavones in relation to ER-defined breast cancer risk. We found no evidence of statistical interaction between alcohol consumption and isoflavone intake in relation to overall, ER+ or ER+PR+ breast cancer risk, at least with regard to dietary sources of isoflavones from natural food products within the range of the general Japanese diet. A recent randomized trial60 suggested that isoflavone supplementation did not modify breast density, and did not act in a hormone replacement therapy-like manner.

Several27, 61–64 studies have suggested the importance of adequate dietary folate intake for the prevention of breast cancer, particularly among heavy drinkers, although we did not found any effect modification by folate intake as well as other previous studies.30, 59, 65 Further investigation for evaluation of effect modification is required, particularly among ER− tumors, because although the hypothesis includes ER− tumors, the present study was unable to evaluate this possibility due to the lack of any clear association between quantitative alcohol consumption and the development of ER− tumors.

Consistent with previous studies, the observed positive association between alcohol intake and overall breast cancer risk was not modified by BMI.32, 59 Our results for ER+ tumors in overweight women (BMI ≥ 25) do not support a previous report that the positive association between alcohol consumption and the development of ER+ tumors was confined to women with BMI <25.66

Our results showed that the effect of alcohol on breast carcinogenesis was not modified by alcohol-related facial flushing, despite the likely role of this response as a predictor of an enzyme deficiency of aldehyde dehydrogenase 2 (ALDH2),36, 37 which is responsible for detoxifying acetaldehyde in ethanol metabolism. In this regard, a previous study reported no effect modification by genetic polymorphism of ALDH2.39

Consistent with a previous study,4 we found little evidence of a statistical interaction between smoking status and alcohol intake on the development of breast cancer, although we previously reported a substantial association between smoking and premenopausal breast cancer risk in our cohort.67

The major strength of our study is its prospective population-based cohort design. Our large sample size and repeated measurement of exposure information over a long follow-up period likely aided the precision of our risk estimate. Differential recall bias was ruled out because exposure information was collected before diagnosis. Low percentages of DCN and DCO suggested that misclassification of disease was unlikely, and if present would have tended to be non-differential between cases and non-cases, which would have attenuated the observed risk toward null.

Several limitations of the study also warrant mention. Because alcohol intake and other epidemiological information were estimated on the basis of self-reported questionnaires, a degree of measurement error was inevitable. The misclassification of alcohol consumption and lack of quantitative evaluation of alcohol intake among occasional drinkers in the questionnaire might have lead to fewer drinkers being classified as having a moderate level of alcohol consumption. In consequence, the impact of a moderate level of alcohol consumption on breast cancer risk could not be clearly evaluated, unlike the case of previous studies.57 However, our FFQ-based estimates of alcohol consumption have been validated repeatedly.43, 44 The study population in the current study tended to have a low prevalence of alcohol drinkers and this weakened the statistical power to detect the association. Nevertheless, we found a statistically significant positive association between excessive alcohol intake (>150 g ethanol/week) and an increased risk of breast cancer. We could not rule out the possibility of uncontrolled confounding. For instance, our data provided no information on the duration or type of exogenous estrogens used. However, given the internal consistency of our results across different stratifications, residual confounding may be unlikely. Nevertheless, our subgroup analyses showed largely null findings. The lack of effect modification by the use of exogenous estrogens or other factors may be explained by the lack of statistical power in the present study due to the small number of cases. Further, our results may have been affected by selection bias due to unknown receptor status of approximately half of the cases or by chance due to the large number of subgroup analyses. Although our results for ER/PR unknown tumors showed a similar positive trend to the result of overall and ER+ tumors, our results, particularly for ER-negative tumors among past drinkers, should be carefully interpreted.

In conclusion, this population-based prospective study found that alcohol consumption was positively associated with an increased risk of breast cancer in a Japanese population, as it is among Western populations. From a public health point of view, our findings are important because extreme alcohol drinking is avoidable. The impact of alcohol on breast cancer risk was not modified by menopausal status, exogenous estrogen use, intake of dietary isoflavones or folate, BMI, alcohol-induced facial flushing or smoking status. With regard to isoflavones, however, our present data took no account of supplement use, but rather only the intake obtainable from natural food products in the general Japanese diet. Further research needs to determine the generalizability of our results to other populations, with particular attention to the supplementation of isoflavones as a substitute for exogenous hormones, and to clarify the mechanisms of alcohol-mediated carcinogenesis in consideration of tumor receptor status as well as genetic variation.

Acknowledgements

The authors thank all staff members in each study area and in the central offices for their cooperation and technical assistance. They also thank the Iwate, Aomori, Ibaraki, Niigata, Osaka, Kochi, Nagasaki and Okinawa Cancer Registries for their provision of incidence data. R.S. received a research resident fellowship from the Foundation for Promotion of Cancer Research (Japan) for the 3rd term Comprehensive 10-year Strategy for Cancer Control.