Znt. J . Cancer: 48, 812-815 (1991) Publication of the International Union Against Cancer Publication de I‘Union lnternationale Contre le Cancer 0 1991 Wiley-Liss, Inc.

ALCOHOL CONSUMPTION AND RISK OF BREAST CANCER M.J. SNEYD, Charlotte PAUL’, G.F.S. SPEARS and D.C.G. SKEGG Department of Preventive and Social Medicine and the Hugh Adam Cancer Epidemiology Unit, University of Otago, Dunedin, New Zealand.

In a national case-control study, 891 New Zealand women aged 25 to 54 with newly diagnosed breast cancer were com- pared with 1,864 control subjects selected at random from the electoral rolls. The relative risk of breast cancer for current drinkers of alcohol, compared with women who had never drunk alcohol, was 1.0 (95% confidence interval 0.64 to 1.7). For ex-drinkers the relative risk was 1.3 (95% confidence in- terval 0.74 to 2.5). Women drinking up to 14 drinks per week had no increase in risk, while the relative risk in those con- suming more than 14 drinks per week was 1.8 (95% confi- dence interval 0.87 to 3.8). There was no evidence of effect modification by age at diagnosis, menopausal status, body mass index, or any of the other variables examined. While these results provide little support for the hypothesis that moderate alcohol consumption increases the risk of breast cancer, they are not inconsistent with the weak associations that have been found in many other studies. Possible expla- nations for such a relationship are considered.

There is strong epidemiological evidence linking alcohol consumption with an increased risk of cancers of the mouth, pharynx, larynx, and oesophagus (Doll and Peto, 1981), but the association between alcohol and breast cancer remains con- troversial. Breast cancer is a major cause of mortality in most developed countries, in which alcohol consumption tends to be common. It is therefore important to clarify any relationship between alcohol and breast cancer.

In the first study linking alcohol and breast cancer, Williams and Horm (1977) found that women with breast cancer drank more alcohol than patients with other malignancies. More than 20 cohort and case-control studies have subsequently evaluated the relationship between alcohol and breast cancer (Long- necker et al., 1988; Willett et al., 1989). Four cohort studies showed a positive association (Hiatt and Bawol, 1984; Hiatt et al., 1988; Schatzkin et al., 1987; Willett et al., 1987), but the Framingham study was negative (Schatzkin et al., 1989). The large number of case-control studies have shown less consis- tency, although a meta-analysis supported a weak dose- response relationship (Longnecker et al., 1988). In most of these studies controls were selected from women admitted to hospital with other diseases, and it has been suggested that such women might drink less than the population in general (Anonymous, 1985).

We studied the association between alcohol and breast can- cer in a large, national case-control study in which controls were selected from the population.

SUBJECTS AND METHODS

In this population-based study, all New Zealand women aged 25 to 54 in whom breast cancer had been diagnosed were considered for inclusion as cases; control subjects were se- lected at random from the electoral rolls. The study population was confined to women whose names were in a current elec- toral roll and whose telephone number could be found.

Details of the methods used to select case and control sub- jects have been described elsewhere (Paul et al., 1990). Briefly, the National Cancer Registry (or, in the Auckland region, the Auckland Breast Cancer Study Group) identified 1,452 women aged 25 to 54 in whom a diagnosis of breast cancer was confirmed histologically between 1 July 1983 and

30 June 1987. We excluded women who had had breast cancer diagnosed previously, those whose names were not in a current electoral roll, and those whose telephone number could not be found. After these exclusions 1,126 cases remained. We also required that patients be interviewed between 4 and 8 months after diagnosis, extended in the 3rd and 4th years to 12 months after diagnosis. Interviews were completed with 901 (95%) of the 944 suitable women approached. The other 182 potentially eligible cases were not interviewed because of death (17), be- cause they were too ill (8), or were identified too late (87), or because the doctor refused permission (51), and for miscella- neous reasons (19). Most interviews (92%) were conducted within 8 months of diagnosis (mean 6.0 months). After inter- view, a further 10 subjects were excluded because they had had breast cancer diagnosed previously, or because the final diag- nosis was not breast cancer. Thus we finally studied 891 cases. Control women were selected randomly from the electoral rolls, and women whose telephone numbers could not be found were excluded. As age was not specified in the electoral rolls, we had to choose more women than were required for the study. We wrote to 4,725 women between 1 November 1983 and 31 October 1987, and information regarding age was ob- tained from all except 65. A total of 1,898 were found to be outside the age range of 25 to 55. We randomly excluded half of those under age 35 (560 women) to approximate more closely the age distribution of the cases. Interviews were com- pleted with 1,915 (90%) of the 2,134 suitable women ap- proached. A further 133 potentially eligible control subjects were not interviewed because they had died (9), or were too ill (8), were not traced or were overseas (94), or had language difficulties (22). About half of the untraced women would have been outside the age range for the study. For each control a “reference date” was calculated by subtracting 6 months from the date of interview to correspond to the mean time between diagnosis and interview in the cases. After their interview 51 women were excluded because they had a history of breast cancer or because they were outside the age range (25 to 54) on the reference date. The final control group thus comprised 1,864 women.

Data collection Information was obtained using identical methods for cases

and controls. Women were interviewed by telephone after an initial approach by letter. All interviews were conducted by 2 nurse interviewers or one of us (C.P.). Most began with the interviewer not knowing whether the subject was a case or a control. Although cases generally reported their illness during the interview, this usually occurred after the alcohol history had been taken.

Information on alcohol consumption was collected by asking women whether they had ever drunk alcohol, and if so, whether they had drunk alcohol in the previous 5 years. Those

ITo whom correspondence and reprint requests should be addressed, at Department of Preventive and Social Medicine, University of Otago Med- ical School, Dunedin, New Zealand.

Received: January 5 , 1991 and in revised form March 22, 1991.

ALCOHOL A N D BREAST CANCER 813

women reporting alcohol consumption within the last 5 years were categorized as current drinkers; those who reported alco- hol consumption only before the last 5 years were categorized as ex-drinkers. Subjects who were current drinkers were asked about their average frequency of alcohol consumption (number of days per week) over the previous 5 years. Those reporting that they consumed alcohol once a week or more frequently, were asked about the number of glasses of beer, sherry, wine or spirits they consumed on those occasions. For each woman, we then calculated the average number of drinks per week, on the basis of the near equivalence of alcohol content among the standard servings of beer, sherry, wine and spirits (O’Hagan et al., 1982). Statistical analyses

The effect of alcohol consumption on breast cancer risk was estimated by calculating odds ratios, adjusted for age in 5-year groups, by the method of Mantel and Haenszel (1959). Con- fidence intervals were estimated by the procedure of Cornfield (1956), as programmed by Thomas (1975). When simulta- neously adjusting for several risk factors we applied the mul- tiple logistic model (Breslow and Day, 1980) using the com- puter program GLIM (Baker and Nelder, 1978).

To address the possibility that any change in the relative risk estimate might be accounted for by an association of alcohol intake with known risk factors for breast cancer, potential con- founding variables were assessed by adding them to the logistic model, one at a time with age. Factors tested in that way were age at menarche, age at first full-term pregnancy, parity, menopausal status, family history of breast cancer, history of benign breast cancer, body mass index, ethnic group, social class, years of education, geographical area, smoking history, hypertension, diabetes, gall-bladder disease, and ever use of oral contraceptives or depot medroxyprogesterone acetate. Factors which were potentially confounding and were included in the final model were age (in 5-year groups), parity, social class, smoking history (never, ex, and current smoker), and age at menarche.

RESULTS

The proportions of cases and controls who were current drinkers were 91% (8141891) and 93% (1729/1864) respec- tively. (Current drinkers were those women who reported any alcohol consumption in the last 5 years.) Three case subjects and 7 controls had missing information on intensity of alcohol use or other confounding variables, so all subsequent analyses were restricted to 888 cases and 1,857 control women. The age-adjusted relative risk of breast cancer for current drinkers compared with never drinkers was 1 . 1 (95% confidence inter- val 0.67 to 1.7). Among ex-drinkers the age-adjusted relative risk was 1.4 (95% confidence interval 0.75 to 2.5). These point estimates were changed only slightly after multivariate analysis adjusting for age, parity, social class, smoking, and age at menarche (Table I).

Women who reported consuming alcohol during the last 5 years were categorized according to their intensity of alcohol consumption, as shown in Table 11. Women drinking up to 14 drinks per week had no increase in risk, whereas those con- suming more than 14 drinks per week had an age-adjusted

relative risk of 2.0 (95% confidence interval 0.95 to 4.0). This point estimate was reduced slightly to 1.8 (95% confidence interval 0.87 to 3.8) after multivariate adjustment for age, parity, social class, smoking, and age at menarche. A test for trend was not significant 0, = 0.37).

We examined the separate effects of sherry, wine, beer, and spirits by calculating the relative risk of breast cancer among women who reported consuming only one type of drink (Table 111). In each case the reference group was never-drinkers. There was no increase in risk associated with consumption of any of these different beverages alone.

To examine whether other factors might modify the associ- ation between alcohol consumption and breast cancer, we per- formed cross-tabulations, within GLIM, of alcohol consump- tion and other variables. There was no consistent association between intensity of alcohol consumption and breast cancer within categories of age at diagnosis, body mass index, or menopausal status, as shown in Table IV. Even when we ex- panded the categories of alcohol consumption, including more than 14 drinks per week, no association was found for any of these variables. Nor was any effect modification apparent within categories of age at first-term pregnancy, family history of breast cancer, parity, smoking, age at menarche, or history of surgery for benign breast disease (data not shown).

DISCUSSION

This study found no overall association between alcohol consumption and the risk of breast cancer, although the point estimate of relative risk in ex-drinkers was slightly elevated. Among current drinkers there was no increased risk of breast cancer in women taking up to 14 drinks per week. The relative risk in women drinking more than 14 drinks per week was estimated to be 1.8 (95% confidence interval 0.87 to 3.8). Since the standard serving of alcoholic drinks in New Zealand contains 7 to 1 1 grams of alcohol (O’Hagan et al., 1982), this corresponds to an average intake of more than 14 grams per day.

In contrast to studies using hospital controls, the possibility of selection bias should have been minimized in this study by selecting cases and controls from the entire population of New Zealand by the same criteria. As reported elsewhere, the gen- eral findings of the study were consistent with knowledge about standard risk factors for breast cancer (Paul et al., 1990). Special steps were also taken to minimize information bias: the cases and controls were not told that the study was concerned with cancer; they were interviewed in an identical way; and in most instances the interviewer did not know whether each woman was a case or a control when the alcohol history was taken. Although differential recall of exposure between cases and controls is always possible in studies of this kind, there was very little publicity in New Zealand about the possible association between alcohol and breast cancer until 1987, by which time the study was almost completed.

Detailed information was available about known risk factors for breast cancer, SO possible sources of confounding could be examined. Adjustment for potential confounding variables pro- duced only slight changes in relative risk estimates. The diet of subjects was not evaluated but, in other studies in which di-

TABLE I - RELATIVE RISK OF BREAST CANCER ACCORDING TO ALCOHOL CONSUMPTION

A,cohol consumption Number of Number of Age-adjusted 95% CI %:$%; 95WCI cases controls relative risk

Never drinkers 29 57 1 .o 1 .o Ex-drinkers 48 75 1.4 (0.75-2.5) 1.3 (0.74-2.5) Current drinkers 81 1 1725 1 . 1 (0.67-1.7) 1 .o (0 .641 .7) ‘Adjusted for age, parity, social class, smoking and age at menarche.

814 SNEYD ET A L .

TABLE II - RELATIVE RISK OF BREAST CANCER ACCORDING TO DIFFERENT LEVELS OF ALCOHOL CONSUMPTION (RELATIVE TO NEVER DRINKERS)

Alcohol consumption Number of Number of Age-adjusted (dnnks per week) cases controls relative risk 95w %%eqt; 95%

Never drinkers 29 57 1 .o 1 .o <1 46 1 926 1.1 (0.70-1.8) 1.1 (0.6% 1.8) 1-3 131 346 0.94 (0.57-1.6) 0.88 (0.53-1.5) 4-7 142 310 0.97 (0.59-1.6) 0.91 (0.54-1.5)

>14 25 28 2.0 (0.95-4.0) 1.8 (0.87-3.8) 8-14 52 115 1 .o (0.5C1.8) 0.91 (0.51-1.6)

'Adjusted for age, parity, social class, smoking and age at menarche.

TABLE I11 - RELATIVE RISK OF BREAST CANCER ACCORDING TO TYPE OF ALCOHOL CONSUMED (RELATIVE TO NEVER DRINKERS)

Alcohol Number of Number of Relative 956 c1 consumotion cases controls risk'

Never drinkers 29 57 1 .o Only sherry 41 85 0.78 (0.42-1.5) Only wine 135 330 0.99 (0.57-1.7) Only beer 38 80 1.0 (0.52-1.9) Onlv stirits 101 250 0.78 (0.44-1.4) 'Adjusted for age, panty, social class, smoking and age at menarche.

etary information was collected, adjustment for diet had no appreciable effect on relative risk estimates (Schatzkin et al., 1987; Willett et al . , 1987).

The information about other risk factors also made it possi- ble to examine for effect modification (interaction). Willett et al. (1989) have reviewed studies that reported higher relative risks for alcohol consumption among younger or pre- menopausal women, or among thin women. We found no ev- idence of effect modification by age at diagnosis, menopausal status, body mass index, or any of the other variables exam- ined. It should also be noted that all of the subjects in this study were under the age of 55 .

Separate analyses of the effects of drinking only sherry, wine, beer or spirits showed no increased risks associated with any of these beverages. The data available were not sufficient

to enable investigation of dose-response relationships for par- ticular beverages. We were also not able to examine any effects of drinking alcohol at specific ages. In one study, an increased risk associated with alcohol intake was found only for those who drank at younger ages (under 30), regardless of current consumption (Harvey et al., 1987). Nor were we able to ex- plore the effect of the duration of alcohol consumption.

Our generally negative findings are consistent with those of the Cancer and Steroid Hormone Study, another population- based case-control study (Chu et al., 1989). Negative results have also been reported from population based studies in a US retirement community (Paganini-Hill and Ross, 1983), in Swe- den and Norway (Adami et al., 1988), and in Canada (Rosen- berg et al., 1990). On the other hand, 4 recent studies that used population controls reported positive findings (O'Connell et al., 1987; Harvey et al., 1987; Rohan and McMichael, 1988; Toniolo et al . , 1989).

Longnecker et al. (1988) conducted a meta-analysis of stud- ies on alcohol and breast cancer that were published up to 1987. They found evidence of a dose-response relationship in the pooled data from both case-control and cohort studies. Using the dose-response curves calculated, the relative risk of breast cancer at an alcohol intake of about 2 drinks daily (rel- ative to non-drinkers) was estimated to be 1.4 (95% confidence interval 1.0 to 1.8) in the case-control data and 1.7 (95% confidence interval 1.4 to 2.2) in the data from cohort studies. Although the results of our study are not persuasive when viewed in isolation, they are quite compatible with these esti-

TABLE IV - RELATIVE RISK OF BREAST CANCER ACCORDING TO LEVEL OF ALCOHOL CONSUMPTION WITHIN CATEGORIES OF OTHER VARIABLES (RELATIVE TO THOSE DRINKING NEVER OR LESS THAN ONCE A WEEK).

NUMBERS OF CASES AND CONTROLS SHOWN IN PARENTHESES

Alcohol consumption (drinks per week) 25-34

Age at diagnosis'

35-44 45-54

Never or < I

3 8 Ex-drinkers

1-7

Never or < I

3 8 Ex-drinkers

1-7

1.0 (39,218) 1 .O (182,398) 1 .O (269,367) 0.66 (18,144) 0.67 (97,304) 0.96 (158,208)

0.76 (3, 19) 1.1 (14, 27) 1.4 (31, 29) 1.1 (4, 23) 0.77 (30, 75) 1.2 (43, 45)

Body mass index* s19 0 19 1-23 0 23 1-27 0 >27.0

1.0 (26, 52) 1.0 (205, 403) 1.0 (158, 313) 1.0 (101, 215)

0.80 ( 5 , 11) 0.83 (34, 71) 0.95 (27, 48) 1.9 (11, 13) 0.40 (1, 6) 1.2 (17, 28) 1.4 (16, 23) 1.4 (14. 18)

0.72 (13, 37) 0.68 (126, 337) 0.85 (95, 205) 1.1 (39, 77)

Premenopausal Menopausal status*

Post-menopausal (artificial)

Post-menopausal (natural)

Never or < I 1.0 (319, 765) 1.0 (71, 84) 1.0 (100, 134) 1-7 0.91 (202, 502) 0.58 (30, 59) 0.57 (41, 95) 3 8 1.0 (54, 112) 0.84 (8, 11) 0.96 (15, 20) Ex-drinkers 1.3 (28, 53) 0.64 (6, 12) 1.5 (14, 10) 'Relative risks adjusted for age (within each age at diagnosis category), parity, social class, smoking and age at menarche.

- 2Relative risks adjusted for age, parity, social class, smoking and age at menarche.

ALCOHOL AND BREAST CANCER 815

mates. There was, however, no evidence of a dose-response relationship in our data.

The technique of meta-analysis has chiefly been used to combine the results of randomized controlled trials, and cau- tion is warranted when it is applied to observational studies with major variations in study design and in potential sources of bias. Since many of the studies were not designed to exam- ine the effect of alcohol, the possibility of publication bias must also be considered (Skegg, 1987). Nevertheless, although the consistency of the published findings has sometimes been exaggerated, chance is not a plausible explanation for the many positive studies.

Attempts have been made to explain the results of some case-control studies by invoking potential sources of bias (such as the use of controls with other diseases), but it would be difficult to account for the results of the cohort studies on the same basis, The fact that positive associations have been found with widely differing study designs argues against selection or information bias as an explanation for the findings.

If it is accepted that the association between alcohol con- sumption and breast cancer is likely to be real, this could be due to confounding or to a causal mechanism. The fact that the

association is weak (with most relative risk estimates close to one) makes confounding a plausible explanation. Adjustments have been made in most studies for known confounding vari- ables but, since the main causes of breast cancer are unknown, unsuspected factors could be associated with both a lifestyle involving alcohol and the risk of developing breast cancer. Nevertheless, while a causal relationship has not been estab- lished, it remains a real possibility.

ACKNOWLEDGEMENTS

This investigation received financial support from the Spe- cial Programme of Research, Development and Research Training in Human Reproduction, World Health Organization, and from the Medical Research Council of New Zealand. We thank Mrs. C. Harris and Mrs. J . Thomson (the interviewers); Mrs. J. Smeijers, Mrs. I. Pairman, and Mr. P. Herbison; Mr. J . Fraser and Miss J. Auld of the National Cancer Registry; Professor R. Kay, Miss B. Mason, and Mrs. L. Neave of the Auckland Breast Cancer Study Group; and the women who participated in the study and the surgeons and general practi- tioners who supported it.

REFERENCES

ADAMI, H.O., LUND, E., BERGSTROM, R. and MEIRIK, O., Cigarette smoking, alcohol consumption and risk of breast cancer in young women. Brit. J. Cancer, 58, 832-837 (1988). ANONYMOUS, Does alcohol cause breast cancer? Lancet, I, 131 1-1312 (1985). BAKER, R.J. and NELDER, J.A., The GLZM system: generalized linear interactive modelling. Numerical Algorithms Group, Oxford (1978). BRESLOW. N.E. and DAY. N.E.. Statistical methods in cancer research. Vol. I. The analysis of cawcontrol studies. IARC Scientific Publication 32, IARC, Lyon (1980). CHU, S.Y., LEE, N.C., WINGO, P.A. and WEBSTER, L.A., Alcohol con- sumption and the risk of breast cancer. Amer. J. Epidemiol., 130,867-877 (1989). CORNFIELD, J., A statistical problem arising from retrospective studies. In: J. Neyman (ed.), Proceedings of the 3rd Berkeley Symposium on mathe- matical statistics and probability, Vol. IV, pp. 135-148, University of California Press, Berkeley (1956). DOLL, R. and PETO, R., The causes of cancer, pp. 1224-1226, Oxford University Press, Oxford (1981). HARVEY, E.B., SCHAIRER, C., BRINTON, L.A., HOOVER, R.N. and FRAU- MENI, J.F., Alcohol consumption and breast cancer. J . nut. Cancer Inst.,

HIATT, R.A. and BAWOL, R.D., Alcoholic beverage consumption and breast-cancer incidence. Amer. J. Epidemiol., 120, 676683 (1984). HIATT, R.A., KLATSKY, A.L. and ARMSTRONG, M.A., Alcohol consump- tion and the risk of breast cancer in a pre-paid health plan. Cancer Res., 48, 2284-2287 (1988). LONGNECKER, M.P., BERLIN, J.A., ORZA, M.J. and CHALMERS, T.C. , A meta-analysis of alcohol consumption in relation to breast cancer. J. Amer. medAss., 260, 652-656 (1988). MANTEL, N. and HAENSZEL, W., Statistical aspects of the analysis of data from retrospective studies of disease. J . nut. Cancer Inst., 22, 719-748 ( 1959). O'CONNELL, D.L., HULKA, B .S . , CHAMBLESS, L.E., WILKINSON, W .E. and DEUBNER, D.C., Cigarette smoking, alcohol consumption, and breast- cancer risk. J. nut. Cancer Inst., 78, 229-234 (1987). O'HAGAN, J., WHITESIDE, E. and BIEDER, L., Handbook on alcoholism

78, 657-661 (1987).

for medical practitioners. Alcoholic Liquor Advisory Council, Wellington (1982). PAGANINI-HILL, A. and ROSS, R.K., Breast cancer and alcohol consump- tion. Lancet, 11, 626627 (1983). PAUL, C., SKEGG, D.C.G. and SPEARS, G.F.S., Oral contraceptives and risk of breast cancer. Int. J. Cancer, 46, 366-373 (1990). ROHAN, T.E. and MCMICHAEL, A.J., Alcohol consumption and risk of breast cancer. Int. J . Cancer, 41, 695-699 (1988). ROSENBERG, L., PALMER, J.R., MILLER, D.R., CLARKE, E.A. and SHA- PIRO, S . , A case-control study of alcoholic beverage consumption and breast cancer. Amer. J. Epidemiol., 131, 6-14 (1990). SCHATZKIN, A., CARTER, C.L., GREEN, S.B., KREGER, B.E., SPLANSKY, G.L., ANDERSON, K.M., HELSEL, W.E. and KANNEL, W.B., Is alcohol consumption related to breast cancer? Results from Framingham Heart Study. J. nut. Cancer Inst., 81, 31-35 (1989). SCHATZKIN, A. , JONES, D.Y., HOOVER, R.N., TAYLOR, P.R., BRINTON, L.A., ZIEGLER, R.G., HARVEY, E B., CARTER, C.L., LICITRA, L.M., DUFOUR, M.C. and LARSON, D.B., Alcohol consumption and breast can- cer in the Epidemiologic Follow-up Study of the first National Health and Nutrition Examination Survey. New Engd. J . Med., 316, 1169-1173 (1987). SKEGG, D.C.G., Alcohol, coffee, fat, and breast cancer. Brit. Med. J . , 295, 1011-1012 (1987). THOMAS, D.G., Exact and asymptotic methods for the combination of 2 X 2 tables. Comput. biomed. Res., 8, 423-446 (1975). TONIOLO, P., RIBOLI, E., F'ROTTA, F., CHARREL, M. and CAPPA, A.P.M., Breast cancer and alcohol consumption: a case-control study in northern Italy. Cancer Res., 49, 5203-5206 (1989). WILLETT, W.C., STAMPFER, M.J. and COLDITZ, G.A., Does alcohol con- sumption influence the risk of developing breast cancer? Two views. In: V.T. De Vita, S. Hellman and S.A. Rosenberg (eds.), Important advances in oncology, pp. 267-281, J.B. Lippincott, Philadelphia (1989). WILLETT, W.C., STAMPFER, M.J., COLDITZ, G.A., ROSNER, B.A., HEN- NEKENS, C.H. and SPEIZER, F.E., Moderate alcohol consumption and the risk of breast cancer. New Engl. J. Med., 316, 1174-1180 (1987). WILLIAMS. R.R. and HORM. J.W.. Association of cancer sites with to- bacco and alcohol consumption and socio-economic status of patients: interview study from the third national cancer survey. J. nut. Cancer Inst., 58, 525-547 (1977).