International Journal of Epidemiology Vol. 26, No.

5
© International Epidemiological Association 1997 Printed in Great Britain

Breast Cancer Risk and Alcohol

Consumption: Results from a Large
Case-Control Study
STEVEN J BOWLIN,*,** M CRISTINA LESKE,** ANDRE VARMA,** PHILIP NASCA,† AURA WEINSTEIN‡
AND LEE CAPLAN§

Bowlin S J (Department of Epidemiology and Biostatistics, School of Medicine, Case Western Reserve University, 10900
Euclid Ave, Cleveland, OH 44106–4945, USA), Leske M C, Varma A, Nasca P, Weinstein A and Caplan L. Breast cancer
risk and alcohol consumption: Results from a large case-control study. International Journal of Epidemiology 1997; 26:
915–923.
Background. Alcohol use is associated with breast cancer in many epidemiological studies. Most, however, have meas-
ured risk from recent consumption patterns, and only a few include analyses for duration of drinking or age that a woman
started to drink. The authors studied the effect of these variables, as well as of recent alcohol consumption patterns, on
breast cancer risk.
Methods. Data from a large case-control study conducted in Long Island, New York from 1 January 1984 to 31 December
1986 were used. A total of 1214 women aged 20–79 years with incident breast cancer were interviewed. A control was
selected for each case from driver’s license files, and matched on age and county of residence. Alcohol consumption was
measured as: ever versus never, grams of alcohol per day, age started drinking, and total years drinking.
Results. After adjustment for breast cancer risk factors, the odds ratio for ever versus never drinking was 1.40 (95%
confidence interval [CI] 1.09–1.79); odds ratios for .0–5 and >5 grams of alcohol use per day, as compared to non-
drinkers, were 1.29 (95% CI : 1.00–1.65) and 1.46 (95% CI : 1.13–1.89), respectively. Age when drinking began was not re-
lated to breast cancer risk, but the greater the total years of drinking, up to 40 years (odds ratio 1.48, 95% CI : 1.13–1.93),
the greater the risk. However, when grams per day and duration of drinking were simultaneously included in the multi-
variate model, duration was not important as a risk factor. This suggests that intensity of drinking may be the important
factor for breast cancer risk. After covariate adjustment, risk from alcohol intake did not differ between pre- and
postmenopausal women.
Conclusions. These data support the belief that alcohol is associated with breast cancer risk.
Keywords: alcohol, breast cancer, case-control study

Intense epidemiological research has been directed at early menarche, late menopause, late age at first
understanding the relationship between breast cancer full-term birth and benign breast disease are difficult or
and alcohol consumption.1 Many reports have found impossible to alter. On the other hand, alcohol use can
an association but others have not.1 Most well-defined be modified and is common among American women.2
risk factors such as age, family history of breast cancer, If its association with breast cancer were causal, alco-
hol use would provide a risk factor to target for primary
prevention of breast cancer.
* Department of Epidemiology and Biostatistics, School of Medicine,
Most epidemiological studies of alcohol and breast
Case Western Reserve University, 10900 Euclid Avenue, Cleveland, cancer have relied on risk assessment from recent con-
OH 44106–4945, USA. sumption patterns.1 The effect of duration of alcohol
** Department of Preventive Medicine, University Medical Center at consumption, or age when starting to drink have
Stony Brook, Stony Brook, NY, USA.
† been less studied. These latter variables may add more
School of Public Health and Health Sciences, University of
Massachusetts, Amherst, MA, USA. important information than recent drinking habits in
‡
State of New York Department of Health, Albany, NY, USA. defining risk and establishing causality. This study
§
National Center for Chronic Disease Prevention and Health investigates the association between alcohol use and
Promotion, Centers for Disease Control and Prevention, Atlanta, GE, breast cancer risk in a large case-control study con-
USA.
At the time this study was completed, Dr Bowlin was with the Depart-
ducted in Long Island, New York. Risk was assessed
ment of Preventive Medicine at the University Medical Center at Stony for recent alcohol use, duration of drinking and age
Brook, Stony Brook, NY, USA. when starting to drink.
915
916 INTERNATIONAL JOURNAL OF EPIDEMIOLOGY

MATERIAL AND METHODS TABLE 1 Alcohol questions used in the telephone interview, Long
Case Selection Island Breast Cancer Study, 1 January 1984 to 31 December
1986
This case-control study included women residing in
Nassau and Suffolk counties in Long Island, New York.
The study methodology is described elsewhere.3 Cases 1. Prior to (date of diagnosis) did you ever drink alcoholic beverages?
2. Age started drinking alcoholic beverages?
were defined as all females with incident breast can- 3. Age stopped drinking alcoholic beverages?
cer, histologically confirmed, in both counties from (if still drinking alcohol—code age at diagnosis)
1 January 1984 to 31 December 1986. Women between 4. Did you usually drink beer, wine or liquor?
20 and 79 years old were eligible. Cases were identified 5. Approximately how often do you drink alcoholic beverages?
a) 5–7 days a week
by surveillance of all Long Island hospitals’ tumour
b) 1–4 days a week
registers and medical records. The New York State c) 1–3 days a month
Cancer Registry was used to locate county residents d) ,1/month
with breast cancers diagnosed elsewhere in New York e) never
State. f) unknown
6. How many drinks do you usually have at a time?
A total of 2140 eligible breast cancer cases were a) 1–2 drinks
identified during that time; 90% were found through b) 3–4 drinks
Long Island hospitals; 1616 cases were contacted, of c) .4 drinks
whom 1420 participated, which represented 88% of the d) unknown
contacted women or 67% of the total identified cases.
In all 524 cases could not be contacted because of
doctor refusal (45%), unable to locate (23%), deceased
(14%), and other (18%). Women who were interviewed and had higher incomes; they also reported more
were younger than those who refused (57.5 years versus alcohol use, benign breast disease, and Jewish religion.4
61.1 years, P , 0.001); participation rates among those These differing characteristics may have distorted the
contacted were similar between Nassau and Suffolk association between alcohol intake and breast cancer in
County; also, participating cases tended to be of higher this study. Therefore, only case-control pairs who each
socioeconomic status (SES) (measured by New York possessed a driver’s license were included, leaving
State Gazetteer coding system) than refusals.4 1214 pairs in the analysis.

Control Selection Data Collection

A control was matched to each case by age (±1 year) Data were collected with a standardized questionnaire
and by county of residence. The driver’s license files of using a telephone interview. Variables included demo-
the New York State Department of Motor Vehicles graphics, a lifetime residential history (with emphasis
were used to construct a list of potential controls. After on environmental exposures), medical care, including
a case was interviewed, we attempted to contact and breast cancer screening practices, reproductive history,
recruit a control from a list of age and county-matched hormone use, family history of breast cancer, and life-
women, in sequence, until a participant was found. style factors. Questions on alcohol exposure included
There were 3487 potential controls identified; whether the respondent had ever used alcohol, age
2097 potential controls were contacted, and 1420 were started and stopped, type drank, frequency, and number
matched with cases for a participation rate of 68% of drinks consumed at a time. Table 1 shows the alcohol
among those contacted or 41% among the total identi- questions asked during the interview. For the goals of
fied. It was not possible to contact 1390 women because this report, risk of breast cancer from alcohol was ex-
of inability to locate (88%), deceased (6%), and other plored comparing ever versus never users, and com-
(6%). Controls who were interviewed were younger paring categories of current alcohol intake, measured in
than those who refused (57.4 years versus 63.5 years, grams per day (g/day). We defined one drink as a shot
P , 0.001); participation rates among those contacted (measure) of liquor (spirits), 12 ounces of beer, or 5–6
were similar between Nassau and Suffolk County. No ounces of wine, with one drink approximately equal to
statistical association was found between participation 12.5 grams of ethanol. Grams of alcohol per day were
and SES.4 calculated as frequency of drinking (based on days)
While the 1420 control women had a driver’s license, × number of drinks consumed per drinking occasion ×
206 cases did not. Comparison between cases with and 12.5 grams of alcohol in one drink. In these analyses,
without a license revealed significant differences be- g/day of alcohol were categorized into 0 g/day, .0–5
tween the groups. Those with a license were younger g/day, and >5 g/day. Five g/day was, therefore, ,1⁄2 of
 BREAST CANCER RISK AND ALCOHOL CONSUMPTION 917

TABLE 2 Unmatched number of cases and controls in each alcohol category by menopausal status, Long Island Breast Cancer Study,
1 January 1984 to 31 December1986

Variable Menopausal status

Prea Posta Combined

Case Control Case Control Case Control

Ever use alcohol:

Ever 248 232 676 642 1063 1008
Never 33 49 98 134 148 206
G/day:
0 35 50 101 137 154 211
.0–5 134 131 343 348 549 550
>5 112 100 331 291 510 453
Age started drinking:b
<17 22 21 173 156 217 188
18–24 178 145 401 381 667 610
25+ 48 66 103 105 181 210
Number years used alcohol:
0 33 49 98 134 148 207
.0–,20 77 78 26 30 120 116
20–.40 170 152 284 260 571 528
40+ 1 2 367 352 374 363
Type of drink:
None 33 49 98 134 148 206
Beer 6 8 23 17 29 31
Wine 117 108 192 205 363 367
Liquor 27 40 198 185 251 251
Combination 98 75 264 235 422 358

a
Sample sizes based on case-control pairs concordant on menopausal status.
b
Among drinkers only.

a drink per day. Risk of breast cancer was also for potential confounders included in the model. Unless
evaluated by type of alcohol drank: beer, wine, liquor otherwise indicated, never drinkers were the reference
or a combination. Women were asked about alcohol group for risk calculations.
intake prior to the date of diagnosis of the case. Variables studied as potential confounders included:
Women were asked at what age they first began to religion (Jewish versus other), years of education (<12,
drink and at what age they stopped. Odds ratios (OR) 13–16, 17+ years), income (median annual family
were calculated to determine whether breast cancer was income from the census tracts of the women divided
associated with the age the women began drinking and into quartiles from lowest to highest—while family in-
with the duration of alcohol use. Table 2 shows the un- come was not studied directly, Krieger6 has shown that
matched number of cases and controls for each alcohol social class variables measured at the census tract are
category, by menopausal status. good proxies for individual-level measurements),
marital status (married, single, widowed/separated/
Statistical Analysis divorced), history of benign breast disease (yes versus
The conditional logistic regression model for matched no), age at menarche (7–11, 12–13, 14+ years), ever
case-control studies used in PROC PHREG in the Stat- pregnant (yes versus no), age at first live birth (never
istical Analysis System (SAS) program was used for pregnant, ,25, 25–29, 30+ years), total weeks of
both univariate and multivariate analyses.5 Univariate breastfeeding her children (0, .0–24, 25–49, 50+
analysis was done by including only one independent weeks), family history of breast cancer in first-degree
variable of interest in the model. Multivariate analysis relative (yes versus no), ever smoke cigarettes (yes
was used to study the association between various versus no), pack-years of cigarette smoking (0, .0–14,
measures of alcohol and breast cancer risk, controlling 15–29, 30+), and Quetelet’s Index (0–20, 21–29,
918 INTERNATIONAL JOURNAL OF EPIDEMIOLOGY

TABLE 3 Odds ratios and 95% confidence intervals for risk fac- TABLE 3 continued
tors and breast cancer, univariate analyses,a Long Island Breast
Cancer Study, 1 January 1984 to 31 December 1986
Risk Factor ORb 95% CIb

Risk factor ORb 95% CIb

Quetelet’s Index
0–20 1.00
Religion: 21–29 0.98 (0.77–1.23)
Others 1.00 30+ 1.10 (0.82–1.48)
Jewish 1.05 (0.85–1.28) Trend P , 0.52
Education in years:
<12 1.00 a
13–16 1.00 (0.83–1.20) The minimum number of pairs used in these analyses due to missing
17+ 0.89 (0.72–1.10) data was 1209 for family history of breast cancer.
b
Trend P , 0.32 OR, odds ratio; CI, confidence interval.
c
Income-quartiles, lowest to highest Statistically significant odds ratios.
1st 1.00
2nd 0.89 (0.71–1.12)
3rd 1.02 (0.79–1.31)
4th 1.11 (0.86–1.44)
Trend P , 0.19
Marital status:
30+ kg/m2). Ordinal variables were tested individually
Married 1.00 for trend in risk across categories in the univariate
Single 1.46 (0.95–2.24) analyses. Models were run for all women together and
Widowed/divorced/separated 1.33 (1.09–1.63)c separately for pre- and postmenopausal women. Ana-
Family history of breast cancer:
lyses showed an interaction between alcohol and county
No 1.00
Yes 1.70 (1.34–2.17)c of residence. Therefore, multivariate analyses were done
History of benign breast disease: separately for Nassau and Suffolk Counties. OR and 95%
No 1.00 CI are presented for both univariate and multivariate
Yes 1.47 (1.24–1.74)c results. Missing data were assigned the median for con-
Age of menarche:
7–11 1.00
tinuous data and the mode for categorical data; ,1.5%
12–13 1.01 (0.82–1.26) of the data were missing for any variable included in
14+ 1.06 (0.83–1.35) the analysis.
Trend P , 0.62
Ever pregnant
No 1.00
Yes 0.76 (0.58–0.99)c RESULTS
Age first live birth Table 3 shows OR and 95% CI from matched univariate
Never pregnant 1.00 analyses for potential breast cancer risk factors other
,25 0.72 (0.54–0.96)c than alcohol variables. Religion, years of education,
25–29 0.77 (0.57–1.04)
30+ 0.89 (0.64–1.24)
household income, age of menarche, menopausal status,
Trend P , 0.81 and Quetelet’s Index had no relationship with breast
Total weeks spent breast feeding: cancer. Single women and those who were widowed/
0 1.00 separated/divorced had higher risk than married
.0–24 0.85 (0.70–1.04)
women, but only the latter association was statistically
25–49 0.59 (0.41–0.87)c
50+ 0.52 (0.36–0.74)c significant. Those with a positive family history of
Trend P , 0.0001 breast cancer (OR = 1.70, 95% CI : 1.34–2.17) and of
Menopausal status: benign breast disease (OR = 1.47, 95% CI : 1.24–1.74)
Pre 1.00 had a higher risk of breast cancer than those who did
Post 0.83 (0.60–1.13)
Ever smoke
not give such a history. Ever pregnant (OR = 0.76, 95%
No 1.00 CI : 0.58–0.99) and having a first live birth before 25
Yes 1.17 (1.00–1.37) years of age (OR = 0.72, 95% CI : 0.54–0.96) were
Pack-years of smoking: associated with lower risk of breast cancer than never
0 1.00
being pregnant. As weeks of breastfeeding increased
.0–14 1.31 (1.06–1.61)c
15–29 1.11 (0.87–1.43) there was a statistically significant decreasing trend in
30+ 1.08 (0.87–1.34) breast cancer risk. Ever smoking cigarettes had a
Trend P , 0.41 statistically marginal increased risk for disease, with a
small OR of 1.17 (95% CI : 1.00–1.37).
 BREAST CANCER RISK AND ALCOHOL CONSUMPTION 919

TABLE 4 Odds ratios and 95% confidence intervals for alcohol variables and breast cancer by menopausal status, univariate analyses,
Long Island Breast Cancer Study, 1 January 1984 to 31 December 1986

Menopausal status

Pre Post Combined

ORa 95% CIa OR 95% CI OR 95% CI

Ever versus never 1.65 (1.02–2.69)e 1.43 (1.08–1.90)e 1.47 (1.17–1.84)e

Use of alcohol (281)b (774) (1211)
G/day
0 1.00 1.00 1.00
.0–5 1.52 (0.91–2.52) 1.31 (0.98–1.74) 1.36 (1.07–1.73)e
>5 1.63 (0.98–2.70) 1.50 (1.11–2.01)e 1.52 (1.19–1.93)e
(281) (774) (1214)
Trend P , 0.11 P , 0.01 P , 0.002
Age started drinkingc
<17 0.65 (0.30–1.41) 0.99 (0.69–1.44) 0.79 (0.58–1.08)
18–24 1.17 (0.60–2.29) 1.03 (0.78–1.36) 1.01 (0.79–1.29)
25+ 1.00 1.00 1.00
(205) (558) (883)
Trend P , 0.08 P , 0.97 P , 0.15
Number years used alcohol
0 1.00 1.00 1.00
.0–,20 1.43 (0.79–2.59) 1.16 (0.64–2.11) 1.45 (1.00–2.12)
20–,40 1.80 (1.06–3.05)e 1.47 (1.07–2.03)e 1.53 (1.19–1.98)e
d
40+ 1.39 (1.03–1.89)e 1.41 (1.07–1.86)e
(281) (776) (1214)
Trend P , 0.03 P , 0.02 P , 0.004
Type of drink
None 1.00 1.00 1.00
Beer 1.17 (0.34–4.05) 1.81 (0.92–3.56) 1.31 (0.75–2.28)
Wine 1.62 (0.97–2.71) 1.26 (0.91–1.74) 1.36 (1.05–1.76)e
Liquor 0.98 (0.49–1.96) 1.46 (1.05–2.04)e 1.39 (1.05–1.85)e
Combination 1.92 (1.12–3.29)e 1.50 (1.10–2.03)e 1.61 (1.25–2.07)e
(281) (776) (1214)

a
OR, odds ratio; CI, confidence interval
b
# of case-control pairs; only pairs concordant on menopausal status were used in the pre- and postmenopausal analyses.
c
Among drinkers only.
d
Three pairs in this group were included in the 20–,40 years category.
e
Statistically significant odds ratios.

Table 4 shows the univariate analyses, stratified by not statistically significant in this group, probably
menopausal status, of breast cancer risk from alcohol. because of small numbers of case-control pairs.
Ever drinkers had a statistically significant increased Other variables examined were the total number of
risk of breast cancer compared to the never drinkers years of alcohol consumed, type of drink, and among
in all the menopausal status groups (OR = 1.47, 95% women who ever drank, age when alcohol was first
CI : 1.17–1.84 for the combined group). Risks were used. Table 4 shows that age at which women started
increased both in premenopausal (OR = 1.65, 95% drinking was not associated with breast cancer. Risk of
CI : 1.02–2.69) and in postmenopausal (OR = 1.43, disease increased with increasing number of years of
95% CI : 1.08–1.90) women. Risk also increased with alcohol consumption, and the test for trend was stat-
increasing g/day of alcohol consumed. In the combined istically significant in all menopausal groups. As
group, the OR was 1.36 (95% CI : 1.07–1.73) for women before, premenopausal women had higher risk than
drinking ,5 g/day, and was 1.52 (95% CI : 1.19–1.93) postmenopausal women with those having 20 to ,40
for those drinking >5 g/day. Risk was higher in pre- years of drinking showing the greatest OR (1.80,
menopausal women, but the OR and tests of trend were 95% CI : 1.06–3.05). No consistent pattern was evident
920 INTERNATIONAL JOURNAL OF EPIDEMIOLOGY

TABLE 5 Odds ratios and 95% confidence intervals for alcohol variables and breast cancer by menopausal status, multivariate analyses,a
Long Island Breast Cancer Study, 1 January 1984 to 31 December 1986

Menopausal status

Pre Post Combined

ORb 95% CIb OR 95% CI OR 95% CI

Ever versus never 1.45 (0.84–2.50) 1.43 (1.06–1.94)c 1.40 (1.09–1.79)c

G/day
0 1.00 1.00 1.00
.0–5 1.26 (0.71–2.22) 1.32 (0.97–1.80) 1.29 (1.00–1.65)
>5 1.54 (0.87–2.74) 1.51 (1.09–2.08)c 1.46 (1.13–1.89)c
Number years used alcohol:
0 1.00 1.00 1.00
.0–,20 1.32 (0.67–2.60) 1.04 (0.56–1.94) 1.33 (0.89–1.98)
20–,40 1.69 (0.94–3.03) 1.57 (1.12–2.21)c 1.48 (1.13–1.93)c
d
40+ 1.37 (0.99–1.90) 1.32 (0.98–1.77)
Type of drink:
None 1.00 1.00 1.00
Beer 1.02 (0.24–4.33) 1.92 (0.95–3.89) 1.28 (0.72–2.29)
Wine 1.46 (0.82–2.61) 1.32 (0.94–1.85) 1.32 (1.00–1.72)
Liquor 0.72 (0.33–1.58) 1.44 (1.01–2.07)c 1.30 (0.96–1.75)
Combo 1.79 (0.97–3.27) 1.52 (1.09–2.12)c 1.56 (1.19–2.04)c

a
A separate conditional logistic regression model was developed for each menopausal group and alcohol variable. All models included, in addition to
the alcohol variable, the following covariates: religion, income status, marital status, family history of breast cancer, history of benign breast disease,
age of menarche, ever pregnant, age first live birth, total weeks spent breastfeeding, and ever smoked.
b
OR, odds ratio; CI, confidence interval.
c
Statistically significant odds ratios.
d
Three pairs in this group were included in the 20–,40 years category.

showing which type of alcoholic beverage increased to the smaller sample size. After covariate adjustment,
risk. All types had elevated OR, with the combination the risk of breast cancer from alcohol intake is not
of beverages generally showing the highest risk. altered by menopausal status. Multivariate analysis for
Table 5 shows the multivariate analysis results for duration of drinking and type of drink and breast cancer
ever drinking, g/day, total years of alcohol consumption, risk was similar as in unvariate analysis.
and type of drink. Separate models were developed for To study how intensity (i.e. g/day) and duration
each alcohol factor and for pre- and postmenopausal (i.e. total years of consumption) of alcohol exposure
women. Included as covariates for each model were may interrelate, we simultaneously included both
religion, income status, marital status, family history of of these factors as ordinal variables in a multivariate
breast cancer, history of benign breast disease, age of model for the combined group of women. Duration of
menarche, ever pregnant, age at first live birth, total drinking was no longer significantly related to breast
weeks breastfeeding, and ever smoked. In the combined cancer risk (P = 0.54). Grams per day was of borderline
and postmenopausal groups, little change in risk from statistical significance (P = 0.08) with OR of 1.15 and
alcohol consumption was seen after controlling for 1.33 for .0 to ,5 and >5 g/day, respectively. Grams
potential confounding; risk was elevated by 40% in per day and total years of drinking, however, were
ever drinkers, 50% in those drinking >5 g/day. Most moderately to strongly correlated (r = 0.60), which may
OR retained their statistical significance, and dose- explain why both factors lost statistical significance
response relationships were seen in g/day. In premeno- when placed in the model together. The data suggest
pausal women, most OR were reduced after controlling that intensity of drinking, and not duration, may be the
for confounding and became similar in magnitude to important risk factor for breast cancer.
those of postmenopausal women. No OR were statistic- The association between alcohol and breast cancer
ally significant in premenopausal women, possibly due risk appeared stronger in Suffolk County than in Nassau
 BREAST CANCER RISK AND ALCOHOL CONSUMPTION 921

TABLE 6 Odds ratios and 95% confidence intervals for alcohol Most reports on breast cancer and alcohol consump-
variables and breast cancer by county of residence, multivariate tion lacked data on duration of drinking, and even data
analyses,a Long Island Breast Cancer Study, 1 January 1984 to
31 December 1986
on age when alcohol use first began. These variables may
add more important information for breast cancer risk
than only recent consumption patterns. In this paper, we
ORb 95% CIb
studied risk by age a woman first began to drink and
duration of drinking.
Nassauc Among the studies which examined the age a woman
Ever versus never 1.16 (0.81–1.65)
G/day:
first started to drink, two found no association between
0 1.00 alcohol intake and breast cancer risk.9,10 Several found
.0–5 1.17 (0.81–1.69) higher risks among those who started drinking at younger
>5 1.19 (0.82–1.73) ages.11–14 Nasca et al. and Holmberg et al. found the
Number years used alcohol:
greatest risk in those who began drinking later in life.15,16
0 1.00
.0–,20 0.99 (0.55–1.79) Several investigators, however, discovered no relation-
20–,40 1.25 (0.85–1.84) ship between age of first starting to drink or early age
40+ 1.10 (0.72–1.67) of alcohol consumption and breast cancer.17–20 Our study
Suffolkc agreed with the latter findings. We cannot rule out, how-
Ever versus never 1.66 (1.16–2.38)d
G/day:
ever, that heavy drinking at an early age is associated
0 1.00 with breast cancer, because we did not measure the
.0–5 1.38 (0.96–1.99) amount of drinking at various ages. The disparity of
>5 1.78 (1.21–2.63)d these studies show further research is needed to under-
Number years used alcohol:
stand how age starting to drink relates to breast cancer.
0 1.00
.0–,20 1.67 (0.93–2.98) A few studies have evaluated the association between
20–,40 1.75 (1.17–2.64)d duration of alcohol use and breast cancer.15,17,19 All
40+ 1.54 (0.99–2.40) failed to show an association. In the present study, the
risk of breast cancer increased as total years of drinking
a
Each model included, in addition to the alcohol variable, the follow- increased; after 40 years, however, no further increase
ing covariates: religion, income status, marital status, family history of was seen. While first age at drinking was not important
breast cancer, history of benign breast disease, age of menarche, ever
for risk, our results on duration of drinking suggested
pregnant, age first live birth, total weeks spent breastfeeding, and ever
smoked. that risk increased as the years of alcohol consumed dur-
b
OR, odds ratio; CI, confidence interval. ing a woman’s life increased. However, when g/day of
c
661 pairs analysed for Nassau County and 545 for Suffolk County. drinking and duration were simultaneously included in
d
Statistically significant odds ratios.
the multivariate model, duration was no longer statistic-
ally significant. This may have occurred because the two
variables were highly correlated (r = 0.60), or because
County. This is seen in Table 6, which shows adjusted intensity of drinking (i.e. g/day) was the important
OR and 95% CI for alcohol measured as ever drinking, contributor to risk, and not duration. Further studies
g/day, and total years of alcohol consumed for each specifically designed to separate the effect of intensity
county separately. and duration of drinking on breast cancer risk are
needed.
We did not expect an interaction of alcohol and
DISCUSSION breast cancer risk by county of residence. Suffolk
There have been several dozen published epidemio- County had higher risks from alcohol consumption than
logical studies of alcohol and breast cancer. Two meta Nassau County. This pattern was also found from this
analyses7,8 and a review1 have recently been published data set in the association of oral contraceptive use and
summarizing these reports. In the present study, risk of breast cancer.21 Nassau and Suffolk County participants
breast cancer was increased by 40–45% for women ever did not differ in their distribution of ever drinking,
drinking alcohol. There was a dose-response relation- g/day, or type of drinks. Intercounty drinking pattern
ship with increasing g/day of alcohol consumed; differences would not explain the county interaction.
women drinking >5 g/day had a 50% increased risk. Other breast cancer risk factors that differ by county are
Five g/day was less than one-half of an alcoholic drink known. Based on control data, Nassau county women
per day, indicating that moderate levels of consumption have a higher educational and income level, higher
increased risk. proportion who are Jewish, higher rates of benign
922 INTERNATIONAL JOURNAL OF EPIDEMIOLOGY

breast disease, and an older age at first live birth than We controlled for the major risk factors for breast
Suffolk county women. How these differences may cancer. Other potential biases in case-control studies
explain why the breast cancer and alcohol association are unmeasured confounders and unmeasurable factors
was stronger in Suffolk than in Nassau County is not contributing to selection bias. We attempted to measure
clear. most known and suspected variables related to breast
Strengths of this study are its size and the measures cancer. While these biases may exist, the possibility
of alcohol consumption used in the analysis. However, was minimized by a careful study design.
biases particular to this study and common to case- Many case-control and cohort studies have found an
control studies may be present. Our participation rate association between alcohol use and breast cancer.
was lower for controls than for cases (see MATERIAL Assuming the relationship is real and causal, an im-
AND METHODS), and we had only a few variables to portant question is how much breast cancer in women
assess fully the selection biases. Participating cases and similar to those in our study is attributable to alcohol
controls were younger than those who did not parti- consumption. This study allows this calculation. We
cipate. The mean difference in age between those who assume an OR of 1.4 for women ever drinking alcohol
were interviewed and those who refused was small (4–6 compared to never drinkers, and a prevalence of ever
years). This difference, while statistically significant, drinking in a population of women similar to our
was probably not large enough to cause a severe selec- sample (estimated from the control group) of 83%.
tion bias, and being non-differential for case-control Applying formulas given by Kelsey,28 results show that
status, did not bias the association of alcohol and breast 25% of breast cancer among these women between 20
cancer. Among those contacted, the participating cases and 79 years old is attributable to ever drinking alcohol.
tended to be of higher SES than those who refused, but Alcohol use is a common exposure and breast cancer is
for controls no differences were seen. Higher SES common among women. If the alcohol-breast cancer
groups tend to use alcohol more often than lower SES hypothesis is correct, many breast cancers would be
groups.2 Alcohol use in the cases may have been greater preventable by reducing alcohol consumption.
compared to controls because of this selection bias
resulting in overestimating the OR. Although statistic-
ally significant, the difference in SES between parti- REFERENCES
1
cipating and non-participating cases was small and Rosenberg L, Metzger L S, Palmer J R. Alcohol consumption
probably only slightly biased the OR. and risk of breast cancer: A review of the epidemiologic
evidence. Epidemiol Rev 1993; 15: 133–44.
While the selection biases for these variables appear 2
National Center for Health Statistics. National Center for Health
to be small, the possibility of biases concerning other Statistics. Advanced data from Vital and Health Statistics.
characteristics cannot be excluded, especially among Vital Health Stat 1993; 16: 111–20.
3
controls. The influence of other potential selection biases The Long Island Breast Cancer Study, Report Number 1. New
is not measurable or known. The results of our study, York State Department of Health, June 1988.
4
The Long Island Breast Cancer Study, Report Number 2. New
however, agree with those of many other studies with
York State Department of Health, April 1990.
high participation rates,13,17,22–26 thus suggesting that 5
SAS Technical Report P-217, SAS/STAT Software: PHREG
these biases had modest effects. Procedure, Version 6, Cary, NC: SAS Institute, 1991,
Cases may have recalled alcohol consumption more 63pp.
6
accurately than controls. Giovannucci et al. however, Krieger N. Overcoming the absence of socioeconomic data
medical records: validation and application of a census-
showed that retrospective data collection and any
based methodology. Am J Public Health 1992; 92: 703–10.
associated recall bias for alcohol consumption had only 7
Longnecker M P, Berlin J A, Orza M J, Chalmers T C. A meta-
minor effects on alcohol reporting and risk estimates, analysis of alcohol consumption in relation to risk of breast
as compared to prospectively collected data.27 Inter- cancer. JAMA 1988; 260: 652–56.
8
viewers were not masked to case-control status, and Howe G, Rohan T, Decarli A et al. The association between
alcohol and breast cancer risk: evidence from the combined
may have differentially assessed alcohol exposure be-
analysis of six dietary case-control studies. Int J Cancer
tween the two groups. However, the alcohol questions 1991; 47: 707–10.
were part of an extensive and lengthy interview 9
Smith S J, Deacon J M, Chilvers C E D. Alcohol, smoking,
measuring many potential risk factors for breast cancer. passive smoking and caffeine in relation to breast cancer
At the time data were collected, the association between risk in young women. Br J Cancer 1994; 70: 112–19.
10
alcohol and breast cancer was not well recognized, and Freudenheim J L, Marshall J R, Graham S et al. Lifetime alcohol
consumption and risk of breast cancer. Nutrit Cancer 1995;
both subjects and interviewers were not aware of the 23: 1–11.
hypothesis of this report. These considerations lessen 11
Harvey E B, Schairer C, Briton L A et al. Alcohol consumption
the likelihood of recall or interviewer bias. and breast cancer. J Natl Cancer Inst 1987; 78: 657–61.
 BREAST CANCER RISK AND ALCOHOL CONSUMPTION 923
12 21
Young T B. A case-control study of breast cancer and alcohol Weinstein A L, Mahoney M C, Nasca P C et al. Breast cancer
consumption habits. Cancer 1989; 64: 552–58. risk and oral contraceptive use: results from a large case-
13
Schatzkin A, Jones D Y, Hoover R N et al. Alcohol consumption control study. Epidemiology 1991; 2: 353–58.
22
and breast cancer in the epidemiologic follow-up study of Rosenberg L, Slone D, Shapiro S et al. Breast cancer and
the First National Health and Nutrition Examination alcoholic-beverage consumption. Lancet 1982; i: 267–70.
23
Survey. N Engl J Med 1987; 316: 1169–73. Talamini R, La Vecchia C, Decarli A et al. Social factors, diet
14
Van’t Veer P, Kok F J, Hermus R J, Sturmans F. Alcohol dose, and breast cancer in a northern Italian population. Br J
frequency and age at first exposure in relation to the risk of Cancer 1984; 49: 723–29.
24
breast cancer. Int J Epidemiol 1989; 18: 511–17. Willet W C, Stampfer M J, Colditz G A et al. Moderate alcohol
15
Nasca P, Baptiste M S, Field N A et al. An epidemiological consumption and the risk of breast cancer. N Engl J Med
case-control study of breast cancer and alcohol consump- 1987; 316: 1174–80.
25
tion. Int J Epidemiol 1990; 19: 532–38. O’Connel D L, Hulka B S, Chambless L E et al. Cigarette
16
Holmberg L, Baron J A, Byers T et al. Alcohol intake and breast smoking, alcohol consumption, and breast cancer risk.
cancer risk: Effect of exposure from 15 years of age. J Natl Cancer Inst 1987; 78: 229–34.
26
Cancer Epidemiol Biomark Prev 1995; 4: 843–47. Richardson S, de Vincenzi I, Pujol H, Gerber M. Alcohol
17
La Vecchia C, Negri E, Parazzini F et al. Alcohol and breast consumption in a case-control study of breast cancer in
cancer: update from an Italian case-control study. Eur J southern France. Int J Cancer 1989; 44: 84–89.
27
Cancer Clin Oncol 1989; 25: 1711–17. Giovannucci E, Stampfer M J, Colditz G A et al. Recall and
18
Longnecker M P, Newcomb P A, Mittendorf R et al. Risk of selection bias in reporting past alcohol consumption among
breast cancer in relation to lifetime alcohol consumption. breast cancer cases. Cancer Causes Control 1993; 4:
J Natl Cancer Inst 1995; 87: 923–29. 441–48.
19 28
Katsouyanni K, Trichopoulou A, Stuver S et al. Ethanol and Kelsey J L, Whittemore A S, Thompson W D, Evans A S.
breast cancer: An association that may be both confounded Biologic and Statistical Concepts. In: Kelsey J L,
and causal. Int J Cancer 1994; 58: 356–61. Whittemore A S, Thompson W D, Evans A S, (eds.).
20
Longnecker M P, Paganini-Hill A, Ross R K. Lifetime alcohol Methods in Observational Epidemiology, 2nd edn. New
consumption and breast cancer risk among postmenopausal York: Oxford University Press, 1996: 22–44.
women in Los Angeles. Cancer Epidemiol Biomark Prev
1995; 4: 721–25. (Revised version received March 1997)