American Journal of Epidemiology

Copyright O 1999 by The Johns Hopkins University School of Hygiene and Public Health
All rights reserved
\fol. 150, No. 6
Printed in U.S.A.
Alcohol Consumption and Cognitive Performance in the Framingham Heart
Study
Penelope K. Elias,
1
Merrill F. Elias,
2
Ralph B. D'Agostino,
1
Halit Silbershatz,
1
and Philip A. Wolf
3
Recent studies have indicated that moderate alcohol intake may be beneficial to cognitive functioning in
women, although not necessarily in men. Data from the Framingham Heart Study, a large, prospective study of
cardiovascular disease in Framingham, Massachusetts, were used to examine the relation between alcohol
consumption and cognitive ability. The major research question was whether a different alcohol-cognition
relation would be found for male and female drinkers. Men (n = 733) and women (n = 1,053), aged 55-88 years,
were queried as to their weekly intake of alcohol, and these data were used to construct groups of abstainers,
very light, light, moderate, and heavy drinkers. Data from earlier reports of alcohol consumption were also
examined. Participants were administered eight tests which reflect performance in the domains of verbal
memory, learning, visual organization and memory, attention, abstract reasoning, and concept formation.
Multivariate linear regression analyses were used with statistical adjustment for age, education, occupation,
cardiovascular disease, and associated risk factors. Women who drank moderately (2-4 drinks/day) showed
superior performance in many cognitive domains relative to abstainers. For men, superior performance was
found within the range of 4-8 drinks/day, although fewer significant relations were observed. These results were
confirmed by prospective analyses of 24-year drinking history. Am J Epidemiol 1999; 150:580-9.
alcohol drinking; cardiovascular diseases; cognition; men; risk factors; women
Early studies of the relation between moderate alco-
hol consumption and cognitive functioning (1-5) sup-
ported the notion that drinking at any level was associ-
ated with poorer performance on cognitive tests. This
research was conducted primarily with relatively small
samples of young to middle-aged male social drinkers.
However, two studies (4, 6) corroborated these results
for a subsample of women whose drinking patterns
(amount and frequency of consumption) were similar
to their male counterparts.
Subsequent research with male and female college
students (7-9) and elderly men (10) led to the conclu-
sion that no significant negative relation exists
Received for publication March 2, 1998, and accepted for publi-
cation January 22, 1999.
Abbreviations: AC, attention and concentration composite; BMI,
body mass Index; Cl, confidence interval; CVD, cardiovascular dis-
ease; HDL-C, high density lipoprotein cholesterol; K-A, Kaplan-
Albert; LJM, Learning and Immediate Memory composite; OR, odds
ratio.
1
Department of Mathematics, Statistics and Consulting Unit,
Boston University, Boston, MA.
2
Department of Psychology, University of Maine, Orono, ME.
3
Department of Neurology, Boston University School of Medicine,
and the Section of Preventive Medicine and Epidemiology, Evans
Memorial Department of Clinical Research and Department of
Medicine, Boston Medical Center, Boston, MA.
Reprint requests to Dr. Penelope K. Elias, Department of
Mathematics, Statistics and Consulting Unit, Boston University, 111
Cummington St., Boston, MA 02215.
between social drinking and level of cognitive func-
tioning. In fact, Bates and Tracy (11) reported that
increasing levels of alcohol use among young female
social drinkers were related to better performance on
many cognitive tests.
More recent studies with older samples have indi-
cated that a U- or J-shaped curve may best describe the
relation between level of alcohol consumption and
cognitive performance (12-16). Drinkers of low to
moderate amounts of alcohol performed at a higher
level than abstainers or heavy drinkers. However, in
one report (12), statistical control for age, income,
education, and gender rendered the findings nonsignif-
icant. In another study (16), the protective effects of
moderate consumption were limited only to those par-
ticipants who exhibited chnical conditions associated
with atherosclerosis.
It is especially important to examine data for men
and women separately when alcohol consumption is a
predictor variable, because their consumption levels
are very different. In virtually every study which
included both sexes, women consumed alcohol less
frequently and in smaller amounts than men (4, 6-8,
11-15, 17). The issue of sex differences in the relation
of alcohol consumption and cognitive performance
was addressed by Dufouil et al. (17), using data from
the Epidemiology of Vascular Aging (EVA) study. This
580

b
y

g
u
e
s
t

o
n

J
u
l
y

1
5
,

2
0
1
4
h
t
t
p
:
/
/
a
j
e
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

Alcohol Consumption and Cognitive Performance 581
research included 574 men and 815 women, aged
59-71 years. No association between drinking and
cognition was found for the male participants; for the
female participants, moderate alcohol consumption
was associated with better performance on seven often
cognitive tests and an overall composite score.
Indeed, physiologic evidence also supports the strat-
egy of a priori gender stratification. Women are more
vulnerable than men to the adverse effects of heavy
alcohol consumption in terms of brain shrinkage (18).
At six drinks per day, women are at higher risk of
death from non-cardiovascular disease causes than
men (19). Moderate levels of alcohol consumption
also have decidedly different effects in men and
women, including effects on high density lipoprotein
cholesterol (HDL-C) levels (20, 21), hypertensive
blood pressures (22), and degree of peripheral arterial
disease (23). Regarding all-cause mortality, the benefi-
cial effects of alcohol occur at approximately two stan-
dard drinks per day less in women than in men (24).
The present study involved a large prospective, com-
munity-based sample of men and women who partici-
pated in the Framingham Heart Study from 1948 to
1976. From study inception, biennial physical examina-
tions were conducted for the purpose of assessing mor-
bidity and risk factors associated with cardiovascular
disease. At Exams 14/15, a subset of study participants
was given a series of neuropsychological tests, the
Kaplan-Albert (K-A) battery. Previous studies with this
sample (25,26) indicated a positive association between
current amount of alcohol consumed on a weekly basis
and performance on the neuropsychological tests. The
present study was designed to explore more fully the
alcohol-cognition relation in the Framingham Heart
Study cohort Most importantly, we wished to examine
gender differences in the relation between current and
past drinking patterns and cognitive test performance.
MATERIALS AND METHODS
Study design
During Exams 14 or 15 (19741976), participants in
the Framingham Heart Study were administered the
Kaplan-Albert (K-A) neuropsychological test battery
(27-30). The study cohort consisted of 2,123 subjects,
aged 55-89 years, who underwent neuropsychological
testing (75 percent of the potential subject pool). The
subjects who were not tested did not differ from the
study cohort in terms of age, sex, or education (25).
All participants were screened for cardiovascular
risk factors and events at the biennial examinations
and extensive medical histories were taken. Of the
2,123 eligible participants, 317 persons were excluded
based on the following criteria: 1) diagnosis of stroke
{n = 122); 2) diagnosis of dementia (n = 12); 3) com-
pleted fewer than the eight subtests of the K-A battery
(n = 173); and 4) missing data for the predictor or con-
trol variables (n = 10). The present study sample con-
sisted of 1,806 individuals.
Measurement of alcohol consumption
At Exams 2, 7, 12, 13, 14, and 15, participants were
asked the following questions as part of the medical
history interview: 1) "How many bottles, cans, or
glasses of beer do you consume in a week?" 2) "How
many glasses of wine do you consume in a week?" 3)
"How many cocktails, highballs, or straight drinks do
you consume in a week?" This information was used
to create a drinking variable measured in ounces of
pure alcohol consumed per week based on the follow-
ing conversion equation:
Drink (oz/week) = (0.57 x no. of cocktails) + (0.444
x no. of beers) + (0.4 x no. of glasses of wine).
The drink variable was used to define groups charac-
terized by level of current drinking: very light
drinkers, 3.5 oz/week (<105 ml); light drinkers, 3.6-7
oz/week (106-211 ml); moderate drinkers (7.1-14
oz/week (212^22 ml); heavy drinkers, 14.1-28
oz/week (423-933 ml); and very heavy drinkers, >28
oz/week (>933 ml). Based on criteria established by
the National Institute on Alcohol Abuse and
Alcoholism (NIAAA) (31), a "drink" was defined as
12 oz (355 ml) of beer, 5 oz (148 ml) of wine, or 1.5
oz (44 ml) of distilled spirits, each of which contains
approximately 0.5 oz (15 ml) of pure alcohol. Thus,
these groups may be described in terms of drinks/day:
very light drinkers, ^1 drink/day; light drinkers, 1-2
drinks/day; moderate drinkers (>2 and <A drinks/day);
heavy drinkers (>4 and <8 drinks/day); and very heavy
drinkers, >8 drinks/day.
As in previous studies (4, 6-8, 11-15, 17), the dis-
tributions of male and female drinkers differed consid-
erably. Amounts of alcohol consumed in mean
oz/week were 5.0 6.5 for men (range, 0-45) and 1.8
3.0 for women (range, 0-23). Both distributions
were positively skewed, with only 10 men and no
women falling into the highest drinking category (very
heavy drinkers). Only nine women fell into the heavy
drinking category. It was decided to eliminate the very
heavy drinking category entirely, and, for the analyses
involving women, to also eliminate the heavy drinking
category. Thus, the final study sample consisted of 733
men and 1,053 women.
Finally, because previous research (32) had demon-
strated that recently abstinent alcohol abusers showed
Am J Epidemiol Vol. 150, No. 6, 1999

b
y

g
u
e
s
t

o
n

J
u
l
y

1
5
,

2
0
1
4
h
t
t
p
:
/
/
a
j
e
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

582 Elias et al.
cognitive impairment on some of the tests in the K-A
battery, we were concerned that the abstainer group
may have included some ex-heavy drinkers who might
bias our results. However, inspection of the data from
the previous examination (Exam 13, 2 years earlier)
revealed that in no case did a current abstainer report
heavy drinking at that time. Rather, all abstainers in
our cross-sectional analyses either never drank or
drank rightly to moderately in the 4 years prior to neu-
ropsychological testing.
Neuropsychological test battery
The K-A battery consists of eight subtests from the
Wechsler Adult Intelligence Scale (WAIS), the
Wechsler Memory Scale (WMS), and the Multilingual
Aphasia Examination (27). The tests were chosen to
sample cognitive performance in multiple domains,
including immediate and long-term verbal memory,
paired associates learning, visual organization and
memory, attention, abstract reasoning, and concept
formation (33). The K-A battery is sensitive to cogni-
tive decline with age, education, and cardiovascular
disease (27-30). Tests were presented in the following
standard sequence: Logical Memory-Immediate
Recall, Visual Reproductions, Paired Associates
Learning, Digit Span Forward, Digit Span Backward,
Word Fluency, Similarities, and Logical Memory-
Delayed Recall. The K-A battery takes approximately
30 minutes to administer.
The full sample means and standard deviations were
used to transform raw test scores to standardized z
scores. A total composite score, consisting of the aver-
age of the z scores obtained for each of the eight tests,
was also calculated for each individual. This score
reflected global test competence (27) and has been
used as the primary variable in studies of the effects of
age, blood pressure, and type II diabetes on cognitive
performance in the Framingham Heart Study cohort
(25, 29, 30).
For the purpose of variable reduction and based on
the results of an earlier factor analysis of the K-A bat-
tery (28), two additional composites were constructed.
The first composite is labeled Learning and
Immediate Memory (LIM) and is the average of the z
scores for Logical Memory-Immediate Recall +
Visual Reproductions + Paired Associates. This com-
posite reflects both immediate learning and secondary
memory processes (28). The second composite is
labeled Attention and Concentration (AC) and is the
average of Digit Span Forward + Digit Span
Backward. Although the AC composite involves
memory processes, it primarily reflects immediate
registration of information, attention, and passive
span of apprehension, as well as "double tracking" in
which memory and digit reversing are accomplished
simultaneously (28, 33).
Statistical control variables
Results of linear regression analyses were adjusted
for five statistical control variables. Three of these vari-
ables were coded as is usual in studies involving the
Framingham data: age (in years at time of neuropsy-
chological testing); education (on a scale of 1-4, with
0-8 years = 1; 9-11 years = 2; 12 years = 3; and >12
years = 4); occupation (on a scale of 1-7, with execu-
tive/professional = 1; managerial = 2; administrative
personnel = 3; clerical and sales = 4; skilled manual
employee = 5; semi-skilled employee = 6; and
unskilled employee = 7). Gender was not included as a
control variable, because all analyses were done for
men and women separately, as explained below.
Because each participant in the Framingham Heart
Study had undergone a thorough physical examination
biennially since 1948, we were able to accurately
ascertain the presence of five risk factors for cardio-
vascular disease and six initial cardiovascular events.
Rather than test statistical models which included a
total of 14 control variables, we simplified the models
by designing two control variables which reflect the
extent of risk for cardiovascular disease and the occur-
rence of previous cardiovascular events.
A risk factor index was constructed as a scale (0-5)
with each risk factor coded 1 (present) or 0 (absent).
Each individual received a risk factor score based on
the sum of the five codes. The risk factors were deter-
mined as follows: 1) hypertension, defined as present
if systolic blood pressure was >140 mmHg or diastolic
blood pressure was ^90 mmHg on each of two succes-
sive readings obtained by the clinic physician or the
individual was on antihypertensive medication; 2) dia-
betes mellitus, defined as present if non-fasting glu-
cose level was at least 11.11 mmol/liter (200 mg/dL) or
the individual was using insulin or an oral hypo-
glycemic agent on or before the K-A test; 3) smoking,
defined as present if the individual reported smoking at
least one cigarette per day during the past year, 4) high
body mass index (BMI), defined as present if BMI
(weight (kg)/height (m)
2
) was >29; 5) high total serum
cholesterol, defined as present if cholesterol levels
were >200 mg/dL (>5.20 mmol/liter).
Lastly, a statistical control variable was constructed
based on the presence or absence of cardiovascular
disease prior to K-A testing. Presence of cardiovascu-
lar disease was determined by a panel of three physi-
cians utilizing criteria published elsewhere (34). The
diseases represented by this variable included myocar-
dial infarction, angina pectoris, coronary insufficiency,
intermittent claudication, congestive heart failure, and
Am J Epidemiol Vol. 150, No. 6, 1999

b
y

g
u
e
s
t

o
n

J
u
l
y

1
5
,

2
0
1
4
h
t
t
p
:
/
/
a
j
e
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

Alcohol Consumption and Cognitive Performance 583
transient ischemic attack. The disease variable was
coded 1 (any of the diseases present) or 0 (all diseases
absent). It differed from the risk factor index in that it
was not cumulative. Thus, an individual who suffered
one cardiovascular event received a score of 1 as did
an individual who suffered >2 events.
Statistical analyses
The distributions for alcohol consumption variables
differed substantially for men and women, the former
group drinking at a much higher level than the latter.
Thus, an a priori decision was made to analyze the
data for men and women separately. All analyses
employed a multivariate linear regression model.
Dependent variables included the three composite
scores (LIM, AC, and Total) and the three remaining
tests (Similarities, Word Fluency, and Delayed
Memory). Data were analyzed using the general linear
models procedure of the SAS statistical software
package (35).
For the analyses of current alcohol consumption
and cognitive performance, the independent variable
was level of drinking (drinking group). These analy-
ses closely paralleled methods used in recent studies
of alcohol consumption and cognition (e.g., 14, 17)
or cardiovascular disease (e.g., 44). Dummy vari-
ables were created to represent the alcohol consump-
tion groups for the men (very light, light, moderate,
and heavy) and women (very light, light, and moder-
ate). Each individual was coded 0 or 1 to indicate his
or her group membership. In one set of analyses, the
referent group was abstainers; in another set of
analyses, abstainers were excluded and the referent
group was very light drinkers. All drinking level
groups were simultaneously entered into the regres-
sion model, along with the five control variables.
Thus, each drinking group was compared with the
abstainer or very light drinker groups with statistical
control for the other drinking groups and the control
variables.
For the analyses of past drinking levels and current
cognitive performance, the independent variable was
mean ounces of alcohol consumed per week, defined
as the sum of oz/week reported at Exams 2, 7, 12, and
13, divided by 4. Thus, this variable represents the
average oz of alcohol consumed per week over 24
years. Because neuropsychological testing began at
Exam 14, at least 2 years intervened between the final
alcohol measurement and cognitive assessment.
Finally, we conducted linear regression analyses
with each risk factor and cardiovascular disease vari-
able entered simultaneously and independently into
the statistical models rather than being included in
their respective disease indices. This procedure
yielded identical results to analyses employing the
indices and thus we report only the latter below.
RESULTS
Demographic characteristics for the men and
women are shown in table 1. The women were slightly
older and had a higher mean risk factor index com-
pared with men, specifically because of a higher inci-
dence of hypertension and higher cholesterol levels.
However, the proportion of the sample with type n
diabetes was lower among the women than among the
men. The women also had significantly less pre-exist-
ing cardiovascular disease than the men.
Table 2 shows the demographic characteristics for
each of the drinking groups for men and women sepa-
rately. For both men and women, mean level of drink-
ing decreased as age increased. Educational level was
higher among the groups that drank more heavily,
TABLE 1. Demographic characteristics of men and women In
the Framlngham Heart Study who took the Kaplan-Albert
Battery, Framlngham, Massachusetts, 1974-1976
Characteristic Men (n = 733)
Women
n = 1,053)
p value*
Age (years), mean (SDt) 66.4(7.2) 67.5(7.4) <0.001
Education, mean (SD)$ 2.8(1.1) 2.8(1.1) NSt
Occupation, mean (SD) 4.0(1.6) 4.1(1.6) NS
Risk factors, mean (SD)H 1.7 (0.9) 1.9 (0.9) <0.001
Hypertension (%)
Type II diabetes (%)
Cigarette smoking (%)
High BMIt (%)
High cholesterol (%)
Pre-existing CVDf (%)#
49.9
9.5
23.7
24.0
68.1
23.6
57.0
6.6
23.5
22.4
84.3
14.7
<0.01
<0.02
NS
NS
<0.001
<0.001
Based on Mest for continuous variables and chi-square test for
qualitative variables.
t SD, standard deviation; NS, not significant; CVD, cardio-
vascular disease; BMI, body mass index.
$ Based on a scale of 1-4 with 0-8 years = 1; 9-12 years = 2;12
years (completed high school) = 3; and >12 years = 4.
Based on a scale of 1-7 with executive/professionaJ = 1;
managerial = 2; administrative personnel = 3; clerical and sales = 4;
skilled manual employee = 5; semi-skilled employee = 6; unskilled
employee = 7.
H Based on a scale of 0-5 wrth each risk factor coded 1
(present) or 0 (absent): hypertension, systolic blood pressure 140
mmHg and/or diastoiic Wood pressure 90 mmHg or subject was on
antihypertensive medication; diabetes, non-fasting glucose level
was 11.11 mmol per liter (200 mg/dl) or the subject was using
insulin or an oral hypogtycemic agent; smoking, subject reported
smoking at least one cigarette per day during the past year; high
BMI, BMI >29 kg/m
2
; high cholesterol, cholesterol >2O0 mg/dL
(>5.20 mmol/liter).
# Pre-existing CVD includes myocardial infarction, angina
pectoris, intermittent claudication, coronary insufficiency, transient
ischemic attack, and congestive heart failure.
Am J Epidemiol Vol. 150, No. 6, 1999

b
y

g
u
e
s
t

o
n

J
u
l
y

1
5
,

2
0
1
4
h
t
t
p
:
/
/
a
j
e
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

584 Elias et al.
TABLE 2. Covartates of alcohol consumption among men and women In the Framlngham Heart Study who took the Kaplan-Albert
Battery, Framlngham, Massachusetts, 1974-1976
Men
Age (years), mean (SDt)
Education, mean (SD)$
Occupation, mean (SD)
Alcohol oz/week, U mean (SD)
Risk factors, mean (SD)#
Hypertension (%)
Type II diabetes (%)
Cigarette smoking (%)
High BMIt (%)
High cholesterol (%)
Pre-existing CVDf (%)tt
Women
Age (years), mean (SD)
Education, mean (SD)}:
Occupation, mean (SD)
Alcohol oz/week, mean (SD)
Risk factors, mean (SD)#
Hypertension (%)
Type II diabetes (%)
Cigarette smoking (%)
High BMI (%)
High cholesterol (%)
Pre-existing CVD (%)ft
Abstain
r =158
67.7 (8.0)
2.5(1.0)
4.4(1.6)
1.7(1.0)
46.8
10.1
26.6
20.3
65.2
29.7
n = 407
69.3 (7.7)
2.5(1.2)
4.4(1.5)
2.0 (1.0)
63.1
10.6
17.4
28.7
83.8
17.7
Very light
n = 244
66.6 (7.3)
2.9(1.0)
3.9(1.6)
1.4(0.9)
. 1.6(0.9)
46.7
9.0
18.0
25.0
64.8
20.9
n = 462
66.9 (7.0)
2.8(1.1)
4.0 (1.5)
1.3(0.9)
1.8(0.9)
54.1
4.8
22.5
21.4
85.1
13.4
Light
n= 137
66.8 (7.0)
2.9(1.1)
3.9(1.7)
5.1(1.1)
1.8(1.0)
48.2
12.4
20.4
27.7
70.8
24.1
n=99
66.3 (7.4)
3.2(1.0)
3.3(1.6)
4.7 (0.9)
1.8(0.8)
48.4
5.1
31.3
13.1
83.8
15.2
Moderate
n= 137
64.8 (6.6)
2.9(1.0)
3.8(1.7)
9.6(1.8)
1.9 (0.9)
55.5
8.8
30.7
24.8
73.7
21.9
n = 85
64.1 (5.7)
3.3(1.0)
3.8(1.5)
8.8(1.5)
1.9(0.8)
52.9
0.0
48.2
7.1
83.5
7.1
Heavy
n = 57
64.8 (5.6)
2.7(1.1)
4.2(1.5)
18.7(3.7)
1.9(0.1)
63.2
5.3
31.6
19.3
70.2
21.1
p value*
<0.01
<0.01
<0.02
<0.02
NSf
NS
<0.02
NS
NS
NS
<0.001
<0.001
<0.001
<0.04
<0.01
<0.01
<0.01
<0.01
NS
<0.06
* Based on /-test for continuous variables and chi-square test for qualitative variables.
t SD, standard deviation; NS, not significant; BMI, body mass Index; CVD, cardiovascular disease.
X Based on a scale of 1-4 with 0-8 years = 1; 9-12 years = 2; 12 years (completed high school) = 3; and >12 years = 4.
Based on a scale of 1-7 with executive/professional = 1; managerial = 2; administrative personnel = 3; clerical and sales = 4; skilled
manual employee = 5; semi-skilled employee = 6; unskilled employee = 7.
U 1 oz = 29.6 ml.
# Based on a scale of 0-5 with each risk factor coded 1 (present) or 0 (absent): hypertension, systolic Wood pressure 140 mmHg and/or
diastoiic blood pressure 90 mmHg or subject was on antihypertensive medication; diabetes, non-fasting glucose level was at least 11.11 mmol
per liter (200 mg/dl) or the subject was using insulin or an oral hypoglycemic agent; smoking, subject reported smoking at least one cigarette
per day during the past year; high BMI, BMI >29 kg/m
2
; high cholesterol, cholesterol >200 mg/dL (>5.20 mmol/liter).
XX Pre-existing cardiovascular disease includes myocardial infarction, angina pectoris, intermittent claudication, coronary insufficiency,
transient ischemic attack, and congestive heart failure.
although this was more evident for the women than for
the men. Among the men, very light, light, and moder-
ate drinkers exhibited higher occupational levels than
abstainers and heavy drinkers. The women's occupa-
tional level was highest for the light drinkers and low-
est for the abstainers.
For the men, the risk factor index was higher among
the heaviest drinkers. This finding was related primar-
ily to their greater cigarette consumption. For women,
the proportions of persons with hypertension, type II
diabetes, and high BMI were highest among the
abstainers. As with the men, cigarette consumption
increased substantially as drinking level increased.
For the women, there was a slight tendency for more
cardiovascular disease to be present among the
abstainers, very light, and light drinkers, relative to
the moderate drinkers (p < 0.06). This trend was not
evident for the men.
Cross-sectional analyses of drinking level and
cognition
The results of the multiple linear regression analyses
of drinking groups and Kaplan-Albert test scores are
shown in table 3. Each regression coefficient represents
the change in standardized score for the drinking group
relative to the abstainers. All regression coefficients
were adjusted for age, education, occupation, risk factor
index, and pre-existing cardiovascular disease index.
For the men, no differences were found between the
very light or light drinkers and the abstainers. Test per-
formance of the moderate drinkers was superior to that
of the abstainers only for Logical Memory-Delayed
Recall. The heavy consumption group performed sig-
nificantly better than the abstainers on the Logical
Memory-Delayed Recall test and also on the AC and
Total composite scores.
Am J Epidemiol Vol. 150, No. 6, 1999

b
y

g
u
e
s
t

o
n

J
u
l
y

1
5
,

2
0
1
4
h
t
t
p
:
/
/
a
j
e
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

Alcohol Consumption and Cognitive Performance 585
TABLE 3. Results of linear regression analyses describing the relation between current level of alcohol consumption and
neuropsychological test performance on the Kaplan-Albert battery for participants In the Framlngham Heart Study, Framlngham,
Massachusetts, 1974-19761
Sex and test on
Kaplan-Albert battery
Men
Delayed Memory
Word Fluency
Similarities
LIM composite^
AC composite
Total composite^]
Women
Delayed Memory
Word Fluency
Similarities
LIM composite
AC composite
Total composite
Very light
3.5 oz/wk
P
0.0319
-0.0800
-0.0853
0.0727
-0.0945
-0.0122
0.0580
0.1072
0.1030
0.0454
0.0963
0.0619
SE
0.0949
0.0865
0.0822
0.0878
0.0972
0.0533
0.0603
0.0601
0.0552
0.0572
0.0629
0.0345
Ught
3.6-7 oz/wk
P
0.1278
0.0379
0.0489
-0.1113
-0.1479
0.0254
0.2044*
0.0926
0.1022
0.2543**
0.0590
0.1236*
SE
0.1079
0.0984
0.0935
0.0999
0.1105
0.0607
0.1000
0.0997
0.0914
0.0572
0.1043
0.0571
Moderate
7.1-14 oz/wk
P
0.2705
0.0984
0.0297
0.1902
-0.0279
0.0999
SE
* 0.1093
0.0997
0.0947
0.1012
0.1120
0.0615
0.3318** 0.1069
0.3839
0.2044
0.3414
0.2665
0.2512
* 0.1066
* 0.0978
** 0.1034
0.1115
** 0.0611
14.1
P
0.3423
0.1490
0.0596
0.1467
0.3085
0.1925
Heavy
-28 oz/wk
SE
0.1433
0.1307
0.1241
0.1326
0.1326
* 0.0805
*p<0.05, **p<0.01.
t Regression coefficients (P) for the outcome variables (Kaplan-Albert tests) are expressed in standard score (z) units and were adjusted
for age, education, occupation, risk factors (range = 0-5), and pre-existing cardiovascular disease (0 = absent, 1 = present). Each drinking group
was compared with abstainers. 1 oz = 29.6 ml. SE, standard error.
X Learning and Immediate Memory (LIM) composite consists of the mean z score for Logical Memory, Visual Reproductions, and Paired
Associates Learning.
Attention and Concentration (AC) composite consists of the mean z score for Logical Memory, Visual Reproductions, and Paired
Associates Learning.
H Total composite consists of the mean z score for all eight Kaplan-Albert tests.
In comparison to the men, women drinkers showed
a more consistent pattern of superior performance, rel-
ative to abstainers. No differences were found between
the very light drinkers and abstainers. Light drinkers
performed in a superior manner, compared with
abstainers, on the Logical Memory-Delayed Recall
test and the LIM and Total composites. However, the
moderate drinking group performed substantially bet-
ter than the abstainers. Significant differences between
the two groups were found for all measures. Thus, rel-
ative to abstainers, increasing alcohol consumption
appears to be positively related to cognitive test per-
formance among women.
The abstainers in our sample tended to exhibit
higher levels of disease which may have been partially
responsible for their avoidance of alcohol. Although
the initial analyses included statistical control for pre-
existing cardiovascular disease and concomitant risk
factors, we also conducted a set of analyses in which
the abstainer group was eliminated and the very light
drinkers were the referent group. The results are pre-
sented in table 4.
Basically the same patterns of results were obtained
as in the analyses for which abstainers were the refer-
ent group. For males, no differences were found
between very light and light drinkers. Moderate and
heavy alcohol consumers differed from very light
drinkers on precisely the same tests as they differed
from abstainers.
For women, alcohol-test performance relations were
attenuated somewhat, although the same trend for
increasing alcohol consumption to be associated with
higher levels of cognitive functioning is noted. Light
drinkers demonstrated superior performance, relative
to very light drinkers, for the LIM composite score.
Moderate drinkers demonstrated superior perfor-
mance, relative to very light drinkers, on the Logical
Memory-Delayed Recall and Word Fluency tests and
the LIM and AC composite scores.
Analysis of drinking history and cognitive perfor-
mance
Although it might be expected that current drinking
habits would be related to cognition, it is likely that
past consumption levels might also contribute signifi-
cantly. Thus, we examined the mean levels of alcohol
consumption over 24 years prior to neuropsychologi-
cal testing in relation to cognitive performance. The
results are shown in table 5. The parameter estimates
(P) represent the change in z-scores on the K-A mea-
sures for each mean oz of alcohol consumed, calcu-
Am J Epidemiol Vol. 150, No. 6, 1999

b
y

g
u
e
s
t

o
n

J
u
l
y

1
5
,

2
0
1
4
h
t
t
p
:
/
/
a
j
e
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

586 Elias et al.
TABLE 4. Results of linear regression analyses describing the relation between current level of alcohol
consumption and neuropsychologlcal test performance on the Kaplan-Albert Battery for non-abstaining
participants In the Framlngham Heart Study, Framlngham, Massachusetts, 1974-19761
Sex and test on
Kaplan-Albert battery
Men
Memory Delayed
Word Fluency
Similarities
UM composite:}:
AC composite
Total compositefl
Women
Memory Delayed
Word Fluency
Similarities
LIM composite
AC composite
Total composite
Light
3.6-7 oz/wk
0.0928
-0.1093
0.1282
0.0322
-0.0570
0.0331
0.1550
0.0004
0.0042
0.2134*
-0.0451
0.0657
SE
0.0992
0.0903
0.0849
0.0914
0.1089
0.0553
0.1013
0.0992
0.0859
0.0970
0.1031
0.0565
Moderate
7.1-14 oz/wk
P
0.2331 *
0.1645
0.1003
0.1070
0.0557
0.1033
0.2969**
0.3050**
0.1120
0.3118**
0.1703
0.2064**
SE
0.1004
0.0914
0.0859
0.0925
0.1021
0.0560
0.1084
0.1063
0.0920
0.1038
0.1104
0.0605
Heavy
14.1-28 oz/wk
P SE
0.2382* 0.1373
0.1932 0.1249
0.1019 0.1175
0.0672 0.1275
0.3532** 0.1021
0.1787* 0.0765
*p<0.05, **p<0.01.
t Regression coefficients (P) for the outcome variables (Kaplan-Albert tests) are expressed in standard score
(z) units and were adjusted for age, education, occupation, risk factors (range = 0-5), and pre-existing
cardiovascular disease (0 = absent, 1= present). Each drinking group was compared with very light drinkers (3.5
oz/wk). 1 oz = 29.6 ml. SE, standard error.
X Learning and Immediate Memory (LIM) composite consists of the mean z score for Logical Memory, Visual
Reproductions, and Paired Associates Learning.
Attention and Concentration (AC) composite consists of the mean z score for Logical Memory, Visual
Reproductions, and Paired Associates Learning.
Tl Total composite consists of the mean z score for all eight Kaplan-Albert tests.
lated as the average consumption reported from Exams
2, 7, 12, and 13.
For men, increasing alcohol consumption was asso-
ciated with better performance only for Logical
Memory-Delayed Recall. For women, however, all
measures, with the exception of Similarities, were
associated with better performance as consumption
level increased.
DISCUSSION
Our results are in general agreement with those of
Dufouil et al. (17). In cross-sectional analyses, light to
moderate alcohol consumption among women was
positively related to many domains of cognitive per-
formance. However, contrary to their findings, we also
found a smaller, more limited, positive relation for
men at relatively high drinking levels (4-8
drinks/day). We found no instance of a significant neg-
ative association between current alcohol consumption
and cognitive performance for either sex. Thus, our
results support the hypothesis that, for social drinkers,
a J-shaped curve best describes the alcohol-cognition
relation (14). Nonetheless, it is certain that this relation
cannot be sustained at very high levels of alcohol con-
sumption as is evident from the large literature on
alcohol abuse and cognitive performance (see Delin
and Lee (36), Parsons (37), and Parsons and Nixon
(38) for reviews).
We also examined effects of drinking history over
the 24 years prior to neuropsychological testing using
mean oz of alcohol consumed per week as the inde-
pendent variable. As with our cross-sectional analyses,
we found a significant positive effect for women, i.e.,
greater alcohol consumption resulted in higher scores
on all cognitive tests, with the exception of
Similarities. For men, only one significant positive
relation emerged. Thus, past, as well as current alcohol
use, appears to have a substantially more protective
effect on cognitive test performance in women than in
men.
The positive association between alcohol consump-
tion and cognition is strikingly reminiscent of a similar
relation found between alcohol consumption and
improvement in cardiovascular disease mortality rates
(39, 40), all-cause mortality (41, 42), and cardiovascu-
lar diseases (43-^45). Conclusions drawn from these
studies differ only in the specification of upper limits for
the protective effect of moderate alcohol consumption.
Rehm et al. (44), using a very large sample of men and
Am J Epidemiol Vol. 150, No. 6, 1999

b
y

g
u
e
s
t

o
n

J
u
l
y

1
5
,

2
0
1
4
h
t
t
p
:
/
/
a
j
e
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

Alcohol Consumption and Cognitive Performance 587
TABLE 5. Results of linear regression analyses describing the
relation between drinking history (mean oz/week over 24
years) and neuropsychologlcal test performance on the
Kaplan-Albert Battery for participants in the Framlngham
Heart Study, Framlngham, Massachusetts, 1974-1976*
Sex and test on
Kaplan-Albert Battery
Men (n = 741)
Delayed Memory
Word Fluency
Similarities
LIM composite^
AC composite!
Total compositeH
Women ( n= 1,059)
Delayed Memory
Word Fluency
Similarities
LIM composite
AC composite
Total composite
P
0.0127
0.0071
-0.0018
-0.0031
0.0035
0.0022
0.0185
0.0298
0.0084
0.0215
0.0254
0.0165
SE
0.0058
0.0052
0.0050
0.0053
0.0059
0.0032
0.0096
0.0096
0.0088
0.0091
0.0100
0.0055
p value
<0.03
NSt
NS
NS
NS
NS
<0.05
<0.01
NS
<0.02
<0.01
<0.01
Regression coefficients (P) for the outcome variables (Kaplan-
Albert tests) are expressed in standard score (z) units and were
adjusted for age, education, occupation, risk factors (range = 0-5),
and pre-existing cardiovascular disease (0 = absent, 1= present).
Each drinking group was compared with abstainers, with all other
drinking groups in the model. SE, standard error.
t NS, not significant.
t Learning and Immediate Memory (LIM) composite consists of
the mean z score for Logical Memory, Visual Reproductions, and
Paired Associates Learning.
Attention and Concentration (AC) composite consists of the
mean z score for Digit Span Forward and Digit Span Backward.
H Total composite consists of the mean z score for all eight
Kaplan-Albert tests.
women, reported results of an analysis of the relation
between alcohol consumption and coronary heart dis-
ease and mortality. An upper limit of the protective
effect of alcohol was found at 28 drinks/week for
women, although no upper limit could be specified for
men. Our results suggest a parallel relation for alcohol
consumption and cognitive performance. Unfortunately,
we had only a very small number of women (n = 9) who
reported alcohol intake greater than 28 drinks/week in
our sample, and thus we could not examine the possi-
bility of cognitive decline above this level.
Several possible mechanisms may be responsible
for the positive relation between alcohol consump-
tion and cognitive performance. Delin and Lee (36)
have summarized the beneficial effects of social
drinking and have concluded that moderate drinkers
enjoy better overall psychological health than
abstainers or heavy drinkers, specifically in terms of
a reduction in stress, anxiety, and depression and an
increase in general well-being. All of these factors
have been shown to be related to better cognitive
functioning (46).
Physiologic factors may also be implicated. For
example, the proportion of women with hypertension
decreased as drinking level increased, and hyperten-
sion has been shown to be an independent predictor of
poor cognitive performance (18, 23). Unfortunately,
the relation of blood pressure and alcohol consump-
tion per se is currently unclear. Some researchers have
found a positive linear association between amount of
alcohol consumed and blood pressure (47), while oth-
ers have found a negative relation between light to
moderate consumption levels and blood pressure (48).
The latter relation appears to be more characteristic of
the present study sample. Earlier studies in the
Framingham population (25, 26, 28) demonstrated
that chronicity of hypertension was associated with
poorer cognitive performance. Thus, the lower pro-
portions of hypertensive disease among the light to
moderate female alcohol consumers in our study may
represent an intervening factor in their better cogni-
tive performance.
To further pursue this possibility, we performed sec-
ondary analyses in which each risk factor (including
hypertension) and pre-existing cardiovascular disease
variable was placed into the regression model as an
independent covariate. For both men and women, the
results were identical to the analyses employing the
risk factor and pre-existing disease indices, that is, all
significant relations obtained in the original analyses
remained so. Therefore, it is unlikely that, among alco-
hol consumers, a lower incidence of hypertension, dia-
betes, or any other risk factor or cardiovascular event
measured in this study could account entirely for the
positive relations between alcohol consumption and
performance. Still, low to moderate alcohol consump-
tion is associated with lower levels of several vascular
risk factors (49) for which we did not have measures.
The protective effects of alcohol on cognitive func-
tioning may operate via reduction in one or more of
these risk factors, for example, increased HDL-C con-
centrations and/or improved insulin sensitivity.
With respect to study limitations, the present study,
similar to virtually every study of drinking and cogni-
tion, included only self-reported measures of alcohol
consumption. Thus, a legitimate question may arise as
to bias resulting from inaccurate recall and intentional
misreporting. In the Framingham Heart Study, partici-
pants were asked only about their consumption over
the previous 2 years and thus no extremely long-term
recall was required. Moreover, the alcohol questions
were a very small part of the overall interview which
focused primarily on cardiovascular disease symptoms
rather than dietary concerns, and there was no obvious
risk to the participants which might have led to less
than candid responses (50).
Am J Epidemiol Vol. 150, No. 6, 1999

b
y

g
u
e
s
t

o
n

J
u
l
y

1
5
,

2
0
1
4
h
t
t
p
:
/
/
a
j
e
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

588 Elias et al.
It is possible that the cross-sectional analysis of
alcohol consumption and cognitive performance
among older adults may have resulted in an abstainer
group which included individuals who, because of
their lower cognitive abilities, were less able to accu-
rately recall their recent drinking habits. For women,
the positive relations between 24-year drinking history
and current cognitive test performance found in our
prospective analyses argue against this possibility. For
men, however, only one cognitive measure was found
to be positively associated with previous alcohol con-
sumption, and thus, for them, we cannot rule out the
possibility of inaccurate recall in the abstainer group
due to poorer cognitive abilities.
In this study, abstainers, as a group, were older and
had lower education and occupation levels than
drinkers. The increase in number of abstainers with
age is common among older cohorts (51). Although we
controlled statistically for these variables, age and the
socioeconomic indicators may have resulted in resid-
ual confounding effects, for example, poorer nutrition
may be associated with increasing age or lower educa-
tion and occupation levels and may also result in an
increase in disease and poorer cognitive performance.
The overall results of this study may best be
described with reference to the three composite scores
because these scores represent the most reliable
indices of cognitive functioning. From this perspec-
tive, we conclude that, for women, moderate levels of
past and current alcohol consumption appear to have
beneficial effects on cognitive test performance. While
it remains possible that this conclusion may also be
applied to men, the effects are likely to be weaker and
less global with regard to overall cognitive ability than
for women.
ACKNOWLEDGMENTS
Supported in part by Special Career Emphasis Award no.
K01-AG0646-04 from the National Institute on Aging to Dr.
Penelope K. Elias, Research Grant no. 5-R37-AGO3055-15
from the National Institute on Aging to Dr. Merrill F. Elias,
and Research Grant no. 2-R01-AG08122-09 from the
National Institute on Aging and Contract no. N01-HC-
38038 from the National Heart, Lung, and Blood Institute's
Framingham Heart Study, National Institutes of Health, to
Dr. Philip A. Wolf.
REFERENCES
1. Parker ES, Noble EP. Alcohol consumption and cognitive
functioning in social drinkers. J Stud Alcohol 1977;38:
1224-32.
2. Parker ES, Noble EP. Alcohol and the aging process in social
drinkers. J Stud Alcohol 1980;41:170-8.
3. Parker ES, Parker DA, Brody JA, et al. Cognitive patterns
resembling premature aging in male social drinkers. Alcohol
Clin Exp Res 1982;6:46-52.
4. Parker DA, Parker ES, Brody JA, et al. Alcohol use and cog-
nitive loss among employed men and women. Am J Public
Health 1983:73:521-6.
5. Parker ES. Alcohol and cognition. Psychopharmacol Bull
1984;20:494-6.
6. Parker ES, Parker DA, Harford TC. Specifying the relationship
between alcohol use and cognitive loss: the effects of fre-
quency of consumption and psychological distress. J Stud
Alcohol 1991:53:366-73.
7. Jones-Saumty DJ, Zeiner AR. Psychological correlates of
drinking behavior in social drinker college students. Alcohol
Clin Exp Res 1985,9:158-63.
8. Hannon R, Butler CP, Day CL, et al. Alcohol use and cognitive
functioning in men and women college students. In: Galanter
M, ed. Recent developments in alcoholism. Vol 1. New York:
Plenum Press, 1985:241-52.
9. Carey KB, Maisto SA. Effect of a change in drinking pattern
on the cognitive function of female social drinkers. J Stud
Alcohol 1987;48:236-42.
10. Dent OF, Sulway MR, Broe GA, et al. Alcohol consumption
and cognitive performance in a random sample of Australian
soldiers who served in the Second World War. Br J Med
1997;314:l655-7.
11. Bates ME, Tracy JI. Cognitive functioning in young "social
drinkers": is there impairment to detect? J Abnorm Psychol
1990:99:242-9.
12. Goodwin JS, Sanchez CJ, Thomas P, et al. Alcohol intake
in a healthy elderly population. Am J Public Health 1987;
77:173-7.
13. Hebert LE, Scherr PA, Beckett LA, et al. Relation of smoking
and low-to-moderate alcohol consumption to change in cogni-
tive function: a longitudinal study in a defined community of
older persons. Am J Epidemiol 1993;137:881-91.
14. Hendrie HC, Gau S, Hall KS, et al. The relationship between
alcohol consumption, cognitive performance, and daily func-
tioning in an urban sample of older Black Americans. J Am
GeriatrSoc 1996;44:1158-65.
15. Christian JC, Reed T, Carmelli D, et al. Self-reported alcohol
intake and cognition in aging twins. J Stud Alcohol
1995;56:414-16.
16. Launer LJ, Feskens EJM, Kalmijn S, et al. Smoking, drinking,
and thinking. The Zutphen Elderly Study. Am J Epidemiol
1996:143:219-27.
17. Dufouil C, Ducimetiere P, Alpe'rovitch A. Sex differences in
the association between alcohol consumption and cognitive
performance. Am J Epidemiol 1997;146:405-12.
18. Mann K, Batra A, Gunther A, et al. Do women develop alco-
holic brain damage more readily than men? Alcohol Clin Exp
Res 1992:16:1052-6.
19. Klatsky AL, Armstrong MA, Friedman GD. Alcohol and mor-
tality. Ann Intern Med 1992;117:646-54.
20. Rossouw JE, Lai-Tung MT, Jooste PL, et al. Alcohol intake in
relation to lipids, lipoproteins and blood pressure. S Afr Med J
1992:82:246-50.
21. Weidner G, Connor SL, Chesney MA, et al. Sex differences in
high density lipoprotein cholesterol among low-level alcohol
consumers. Circulation 1991;83:176-80.
22. Laforge R, Williams GD, Dufour MC. Alcohol consumption,
gender, and self-reported hypertension. Drug Alcohol Depend
1990;26:235-49.
23. Jepson RG, Fowkes FG, Donnan PT, et al. Alcohol intake
as a risk factor for peripheral arterial disease in the general
population in the Edinburgh Artery Study. Eur J Epidemiol
1995;11:9-14.
24. Holman CD, English KR, Milne E, et al. Meta-analysis of
alcohol and all-cause mortality: a validation of NHMRC rec-
ommendations. Med J Aust 1996;164:141-5.
25. Elias MF, Wolf PA, D'Agostino RB, et al. Untreated blood
Am J Epidemiol Vol. 150, No. 6, 1999

b
y

g
u
e
s
t

o
n

J
u
l
y

1
5
,

2
0
1
4
h
t
t
p
:
/
/
a
j
e
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m

Alcohol Consumption and Cognitive Performance 589
pressure level is inversely related to cognitive functioning: the
Framingham Study. Am J Epidemiol 1993;138:353-64.
26. Elias PK, D'Agostino RB, Elias MF, et al. Blood pressure,
hypertension, and age as risk factors for poor cognitive perfor-
mance. Exp Aging Res 1995;21:393-417.
27. Farmer ME, White LR, Kittner SJ, et al. Neuropsychological
test performance in Framingham: a descriptive study. Psychol
Rep 1987;60:1023-40.
28. Elias MF, Elias PK, D'Agostino RB, et al. Role of age, educa-
tion, and gender on cognitive performance in the Framingham
Heart Study: community-based norms. Exp Aging Res 1997;
23:201-25.
29. Elias MF, D'Agostino RB, Elias PK, et al. Neuropsychological
test performance, cognitive functioning, blood pressure and
age: the Framingham Heart Study. Exp Aging Res 1995;21:
369-91.
30. Elias PK, Elias MF, D'Agostino RB, et al. NIDDM and blood
pressure as risk factors for poor performance: the Framingham
Study. Diabetes Care 1997;20:1388-95.
31. National Institute on Alcohol Abuse and Alcoholism. Sixth
special report to Congress on alcohol and health. Rockville,
MD: NIAAA, 1987. (NIAAA report no. ADM 87-1519).
32. O'Mahony JF, Doherty B. Intellectual impairment among
recently abstinent alcohol abusers. Br J Clin Psychol 1996;
35:77-83.
33. Lezak MD. Neuropsychological assessment. New York:
Oxford University Press, 1983.
34. Kannel WB, Wolf PA, Garrison RJ. Section 34: Some risk fac-
tors related to the annual incidence of cardiovascular disease
and death using pooled repeated biennial measurements:
Framingham Heart Study. 30-year follow-up. Springfield, VA:
National Technical Information Service, 1987.
35. SAS Institute, Inc. SAS user's guide, version 6.11. Cary, NC:
SAS Institute, Inc, 1991.
36. Delin CR, Lee TH. Drinking and the brain: current evidence.
Alcohol Alcohol 1992;27:117-26.
37. Fuchs CS, Stampfer MJ, Colditz GA, et al. Alcohol consump-
tion and mortality among women. N Engl J Med 1995;332:
1245-50.
38. Parsons OA. Neurocognitive deficits in alcoholics and social
drinkers: a continuum? Alcohol Clin Exp Res 1998;22:954-61.
39. Parsons OA, Nixon SJ. Cognitive functioning in sober social
drinkers: a review of the research since 1986. J Stud Alcohol
1998;59:180-90.
40. Camargo CA, Hennekens CH, Gaziano JM, et al. Prospective
study of moderate alcohol consumption and mortality in US
male physicians. Arch Intern Med 1997;157:79-85.
41. Scherr PA, LaCroix AZ, Wallace RB, et al. Light to moderate
alcohol consumption and mortality in the elderly. J Am Geriatr
Soc 1992,40:651-70.
42. Doll R, Peto R, Hall E, et al. Mortality in relation to consump-
tion of alcohol: 13 years' observations on male British doctors.
BrJMedl994;309:911-18.
43. Rimm EB, Klatsky A, Grobbee D, et al. Review of moderate
alcohol consumption and reduced risk of coronary heart dis-
ease: is the effect due to beer, wine, or spirits? Br J Med 1996;
312:731-6.
44. Rehm JT, Bondy SJ, Sempos CT, et al. Alcohol consumption
and coronary heart disease morbidity and mortality. Am J
Epidemiol 1997;146:495-501.
45. Kannel WB, Ellison RC. Alcohol and coronary heart disease:
the evidence for a protective effect. Clin Chim Acta 1996;
246:59-76.
46. Elias MF, Elias PK, Elias JW. Basic processes in adult devel-
opmental psychology. St Louis, MO: CV Mosby, 1977.
47. Choudhury SR, Okayama A, Kita Y, et al. The associations
between alcohol drinking and dietary habits and blood pres-
sure in Japanese men. J Hypertension 1995;13:587-93.
48. Rabbia F, Veglio F, Russo R, et al. Role of alcoholic beverages
in essential hypertensive patients. Alcohol Alcohol 1995;
30:433-9.
49. Kiechl S, Willeit J, Poewe W, et al. Insulin sensitivity and reg-
ular alcohol consumption: large, prospective, cross-sectional
population study (Bruneck study). Br J Med 1996;313:104(M.
50. Gordon T, Kannel WB. Drinking and its relation to smoking,
BP, blood lipids, and uric acid. Arch Intern Med 1983;143:
1366-74.
51. Adams WL, Garry PJ, Rhyne R, et al. Alcohol intake in the
healthy elderly. Changes in age in a cross-sectional and longi-
tudinal study. J Am Geriatric Soc 1990;38:211-16.
Am J Epidemiol Vol. 150, No. 6, 1999

b
y

g
u
e
s
t

o
n

J
u
l
y

1
5
,

2
0
1
4
h
t
t
p
:
/
/
a
j
e
.
o
x
f
o
r
d
j
o
u
r
n
a
l
s
.
o
r
g
/
D
o
w
n
l
o
a
d
e
d

f
r
o
m