Vol. 161, No. 9
Printed in U.S.A.
DOI: 10.1093/aje/kwi111
American Journal of Epidemiology
Copyright ª 2005 by the Johns Hopkins Bloomberg School of Public Health
All rights reserved
Alcohol Intake in Middle Age and Risk of Cardiovascular Disease and Mortality:
Accounting for Intake Variation over Time
J. R. Emberson1, A. G. Shaper1, S. G. Wannamethee1, R. W. Morris1, and P. H. Whincup2
1
Department of Primary Care and Population Sciences, Royal Free and University College Medical School, London,
United Kingdom.
2
Department of Community Health Sciences, St. George’s Hospital Medical School, London, United Kingdom.
Received for publication September 17, 2004; accepted for publication December 21, 2004.
Moderate alcohol consumption is associated with a decreased risk of cardiovascular disease. However, the
impact of variation in alcohol intake over time on estimated risk relations has not been adequately addressed. In
this study, 6,544 middle-aged British men without previous cardiovascular disease were followed for
cardiovascular events and all-cause mortality over 20 years from 1978/1980 to 1998/2000. Alcohol intake was
ascertained at regular points throughout the study. A total of 922 men had a major coronary event within 20 years,
352 men had a stroke, and 1,552 men died of all causes. Baseline alcohol intake displayed U-shaped relations
with cardiovascular disease and all-cause mortality, with light drinkers having the lowest risks and nondrinkers
and heavy drinkers having similarly high risks. However, the nature of these relations changed after adjustment
for intake variation; risks associated with nondrinking were lowered, and risks associated with moderate and
heavy drinking increased. Regular heavy drinkers had a 74% higher risk of a major coronary event, a 133% higher
risk of stroke, and a 127% higher risk of all-cause mortality than did occasional drinkers (these estimates were
8%, 54%, and 44% before adjustment for intake variation). The findings suggest that considerable caution may be
needed before any recommendations regarding acceptable limits of alcohol consumption are made.
alcohol drinking; cerebrovascular accident; coronary disease; mortality
Epidemiologic studies have consistently demonstrated
that light to moderate levels of alcohol intake are associated
with reduced risks of cardiovascular disease and all-cause
mortality (1–4), possibly because of beneficial effects of
alcohol on blood lipids (especially high density lipoprotein
cholesterol) and clotting factors (5). Heavy alcohol consumption, on the other hand, is associated with increased
risks of cardiovascular disease and all-cause mortality, though
often the disease risks of heavy drinkers are observed to
be only marginally greater than, if not similar to, those of
nondrinkers. While the epidemiologic evidence on the
relation between alcohol and different diseases is now
extensive (3), almost all cohort studies that have estimated
these relations have used baseline measures of alcohol intake
in analyses. However, because of changing drinking habits in
individuals over time and random measurement errors,
assessments of alcohol intake taken at baseline may not
reflect true usual drinking practices of study participants over
the period of risk being studied. Systematic changes and
errors in reported alcohol consumption have the potential to
change the magnitude and even direction of estimated doseresponse relations. While this phenomenon is recognized (6),
few epidemiologic studies have been able to directly estimate
the impact of these influences on estimated risk associations.
In this paper, data from a prospective study of cardiovascular disease among middle-aged British men (the British
Regional Heart Study) are used to examine associations
between alcohol intake and the 20-year risk of coronary
heart disease, stroke, and all-cause mortality. Using repeated
measurements of alcohol intake collected after approximately 5, 13, 17, and 20 years of follow-up, we assessed the
impact of individual variation in reported alcohol intake on
estimated disease relations.
MATERIALS AND METHODS
The British Regional Heart Study is a prospective study
of cardiovascular disease in one general practice in each of
Correspondence to J. R. Emberson, Clinical Trial Service Unit, Harkness Building, Radcliffe Infirmary, Oxford OX2 6HE, United Kingdom
(e-mail: [email protected]).
856
Am J Epidemiol 2005;161:856–863
Variation in Alcohol Intake over Time and Disease Outcome
24 British towns (7). In 1978–1980, 7,735 men aged 40–59
years were recruited into the study. Since the baseline
assessment (which included a nurse-administered questionnaire, available at www.ucl.ac.uk/primcare-popsci/brhs),
men have been followed for all-cause mortality and
cardiovascular morbidity, with fewer than 1 percent being
lost to follow-up (8). Postal questionnaires were completed
by surviving study participants in 1983–1985 (on the fifth
anniversary of each participant’s entry into the study), in
1992, and in 1996. Between 1998 and 2000, surviving men
were invited to attend a rescreening, at which a further
questionnaire was completed. Response rates to the postal
questionnaires were high, ranging from 98 percent to 88
percent, while 77 percent of surviving men attended the
screening between 1998 and 2000 and completed the
questionnaire.
Baseline evidence of cardiovascular disease
Participants completed a World Health Organization
(Rose) chest pain questionnaire and were asked about recall
of any physician diagnosis of myocardial infarction, stroke,
or angina and whether they had ever had a history of severe
chest pain lasting half an hour or more that caused them to
consult a physician. Individuals with recall of myocardial
infarction, angina, or stroke, or with a history of severe chest
pain or Rose angina questionnaire evidence of definite or
possible angina (or any combination of these), were
excluded from analyses.
Baseline assessment of alcohol intake
Alcohol intake, ascertained from questions inquiring
about frequency, quantity, and type of alcoholic beverage
consumed, was categorized into five groups: 1) nondrinkers;
2) occasional drinkers (1–2 times/month or on special
occasions); 3) light drinkers (1–2 drinks/day or ‘‘weekend
only’’ drinkers (1–6 drinks/day)); 4) moderate drinkers (3–6
drinks/day or weekend only drinkers (>6 drinks/day)); and
5) heavy drinkers (>6 drinks every day) (9). Twenty-five
biochemical and hematologic factors analyzed from a blood
sample taken at the time the questionnaire was completed
indicated that the reported levels of alcohol consumption
were valid on a group basis (10).
Follow-up questionnaires and ‘‘average’’ alcohol intake
At each of the four follow-up questionnaires, surviving
participants were asked about their drinking habits, and
alcohol intake was recategorized according to the baseline
definitions. Using these data, we calculated ‘‘average’’
alcohol intake over the study period (or until the time to
first myocardial infarction or stroke, if observed) for each
individual. A five-point scale from zero (none) to four
(heavy) was used to denote the alcohol intake level at the
baseline assessment and at each of the follow-up assessments. Changes in consumption reported between consecutive questionnaires were assumed to occur linearly over the
intervening period, unless a major coronary heart disease
event or stroke occurred during that period, in which case it
was assumed that the individual continued to drink at the
Am J Epidemiol 2005;161:856–863
857
same rate as reported at the most recent questionnaire until
the date of that event. From these average exposure levels,
each individual was reclassified on the original scale (an
average exposure of <0.5 was defined as ‘‘none,’’ 0.5–1.49
was defined as ‘‘occasional,’’ 1.5–2.49 was defined as
‘‘light,’’ 2.5–3.49 was defined as ‘‘moderate,’’ and 3.5
was defined as ‘‘heavy’’).
Assessment of incident cardiovascular morbidity and
all-cause mortality
Information on incident mortality was collected through
the established ‘‘tagging’’ procedures provided by the
National Health Service central registers. Fatal coronary
events were defined as deaths from ischemic heart disease,
including sudden death of presumed cardiac origin
(International Classification of Diseases, Ninth Revision,
codes 410–414) and fatal strokes as deaths with samesource codes 430–438. Nonfatal heart attacks and strokes
were ascertained from general practitioner reports supplemented by systematic reviews of patient records (including
hospital and clinic correspondence) every 2 years throughout the study period (11). Nonfatal heart attacks were
diagnosed according to established World Health Organization criteria, while nonfatal strokes were defined as all
cerebrovascular events that produced a neurologic deficit
present for more than 24 hours (12). In this paper, major
coronary heart disease events include death from coronary
heart disease and nonfatal myocardial infarction.
Statistical analysis
Major coronary heart disease, stroke, and all-cause
mortality rates were calculated per 1,000 person years of
follow-up and directly standardized to the age distribution
of the cohort. Cox proportional hazards regression was used
to estimate the age-adjusted hazard ratio of major coronary
heart disease, stroke, and all-cause mortality over 20 years
by alcohol intake level, from both baseline data (before
adjustment for intake variation) and ‘‘average’’ alcohol
intake (after adjustment for intake variation). For presentation purposes, hazard ratios were calculated as ‘‘floating
absolute risks’’ (13), and inverse variance-weighted quadratic curves were fitted through the values (figure 1).
‘‘Occasional’’ drinkers were used as the referent category in
preference to ‘‘nondrinkers’’ because of the problems in
separating former drinkers from lifelong teetotalers in the
latter group. The ‘‘relative informativeness’’ of baseline
versus average alcohol intake was evaluated by examining
the contributions made by each measure to the v2 likelihood
ratio statistic in the Cox regression model (14). Analyses
were subsequently repeated after adjustment for cigarette
smoking, physical activity, and body mass index, to assess
the independent effects of alcohol on cardiovascular and allcause mortality risk taking into account other important
lifestyle characteristics. Analyses were not adjusted for
blood pressure or high density lipoprotein cholesterol, as
these factors are likely to mediate some of the alcoholdisease relation. The effect of restricting analyses to men
who completed all four follow-up questionnaires (or else all
858 Emberson et al.
questionnaires prior to their date of first major cardiovascular disease event or death) was assessed in a sensitivity
analysis.
RESULTS
Of the 7,735 men recruited into the study, 1,186 (15.3
percent) had baseline evidence of cardiovascular disease
(see Materials and Methods) and were excluded from
analyses. Of the remaining men, five had incomplete
baseline data on alcohol consumption, leaving 6,544 men
for analysis.
Alcohol intake over the study period
FIGURE 1. Relative hazard of major coronary heart disease (CHD)
(coronary death and nonfatal myocardial infarction), stroke, and allcause mortality by alcohol intake, among British Regional Heart Study
men originally free from cardiovascular disease followed from 1978/
1980 to 1998/2000. The black circles and solid line correspond to
baseline alcohol intake levels, and the white circles and dashed line
correspond to ‘‘usual’’ alcohol intake levels obtained after adjustment
for individual variation in alcohol intake. The size of each plotting
symbol indicates the amount of statistical information on which each
estimate is based; the area is inversely proportional to the variance of
the log floating absolute risk. The vertical lines show 95% confidence
intervals for the floating absolute risks.
Information on alcohol intake was obtained for 6,544 men
at baseline, for 6,127 men in 1983–1985, for 4,916 men in
1992, for 4,433 men in 1996, and for 3,706 men at the
20-year screening between 1998 and 2000 (table 1). At
baseline, 358 men (5.5 percent) were classified as nondrinkers and 696 (10.6 percent) as heavy drinkers. The
modal category was ‘‘light’’ drinking, comprising one third
of the study cohort, with ‘‘occasional’’ and ‘‘moderate’’
drinkers each contributing approximately one quarter of the
men. During the follow-up period, there was a steady
downward trend in the amount of alcohol consumption
reported by surviving study participants (probably because
of a combination of the cohort’s becoming older and
drinking less, possible secular changes in alcohol consumption with time, and survival ‘‘selection’’ effects). The result
of this was that, by 20 years of follow-up, only 2.8 percent of
the surviving men (who responded to the questionnaire)
were classified as heavy drinkers while 10.0 percent were
classified as nondrinkers.
For each subject, average alcohol intake during his
‘‘exposure period’’ (time until censoring or first cardiovascular event, whichever is lowest) was calculated, and men
were reclassified into the original categories. Table 2 shows
how the baseline classifications compare with those derived
from ‘‘average’’ levels of alcohol consumption. The extent
to which men were classified differently based on their
baseline and average alcohol intake levels indicates the
amount of individual variation in alcohol intake present.
Intake variation can be seen to increase with increasing
alcohol consumption, as reflected by the observation that 78
percent of those originally defined as nondrinkers were
defined the same when based on average levels, compared
with 65 percent of those originally defined as light drinkers
and only 28 percent of those originally defined as heavy
drinkers. Baseline assessment of alcohol exposure resulted
in substantial overestimation of the number of men who
were truly heavy drinkers during the study period and
underestimation of the number of men who, on average,
could be classified as nondrinkers.
Alcohol intake and the risk of major coronary heart
disease, stroke, and all-cause mortality
The relations between alcohol intake (before and after
adjustment for intake variation) and the risk of major
Am J Epidemiol 2005;161:856–863
Variation in Alcohol Intake over Time and Disease Outcome
859
TABLE 1. Responses to alcohol intake at baseline and after approximately 5, 13, 17, and 20 years of follow-up, British Regional
Heart Study, 1978/1980–1998/2000
Follow-up questionnaires of surviving men
Baseline
Reported alcohol intake
No.
None
Occasional
5 years
%
No.
13 years
%
No.
17 years
%
No.
20 years
%
No.
%
358
5.5
565
9.2
820
16.7
699
15.8
370
10.0
1,556
23.8
1,826
29.8
1,188
24.2
963
21.7
1,000
27.0
Light
2,189
33.5
2,294
37.4
2,032
41.3
1,981
44.7
1,636
44.1
Moderate
1,745
26.7
1,179
19.2
705
14.3
646
14.6
598
16.1
Heavy
Total
696
10.6
264
4.3
171
3.5
144
3.3
102
2.8
6,544
100.0
6,127
100.0
4,916
100.0
4,433
100.0
3,706
100.0
coronary heart disease, stroke, and all-cause mortality over
20 years are shown in figure 1 and table 3. Estimated hazard
ratios and confidence intervals are shown relative to
occasional drinkers.
variation, however. Relative to occasional drinkers, regular
heavy drinkers had a 74 percent higher risk of major
coronary heart disease and a 133 percent higher risk of
stroke (these estimates were 8 percent and 54 percent,
respectively, when derived from baseline exposure). The
contribution made by alcohol intake to the likelihood ratio
statistics for the prediction of major coronary heart disease
and stroke increased markedly after adjustment for intake
variation (from 9.4 to 37.0 for major coronary heart disease
and from 10.6 to 20.1 for stroke), indicating substantial
increases in the ‘‘relative informativeness’’ of alcohol intake
for both diseases.
Alcohol and the risk of major coronary heart disease
and stroke
A total of 922 men (14.1 percent) experienced a major
coronary heart disease event within 20 years, and 352 men
(5.4 percent) experienced a stroke (age-standardized rates of
8.4 and 3.2 per 1,000 person-years, respectively). Before
adjustment for intake variation, a shallow U-shaped relation
between alcohol intake and major coronary heart disease
risk was observed, with ‘‘light’’ drinkers having the lowest
risk, 19 percent lower than for occasional drinkers. For
stroke, an apparently stronger U-shaped relation was
observed (though this could equally be a chance observation), with nondrinkers and heavy drinkers experiencing
similarly high stroke risks and light drinkers again experiencing the lowest risk. Adjustment for intake variation
increased the estimated benefits for major coronary heart
disease risk associated with light drinking to 26 percent,
whereas for stroke, the relative differences among nondrinking, occasional drinking, and light drinking groups
became small and insignificant. Risks associated with heavy
drinking were greatly increased after adjustment for intake
Alcohol and all-cause mortality
A total of 1,552 men (23.7 percent) died of all causes
within 20 years (age-standardized death rate of 13.7 per
1,000 person-years). Again, a U-shaped relation was
observed with baseline alcohol intake. However, taking
variation in alcohol intake over time into account caused an
increase in the all-cause mortality risks associated with
moderate and heavy drinking. Regular heavy drinking was
associated with a 127 percent increase in all-cause mortality
risk relative to occasional drinkers (44 percent before
adjustment for intake variation). Light drinking was associated with the lowest all-cause mortality risk, 18 percent
lower than for occasional drinkers. The ‘‘relative
TABLE 2. Comparison of alcohol intake measured at baseline with ‘‘average’’ alcohol intake over the follow-up period, British
Regional Heart Study, 1978/1980–1998/2000
Average alcohol intake level during the study
Baseline assessment of
alcohol intake
None
Occasional
%
None
280
78
66
18
Occasional
264
17
974
41
2
562
9
1
Light
Moderate
Heavy
Total
594
9
Am J Epidemiol 2005;161:856–863
No.
Light
No.
%
No.
Moderate
%
11
3
63
307
26
1,421
156
9
19
1,777
No.
Heavy
%
No.
Total
%
No.
%
1
0
358
100
20
11
1
65
164
7
1
0
1,556
100
2,189
842
48
689
39
49
3
1,745
100
100
3
113
16
366
52
198
28
696
100
27
2,694
41
1,231
19
248
4
6,544
100
1.84, 2.81
1.15, 1.52
0.72, 0.93
0.77, 1.12
95%
confidence
interval
2.27
Sensitivity analysis
1.06, 2.22
1.54
* Estimated hazard ratios and 95% confidence intervals are shown relative to occasional drinkers.
1.32
1.44
1.21, 1.72
1.12
1.46, 3.71
0.88, 1.61
1.19
1.31, 2.33
1.74
0.86, 1.35
1.08
1.01
0.78, 1.12
0.94
Heavy
Moderate
0.84, 1.21
0.72, 1.31
0.97
0.74
0.68, 0.96
0.81
Light
0.63, 0.87
1.02, 2.44
1.00
1.58
1.00
Occasional
0.91
0.75, 1.34
1.00
1.00
None
0.72, 1.15
95%
confidence
interval
Hazard
ratio
95%
confidence
interval
Hazard
ratio
95%
confidence
interval
Hazard
ratio
Level of
alcohol intake
The primary analyses were repeated after adjustment for
body mass index, physical activity, and cigarette smoking,
factors known to be related to both alcohol consumption and
the risk of cardiovascular disease and all-cause mortality
(15). The shape of the disease relations with usual alcohol
intake was essentially unchanged by adjustment for these
factors, though the excess risks from moderate and heavy
drinking were reduced. After adjustment for body mass
index, physical activity, and cigarette smoking, but before
adjustment for variation in alcohol intake over time, heavy
drinkers appeared to have a 17 percent lower risk of major
coronary heart disease, a 27 percent higher risk of stroke,
and a 15 percent higher risk of all-cause mortality compared
with occasional drinkers. After adjustment for variation in
alcohol intake, however, risks among heavy drinkers were,
respectively, 32 percent, 86 percent, and 70 percent higher
than for occasional drinkers.
2.33
1.45
1.08, 1.96
0.98, 1.29
0.82
0.88
0.93
0.71, 1.22
0.77, 1.01
1.00
0.93
0.98, 1.52
1.00
1.22
0.73, 1.60
1.08
1.00
Hazard
ratio
95%
confidence
interval
Hazard
ratio
95%
confidence
interval
Hazard
ratio
Usual exposure
All-cause mortality
Baseline exposure
Usual exposure
informativeness’’ increased nearly threefold by taking intake variation into account (the likelihood ratio contribution
increased from 38.2 to 101.3).
Effect of adjustment for other lifestyle risk factors
Stroke
Baseline exposure
Usual exposure
Major coronary heart disease
Baseline exposure
TABLE 3. Relative hazard of major coronary heart disease, stroke, and all-cause mortality by alcohol intake, both before (baseline exposure) and after (usual exposure) taking
account of intake variation over time, British Regional Heart Study, 1978/1980–1998/2000*
860 Emberson et al.
When analyses were restricted to the 4,536 men (69.3
percent) who completed all four follow-up questionnaires
(or all questionnaires up to their first major cardiovascular
disease event or death), the estimated disease risks for heavy
drinkers relative to occasional drinkers increased still
further, though confidence intervals were, of course, wider
(results available from authors on request).
DISCUSSION
In middle-aged British men with no previous evidence
of cardiovascular disease, after adjustment for individual
variation in intake, regular light alcohol consumption was
associated with a statistically significant 26 percent reduced
risk of coronary heart disease and 18 percent reduced risk of
all-cause mortality, as well as a statistically insignificant
7 percent reduced risk of stroke (compared with occasional
drinkers). Moderate drinking and heavy drinking, on the
other hand, were associated with substantially increased
risks of stroke, all-cause mortality, and (to a lesser degree)
major coronary heart disease. Stroke and all-cause mortality
risks were more than twice as high among heavy drinkers
than among occasional drinkers, while major coronary heart
disease risk was 74 percent higher in this group (all p’s <
0.001). However, before variation in alcohol intake over
time was taken into account, only moderate differences in
stroke and all-cause mortality risks between heavy drinkers
and occasional drinkers were observed (with no significant
differences in major coronary heart disease risk).
Validity of methods
In this paper, it was possible to relate 20-year cardiovascular and all-cause mortality risks to average alcohol intake
Am J Epidemiol 2005;161:856–863
Variation in Alcohol Intake over Time and Disease Outcome
during the study because, in addition to the baseline
assessment, alcohol intake was reassessed in a high proportion of surviving men after approximately 5, 13, 17, and
20 years of follow-up. These repeated assessments allowed
estimation of ‘‘usual’’ alcohol intake over the study period
for each individual. Crucially, because only information on
alcohol consumption that was recorded while the subject
was still at risk of a first major cardiovascular event was
used in this process, our results are unlikely to be affected by
reverse causality bias (the effect on the alcohol-disease
relation that could otherwise be caused by nonfatal cardiovascular events leading to subsequent change in drinking
pattern). Alternatively, we could have fitted alcohol consumption as a ‘‘time-dependent’’ covariate, that is, a variable
that could change value at the time each questionnaire was
completed. However, because of the fairly long intervals
between successive questionnaires as well as the desire to
use information gained from the 20-year questionnaire in
our analyses (information which would effectively be
redundant in any time-dependent approach), it was decided
to use the information gained from each individual to create
a single measure that best represented his usual alcohol
intake exposure during the study period. Our method of
analysis should therefore provide reliable estimates of the
true relations between long-term ‘‘usual’’ alcohol consumption patterns and the development of cardiovascular disease
and the risk of death. However, it is possible that some men
may have reduced their alcohol intake from ‘‘moderate’’ or
‘‘heavy’’ drinking to ‘‘light’’ drinking because of ill health
short of major cardiovascular disease. If these men were
subsequently classified as ‘‘light’’ drinkers on the basis of
usual levels but retained the risks associated with their
previous drinking habits, then the corrected risk reductions
for ‘‘light’’ drinking shown in figure 1 may still be too
conservative. It could therefore be argued that the true risk
reductions associated with ‘‘light’’ drinking are greater than
the adjusted estimates shown. Similarly, it is possible that,
by being classified as heavy drinkers at one assessment and
as nondrinkers or light drinkers at the next, some ‘‘binge
drinkers’’ may have been incorrectly classified as light or
moderate drinkers on the basis of usual average intake,
leading to artificially low estimates of the benefits associated with these levels of drinking (since binge drinking is
associated with high excess mortality risks (16)). In
addition, misclassification of noncoronary heart disease
deaths as coronary heart disease in heavy drinkers could
have led to the estimated coronary heart disease risks
associated with heavy drinking being artificially high.
However, we feel that each of these potential sources of
bias is unlikely to have had a substantial effect in our
analyses for the following reasons. First, among the 2,694
men classified as ‘‘light’’ drinkers on the basis of ‘‘usual’’
intake, no major (or even minor) differences in major
cardiovascular disease or all-cause mortality rates were
observed between the 1,421 men who were classified as
light drinkers at baseline (men whose drinking habit was
constant over the exposure period) and the 955 men who
were classified as moderate or heavy drinkers at baseline
(men whose drinking habits varied during the exposure
period). Second, while it is possible that our approach led to
Am J Epidemiol 2005;161:856–863
861
some binge drinkers being incorrectly classified into the
light or moderate drinking categories, it should be reiterated
that the use of ‘‘averaged’’ exposures did lead to substantially improved categorization of individuals (as reflected by the increases in the likelihood ratio statistic), an
observation that would not have been expected if they
exhibited a greater degree of misclassification than the
baseline exposures. Misclassification of noncoronary heart
disease deaths as coronary heart disease in heavy drinkers
could potentially affect the validity of both ‘‘baseline’’ and
‘‘usual’’ alcohol-coronary heart disease associations. However, the differences between the risk relations observed
using baseline compared with usual exposure categories
should not have been materially affected by this potential
source of error, as any such biases would be likely to apply
to both categorizations similarly.
Comparison with other studies
Consistent with the estimated risk-relations presented in
this paper before adjustment for individual alcohol intake
variation, systematic reviews of the effects of alcohol on
mortality risk have consistently shown either U- or J-shaped
relations between alcohol consumption and the risk of
cardiovascular disease and all-cause mortality (1–4), with
light to moderate drinkers experiencing 25–35 percent lower
risks than nondrinkers and heavy drinkers experiencing
risks similar to those of nondrinkers (though the evidence
for a protective effect of moderate alcohol on stroke is less
clear) (17). In the Whitehall II study of British civil servants,
for instance, moderate drinkers had approximately half the
11-year risk of death from coronary heart disease and all
causes compared with either nondrinkers or heavy drinkers
(18), while in the British Doctors’ Study (6), coronary heart
disease death rates were 43 percent lower for men who
drank 15–21 units of alcohol a week (compared with
nondrinkers), death rates from stroke were 17 percent lower,
and deaths from all causes were 28 percent lower (men who
drank at least 43 units of alcohol a week had similar death
rates to those of nondrinkers). Several very large US cohort
studies, including the Physicians’ Health Study (19), the
Nurses’ Health Study (20), Cancer Prevention Study II (21),
and the Multiple Risk Factor Intervention Trial (22), have
found similar associations between alcohol intake and the
risk of cardiovascular disease and all-cause mortality, and
they have confirmed that these relations exist in women as
well as men (20, 21). Previous reports from the British
Regional Heart Study have drawn attention to the strong
downward trend with increasing age, from heavy or
moderate drinking to light, occasional, or nondrinking
status, affected to a considerable extent by the accumulation
of ill health over time (23, 24). Reduction in alcohol intake
or giving up drinking is associated with higher rates of new
diagnoses and increased cardiovascular and noncardiovascular mortality than remaining stable (23). Attention has also
been drawn to the disadvantages of using the nondrinkers
(lifelong teetotalers and former drinkers) as a referent group,
as this may lead to further exaggeration of the protective
effect of regular light/moderate drinking (24).
862 Emberson et al.
However, most previous studies have used only baseline
measures of alcohol consumption in analyses and have not
attempted to quantify the role that individual variation in
alcohol intake could play, although in the British Doctors’
Study, a reasonable degree of consistency between alcohol
intake at the beginning and end of the study suggested that
their results may have been fairly robust to these effects (6).
Of the studies that have attempted to take changes in
individual drinking patterns into account, most have used
just two assessments of alcohol intake, and findings have
been inconsistent (25, 26). In the MONICA-Augsburg cohort,
for instance, it was found that the estimated benefits of light
alcohol consumption increased after taking the second
measure of alcohol consumption into account (26), while
in the First National Health and Nutrition Examination
Survey, no elevated mortality was observed when consistent
never drinkers were compared with light drinkers (25). In
the Health Professionals Follow-up Study, assessment of
alcohol intake every 4 years allowed examination of the
effects of changes in alcohol consumption on the 12-year
risk of myocardial infarction. In this large American study,
the beneficial effects of alcohol consumption on the risk of
myocardial infarction estimated using baseline measurements were found to be similar to estimates derived from
analyses that fitted alcohol consumption as a ‘‘timedependent’’ covariate (27). However, three other large
American studies have demonstrated that baseline measures
of drinking groups may be particularly unreliable for younger
samples, longer follow-up, and heavier drinkers (28).
cular and, particularly, all-cause mortality risk from moderate alcohol consumption. Furthermore, it is suggested that
the risks associated with heavy drinking may have been
systematically underestimated in previous prospective studies (because of the small proportion of men who tend to
remain heavy drinkers over a prolonged period of time). The
potential for excess disease risks caused by heavy drinking
to be greatly underestimated (or even completely missed)
when using baseline assessments of alcohol intake in
analyses is particularly worrying. Unless information from
follow-up assessments of alcohol consumption is used in
analyses, it is likely that prospective observational studies,
and in particular those with long follow-up periods, will
continue to systematically underestimate these risks. While
some of the effect sizes presented in this paper may be
exaggerated by the role of chance, the results suggest that
considerable caution may be needed before any recommendations regarding acceptable limits of alcohol consumption
are made.
ACKNOWLEDGMENTS
The British Regional Heart Study is a British Heart
Foundation research group and receives additional support
from the Department of Health (England). J. R. E. was
supported by a British Heart Foundation Junior Research
Fellowship during the period this work was carried out.
The views expressed in this publication are those of the
authors and not necessarily those of the funding agencies.
Implications for epidemiologic studies
Prospective observational studies help to evaluate the
strength and shape of relations between risk factors and the
development of disease. However, by relying solely on risk
factor measurements taken at a single point in time, true
long-term differences in disease risk between risk groups
can become imprecisely assessed because of exposure
variation over time. It is well recognized that within-person
variation in continuous risk factors, such as blood cholesterol and blood pressure, tends to lead to underestimation of
true risk associations (29, 30). Similarly, we have found that
cardiovascular disease risks associated with heavy cigarette
smoking and physical inactivity can also be underestimated
when baseline measures are used in analyses (31). In this
paper, adjustment for intake variation led to an increase in the
risks associated with moderate, and particularly with heavy,
drinking. Epidemiologic studies, and in particular those with
long follow-up periods, should therefore endeavor to determine whether any systematic changes in risk exposure
during the study have occurred. A range of approaches,
including the method described in this paper, may be used to
examine the effect that these changes may have on estimated
disease associations.
Conclusions
The results in this paper suggest that previous prospective
studies may have overestimated the benefits for cardiovas-
REFERENCES
1. Beaglehole R, Jackson R. Alcohol, cardiovascular disease and
all causes of death: a review of the epidemiological evidence.
Drug Alcohol Rev 1992;11:175–89.
2. Beilin LJ, Puddey IB. Alcohol and cardiovascular disease—
more than one paradox to consider. J Cardiovasc Risk
2003;10:1–30.
3. Corrao G, Bagnardi V, Zambon A, et al. A meta-analysis of
alcohol consumption and the risk of 15 diseases. Prev Med
2004;38:613–19.
4. Reynolds K, Lewis B, Nolen JD, et al. Alcohol consumption
and risk of stroke: a meta-analysis. JAMA 2003;289:579–88.
5. Rimm EB, Williams P, Fosher K, et al. Moderate alcohol
intake and lower risk of coronary heart disease: meta-analysis
of effects on lipids and haemostatic factors. BMJ 1999;
319:1523–8.
6. Doll R, Peto R, Hall E, et al. Mortality in relation to consumption of alcohol: 13 years’ observations on male British
doctors. BMJ 1994;309:911–18.
7. Shaper AG, Pocock SJ, Walker M, et al. British Regional Heart
Study: cardiovascular risk factors in middle-aged men in 24
towns. BMJ 1981;283:179–86.
8. Walker M, Shaper AG, Lennon L, et al. Twenty year followup of a cohort based in general practices in 24 British towns.
J Public Health Med 2000;22:479–85.
9. Shaper AG, Wannamethee G, Walker M. Alcohol and
mortality in British men: explaining the U-shaped curve.
Lancet 1988;2:1267–73.
Am J Epidemiol 2005;161:856–863
Variation in Alcohol Intake over Time and Disease Outcome
10. Shaper AG, Pocock SJ, Ashby D, et al. Biochemical and
haematological response to alcohol intake. Ann Clin Biochem
1985;22:50–61.
11. Walker MK, Whincup PH, Shaper AG, et al. Validation of
patient recall of doctor-diagnosed heart attack and stroke:
a postal questionnaire and record review comparison. Am J
Epidemiol 1998;148:355–61.
12. Shaper AG, Phillips AN, Pocock SJ, et al. Risk factors
for stroke in middle aged British men. BMJ 1991;302:
1111–15.
13. Plummer M. Improved estimates of floating absolute risk.
Stat Med 2004;23:93–104.
14. Prospective Studies Collaboration. Age-specific relevance of
usual blood pressure to vascular mortality: a meta-analysis of
individual data for one million adults in 61 prospective studies.
Lancet 2002;360:1903–13.
15. Wannamethee G, Shaper AG. Blood lipids: the relationship
with alcohol intake, smoking, and body weight. J Epidemiol
Community Health 1992;46:197–202.
16. Kauhanen J, Kaplan GA, Goldberg DE, et al. Beer binging and
mortality: results from the Kuopio ischaemic heart disease risk
factor study, a prospective population based study. BMJ
1997;315:846–51.
17. Mazzaglia G, Britton AR, Altmann DR, et al. Exploring the
relationship between alcohol consumption and non-fatal or
fatal stroke: a systematic review. Addiction 2001;96:1743–56.
18. Britton A, Marmot M. Different measures of alcohol consumption and risk of coronary heart disease and all-cause
mortality: 11-year follow-up of the Whitehall II Cohort Study.
Addiction 2004;99:109–16.
19. Gaziano JM, Gaziano TA, Glynn RJ, et al. Light-to-moderate
alcohol consumption and mortality in the Physicians’ Health
Study enrollment cohort. J Am Coll Cardiol 2000;35:96–105.
20. Fuchs CS, Stampfer MJ, Colditz GA, et al. Alcohol consumption and mortality among women. N Engl J Med 1995;
332:1245–50.
21. Thun MJ, Peto R, Lopez AD, et al. Alcohol consumption and
mortality among middle-aged and elderly U.S. adults. N Engl
J Med 1997;337:1705–14.
Am J Epidemiol 2005;161:856–863
863
22. Suh I, Shaten BJ, Cutler JA, et al. Alcohol use and mortality
from coronary heart disease: the role of high-density lipoprotein cholesterol. The Multiple Risk Factor Intervention
Trial Research Group. Ann Intern Med 1992;116:881–7.
23. Shaper AG, Wannamethee SG. The J-shaped curve and
changes in drinking habit. Novartis Found Symp 1998;216:
173–88.
24. Shaper AG, Wannamethee SG. Epidemiological confounders
in the relationship between alcohol and cardiovascular disease.
In: Paoletti R, Klatsky AL, Poli A, et al, eds. Moderate alcohol
consumption and cardiovascular disease. Dordrecht, Netherlands: Kluwer Academic, 2000:105–12.
25. Fillmore KM, Kerr WC, Bostrom A. Changes in drinking
status, serious illness and mortality. J Stud Alcohol 2003;
64:278–85.
26. Wellmann J, Heidrich J, Berger K, et al. Changes in alcohol
intake and risk of coronary heart disease and all-cause
mortality in the MONICA/KORA-Augsburg cohort 1987–97.
Eur J Cardiovasc Prev Rehabil 2004;11:48–55.
27. Mukamal KJ, Conigrave KM, Mittleman MA, et al. Roles of
drinking pattern and type of alcohol consumed in coronary
heart disease in men. N Engl J Med 2003;348:109–18.
28. Kerr WC, Fillmore KM, Bostrom A. Stability of alcohol
consumption over time: evidence from three longitudinal
surveys from the United States. J Stud Alcohol 2002;63:
325–33.
29. Clarke R, Shipley M, Lewington S, et al. Underestimation of
risk associations due to regression dilution in long-term
follow-up of prospective studies. Am J Epidemiol 1999;
150:341–53.
30. MacMahon S, Peto R, Cutler J, et al. Blood pressure,
stroke, and coronary heart disease. Part 1. Prolonged
differences in blood pressure: prospective observational
studies corrected for the regression dilution bias. Lancet
1990;335:765–74.
31. Emberson JR, Whincup PH, Morris RW, et al. Lifestyle and
cardiovascular disease in middle-aged British men: the
effect of adjusting for within-person variation. Eur Heart J
(in press).