Stroke Risk From Alcohol Consumption

1093
Stroke Risk From Alcohol Consumption
Using Different Control Groups
Yoav Ben-Shlomo, MRCP; Hugh Markowe, MFPHM;
Martin Shipley, MSc; and M.G. Marmot, FFPHM
Background and Purpose: Our aim in this study was to investigate the relation between chronic alcohol
consumption and stroke.
Methods: A case-control study was carried out using two hospital-based control groups and the results
of a community-based survey of alcohol consumption. Hospital-based control subjects were chosen either
from “general” medical admissions or a subset of “select” admissions that excluded possible alcoholrelated admissions. Cases were selected from hospital inpatients.
Results: The relative risk for stroke associated with alcohol consumption greater than 300 grams per
week for general control subjects was 0.73 (95% confidence interval [CI], 0.54-3.49) compared with 1.30
(95% CI, 0.42-4.05) for select control subjects. The odds ratio was further increased to 1.93 (95% CI,
0.87-4.28) using data from the community-based survey. None of these estimates were statistically
significant.
Conclusions: These results illustrate how the risk associated with alcohol consumption varies depending
on the choice of control groups and may explain the contradictory results from previous case-control
studies. Because of different biases associated with control selection, we believe that the results of this
study are consistent with those of other studies that demonstrate a modest increased risk for stroke
associated with alcohol consumption.
(Stroke 1992;23:1093-1098)
KEY WORDS • alcohol drinking • risk factors
A lcohol has been recognized as a possible risk
/ factor for stroke from as early as 1725.’ Since
J_ V . that time, at least 62 different epidemiological
studies have examined this association.
2 A case-control
study from Birmingham, UK,
3 was the first to suggest a
markedly raised odds ratio for heavy drinking among
men. The use of elective surgical patients as control
subjects may have overestimated the true odds ratio
because of an underrepresentation of heavy drinkers
(exclusion bias). As different control groups may produce
varying results in case-control studies, we have examined
the effect on the relation between chronic alcohol consumption and stroke using two sets of controls: general
medical admissions plus a subset of admissions excluding
possible alcohol-related conditions, and in addition a
community survey from a general practice.
Subjects and Methods
Patients were recruited from admissions to three
adjoining hospitals. They were identified through daily
From the Department of Epidemiology and Public Health,
University College and the Middlesex School of Medicine (Y.B.-
S.), and the Department of Epidemiology and Population Sciences, London School of Hygiene and Tropical Medicine (H.M.,
M.S., M.G.M.), London.
Supported by a grant from the Chest, Heart and Stroke Association, London, UK. Y.B.-S. is supported by a Wellcome Fellowship Trust in Clinical Epidemiology.
Address for correspondence: Dr. Y. Ben-Shlomo, Department
of Epidemiology and Public Health, University College and the
Middlesex School of Medicine, 66-72 Gower Street, London
WC1E 6EA, UK.
Received August 13, 1991; accepted March 27, 1992.
contact with the admissions office, accident and emergency department, junior medical staff, and, where
available, computed tomography department. Inclusion
criteria for cases were age between 15 and 69 years, no
past history of stroke or transient ischemic attacks, and
diagnosis of possible stroke by admitting medical team.
Cases in which subsequent diagnostic information altered the diagnosis were later excluded.
Hospital controls were chosen from the same sources
as the cases and met the same inclusion criteria but had
an admission diagnosis other than possible or definite
stroke. Two control groups were chosen, and these will
be referred to as the “general” or “select” controls.
Both groups were matched to cases on the basis of sex,
hospital of admission, age (5-year age-matching was
used for select controls whereas 10-years matching was
used for general controls), and time of week of admission (weekend or weekday). Potential control subjects
were excluded if the primary reason for current admission was an acute general surgical admission, a social
admission, an overdose, or mental illness (including
alcoholism); or if the patient was unable to speak
English or was not a UK resident. In addition, the select
group had as grounds for exclusion a further list of
medical conditions as reasons for admission that could
possibly be related to alcohol ingestion: cirrhosis of the
liver, hemochromatosis, and all other diseases of the
liver; acute gastritis; major gastrointestinal disease; myocarditis or cardiomyopathy; neuropathy; epilepsy; or
myocardial infarction.
Cases and controls were interviewed as soon after
admission as possible. As it was not possible to blind
Downloaded from http://ahajournals.org by on October 18, 2019
1094 Stroke Vol 23, No 8 August 1992
interviewers to the case status of subjects, the interview
was carried out using a structured questionnaire on
subjects or, where not possible, relatives. All interviews
were taped and were later assessed by one of the
investigators, unaware of the subject’s diagnosis, for
possible biases in questioning. No subjective differences
in
questioning between cases and controls were found.
When a relative was interviewed on behalf of a case, this
was also done for the controls to ensure that the quality
of information obtained was comparable. Subjects were
informed that the purpose of the study was to investigate the role of lifestyle factors in sickness and health.
They were not aware that the study was specifically
examining cerebrovascular disease and its possible relation with alcohol. Several questions on other healthrelated activities were included to mask the focus on
alcohol. Alcohol consumption was assessed by a quantity/frequency measure based on the General Household Survey. Weekly alcohol intake was calculated and
grams of alcohol per week estimated assuming 1 unit
equals 10 grams. In addition, the CAGE screening
questions
4 were asked; the four simple questions used in
this screening tool have been shown in several studies to
have a sensitivity of 76-93% and specificity of 89-94%
and to be superior to biochemical screening tests in
detection of problem drinkers.
57
All cases and controls had the following information
extracted from their medical notes: clinical history and
examination, biochemical and hematological parameters (specifically, -y-glutamyltranspeptidase, aspartate
transaminase, and mean corpuscular volume), and
other relevant investigations (e.g., lumbar puncture,
computed tomographic [CT] scan, and cerebral angiography). The laboratory data were taken from routine
clinical workup.
Patients were accepted as cases only if they had
objective evidence of a stroke from CT scans, which
were performed in 54% of cases, or from lumbar
puncture/cerebral angiography. The remaining subjects
with a clinical diagnosis of stroke were independently
reviewed by either a consultant neurologist or one of the
investigators and were included if they fulfilled the
World Health Organization criteria for stroke: “A focal
or global disturbance of cerebral function leading to
death or persisting for more than 24 hours, with no
apparent cause other than vascular.” Subdivision of
stroke types into hemorrhage or infarction was based on
either a clinically/radiologically proven diagnosis or the
Allen criteria
8 with an 80% cutoff. This scale uses
routine information from clinical history and physical
examination to allocate cases into an appropriate subtype. Using these two methods, only 8 (4.9%) cases
were unclassifiable. The discharge diagnoses for all
hospital controls were obtained and coded according to
the ninth revision of the International Classification of
Diseases (ICD).
9
As an additional source of data on alcohol consumption, we surveyed patients from a general practice
whose catchment area included two of the three hospitals. The age-sex register for all patients 16-70 years of
age was the sampling frame for the survey. The survey
questionnaire used wording identical to that in the
case-control study for questions concerning alcohol
consumption. Questions on smoking and exercise were
also included because subjects were informed that they
were taking part in a health survey. The questionnaire
was tested on 25 individuals to check for comprehensibility and ambiguity. Because this central London population is known to be very mobile and approximately
25% of registered patients are living in temporary
residences, accurate practice lists are impossible to
maintain. Each letter had a post office return sticker so
that questionnaires not delivered would be returned
with some indication as to why they had failed to reach
the subject. We found 3,382 potential subjects from the
register. However, 14 addresses were duplicated, and 85
subjects had addresses with insufficient detail. The post
office returned 1,264 questionnaires due to either an
inaccurate address or the subject no longer residing at
the address, leaving 2,018 questionnaires.
All data on occupation were classified by the Registrar General’s criteria for social class.
10 This groups
occupations into six categories from social class I (highest) to social class V (lowest). Those subjects who were
retired were classified according to their last full-time
occupation. Women were classified according to their
husband’s occupation if married or classified by their
occupation if unmarried.
The data were analyzed using
SAS/PC and EGRET for
regression analysis. Comparison of proportions was
performed by the
x2 test and mean values by t tests.
Because aspartate transaminase values were not normally distributed and transformations were relatively
ineffective, the Wilcoxon rank sum score was used.
Odds ratios were calculated by logistic regression. Because data on alcohol consumption were missing from
some cases, analysis of the hospital control groups using
individual matching resulted in a reduced sample size
and power. In this study, however, the four matching
criteria for the hospital control subjects give rise to
relatively few strata. In this situation, it has been
shown
11 that a superior analysis is achieved by stratifying the cases and controls according to their matching
criteria and analyzing the data using conditional logistic
regression. In our data set, the size of some of the strata
became so large that conditional logistic regression
became computationally unfeasible. We therefore analyzed the data using unconditional logistic regression
but at the same time ensuring that there were sufficient
terms in the models to control for the matching factors.
12 This, in fact, enabled us to control for age more
precisely than did the original matching criteria. The
community control subjects were not individually
matched and were analyzed in a conventional fashion by
unconditional logistic regression. Possible confounding
variables were chosen for inclusion in the models if they
could be related to both the risk of disease outcome and
possible exposure, e.g., smoking. A history of medication was also included as a broad marker of a chronic
disease process. Social class was used in the early
models but did not alter the odds ratios for the hospital
control groups and was therefore excluded. Tests of
significance were calculated using likelihood ratio tests,
and odds ratios and 95% confidence intervals were
calculated from the logistic regression coefficients and
standard errors.
Results
One hundred sixty-four eligible cases were selected
for the study: 111 (68%) cases of thromboembolic
Downloaded from http://ahajournals.org by on October 18, 2019
Ben-Shlomo et al Alcohol Consumption and Stroke 1095
stroke, 25 (15%) of hemorrhagic stroke, 20 (12%) of
subarachnoid hemorrhage, and 8 (5%) of unclassified
stroke; 165 general control and 115 select control subjects were also recruited. Because select controls had far
more exclusion criteria, they were harder to obtain.
Occasionally, control subjects were interviewed before a
case was confirmed, which resulted in a greater number
of general control subjects than cases and an unequal
number of men and women in each group. Cases and
controls did not differ significantly by age, sex, and
social class. Non-Caucasians were significantly more
common among the cases (p<0.02). The select controls
also had a greater representation of single men
(/7<0.02), although this was not significant if marital
status for males and females was combined.
The three most common diagnoses among the general
controls were ischemic heart disease (ICD 410-414)
(35%), other circulatory diseases excluding ischemic
heart disease (ICD 390-409, 415-459) (20%), and
respiratory diseases (ICD 460-519) (11%). For the
select controls, the most common diagnoses were respiratory diseases excluding acute respiratory diseases
(ICD 467-519) (29%), other circulatory diseases excluding ischemic heart disease (ICD 390-409,415-459)
(16%), and acute respiratory diseases (ICD 460-466)
(15%). This variation reflects the different exclusion
criteria for the select group.
From the total sample of 164 cases, data on alcohol
consumption were only obtained on 115 cases. No data
were obtained for 49 cases because in 29 cases (18%),
the patients either died or were severely dysphasic, and
no history could be obtained from a relative; in nine
cases (5%), the patients were discharged before interview; and in 11 cases (7%), the patients did not give
consent to participate in the study. These cases with
missing alcohol data were compared with the 115 cases
included (Table 1). No significant differences were
found for demographic variables or laboratory results.
Mean aspartate transaminase values were greater for
cases with missing alcohol data, but this was not significant (/?=0.07 by Wilcoxon rank sum score).
The relation between stroke and alcohol consumption
is shown in Table 2 as odds ratios comparing cases with
each control group. There were no significant associations between alcohol consumption or the classification
based on the CAGE questions and risk of stroke. All
analyses using the hospital-based control subjects were
adjusted for age, sex, hospital, and day of admission
(weekday/weekend). When using hospital-based control
subjects, there were no significant associations between
reported alcohol consumption or classification based on
the CAGE questions and risk of stroke. The relation
between stroke and alcohol consumption (grams per
week), adjusted for cigarette smoking, history of hypertension, diabetes, heart disease, and race, is shown in
Table 3. There was no significant interaction between
sex and grams of alcohol consumed on the risk of stroke.
All levels of alcohol consumption were associated with
an increased odds ratio for the select controls, but these
were not significant and did not show a dose-response
relation. For the general control subjects, the highest
level of consumption resulted in an odds ratio of less
than one. Biochemical and hematological markers again
showed no consistent relation, but aspartate transaminase levels showed an inverse relation for only the
TABLE 1. Comparison of Data in Cases With and Without
Alcohol Data
Characteristic
Sex
Males
Females
Age (years)
Mean±SE
Range
Race
Caucasian
Non-Caucasian
Social class
I and II
III
IV and V
Smoking history from doctor’s notes
Smoker
Nonsmoker
Biochemical parameters
Mean corpuscular volume
Mean±SE
Range
Aspartate transaminase
Mean±SE
Range
y-Glutamyltranspeptidase
Mean±SE
Range
Alcohol
Missing
28 (57)
21 (43)
58.7±1.1
37-69
32 (78)
9 (22)
6 (27)
8 (36)
8 (36)
19 (50)
19 (50)
90.7±1.24
74.6-113.0
53.2+11.9
9.0-448.0
60.5 ±18.4
9.0-500.0
data
Available
65 (57)
50 (43)
57.6±0.9
20-69
89 (78)
25 (22)
31 (28)
51 (46)
30 (27)
53 (57)
40 (43)
91.1 ±0.83
66.0-113.0
34.8+3.3
6.0-182.0
57.0±7.3
3.0-362.0
Values in parentheses are percent,
n, Number of cases.
general control group, which indicated that general
control subjects were more likely than stroke patients to
have raised aspartate transaminase levels.
Among the hospital-based control subjects were 22
(8%) with known liver disease who might be expected to
have abnormal liver function tests, compared with only
four (2%) subjects among the cases (^
2, p=0.067).
These 22 subjects were mainly found in the general
control group (21 of 22). The biochemical parameters
were thus reanalyzed after exclusion of all subjects with
a history of liver disease. This further analysis showed
only minor alterations to the odds ratios (data not
shown).
Strokes were subdivided into type and the analyses
repeated for thromboembolic strokes. There were too
few cerebral hemorrhages to permit separate analyses.
Because thromboembolic strokes formed the bulk of the
cases, the reanalysis resulted in only minor differences
from that for all strokes. Because there were small
numbers of non-Caucasian subjects, excluding these did
not materially alter the results.
The analyses in Tables 2 and 3 use nondrinkers as the
baseline reference group. The multivariate analysis was
repeated first excluding ex-drinkers from the nondrinkers and then both ex-drinkers and drinkers who reported drinking much less. This resulted in minor
differences from the odds ratios seen in Table 3.
Downloaded from http://ahajournals.org by on October 18, 2019
1096 Stroke Vol 23, No 8 August 1992
TABLE 2. Relation Between Stroke, Alcohol Consumption, and CAGE Questionnaire
Exposure variable
Current alcohol drinker
No
Yes
Alcohol consumption
(grams per week)
Nondrinker or <10
10-100
100-300
>300
Score of a 2 on CAGE
screening questions
No
Yes
Stroke
cases
to
18
97
43
35
20
17
80
16
(«)
16
68
30
25
16
13
57
9
Hospital control
Select group
Odds ratios
(95% CI)
1.00
1.26 (0.57-2.80)
(x
2=0.33,df=l,/>=0.57)
1.00
1.01 (0.49-2.09)
0.85 (0.37-1.99)
1.16 (0.45-3.00)
(*
2=0.36, df=3,p=0.95)
1.00
1.39 (0.55-3.47)
(*
2=0.50, df=l,p=0.48)
to
14
105
48
29
17
26
88
17
subjects
General group
Odds ratios
(95% CI)
1.00
0.79 (0.36-1.76)
(^=0.32, df=l,p=0.57)
1.00
1.55 (0.77-3.10)
1.39 (0.62-3.13)
0.99 (0.44-2.42)
(A-
2=2.14, df=3,p=0.55)
1.00
1.30 (0.59-2.86)
(*
2=0.41, df=l,p=0.52)
to
106
611
247
307
108
55
518
83
Community survey subjects
Odds ratios
(95% CI)
1.00
1.26 (0.71-2.26)
(*
2=0.67, df=l,p=0.41)
1.00
0.97 (0.58-1.62)
1.58 (0.83-3.00)
2.31 (1.11-4.82)
(*
2=6.85, df=3,p=0.08)
(for trend, *
2=5.26, df=l,p=0.02)
1.00
1.67 (0.86-3.24)
(A-
2=2.20,df=l,p=0.14)
Cases are compared with each control group and the community survey. All models using hospital controls were adjusted for age and sex
and include a term for hospital and day of admission (weekday/weekend). Model comparing cases with community survey was adjusted only
for age and sex.
n, Number of cases or control subjects.
The response rate from the community questionnaires was 37% (752 of 2,018) for all residents, 49%
(684 of 1,401) for permanent residents, and 11% (68 of
617) for temporary residents. The latter were simply
determined by the address (e.g., hotel, YMCA, B&B);
temporary residents registered at a domestic home
address would thus not be identified. Therefore, the
response rate for permanent residents is probably an
underestimate of the true rate because temporary residents were far less likely to respond. The basic demographic details of the community sample were compared with routine census data for the health authority
TABLE 3. Relation Between Stroke, Alcohol Consumption, and CAGE Questionnaire
Exposure variable
Current alcohol drinker
No
Yes
Alcohol consumption
(grams per week)
Nondrinker or <10
10-100
100-300
>300
Score of S 2 on CAGE
screening questions
No
Yes
Stroke
cases
to
16
91
40
33
18
16
73
16
to
14
63
28
22
16
11
53
9
Hospital control
Select group
1.00
2.22
Odds ratios
(95% CI)
(0.83-5.94)
(*
2=2.56, df=l,p=0.11)
1.00
1.39
1.01
1.30
(0.60-3.25)
(0.37-2.76)
(0.42-4.05)
(*
2=0.72, df=3,p=0.87)
1.00
0.99 (0.33-2.96)
(*
2=0.00,df=l,p=0.98)
to
13
98
44
27
17
24
83
15
subjects
General group
Odds ratios
(95% CI)
1.00
0.96 (0.39-2.34)
(*
2=0.08, df=l,/?=0.93)
1.00
1.46 (0.66-3.23)
1.38 (0.54-3.49)
0.73 (0.54-3.49)
(*
2=2.65, df=3,/?=0.45)
1.00
1.07 (0.43-2.65)
(*
2=0.02, df=l,p=0.89)
Community survey subjects
to
90
550
211
282
99
49
456
69
1.00
1.59
Odds ratios
(95% CI)
(0.83-3.05)
(A-
2=2.09,df=l,p=0.15)
1.00
0.95
1.47
1.93
(0.55-1.66)
(0.74-2.95)
(0.87-4.28)
(*
2=3.99,df=3,/>=0.26)
(fort
df=
1.00
1.63
(*
2=
rend, *
2=2.92,
=l,p=0.09)
(0.79-3.34)
1.71, df=l,p=0.19)
Cases are compared with each control group and the community survey by multivariate analysis. All models using hospital controls were
adjusted for age, sex, cigarette smoking, history of hypertension, diabetes, heart disease, and race. For the community survey, adjustment
was for age, sex, social class, cigarette smoking, and history of hypertension,
n, Number of cases or control subjects.
Downloaded from http://ahajournals.org by on October 18, 2019
Ben-Shlomo et al Alcohol Consumption and Stroke 1097
area in which this practice is located. For males there
was a relative underrepresentation of young subjects
(*
2=37.2, df=5,/><0.001), and social classes IV and V
(lower social classes) were also underrepresented in the
survey (*
2=12.2, df=2,/?<0.01).
Using the community group, the odds ratios for
alcohol consumption and stroke were calculated initially
controlling for age and sex (Table 2) and then controlling for age, sex, social class, smoking status, and a
history of hypertension (Table 3). Both the odds ratio
for alcohol consumption greater than 300 grams and the
X2 test for trend were significant (/?=0.02). However,
after adjustment for confounding, both the odds ratio
for consumption greater than 300 grams and the
x2 test
for trend were no longer significant (p=0.09).
Discussion
The results of this study illustrate how the risk
associated with alcohol consumption may vary depending on the choice of controls. The results using the
general medical controls support other studies that
suggest this population has an overrepresentation of
heavy drinkers.
71314 Because this was expected, a select
control group was chosen excluding possible alcoholrelated admissions and this resulted in an increased risk,
odds ratio 1.30 (95% confidence interval [CI], 0.42-
4.05). However, a relation may still exist between alcohol consumption and other medical conditions. In addition, heavy drinkers may still be overrepresented in
the hospital population if, given the same severity of
illness, they have a greater probability of admission than
other people with that illness (Berksonian bias).
The risk associated with alcohol consumption greater
than 300 grams per week was further increased when
comparison was made with the community control
group but was altered after controlling for confounding
variables to an odds ratio of 1.93 (95% CI, 0.87-4.28).
Community-based studies are known to underestimate
population alcohol consumption as estimated from Customs and Excise data.
15 In particular, compared with a
hospital control group, are the problems of nonresponse
and recall bias. Nonresponders are more likely to
contain a greater proportion of heavy drinkers,
1516 and
heavy drinkers are also less likely to be registered with
a general practitioner,
17 thus excluding them from a
general practice sampling frame. The response rate in
this study was low partly because of the highly mobile
population found in central London practices. However,
the proportion of identified heavy drinkers was remarkably similar to another general practitioner-based study
in North London,
18 which had a 75% response rate. This
adds credibility to the findings of the community survey
but does not negate the problem of nonresponse bias,
which may have affected both studies. “Recall bias”
would be a further problem if “healthy” community
control subjects underreported alcohol consumption
more than “sick” hospital control subjects. Little empirical information is available to examine this issue, but a
study comparing dietary and alcohol consumption data
between hospital control subjects and community control subjects
19 does not support this suggestion.
The interpretation of our results are limited by two
factors. Some patients were unable to provide an adequate alcohol history. Missing alcohol data in cases
resulted predominantly from patients dying or being
severely dysphasic and unable to give a history (29 of
49). This is a problem in all case-control studies of
stroke and can be only partially overcome by the use of
proxy information from relatives. Comparison of demographic data, smoking histories, and biochemical parameters from cases with and without alcohol data
showed no significant differences. However, the possibility that patients with fatal and severe cases of stroke
had heavier alcohol intakes cannot be completely excluded and limits the generalizability of our results. The
measurement of alcohol consumption for the community survey was not identical with that for the hospital
control groups; the former completed a postal questionnaire whereas the latter were interviewed. Cutler et al
20
have shown that subjects reported greater alcohol consumption when they were interviewed than when they
completed an identical postal questionnaire. Their
method of assessing alcohol consumption was very
similar to that used in the present study. This bias would
result in an underestimate of alcohol consumption for
the community compared with the hospital group.
There are two possible interpretations for our results.
First, there may be no significant association between
chronic alcohol consumption and risk of ischemic stroke
in a middle-aged European population. This conclusion
is supported by the lack of a relation in the majority of
other studies. However, many of these studies,
2123
including the present study, had insufficient power to
significantly detect a modest increased risk associated
with alcohol consumption. Alternatively, the variation
in odds ratios may help explain why other case-control
studies have shown apparently contradictory results.
Because of the different biases involved in choosing
control groups, the “true risk” may be underestimated
by hospital-based control groups while overestimated by
community-based controls. Our results suggest that
alcohol may increase the risk of stroke by 30-90%. This
estimate is consistent with the results of many other
case-control
21222426 and cohort studies.242728 Could
our findings explain the differences found in other
case-control studies of alcohol consumption and
stroke?
Nine case-control studies
321262930 have reported
findings on alcohol and stroke, although the two studies
from Birmingham, UK, actually used the same cases
with a different control group.
326 Six of these used
hospital inpatient control subjects, one used outpatients, and only two used community-based control
subjects. Four of these studies found a significantly
increased risk with heavy alcohol consumption or a
linear trend with consumption. Both studies that used
community controls found a significant risk, but one
failed to control for confounding variables such as
smoking,
24 and the other, although finding a significant
trend with consumption, reported a nonsignificant odds
ratio of 1.8 (95% CI, 0.8-4.5) for the highest alcohol
consumption group (more than 300 grams of alcohol per
week). The remaining significant study, which also
controlled for potential confounding variables, was a
hospital-based case-control study
3 whose controls were
patients admitted for routine surgical procedures. Interestingly, this study is the only one using only elective
surgical controls as opposed to acute medical and/or
surgical controls. The results showed that an alcohol
intake of at least 300 grams per week was associated
Downloaded from http://ahajournals.org by on October 18, 2019
1098 Stroke Vol 23, No 8 August 1992
with a relative risk of 4.2 (95% CI, 1.7-10.0) for men.
However, the strict selection criteria used for the control group excluded those conditions with a recognized
association with excessive alcohol use (such as trauma,
fractures, and peptic ulcer) or diseases known to alter
liver function, including carcinoma and infection, which
may also have an association with alcohol consumption.
In addition, “healthy” elective admissions may underrepresent the alcohol consumption of the population.
Henrich and Horwitz
23 directly tested the effect of
selecting elective surgical patients in their hospitalbased case-control study. They chose both medical and
surgical admissions and found no significant relation
between alcohol and ischemic stroke. They also reanalyzed their data using only controls chosen according to
the Birmingham criteria, which resulted in a proportional increase in the odds ratio of 50%. This increase in
the estimated odds ratios was interpreted as an effect of
exclusion bias.
We believe that the contradictory results from casecontrol studies of alcohol consumption and stroke may
be explained by methodological differences predominantly in control selection. The variations of risk found
in this study with different comparative groups favors a
hypothesis that alcohol consumption may modestly increase the risk of stroke, despite the lack of a statistically significant finding. Future studies must attempt to
overcome the different biases associated with control
selection and have a sufficiently large sample size to
enable detection of a modest increased risk.
Acknowledgments
We would like to thank Bonita Peachey for her help in
tracing records and coding and Dr. Cohen for information and
use of his general practice register.
References
1. Sedgwick J:A New Treatise on Liquors Wherein the Use and Abuse
of Wine, Malt Drinks, Water etc are Particularly Considered in Many
Diseases, Institutions and Ages With Proper Manner of Using Them
Hot or Cold Either as Physik, Diet or Both.
London, Rivington, 1725
2. Camargo CA Jr: Moderate alcohol consumption and stroke: The
epidemiologic evidence.
Stroke 1989;20:1611-1626
3. Gill JS, Zezuika AV, Shipley MJ, Gill SK, Beevers DG: Stroke and
alcohol consumption.
N EnglJMed 1986;315:1041-1046
4. Mayfield D, McLeod G, Hall P: The CAGE questionnaire: Validation of a new alcoholism screening instrument.
Am J Psychiatr
1974;131:1121-1123
5. Beresford TP, Blow FC, Hill E, Singer K, Lucey MR: Comparison
of CAGE questionnaire and computer-assisted laboratory profiles
in
screening for covert alcoholism.
Lancet 1990;2:482-485
6. Bush B, Shaw S, Cleary P, Delbanco TL, Aronson MD: Screening
for alcohol abuse using the CAGE questionnaire.
Am J Med 1987;
82:231-235
7. Barrison IG, Viola L, Mumford J, Murray RM, Gordon M, Murray-Lyon IM: Detecting excessive drinking among admissions to a
general hospital.
Health Trends 1982; 14:80-83
8. Allen CMC: Clinical diagnosis of the acute stroke syndrome.
Q J
Med
1983;208:515-523
9. World Health Organization:
Manual of the International, Statistical
Classification of Diseases, Injuries, and Causes of Death, 9th Revision.
Geneva, World Health Organization, 1977
10. OPCS: Classification of occupations. London, Her Majesty’s Stationery Office, 1980
11. Brookmeyer R, Liang KY, Linet M: Matched case-control designs
and overmatched analyses.
Am J Epidemiol 1986;124:693-701
12. Schlesselman JJ:
Case-Control Studies: Design, Conduct, Analysis.
New York/Oxford, Oxford University Press, 1982, pp 1-354
13. Jariwalla AG, Adams PH, Hore BD: Alcohol and acute general
medical admissions to hospital.
Health Trends 1979;11:95-97
14. Jarman CMB, Kellet JM: Alcoholism in the general hospital.
Br
Med/1979;2:469-472
15. Wilson P: Improving the methodology of drinking surveys.
The
Statistician
1981;30:159-167
16. Crawford A: Bias in a survey of drinking habits.
Alcohol Alcohol
1987;22:167-179
17. Shanks NJ: Medical provision for the homeless in Manchester.
J R
Coll Gen Pract
1983;33:40-43
18. Wallace P, Haines A: Use of a questionnaire in general practice to
increase the recognition of patients with excessive alcohol consumption.
Br Med J 1985;290:1949-1953
19. Gonzalez CA, Torrent M, Agudo A, Riboli E: Hospital versus
neighbourhood controls in the assessment of dietary factors.
Int J
Epidemiol
1990;19:354-361
20. Cutler SF, Wallace PG, Haines AP: Assessing alcohol consumption in general practice patients: A comparison between questionnaire and interview (findings of the Medical Research Council’s
general practice research framework study on lifestyle and health).
Alcohol Alcohol 1988;23:441-450
21. Herman B, Schmitz PIM, Leyten ACM, Van Lujik JH, Frenken
CWGM, Op de Coul AAW, Schulte BPM: Multivariate logistic
analysis of risk factors for stroke in Tilburg, The Netherlands.
Am
J Epidemiol
1983;118:514-525
22. Gorelick PB, Rodin MB, Langenberg P, Hier DB, Costigan J:
Weekly alcohol consumption, cigarette smoking, and the risk of
ischemic stroke: Results of a case-control study at three urban
medical centers in Chicago, Illinois.
Neurology 1989;39:339-343
23. Henrich JB, Horwitz RI: Evidence against the association between
alcohol use and ischemic stroke risk.
Arch Intern Med 1989;149:
1413-1416
24. Peacock PB, Riley CP, Lampton TD, Raffel SS, Walker JS: The
Birmingham stroke: Epidemiology and rehabilitation study, in
Stewart G (ed):
Trends in Epidemiology: Applications to Health
Service Research and Training.
Springfield, HI, Charles C Thomas
Publishing, 1972, pp 231-345
25. Abu-Zeid HAH, Choi NW, Maini KK, Hsu P-H, Nelson NA:
Relative role of factors associated with cerebral infarction and
cerebral haemorrhage: A matched pair case-control study.
Stroke
1977;8:106-112
26. Gill JS, Shipley MJ, Hornby RH, Gill SK, Beevers DG: A community case-control study of alcohol consumption in stroke.
Int J
Epidemiol
1988;17:542-547
27. Klatsky AL, Friedman GD, Siegelaub AB: Alcohol and mortality:
A ten-year Kaiser-Permanente experience.
Ann Intern Med 1981;
95:139-145
28. Paginini-Hill A, Ross RK, Henderson BE: Postmenopausal oestrogen treatment and stroke: A prospective study.
BrMedJ 1988;297:
519-522
29. Taylor JR, Combs-Orme T, Anderson D, Taylor DA, Koppenol C:
Alcohol, hypertension and stroke.
Alcoholism 1984;8:283-286
30. Von Arbin M, Britton M, De Faire U, Tisell A: Circulatory manifestations and risk factors in patients with acute cerebrovascular
disease and in matched controls.
Ada Med Scand 1985;218:
373-380
Downloaded from http://ahajournals.org by on October 18, 2019