51
The Role of Lifestyle Factors in the
Etiology of Stroke
A Population-Based Case-Control Study in
Perth, Western Australia
Konrad Jamrozik, MBBS, DPhil, FAFPHM; Robyn J. Broadhurst, BA, BSc;
Craig S. Anderson, MBBS, FRACP; Edward G. Stewart-Wynne, MBChB, FCP(SA), FRACP
Background and Purpose We sought to examine risk factors
for all strokes and for ischemic stroke and primary intracere-
bral hemorrhage separately.
Methods This was a population-based case-control study.
Downloaded from http://stroke.ahajournals.org/ by guest on June 5, 2018
Each case subject meeting World Health Organization criteria
for stroke (n=536) from a population-based register of acute
cerebrovascular events compiled in Perth, Western Australia,
in 1989 to 1990 was matched for age and sex with up to five
control subjects drawn from the same geographical area.
Objective confirmation of the type of stroke was available from
computed tomography, magnetic resonance imaging, or
necropsy for 86% of the case subjects. Data on medical history
and lifestyle factors were collected from case and control
subjects by interview of the subject or a proxy informant.
Results Current smoking, consumption of meat more than
four times weekly, and a history of hypertension or intermit-
A lthough mortality from stroke has fallen consid-
/ \ erably in Australia1 and other developed coun-
.Z \ . tries2 over at least the last four decades, there
are few data sets in which explanations for this trend
can be sought. However, it seems likely that a decrease
in the incidence of stroke has contributed to the sus-
tained fall in mortality from cerebrovascular disease.
The incidence of stroke potentially can be reduced
further if some sufficient causes of stroke contain com-
ponent causes that are avoidable and those causes can
be identified and modified or removed. It is therefore
worthwhile to reexamine the etiologic factors for stroke
from time to time to determine the avoidable factors
involved. This should also prompt changes in the em-
phasis given to particular preventive activities to reflect
changes over time in the population-attributable pro-
portions attendant on individual factors. For example, a
decrease in the prevalence of atrial fibrillation as a
consequence of a fall in the frequency of both rheumat-
ic3 and ischemic heart disease4-5 should mean that the
proportion of all strokes related to atrial fibrillation is
lower now in developed countries than ever before.
Received August 17, 1993; final revision received October 4,
1993; accepted October 4, 1993.
From the Department of Public Health, University of Western
Australia (K.J., R.J.B.), and the Department of Neurology, Royal
Perth Hospital (C.S.A., E.G.S.-W.), Perth, Western Australia.
Correspondence to Dr Konrad Jamrozik, University Depart-
ment of Public Health, 2nd Floor, M Block, QEII Medical Centre,
Nedlands, Western Australia 6009.
tent claudication were each associated with increased risk in
multivariate models for all strokes and for all first-ever strokes.
Consumption of 1 to 20 g/d alcohol in the preceding week was
associated with a significant reduction in the risk of all strokes,
all ischemic strokes, and of primary intracerebral hemorrhage,
while eating fish more than two times per month appeared to
protect against first-ever stroke and against primary intrace-
rebral hemorrhage. Diabetes mellitus was associated with a
significantly increased risk of ischemic stroke but a decreased
risk of hemorrhagic stroke.
Conclusions Risk factors for ischemic and hemorrhagic
stroke are not exactly the same. Changes in lifestyle relating to
tobacco and diet might make important contributions to further
reductions in the incidence of stroke. (Stroke. 1994;25:51-59.)
Key Words • Australia • case-control studies • cerebral
ischemia • intracerebral hemorrhage • risk factors
For many years medical thinking about the preven-
tion of stroke has been characterized by an emphasis on
secondary prevention. Acknowledging male sex and
advancing age as risk factors that cannot be changed,
attention has been concentrated on hypertension, dia-
betes mellitus, and atrial fibrillation6 and on the role of
aspirin in those with symptomatic cerebrovascular dis-
ease. 7 The relation between smoking and stroke
emerged much later than the role of smoking in isch-
emic heart disease and peripheral vascular disease,8 and
there is still some uncertainty surrounding the effect of
alcohol on the risk of stroke,9 probably because quali-
tatively different risks attend different levels of con-
sumption. Other aspects of lifestyle, particularly dietary
factors, have received little attention in terms of their
possible effect on the incidence of stroke.
Another feature of the traditional approach to the
prevention of cerebrovascular disease is that "stroke"
has often been considered as a single pathological
entity.1011 Indeed, it is only with the advent of reliable
techniques for the in vivo diagnosis of the pathological
basis of a particular case of stroke that there has been
the possibility of asking whether risk factors for isch-
emic cerebrovascular disease are different from those
for hemorrhagic disease. It is conceivable that control of
certain risk factors for cerebrovascular disease would
lead to a differential decrease in the incidence of one
particular type of stroke.
This article describes a case-control study of stroke
based on a population-based register of acute cere-
 52 Stroke Vol 25, No 1 January 1994
brovascular events compiled in Perth, Western Austra-
lia, aimed at identifying risk factors for ischemic and
hemorrhagic strokes separately.
Subjects and Methods
Case Subjects
Cases of stroke were drawn from the register of acute
cerebrovascular events compiled as part of the Perth Commu-
nity Stroke Study (PCSS). As described elsewhere,12 the PCSS
attempted to identify and document every case of stroke and
transient (cerebral) ischemic attack (TIA) affecting members
of a geographically defined population resident in the northern
suburbs of Perth, the capital city of the State of Western
Australia. Multiple sources were used to ascertain cases,
standard definitions of stroke and TIA were applied, and the
pathological basis of 86% of the cases of stroke was confirmed
by computed tomographic or magnetic resonance imaging scan
or by necropsy. The register operated between February 20,
1989, and August 19,1990, inclusive. First-ever strokes refer to
patients who had no previous history of stroke, that is, patients
who suffered their "first-ever-in-a-lifetime" stroke during this
period. Based on census data, the Australian Bureau of
Statistics estimated the size of the study population on June
Downloaded from http://stroke.ahajournals.org/ by guest on June 5, 2018
30, 1989, to be 138 708 persons.
Control Subjects
Each case subject was matched for sex and 5-year birth
cohort with one or more control subjects drawn from electoral
rolls for the study area of the PCSS. Because enrollment to
vote is compulsory for all Australian citizens aged 18 years or
older, electoral rolls provide a close approximation to a
population-wide, name-identified list of adults. On the basis of
the distribution of pathology in a pilot register compiled for
the same area in 198613 and a desire to have at least 80%
power for detecting relative risks of 2.0 attendant on exposures
shared by 20% of the general population, the matching ratio of
control to case subjects was varied from 1:1 for each case of
occlusive stroke to 5:1 for each case of intracerebral hemor-
rhage. In a few instances an individual invited to be a control
subject was found to have suffered a stroke during the period
of the study that had not previously been registered. These
subjects were then included as case subjects, and new control
subjects were selected.
Ascertainment of Exposure
Case subjects were interviewed by a neurology registrar
(C.S.A.) at home or in the hospital as soon as possible after the
episode of stroke became known to the PCSS. A relative or
other informant was frequently present during this interview
and served as the sole source of information in 20% of
instances when a patient was unable to communicate or death
supervened before an interview with the patient could be
arranged.
Control subjects were initially invited by mail to participate
in the study, and this was followed up by a telephone call.
Those who agreed were interviewed at home by one of three
research nurses or by C.S.A. Potential control subjects who
could not be contacted by telephone or by personal visit after
several attempts, or who had moved outside of the study area
after the date of the stroke for the matching case subject, were
replaced by others. Interviews with control subjects began in
August 1989 and were completed in February 1991.
Exposures of interest were recorded using a semistructured
interview schedule and precoded questionnaire. Because cases
were first seen after the stroke had occurred, only a history of
hypertension is used in the present analysis. Peripheral vascu-
lar disease was judged to be present if there was a clear history
of intermittent claudication, that is, calf pain brought on by
walking and relieved by rest. Diabetes mellitus was accepted
on the basis of a history of that condition except for some
patients with stroke in whom a random blood glucose was
found to exceed 11.1 mmol/L. Self-reported data on smoking
were used to classify subjects as lifelong nonsmokers, ex-
smokers of at least 12 months' standing, current smokers of 1
to 20 cigarettes daily, and those smoking more than 20
cigarettes daily. Consumption of alcohol was assessed via a
retrospective "diary" of drinking on each day of the week
preceding the stroke (case subjects) or the interview (control
subjects) and analyzed in terms of average daily consumption
of absolute alcohol during that period. All subjects were asked
about any history of stroke, TIA, or myocardial infarction,
current frequency of eating meat and fish and of adding salt to
food, whether they habitually used full-fat as opposed to
reduced-fat or skim milk, and whether they usually spread
butter rather than margarine on bread. Culinary habits of
interest included removal of visible fat from meat and of the
skin from poultry before eating it. Each subject was also asked
about all medications being taken at the time of the stroke
(case subjects) or interview (control subjects), although only
the data on aspirin are presented in this report.
Statistical Methods
The data for each case and its matching control(s) were
initially reviewed using SAS software14 to produce frequency
tables. Univariate comparisons and reverse stepwise multivari-
ate conditional logistic regression were performed separately
for cases of intracerebral hemorrhage (excluding subarachnoid
hemorrhage), cases of ischemic stroke (thrombotic stroke,
embolic stroke, lacunar stroke, and boundary zone infarction
combined), and for all strokes (including subarachnoid hem-
orrhage), using EGRET software.15 In general, subjects with
missing information were excluded from all analyses. This
meant that increasing numbers of case-control sets entered a
given model as the number of factors under consideration fell.
However, if a multivariate model would not converge, missing
values were reset to valid values such that the final estimates of
differences between cases and controls were likely to be
conservative. For example, an individual for whom no smoking
or dietary history could be obtained was deemed to be a
lifelong nonsmoker who ate meat infrequently and fish often
and who always used reduced-fat or skim milk.
Population-attributable proportions were calculated by de-
termining the number of strokes of a given pathological type in
each group defined by sex and 5-year age group that were
mathematically attributable to a particular exposure, on the
assumption that the exposure modified the risk of the disease.
The figures for all age-sex cells were added and the total
divided by the total number of strokes of that type that were
observed. Algebraically, this may be expressed as
—, n i P i (OR-l)
where ns is the number of strokes of a given type occurring in
a particular sex and 5-year age group, p, is the proportion of
control subjects in that group reporting exposure to the factor
of interest, and OR is the odds ratio for the factor derived
from a multivariate conditional logistic model. This calculation
was repeated using the lower and upper 95% confidence limits
of each odds ratio to obtain some estimate of the range of the
proportion of strokes mathematically attributable to each
exposure. The prevalences of exposure to individual risk
factors among control subjects were summarized by age stan-
dardization using Segi's "world" population as the external
reference.16
Ethical Clearance
The protocol for the study was approved by the Committee
for Human Rights of the University of Western Australia, the
Ethics Committee of Royal Perth Hospital, and the Confiden-
 Jamrozik et al Etiology of Stroke 53

TABLE 1. Results of Univariate Analyses of Risk Factors for All Strokes Combined

Univariate OR
Risk Factor (501 cases, 931 controls) 95% Cl
Alcohol, g/d*
0 1.0
1-20 0.43 0.32-0.58
21-40 0.67 0.38-1.16
41-60 0.55 0.23-1.28
>61 2.51 1.33-4.74
Tobacco
Nonsmoker 1.0
Ex-smoker 0.88 0.65-1.20
1-20 cigarettes daily 1.58 1.07-2.32
>21 cigarettes d^ly 5.83 3.24-10.5
History of hypertension 2.35 1.83-3.04
Claudication 2.58 1.83-3.62
Downloaded from http://stroke.ahajournals.org/ by guest on June 5, 2018

Diabetes mellitus 2.14 1.49-3.08

Previous myocardial infarction 1.88 1.35-2.62
Regular use of aspirin 1.37 1.04-1.81
Previous stroke or TIA 6.27 4.46-8.81
Use of reduced-fat or skim milk 0.51 0.38-0.69
Consumption of meat > 4 times weekly 1.62 1.19-2.19
Adding salt to food 2.01 1.53-2.63
Consumption of fish >2 times per month 0.77 0.56-1.01
Margarine used "usually" or "always" 1.05 0.81-1.36
Always trimming fat from meat 0.81 0.62-1.06
Always removing skin from poultry 0.85 0.65-1.10
OR indicates odds ratio; Cl, confidence interval; and TIA, transient ischemic attack.
*Refers to average daily consumption during the preceding week.

tiality of Health Information Committee of the Health De-
partment of Western Australia.
Results
The PCSS register includes a total of 536 events
meeting the World Health Organization criteria for
stroke. These events occurred in 492 individuals. At least
some data on exposures were available for all but one of
the case subjects. Of 1441 potential control subjects
initially selected, 169 had died or moved out of the study
area, 75 could not be contacted by telephone or by
personal visit to the address given on the electoral roll, 29
were already included as a case, 237 declined to partici-
pate, and the remainder completed an interview. Thus,
the response rate among those not included as a case and
not known to have died or moved away was 75%.
Risk Factors for All Strokes Combined
Data for 501 cases of stroke and 931 control subjects
included in the univariate analyses of individual risk
factors are presented in Table 1, and the first column
of Table 2 shows the most parsimonious multivariate
model for all strokes combined. Exclusive use of
margarine and regular use of aspirin, along with
regular consumption of fish, removal of fat and skin,
and previous acute myocardial infarction, all fail to
enter the latter model. Although the point estimates
associated with the remaining individual factors in
Table 1 are altered, the findings of the univariate
analysis are qualitatively unchanged, despite the fact
that the multivariate model is based on only 828
subjects (295 cases).
The second set of results in Table 2 represents the
most parsimonious model involving the 728 subjects
(175 cases) who had no previous history of either
stroke or TIA and for whom the data on other
exposures were complete. The same factors enter this
model as that for all strokes except that use of added
salt was not associated with a significant increase in
risk, whereas consumption of fish more than two times
per month appeared to confer protection against a
first-ever stroke. Although there is some variation in
the point estimates for other individual exposures, the
overall patterns of risk and protective factors for all
strokes and for first-ever strokes are quite similar. This
being said, consumption of one to two alcoholic drinks
daily is not clearly associated with protection against a
first-ever stroke, and there is a suggestion that the
increases in risk associated with heavy drinking and
with current smoking are greater for first-ever strokes
than for all strokes combined.
 54 Stroke Vol 25, No 1 January 1994

TABLE 2. Multivariate Analysis of Risk Factors for All Strokes and First-Ever Strokes

All Strokes Combined First-ever Strokes

(295 cases, 533 controls) (175 cases, 553 controls)

Risk Factor OR 95% Cl OR 95% Cl

Alcohol, g/d*
0 1.0 1.0
1-20 0.62 0.40-0.94 0.67 0.40-1.12
21-40 0.83 0.37-1.88 0.98 0.38-2.50
41-60 0.65 0.20-2.17 0.40 0.09-1.89
>61 2.75 0.98-7.73 6.71 1.68-26.8
Tobacco
Nonsmoker 1.0 1.0
Ex-smoker 0.75 0.46-1.24 0.65 0.36-1.18
1 -20 cigarettes daily 1.99 1.04-3.79 2.57 1.12-5.89
>21 cigarettes daily 3.52 1.35-9.14 4.74 1.55-14.5
History of hypertension 2.33 1.57-3.47 2.87 1.74-4.73
Downloaded from http://stroke.ahajournals.org/ by guest on June 5, 2018

Claudication 1.93 1.10-3.39 2.61 1.22-5.55

Previous stroke or TIA 4.81 2.89-8.01 Not relevant
Use of reduced-fat or skim milk 0.49 0.31-0.76 0.44 0.25-0.77
Consumption of meat >4 times weekly 2.17 1.33-3.53 2.30 1.29-4.08
Adding salt to food 1.53 1.01-2.31 Did not enter model
Consumption of fish >2 times per month Did not enter model 0.60 0.36-0.99
OR indicates odds ratio; Cl, confidence interval; and TIA, transient ischemic attack.
*Refers to average daily consumption over the preceding week.

Risk Factors for Ischemic and Hemorrhagic
Strokes Considered Separately
Tables 3 and 4 present two comparisons of risk factors
for ischemic strokes and primary intracerebral hemor-
rhage (PICH). In Table 3 only those factors entering the
most parsimonious model for each type of stroke are
listed. Thus, consumption of alcohol at an average of
greater than six "standard" (10-g) drinks per day is seen
to confer an increased risk only of hemorrhagic stroke,
and exsmokers appear to enjoy a degree of protection
against ischemic but not hemorrhagic stroke. Smokers
of more than 20 cigarettes daily have a significant
increase in risk of either form of stroke, as do individ-
uals with a previous episode of cerebrovascular disease.
Hypertension and diabetes mellitus are associated with
an increased risk of ischemic stroke only, whereas
habitual use of reduced-fat or skim milk appears to
protect against such events. Adding salt to food seems
to increase the risk only of PICH, but regular consump-
tion of fish is associated with a decrease in the risk of
strokes of this type.
Table 4 presents results of separate but identical
multivariate analyses of the risk factors for each form of
stroke, each model consisting of factors that were
associated with a significant variation in risk from the
null value for at least one of the two types of stroke. To
obtain convergence of these statistical models, it was
necessary to recode missing values to valid but conser-
vative levels, as described in "Subjects and Methods."
This led to exclusion of the variable reflecting type of
milk used but inclusion of one describing previous acute
myocardial infarction. Although the stratum-specific
confidence intervals include unity, the results in Table 4
suggest that the dose-response curve for alcohol for the
two forms of stroke may differ, with that for PICH rising
sooner than the curve for ischemic stroke. For both
types of stroke there is an obvious upward trend in risk
with increasing intensity of smoking, although the point
estimates are consistently higher for PICH. Whereas the
risks associated with a history of hypertension or a
previous episode of cerebrovascular disease are effec-
tively indistinguishable, occlusive coronary disease ap-
pears to be selectively associated with ischemic stroke.
The small degree of overlap in the confidence limits
surrounding the point estimates for diabetes mellitus
prompts the suggestion that diabetes has opposite ef-
fects on the incidence of the two forms of stroke. As
expected, the results for salt and fish are consistent with
the pattern evident in Table 3, with a selective protec-
tive effect for PICH of frequent consumption of fish
again being apparent.
Population-attributable proportions for ischemic
stroke, PICH, and all strokes combined are presented in
Table 5. To convey the uncertainty surrounding such
attributions, the figures in this table have been calcu-
lated from the upper and lower confidence limits pre-
sented earlier for each factor. Because these calcula-
tions take into account the prevalence of the individual
exposures (displayed as age-standardized proportions in
Table 6), the apparent importance of particular factors
differs from that suggested by the corresponding odds
ratios. For example, the relative rarity of heavy con-
sumption of alcohol means that it is implicated in at
most one in nine strokes, while between 2% and 30% of
 Jamrozik et al Etiology of Stroke 55

TABLE 3. Significant Associations Identified in Separate Muitivariate Models of Risk Factors for
Ischemic Stroke and Primary Intracerebrat Hemorrhage

Primary Intracerebral
Ischemic Stroke* Hemorrhage
(243 cases, 319 (59 cases, 279
controls) controls)

Risk Factor OR 95% Cl OR 95% Cl

Alcohol, g/dt
0 1.0 1.0
1-20 0.54 0.32-0.92 0.28 0.12-0.67
21-40
41-60
>61 5.88 1.19-29.1
Tobacco
Nonsmoker 1.0 1.0
Ex-smoker 0.41 0.22-0.76
1-20 cigarettes daily 3.33 1.05-10.6
Downloaded from http://stroke.ahajournals.org/ by guest on June 5, 2018

>21 cigarettes daily 4.67 1.10-19.8 9.78 2.25-42.5

History of hypertension 3.56 2.17-5.85
Diabetes mellitus 3.81 1.77-8.23
Previous stroke or TIA 3.91 2.17-7.06 8.48 3.38-21.3
Use of reduced-fat or skim milk 0.43 0.26-0.72
Adding salt to food 3.14 1.52-6.48
Consumption of fish >2 times per month 0.42 0.19-0.90
OR indicates odds ratio; Cl, confidence interval; and TIA, transient ischemic attack. Blanks in the table indicate
that the 95% Cl for that factor included 1.0.
Combination of thrombotic stroke, embolic stroke, boundary zone infarction, and lacunar infarction.
tRefers to average daily consumption during the preceding week.

additional strokes appear to have been avoided by
almost half the community consuming one or two
alcoholic drinks daily. Surprisingly, hypertension and
frequent consumption of meat appear to make approx-
imately equal contributions to the overall burden of
stroke borne by this community, ahead of those related
to salt, diabetes, smoking, and peripheral vascular dis-
ease. For PICH, however, salt ranks ahead of hyperten-
sion, and the maximum likely contribution from diabe-
tes is modest. Nevertheless, diabetes may be implicated
in the genesis of one sixth of all strokes.
Apart from moderate consumption of alcohol, the only
factor associated with an overall protective effect is habit-
ual use of reduced-fat or skim milk. The data suggest that
the burden of stroke experienced by this community has
been reduced by at least 9%, and perhaps by as much as
one third, by the use of these products. Given that more
than two thirds of adults in the study population still use
full-fat milk (Table 6), there appears to be considerable
scope for further reduction of cerebrovascular disease by
changing dietary patterns.
Discussion
Our results are consistent with previous observations
regarding the role of alcohol, tobacco, hypertension,
and clinically evident cerebrovascular disease in the
etiology of stroke,8-911 but they also suggest systematic
differences in the risk factors for ischemic and hemor-
rhagic strokes and suggest that dietary factors may
contribute significantly to the overall burden of cere-
brovascular disease.
Selection bias relating to choice of cases is unlikely to
have been a problem because the study drew upon a
population-based register of acute cerebrovascular
events that was compiled prospectively using multiple
sources of ascertainment and internationally accepted
definitions of stroke and TIA. Objective evidence of the
pathological basis was obtained for 86% of cases of
stroke, and rigid criteria were used to differentiate
hemorrhagic from ischemic strokes. Although the latter
classification was not performed in a blind manner, this
would not affect the results for all strokes combined.
A similarly detailed protocol was used to identify
potential control subjects concurrently from the same
population as that giving rise to the case subjects. The
response rate among control subjects was high but not
perfect, and this, combined with the fact that residents
of the study area who were not Australian citizens are
omitted from electoral rolls, as are some elderly, infirm
persons,17 means that the data for control subjects could
be affected by selection bias. Insofar as nonresponse is
associated with heavy smoking or drinking, for example,
or with established ill health, perhaps secondary to
aspects of lifestyle, this would serve to exaggerate the
differences apparent between case and control subjects.
On the other hand, the inclusion among control subjects
of a small number of individuals who had experienced a
stroke or TIA outside the registration period of the
PCSS would have the opposite effect on our results.
 56 Stroke Vol 25, No 1 January 1994

TABLE 4. Comparison of Identical Multivariate Models of Risk Factors for Ischemic Stroke and
Primary Intracerebral Hemorrhage

Primary Intracerebral
Ischemic Stroke* Hemorrhage
(360 cases, 518 (59 cases, 279
controls) controls)

Risk Factor OR 95% Cl OR 95% Cl

Alcohol, g/dt
0 1.0 1.0
1-20 0.76 0.52-1.12 0.26 0.11-0.63
21-40 0.63 0.26-1.53 0.15 0.02-1.25
41-60 0.49 0.15-1.63 0.35 0.05-2.50
>61 2.15 0.69-6.65 4.42 0.89-21.9
Tobacco
Nonsmoker 1.0 1.0
Ex-smoker 0.49 0.31-0.81 1.11 0.43-2.85
1-20 cigarettes daily 1.45 0.81-2.68 3.17 0.92-11.0
Downloaded from http://stroke.ahajournals.org/ by guest on June 5, 2018

>21 cigarettes daily 4.92 1.90-12.7 9.84 2.09-46.4

History of hypertension 2.63 1.79-3.86 2.04 0.96-4.32
Diabetes mellitus 2.30 1.31-4.00 0.17 0.02-1.68
Previous stroke or TIA 6.33 3.89-10.3 10.6 3.97-28.2
Previous myocardial infarction 1.85 1.12-3.03 0.41 0.09-1.83
Adding salt to food 1.04 0.72-1.51 3.46 1.58-7.58
Consumption of fish >2 times per month 0.90 0.60-1.36 0.43 0.20-0.96
OR indicates odds ratio; Cl, confidence interval; and TIA, transient ischemic attack.
Combination of thrombotic stroke, embolic stroke, boundary zone infarction, and lacunar infarction.
tRefers to average daily consumption during the preceding week.

Potentially greater difficulties surround the collection
and interpretation of the data concerning the exposures
of interest. Interviews with case and control subjects
were conducted by different individuals, making blind-
ing impossible, and the data are largely based on
self-report. On the other hand, the items included in the
present analysis formed only a small part of longer
structured interviews for both case and control subjects,
and a close relative or other informant was frequently
present and could have verified or contradicted re-
sponses from the interviewee. Recall bias remains a
potential problem, but the questions did relate largely
to usual behaviors, to the very recent past, or to medical
conditions that were likely to be ongoing.
We expected that each of hypertension, diabetes, and
previous stroke or TIA would be associated with a
significant increase in the risk of ischemic stroke in this
population, while the relation between consumption of
alcohol and stroke is better defined than in previous
studies. Heavy consumption of alcohol has been associ-
ated with ischemic stroke in a few studies, 918 but only
among men. Reports on the effects of moderate intake
of alcohol are also contradictory. In the current study,
the relation between alcohol and risk of ischemic stroke
is not linear, but there is no conclusive evidence of
increased risk at levels of recent consumption exceeding
60 g/d of alcohol. In contrast, consumption within the
range recommended by the Australian National Health
and Medical Research Council19 appears to have a
protective effect.
Our results also support accumulating epidemiologic
evidence indicating that cigarette smoking is a risk
factor for ischemic stroke, 820 although the effect ap-
pears less marked than for ischemic heart disease and
peripheral vascular disease. That ex-smokers have a
significantly reduced risk compared with self-reported
never-smokers was an unexpected finding, although
some decrease in risk was also reported recently in
another Australian study.21 Other investigations8'9'20
have shown that the relative risk of ischemic stroke
approaches that of never-smokers within several years
of stopping smoking. These findings are not totally
consistent with the theory that smoking exerts its dele-
terious effects by accelerating the age-related develop-
ment of cerebral and carotid atherosclerosis.22-23
Rather, they suggest that smoking is also involved in the
sequence of events that precipitate the clinical episode,
for example, by increasing levels of fibrinogen or com-
plicating the formation of atheroma by adversely affect-
ing platelet function. Part of the reason for our finding
of a reduced risk among ex-smokers could lie in a
tendency for proxy-informants for dead, unconscious, or
aphasic case subjects who were ex-smokers to give
answers indicating that they were lifelong nonsmokers.
It is also conceivable that some individuals who give up
smoking make additional changes in their lifestyle that
reduce the risk of stroke even further.
Of the factors included in the present analysis, only
previous stroke or TIA, smoking, use of added salt, and
consumption of fish and alcohol remain in the final
 Jamrozik et al Etiology of Stroke 57

TABLE 5. Estimated Proportions of Strokes Attributable to or Prevented by Particular Exposures, With

Values Calculated From the Upper and Lower Confidence Limits of the Odds Ratios in Tables 2 and 4

Intracerebral
Ischemic Strokes Hemorrhage All Strokes*
Risk Factor (n=382) (n=60) (n=536)
Alcohol, g/dt
1-20 -22, 4 -59, -17 -30, - 2
21-40 -5, 3 -7, 1 -4, 5
41-60 -3, 2 -5, 5 -3, 3
>61 - 1 , 10 0, 17 0, 11
Tobacco
Ex-smokers -51, 9 -35, 41 -33, 9
1-20 cigarettes daily - 2 , 15 -1, 46 1, 22
21 + cigarettes daily 3, 19 3, 36 1, 15
History of hypertension 26,55 -2, 57 20, 51
Claudication Not in model Not in model 0.8, 16
Previous stroke or TIA 21,42 19, 59 15, 36
Downloaded from http://stroke.ahajournals.org/ by guest on June 5, 2018

Previous myocardial infarction 1, 17 -10, 7 Not in model

Diabetes mellitus 3,20 -8, 5 0, 18
Use of reduced-fat or skim milk Not in model Not in model -33, - 9
Consumption of meat >4 times weekly Not in model Not in model 19, 64
Adding salt to food -16,20 21, 73 1, 38
Consumption of fish >2 times per month - 3 9 , 20 -126, -3 Not in model
Values are attributable proportions (%); positive values indicate an increased risk of stroke; negative, a decreased risk.
See "Subjects and Methods" for details of calculations.
Includes subarachnoid hemorrhage and strokes with uncertain pathology.
tRefers to average daily consumption during the preceding week.

statistical model of risk factors for hemorrhagic stroke.
Again, moderate drinking is associated with a protective
effect, but there is a significant excess risk at levels of
recent drinking above 60 g/d. This is consistent with
earlier reports of a selective relation between heavy
drinking and PICH, 2425 and the same trend is apparent
in the data for drinking during an average week (not
shown here). While there are a number of mechanisms
by which alcohol could contribute to ischemic stroke, its
effects on blood pressure, platelets, and the fibrinolytic
system may explain the increased risk of intracerebral
hemorrhage.26 The most plausible explanation for an
overall protective effect of moderate consumption of
alcohol is that it elevates high-density lipoprotein cho-
lesterol, while, at this level of intake, its effect on
platelets and clotting may be beneficial.
An overview of the relation between smoking and
stroke8 has suggested that any effect of smoking on
PICH specifically is probably small and may even be
protective. This conclusion was based on four studies,
and the authors of the overview commented that the
results were dominated by one particular study that had
been conducted in specialist neurological hospitals. In
contrast, our data clearly show excess risks of PICH
associated with any current smoking. That our cases of
PICH were collected by population-wide ascertainment
and were all objectively confirmed lends credibility to
the present results.
Despite matching up to five control subjects to each
case of PICH, this aspect of the study lacked statistical
power because it was based on only 60 cases. Thus, as may
be seen from Table 4, the 95% confidence limits surround-
ing the odds ratio for a history of hypertension exclude 1.0,
but only by a small margin. The same table also suggests
that the effect of diabetes mellitus on the risk of PICH is
qualitatively opposite to its effect on the risk of ischemic
stroke. In other studies, the types of strokes suffered by
patients have often been poorly defined, but the ratio of
ischemic to hemorrhagic strokes appears to be higher in
diabetics than nondiabetics.27
Even with the limited statistical power, adding salt to
food before it is eaten emerged as a strong risk factor for
PICH, while frequent consumption of fish had a signifi-
cant protective effect. The influence of these factors was
not entirely explained by confounding by consumption of
alcohol or a history of hypertension. Without a complete
dietary survey, one cannot conclude that salt and fish per
se are directly implicated in the etiology of strokes of any
kind. Had we undertaken such a survey, confounding
with other behaviors would remain possible, and it still
would not necessarily be clear which nutrient, if any,
altered the risk of stroke. However, our findings do lend
credence to ecological studies linking changes in the
consumption of foods with reductions in blood pressure
or serum cholesterol. For example, a significant decline
in mean blood pressure and the incidence of stroke in
northeast Japan between 1953 to 1966 and 1980 to 1983
was linked to changes in diet that included increased
intake of animal protein and decreased intake of salt.28
More recently, Law, Frost, and Wald29 combined data on
 58 Stroke Vol 25, No 1 January 1994

TABLE 6. Age-Standardized Estimated Proportions of might result in additional sizeable reductions in the bur-
Adults in the Study Population Exposed to den of fatal and disabling cerebrovascular disease borne
Particular Factors by the community.
Age-Standardized
Prevalence of Acknowledgments
Risk Factor Exposure, %* This study was supported by grants from the National
Health and Medical Research Council, the Australian Brain
Alcohol, g/d
Foundation, the Medical Research Foundation of Royal Perth
1-20 45.1 Hospital, and the Western Australian Health Promotion
21-40 5.3 Foundation. Dr Jamrozik drafted the manuscript during his
tenure of a Rowden White Overseas Travelling Fellowship of
41-60 6.3 the Royal Australasian College of Physicians.
261 4.7 The authors are indebted to Joyce Lim, Jenny Linto, and
Sue Forbes, to Floris Gout, and to Jan Forrester and Raywin
Tobacco
Tuohy for assistance with collection, storage, and entry of the
Ex-smokers 31.7 data, respectively. The study would not have been possible
without cooperation received from the patients and control
1-20 cigarettes daily 20.1
subjects and the doctors of both, and from the Registrar-
>21 cigarettes daily 5.9 General of Births, Deaths, and Marriages, the Australian
Bureau of Statistics, and the Health Department of Western
History of hypertension 19.9
Australia. The authors are grateful to Cathy Harwood for
Claudication 2.6 preparing the manuscript and to Sir Richard Doll and Dr
Downloaded from http://stroke.ahajournals.org/ by guest on June 5, 2018

Diabetes mellitus 2.6 Graeme Hankey for critical comment on it.

Previous stroke or TIA 1.6 References

Use of reduced-fat or skim milk 30.3
Consumption of meat >4 times weekly 74.3
Use of added salt 32.3
Consumption of fish >2 times per
month 67.5
TIA indicates transient ischemic attack.
* Based on data for control subjects aged >20 years and
age-standardized to the "world" population.
blood pressure and intake of salt from a number of
observational studies of subjects from the same and
different populations, as well as several intervention
trials. They estimated that a modest 5 mm Hg reduction
in the mean diastolic blood pressure of the population,
achievable by reducing the average daily intake of salt by
50 mmol (through avoiding salty foods and not adding
salt to food during cooking), would reduce the incidence
of stroke by one fifth and that of ischemic heart disease
by one sixth. Implementing changes across the whole
population to reduce salt in processed food could pro-
duce a decrease in mean intake of salt of 100 mmol daily
and might reduce the incidence of and mortality from
stroke by 38%.
The need for caution in interpreting our results is
emphasized further by our failure to demonstrate any
significant benefits associated with recommended dietary
behaviors such as trimming fat from meat or skin from
poultry and avoidance of butter. In addition, habitual use
of reduced-fat or skim milk was associated with a clear
decrease in risk only in models for ischemic strokes and for
all strokes combined. Nevertheless, seen in the light of
historical trends in Australia toward lower average con-
sumption of meat30 and increasing prevalence of regular
consumption of alcohol30 and of avoidance of salt,31 our
findings provide broad support for a hypothesis that the
downward trend in mortality from cerebrovascular disease
is due in part to decline in the incidence of stroke, and
some evidence that would support a selective decline in
primary intracerebral hemorrhage. They also suggest that
further changes in lifestyle relating to tobacco and diet
1. National Heart Foundation of Australia. Heart Facts: 1989.
Canberra, Australia: National Heart Foundation; 1991.
2. Bonita R. Epidemiology of stroke. Lancet. 1992;339:342-344.
3. Kaplan EL. Epidemiological approaches to understanding the
pathogenesis of rheumatic fever. Int J Epidemiol. 1985;14:499-501.
4. Beaglehole R. International trends in coronary heart disease mor-
tality, morbidity, and risk factors. Epidemiol Rev. 1990;12:l-15.
5. Thompson PL, Hobbs MST, Martin CA. The rise and fall of
ischemic heart disease in Australia. Aust N Z J Med. 1988;18:
337-345.
6. Warlow CP. Cerebrovascular disease. In: Weatherall DJ, Led-
ingham JGG, Warrell DA, eds. Oxford Textbook of Medicine.
Oxford, England: Oxford University Press, Inc; 1983:21.65.
7. Antiplatelet Trialists' Collaboration. Secondary prevention of
vascular disease by prolonged antiplatelet treatment. Br Med J.
1988;296:320-331.
8. Shinton R, Beevers G. Meta-analysis of relation between cigarette
smoking and stroke. Br Med J. 1989;298:789-794.
9. Wolf PA. Cigarettes, alcohol, and stroke. N EnglJ Med. 1987;316:
1087-1089.
10. Yates PO. A change in the pattern of cerebrovascular disease.
Lancet. 1964;l:65-69.
11. Marmot MG, Poulter NR. Primary prevention of stroke. Lancet.
1992;339:344-347.
12. Anderson CS, Jamrozik K, Burvill PW, Johnson GA, Chakera TM,
Stewart-Wynne EG. Ascertaining the true incidence of stroke:
experience from the Perth Community Stroke Study, 1989-1990.
Med J Aust. 1993;158:80-84.
13. Ward GR, Jamrozik K, Stewart-Wynne EG. Incidence and
outcome of cerebrovascular disease in Western Australia. Stroke.
1988;19:1501-1506.
14. SAS Users' Guide, Version 6.7. Cary, NC: SAS Institute; 1991.
15. EGRET Statistical Package. Seattle, Wash: Statistics and Epide-
miology Research Corporation; 1991.
16. Waterhouse J, Muir C, Correa P, Powell J. Cancer Incidence in Five
Continents. Lyon, France: IARC; 1976:456.
17. Australian Electoral Office. A Quantitative Assessment of Electoral
Enrollment in Australia: Research Report. Canberra, Australia: Aus-
tralian Electoral Office; 1983.
18. Gill JS, Zezulka AV, Shipley MJ, Gill SK, Beevers DG. Stroke and
alcohol consumption. N EnglJ Med. 1986;315:1041-1046.
19. National Health and Medical Research Council. Is There a Safe
Level of Daily Consumption of Alcohol for Men and Women?
Canberra, Australia: Australian Government Publishing Service;
1992.
20. Donnan GA, Adena MA, O'Malley HM, McNeil JJ, Doyle AE,
Neill GC. Smoking as a risk factor for cerebral ischaemia. Lancet.
1989;2:643-647.
21. Al-Roomi K, Heller RF, Holland T, Wlodarczyk J. The importance
of hypertension in the aetiology of infarctive and haemorrhagic
stroke. Med J Aust. 1992;157:452-455.

22. Whisnant JP, Homer D, Ingall TJ, Baker HL Jr, O'Fallon WM,
Wiebers DO. Duration of cigarette smoking is the strongest pre-
dictor of severe extracranial carotid artery atherosclerosis. Stroke.
1990;21:707-714.
23. Dempsey RJ, Moore RW. Amount of smoking independently
predicts carotid artery atherosclerosis severity. Stroke. 1992;23:
693-696.
24. Blackwelder WC, Yano K, Rhoads GG, Kagan K, Gordon T,
Palesch Y. Alcohol and mortality: the Honolulu Heart Study. Am
JMed. 1980;68:164-169.
25. Donahue RP, Abbott RD, Reed DM, Yano K. Alcohol and hem-
orrhagic stroke: the Honolulu Heart Program. JAMA. 1986;255:
2311-2314.
26. Gorelick PB. The status of alcohol as a risk factor for stroke.
Stroke. 1989;20:1607-1610.
Downloaded from http://stroke.ahajournals.org/ by guest on June 5, 2018
Jamrozik et al Etiology of Stroke 59
27. Abbott RD, Donahue RP, MacMahon SW, Reed DM, Yano K.
Diabetes and the risk of stroke: the Honolulu Heart Program.
JAMA. 1987;257:949-952.
28. Shimamoto T, Komachi Y, Inada H, Mitsunori D, Iso H, Sato S,
Kitamura A, Iida M, Konishi M, Nakanishi N, Terao A, Naito Y,
Kojima S. Trends for coronary heart disease and stroke and their
risk factors in Japan. Circulation. 1989;79:503-515.
29. Law MR, Frost CD, Wald NJ. By how much does dietary salt
reduction lower blood pressure? HI: analysis of data from trials of
salt reduction. Br Med J. 1991;302:819-824.
30. Dwyer T, Hetzel BS. A comparison of trends of coronary heart
disease mortality in Australia, USA and England and Wales with
reference to three major factors — hypertension, cigarette smoking
and diet. Int J Epidemiol. 1980;9:65-71.
31. Jamrozik K, Hockey RL. Trends in risk factors for vascular
disease. Med J Aust. 1989;150:l-4.
 The role of lifestyle factors in the etiology of stroke. A population-based case-control study in
Perth, Western Australia.
K Jamrozik, R J Broadhurst, C S Anderson and E G Stewart-Wynne

Stroke. 1994;25:51-59
doi: 10.1161/01.STR.25.1.51
Stroke is published by the American Heart Association, 7272 Greenville Avenue, Dallas, TX 75231
Downloaded from http://stroke.ahajournals.org/ by guest on June 5, 2018

Copyright © 1994 American Heart Association, Inc. All rights reserved.

Print ISSN: 0039-2499. Online ISSN: 1524-4628

The online version of this article, along with updated information and services, is located on the
World Wide Web at:
http://stroke.ahajournals.org/content/25/1/51

Permissions: Requests for permissions to reproduce figures, tables, or portions of articles originally published in
Stroke can be obtained via RightsLink, a service of the Copyright Clearance Center, not the Editorial Office.
Once the online version of the published article for which permission is being requested is located, click Request
Permissions in the middle column of the Web page under Services. Further information about this process is
available in the Permissions and Rights Question and Answer document.

Reprints: Information about reprints can be found online at:

http://www.lww.com/reprints

Subscriptions: Information about subscribing to Stroke is online at:

http://stroke.ahajournals.org//subscriptions/