THE LANCET

Epidemiology of stroke

C P Warlow

W h a t is a s t r o k e ? least mortality gives some idea of the burden. Stroke is

A stroke is a clinically defined syndrome of rapidly the third commonest cause of death after coronary heart
developing symptoms or signs of focal loss of cerebral disease and all cancers, not only in developed countries,
function with no apparent cause other than that of but world wide. 2
vascular origin, but the loss of function can at times be Stroke incidence, based on representative community
global (applied to patients in deep coma and to those samples, provides far more detail on stroke burden in
with subarachnoid haemorhage). Symptoms last more populations than does mortality. However, community-
than 24 h or lead to death/ The syndrome varies in based studies require considerable resources and rigorous
severity from recovery in a day, through incomplete methods, and it is difficult to be sure that all the cases
recovery, to severe disability, to death. Whether it is for have been found and accurately diagnosed. Also, sample
clinical or epidemiological purposes, the definition has sizes have nowhere been all that large, which makes
not changed for over 20 years, and there is no reason to comparisons of incidence by time and place imprecise.
change it in the foreseeable future. If the definition Moreover, incidence has been studied reliably only in
depended heavily on technology, such as brain imaging, various white populations, among whom it is more or less
it would be useless to epidemiologists because the similar (about two first-ever-in-a-lifetime strokes per 1000
technology may be unavailable in many parts of the world per a n n u m overall, about four in people aged 45-84),
where stroke is a major burden. Even if the technology except in Russia, where the incidence seems
were available, it may evolve rapidly, so that accurate to be high and in France where it is low (figure 1).
comparison of the burden of stroke in different places and The prevalence of stroke survivors in a community
at different times becomes impossible. Like other clinical depends on both incidence and case-fatality, which are
syndromes, such as pneumonia or meningitis, stroke is not influenced by the same factors. Thus, for any
highly heterogeneous, and its numerous causes influence interesting variations by time or place to be related to
the prognosis, the type of treatment required, and the likely explanations, information on incidence is still
preventive strategies. For the definition of many of the needed. Also, prevalent cases are more biased by
types and subtypes of stroke, technology is certainly underrepresentation of fatal cases than are incident cases
necessary (see page 6). in, for example, case-control studies. Furthermore,
The epidemiology of "all strokes" lumped together is prevalence is tiresome and cumbersome to measure,
one matter, whereas the epidemiology of the various because it requires a large population to be surveyed for
stroke types and subtypes may be quite another. Ideally, all strokes in the past, with inevitable difficulty in
all require study. But, until the quite recent past, most knowing whether a stroke has really occurred and what
epidemiological studies and clinical trials were based type it was. Prevalence can of course be estimated from
mainly on all strokes, sometimes with separate analysis of incidence and case-fatality, in which case incidence has
the rather distinct clinical syndrome of spontaneous to be measured in the first place. Finally, measurement
subarachnoid haemorrhage. Even nowadays the most of the prevalence of stroke-related disability, which is
fundamental classification of strokes as primary of obvious interest to health-care planners, is all
intracerebral haemorrhage (PICH) or ischaemic stroke but impossible because of overlapping disabilities caused
may not be practicable because computed tomographic by disorders such as osteoarthritis, claudication, and
scanning is unavailable. Moreover, even where computed dementia. In practice the prevalence of overall disability is
tomography is available, it may be difficult to get scans on probably of more relevance, anyway, than is prevalence of
all the patients, and quickly enough to exclude PICH, specific disabilities.
especially in community-based studies. The long-term outcome after stroke can be measured in
several ways, with various degrees of accuracy. 1 Overall,
Burden of stroke about 20% of patients having their first strokes are dead
The various facets of the burden of stroke include in a month (early case-fatality), and of those alive at 6
mortality, incidence, prevalence, long-term outcome, and months about a third are dependent on others for
cost. activities of daily living. But these averages conceal huge
Death certificate data provide easily available, but not individual variations, which depend not just on stroke
very accurate, information on stroke mortality, obviously type and subtype but on several other features, including
omit the milder forms of stroke that are seldom fatal (eg, initial stroke severity and pre-stroke disability. The
lacunar stroke), and cannot reliably distinguish even the stroke-recurrence rate is about 5% per annum, but tends
most basic pathological types of stroke (eg, ischaemic to be higher in the first few weeks and months, especially
stroke vs PICH). Accuracy can also be compromised by if the stroke is due to severe carotid stenosis. 1'4 Also,
variations in how stroke is recorded and coded, especially because stroke survivors usually have vascular disease
if it appears in part two of the death certificate. But at affecting all their arterial systems and not just the cerebral
circulation, they are at high risk of serious coronary
Lancet 1998; 352: (suppl III) 1-4 events--about 3% per annum. 1
Department of Clinical Neurosciences,Western General Hospital, Although the cost of stroke is clearly an important public
Edinburgh EH4 2XU, UK (C P Warlow FRCP) health and indeed political issue, it is very difficult to

Stroke • Vo1352 • October • 1998 sial

 THE LANCET

35,

238

Figure 1: Annual age-standardised incidence of first-ever-in-a-lifetimestroke (all types combined) per 100 000 population in the
1980s and 1990s for people aged 45-84, from 11 comparable community-based studies
Error bars are 95% Cl.
From ref 3 with permission.

assess because it depends so much on exactly what is number at high risk? 4 Maybe the main reason is that the
included. In Scotland the cost of each stroke to the prevalence of causative risk factors (eg, diabetes), or their
National Heath Service in 1988 was about £6000, but population mean (eg, blood pressure), has been
this estimate did not include anything other than hospital declining, and this change may partly reflect a birth-
and some family-doctor costs? Add in community, social cohort effect due to improvement in fetal health many
service, and family costs plus the indirect cost of loss of years ago.
productivity, and the total may be more like £70 000, at
least in the U S A at 1990 prices. 6 Yet more complexities Problem of stroke heterogeneity
arise when the cost of looking after a stroke patient is not Because most observational epidemiological studies of
entirely due to the stroke itself; there can be confounding stroke, as well as randomised trials in stroke prevention,
with non-stroke dementia, non-stroke mobility problems have perforce had to lump all strokes together, important
such as arthritis, and other non-stroke causes of disability differences in not just outcome and appropriate treatment
and dependency. policies, but also in cause, may have been obscured in
specific stroke subtypes. After all, the cause of an
Is s t r o k e b u r d e n c h a n g i n g w i t h t i m e ? intracerebral haemorrhage is in general unlikely to be the
Although stroke mortality has declined in many countries same as the cause of an ischaemic stroke (although there
in the past 50 years or so, particularly in the west and in are rare exceptions, such as cerebral vasculitis). Likewise,
Japan, it has increased in others, especially in Eastem the cause of atheroma affecting the cerebral circulation is
Europe. 7 But whether this variation reflects just changing surely different from the cause of carotid dissection in the
fashion in death certification or a real change in stroke young, yet both cause ischaemic stroke. Increasingly,
incidence or case-fatality is unclear. If measuring stroke therefore, epidemiology is following clinical practice, and
incidence accurately merely once is difficult, monitoring as well as considering all strokes together as the necessary
incidence accurately over time is even harder, and has first step, divides them into the three principal
hardly been achieved anywhere even though it must be pathological types (ischaemic stroke, P I C H , and
desirable to do so, at the very least to monitor the effects subarachnoid haemorrhage), and then into various
of "mass" and "high-risk" prevention programmes. It subtypes (such as ischaemic stroke caused by intracranial
does seem, however, that incidence has fallen in small-vessel disease, large-vessel atheroma, and embolism
Rochester, Minnesota, but is now rising againfl has risen from the heart). The trouble with this approach of
in Siberiafl and Sweden, 1° but has not changed in New "splitting" strokes into various subtypes is that it depends
Zealand. H Where stroke incidence has changed, the so much on diagnostic skills and technology that will not
reasons must almost certainly be environmental rather be available everywhere. Furthermore, there are constant
than genetic, and so presumably amenable to arguments about appropriate diagnostic criteria. Exactly
modification. But, any decline in incidence cannot be what characterises an ischaemic stroke definitely due to
very much to do either with the medical management of embolism from the heart, for example, let alone one due
hypertension ~2 or of transient ischaemic attacks, I3 largely to a specific cardiac abnormality such as a patent foramen
because most strokes arise in the much larger number of ovale? And what if there is more than one possible cause
people at moderate risk of stroke than in the smaller of stroke in an individual, which is so often the case?

sin2 Stroke • Vol 352 ° October • 1998

 THE LANCET

stroke seems to reduce their stroke risk, even though the

association between stroke and plasma cholesterol is so
weak '9 (as yet no trials of cholesterol lowering in stroke
survivors are available).

W h y is s t r o k e d i f f e r e n t f r o m m y o c a r d i a l
infarction?
If 80% or so of strokes are ischaemic9 and if the vast
majority of those ischaemic strokes are due to
degenerative vascular disease, as are the vast majority of
myocardial infarctions, it is odd that there is such a
quantitative, if not qualitative~ difference in their
important risk factors. Compared with myocardial
infarction, stroke patients are at least 10 years older, there
is not such a male excess in middle age, and increasing
blood pressure is more strongly, and increasing plasma
cholesterol is less strongly, associated with stroke. This
Figure 2: Relative risk of stroke related to usual diastolic difference is all the more surprising since about a third of
blood pressure (DBP) in five categories defined by baseline patients with ischaemic stroke already have clinical
DBP manifestations of coronary-heart disease, such as angina
From an overview analysis of seven prospective observational studies. or a past myocardial infarct, s° Part of the reason is
Approximate mean usual DBP estimated from later remeasurements in
Framingham study. Error bars are 95% CI. presumably that so many epidemiological studies have
From ref 18 with permission lumped all strokes together, thus weakening causal
associations with ischaemic stroke if they are not also
Also, for many stroke subtypes, the sample sizes can be associations with haemorrhagic stroke. However, this
too small to provide precise answers. But splitting has explanation cannot be the whole story because most
thrown up some interesting possibilities--that ischaemic strokes are ischaemic. Furthermore, although about a
stroke might be related to increasing plasma cholesterol, fifth of ischaemic strokes are probably due to intracranial
whereas haemorrhagic stroke is related more to low small-vessel disease, whose pathological features are
plasma cholesterol; is that intracranial small-vessel disease different from those of atheroma, the risk factors are very
does not seem to be associated with carotid atheroma; ~6 similar, 2~ so these risk factors should be similar in
and that the risk of stroke seems much smaller in patients myocardial infarction, where atheroma is by far the most
with transient ischaemia in the eye than in those with common underlying vascular abnormality
ischaemia in the brain, even when they have similar
severity of arterial disease in the neckJ 7 The f u t u r e
Good stroke epidemiology can commonly be improved
"Risk factorology" by combining it with excellent clinical and diagnostic
Observational epidemiological studies, whether cohort or skills so that in observational and experimental
case-control, have revealed an amazing number of epidemiological studies what is called a stroke really is a
associations (risk factors) with stroke, most of which stroke, and the different types of stroke can be studied
cannot possibly b e on the causal pathway (claudication, separately. There is a real need for accurate comparisons
for example). In any event, new and eagerly reported of stroke incidence, in total and more importantly by
associations are commonly not confirmed in later but subtype, in different parts of the world, especially in
usually less prominent studies, or they turn out to be due different racial groups, to identify differences that may
to confounding. At the end of the day, the whole point of provide aetiological clues. No doubt new risk factors will
these studies is to understand causation and, so far, they emerge, some of which will turn out to be on the causal
have identified few causes of stroke: increasing blood pathway. Genetic risk factors are already emerging but
pressure, cigarette smoking, atrial fibrillation, and have not yet led to any really new insights into stroke
diabetes mellitus are widely accepted as being causal, prevention.
whereas increasing plasma cholesterol, fibrinogen, and
homocysteine may be. Although the hope is that I thank Cathie Sudlow for her help in preparation of this paper.
randomised trials of removal of a risk factor would reduce References
the risk of stroke if it were really causal, this result cannot 1 Warlow CP, Dennis MS, van Gijn J, et al. Stroke: a practical guide to
necessarily be expected. It may be too late, because a management. Oxford: Blackwell Scientific, 1996.
disease process is already primed to cause stroke; to 2 Murray CJL, Lopez AD. Mortality by cause for eight regions of the
change disease incidence it may be necessary to intervene world: global burden of disease study. Lancet 1997; 349:1269-76.
3 Sudlow CLM, Warlow CP, for the International Incidence
for far longer than the patience of most clinical trialists; Collaboration. Comparable studies on the incidence of stroke and its
the treatment might cause as well as prevent stroke (eg, pathological types: results from an international collaboration. Stroke
warfarin); and trials may simply be impracticable (eg, 1997; 2 8 : 4 9 1 99.
those of giving up smoking). Fortunately, the treatment 4 European Carotid Surgery Trialists' Collaborative Group.
Randomised trial of endarterectomy for recently symptomatic carotid
of hypertension seems to reduce stroke risk rapidly, so stenosis: final results of the MRC European Carotid Surgery Trial
increasing blood pressure is readily accepted as causal for (ECST). Lancet 1998; 351: 1379-87.
stroke in general (figure 2), even though the various 5 Isard PA, Forbes JF. The cost of stroke to the National Health Service
in Scotland. Gerebrovasc Dis 1992; 2: 47-50.
subtypes were never distinguished in the trials or in the
6 Taylor T N , Davis PH, Torner JC, Holmes J, Meyer JW,
cohort studies. Rather surprisingly, the lowering of Jacobson MF. Lifetime cost of stroke in the United States. Stroke
plasma cholesterol in patients who have not yet had a 1996; 27: 1459-66.

Stroke • Vol 352 • October • 1998 sin3

THE LANCET

7 Bonita R, Stewart A, Beaglehole R. International trends in stroke Group. Blood pressure, cholesterol and stroke in Eastern Asia. Lancet
mortality: 1970-1985. Stroke 1990; 21: 989-92. (in press)
8 Brown RD, Whisnant JP~ Sicks JD, O'Fallon WM~ Wiebers DO. 16 Boiten J, Rothwell P, Slattery J, Warlow C, for the European Carotid
Stroke incidence, prevalence, and survival: secular trends in Surgery Trialist's Collaborative Group. Frequency and degree of
Rochester, Minnesota, through 1989. Stroke 1996; 27: 373-80. carotid stenosis in small centrum ovale infarcts as compared to lacunar
9 Feigin VL, Wiebers D O , Whisnant JP, O'Fallon M. Stroke incidence infarcts. CerebrovascDis 1997; 7: 138-43.
and 30-day case-fatality rates in Novisibirsk~ Russia, 1982 through 17 Streifler JY, Elaisziw M, Benavente R, et al. T h e risk of stroke in
1992. Stroke 1995; 26: 924-29. patients with first-ever retinal vs hemispheric transient ischaemic
10 Terent A. Increasing incidence of stroke among Swedish women. attacks and high-grade carotid stenosis. Arch Neuro11995; 52: 246-49.
Stroke 1988; 19: 598-603. 18 M a c M a h o n S, Peto R, Cutler]', et al. Blood pressure, stroke, and
coronary heart disease. Part 1, prolonged differences in blood
11 Bonita R, Broad JB, Beaglehole R. Changes in stroke incidence and
pressure: prospective observational studies corrected for the regression
case-fatality in Auckland, New Zealand, 1981-1991. Lancet 1993;
dilution bias. Lancet 1990; 335: 765-74.
342: 1470-73.
19 Hebert PR, Gaziano JM, C h a n KS, Hennekens CH. Cholesterol
12 Bonita R, Beaglehole R. Increased treatment of hypertension does not lowering with statin drugs, risk of stroke, and total mortality: an
explain the decline in stroke mortality in the United States, overview of randomised trials. JAMA 1997; 278: 313-21.
1979-1980. Hypertension 1989; 13: 1-69-73.
20 Sandercock P A G , Warlow CP, Jones LN, Starkey I. Predisposing
13 Warlow CP. C a n neurologists influence stroke incidence and do they? factors for cerebral infarction: the Oxfordshire Community Stroke
J R Gall Physicians (in press). Project. BMJ 1989; 298: 75-80.
14 Rose G. T h e strategy of preventive medicine. Oxford: Oxford 21 Lodder J, Bamford JM, Sandercock P A G , Jones LN, Warlow CP. Are
University Press, 1992. hypertension or cardiac embolism likely causes of lacunar infarction?
15 Eastern Stroke and Coronary Heart Disease Collaborative Research Stroke 1990; 21: 375-81.

sin4 Stroke • Vol 352 • October • 1998