Abstract
In this paper we review the evidence that migraine is associated with ischaemic stroke in young women, emphasizing potential biases, factors that may influence the association, potential mechanisms, and potential public health impact. Consistency of case-control findings from several countries and supporting evidence from prospective data suggest that the association is not an artifact of study design or execution, although, due to methodological limitations, none of the studies mentioned can be considered definite proof of the association. However, it is less clear whether migraine without aura is associated with stroke or whether the association is restricted to migraine with aura. Similarly, there are few data examining the magnitude of the association among non-users of oral contraceptives compared with those who use low oestrogen oral contraceptives. As a consequence, there is a lack of data concurrently stratifying both by the presence vs. the absence of aura and by the use of low oestrogen oral contraceptives vs. non-use of oral contraceptives. Moreover, there is still no convincing evidence on the mechanisms that would be implied and on the groups of migraineurs really at risk of ischaemic stroke. Despite the considerable advances in our understanding of the relationship between migraine and stroke, there are many gaps in the data needed for public health recommendations.
Introduction
In recent years, several studies have been published on the association between migraine and stroke in young women (i.e. under 45 years of age) (1–5). Overall, these studies raise the possibility that migraine is a risk factor for ischaemic stroke in this age and sex group. As migraine could affect up to one-third of women under 45 (6), these results are of great concern for young women with migraine and for their physicians.
In this paper we will review the evidence that migraine is associated with ischaemic stroke in young women and evaluate the likelihood that this is a causal association. We will also review some of the hypotheses raised by this association.
Characteristics of the studies on migraine and stroke in young women
Few studies on the association between migraine and stroke in young women were performed before the publication of the International Headache Society (IHS) criteria in 1988 (7). One of the most important is the Collaborative Group for the Study of Stroke in Young Women published in 1975 (8), in which 140 women, 15–44 years of age, who had had a stroke were compared to two groups of controls. The results for ischaemic stroke were somewhat conflicting: when cases were compared to hospitalized controls, the risk of stroke was not significantly increased in migraineurs (odds ratio (OR∗= 1.1 [0.6; 1.9]), whereas with neighbour controls, the increase in risk was significant (OR∗= 1.7 [1.1; 2.6]). The risk estimates were similar when only non-users were considered.
It could be argued that there was no major difference in risk estimates since both estimates were above 1 and the confidence intervals from the two control groups largely overlapped. However, the association was not specific to ischaemic stroke. A similar pattern of association was found for haemorrhagic stroke. Among non-users of oral contraceptives, when cases were compared to hospitalized controls, the risk of stroke was not significantly increased in migraineurs (OR = 1.2 [0.8; 1.8]), whereas with neighbour controls, the increase in risk was significant (OR = 1.8 [1.2; 2.7]).
In conclusion, this first study showed a weak evidence of an increased risk of ischaemic stroke in young women with migraine due to inconsistent findings for the two control groups and a lack of specificity to either ischaemic or haemorrhagic stroke. With the publication of the IHS criteria (7), the debate was revitalized and, since then, several studies using these criteria have been published (Table 1). These recent studies have consistently shown an increased risk of ischaemic stroke in young women with migraine (Table 1).
Recent case-control studies on the association between migraine and stroke in young women
Adjusted for age, hypertension, diabetes.
Adjusted for age, history of hypertension, smoking, use of oral contraceptives.
Adjusted for age.
Adjusted for high bloodpressure, education, smoking, family history of migraine, alcohol consumption and social class.
Study design
Case-control studies (Table 1)
Setting
In these studies, patients with a stroke were compared to patients without stroke. Four of these studies were hospital-based, with the number of cases varying from 20 to 146 (1, 3–5). In most studies, controls were hospitalized in the same centres as cases. One study used two groups of controls, one from the hospital and one from railway system personnel. As the prevalence of migraine did not differ significantly in both groups, they were merged into a unique control group (4). In the largest study, 497 cases were recruited from Danish population-based registries in which all hospitalized cases are registered. Controls were also selected through a national register and the participation rate was high for both cases and controls (85 and 88%, respectively) (2).
Diagnosis of stroke
Confirmatory CT scans were performed in most of the case-control studies, although in the World Health Organization (WHO) study there was no mention in the main paper of the percentage of cases with CT scans performed (5).
All studies included cases with an ischaemic stroke with a permanent deficit, although patients with transient ischaemic attacks (TIA), generally defined as symptoms lasting less than 24 h, were also enrolled in two studies (2, 4). In the Danish study, TIA represented 27% of cases and no separate analysis was performed (2). In the Italian study, TIA represented 47% of the inclusions in women. Separate analyses showed that migraine was strongly associated with the risk of TIA (odds ratio 2.5 [1.2–4.9]) but not with that of ischaemic stroke (odds ratio 1.3 [0.7–2.4]) (4).
The association between migraine and haemorrhagic stroke was studied in only the earlier Collaborative Study (8) and two recent studies. As noted earlier, the Collaborative Study found that migraine was similarly associated with both ischaemic and haemorrhagic stroke. In the recent European WHO study, no association (odds ratio 1.10 [0.63–1.94]) with migraine was found (5). In the pooled analysis of data from two recent United States of America (USA) population-based case-control studies conducted by Kaiser Permanente and the University of Washington, the crude odds ratio was 1.45 (1.01–2.08]. For all three reports, most haemorrhagic stroke cases were subarachnoid haemorrhage.
Diagnosis of migraine
In all the studies mentioned, migraine was diagnosed retrospectively, that is after the stroke had occurred in cases and after or during hospitalization in hospitalized controls. In some studies the assessment was made by a neurologist using a structured questionnaire reproducing the IHS criteria (3, 4). In two studies the diagnosis of migraine was derived through the answers to a written questionnaire. In the population-based study from Lidegaard et al. the IHS criteria were not used: only simple questions about migraine and the frequency of attacks were asked (2). In the WHO study by Chang et al. the assessment of migraine was performed through a written questionnaire reproducing partly the IHS criteria (5).
Sub-types of migraine, with and without aura, were assessed in three studies (3–5). In all, migraine with aura was associated with a higher risk of ischaemic stroke than migraine without aura. In two of these studies, the risk associated with migraine without aura alone did not reach significance (4, 5).
Prospective studies
Although cohort studies of young women specifically designed to study the issue of the association between migraine and stroke risk would be so large and expensive as to be impractical, there is much benefit to examining the available prospective data, even in populations such as older men. In the Physician's Health Study (9), 22 071 USA male physicians aged 40–84 years who had no prior history of cardiovascular disease were enrolled in a randomized controlled trial. In the follow-up questionnaire at 6 months and yearly thereafter, participants were asked if they had experienced a migraine headache since the last questionnaire. No instructions were given as to what headaches they should call migraine. If a physician reported migraine on at least one of his follow-up responses, then he was classified as having migraine. A sub-study confirmed that persons so classified had had their first attack long before randomization into the study and were reflective of prevalent migraine rather than incident migraine. A separate question asked whether they had experienced ordinary (non-migraine) headaches since the last questionnaire.
Ischaemic or haemorrhagic stroke endpoints were determined for the clinical trial by the study neurologist, based on medical record review. Non-migraine-type headaches were not associated with stroke. In multiple risk factor adjusted analyses, which varied little from the crude results, migraine was associated with ischaemic stroke (OR = 2.0, 95% CI 1.1–3.6). This risk was slightly but not significantly lower in the presence of other risk factors such as hypertension, diabetes mellitus and cigarette smoking, although the numbers of strokes among migraineurs were small and the power to test interactions was small. Data for haemorrhagic stroke was too sparse for separate analysis. The severity of the strokes was similar for those with and without migraine, reducing the likelihood that a migraine event itself would have been misclassified as a stroke.
Analyses based on the National Health and Nutrition Examination Survey (10) found that both physician-diagnosed migraine and analgesic use for headaches were associated with stroke of all types after adjustment for established stroke risk factors. Interestingly, the association was stronger at younger ages and after controlling for other risk factors. At age 40, the estimated hazard ratio was 2.81 (95% CI 1.45–5.43) for migraine and 4.6 (95% CI 2.13–9.96) for headache. All cases (number unspecified) of stroke associated with migraine in women under age 45 occurred in oral contraceptive users. A limitation of this study was that the ascertainment of incident stroke was by self-report, proxy report or death certificate, with no independent validation. A further limitation is that migraine history and age of migraine onset was ascertained at follow-up concurrently with stroke history, although headache history was ascertained as baseline.
A common limitation of these prospective studies is that none used the IHS criteria for the diagnosis of migraine. This is particularly true for the second study, for which only physician-diagnosed migraine was available (10). This would result in misclassification (see below).
Could the observed association be explained by biases?
From this review of the literature, the overall pattern of results is consistent, as all studies found an association between migraine and ischaemic stroke in young women. However, all these studies are observational studies and therefore the association could be an artifact of potential biases due to study design or study execution. The question therefore arises whether the association that was observed could be explained by a bias or a conjunction of biases common to the various studies. Several potential biases could be considered. We will discuss only the most plausible.
Selection of patients
Patient selection could occur at various levels, but the association observed between migraine and ischaemic stroke would be explained by selection biases only if migraineurs were over-represented in cases or under-represented in controls.
First, a referral bias might exist: young women with stroke and migraine could be considered as more problematic cases and would be referred more frequently to academic centres than cases without migraine. Although possible, this bias seems unlikely since most were multicentre studies and included academic and non-academic centres. In one study, separate analysis showed that the prevalence of migraine was similar for cases included in academic and non-academic centres (3). Furthermore, the association was also found in the population-based study (2).
Selection biases could also occur within the centres if the investigators more readily included cases with migraine; however, no such selection occurred for controls. This would be the case, for example, if patients with migraine had less severe strokes than non-migraineurs and if the investigators more frequently asked patients with minor strokes to enter the study. This bias is unlikely to have been a major factors in the case-control studies, as in four studies the recruitment was made of consecutive cases (1, 3–5) and in one study it was population-based (2).
Controls could also be selected in a way that migraine would be under-represented in this group, which would lead to the false conclusion that migraine is comparatively over-represented in cases. This could occur in hospital-based case-control studies (1, 3, 5) if controls were hospitalized for diseases in which migraine is particularly infrequent. However, even if such diseases exist, which is not known at the present time, controls were usually recruited from several medical departments, which would reduce the importance of this potential bias. A reduced prevalence of migraine in controls could also be due to a high non-response rate of controls with migraine: because of their headaches, controls with migraine could be less willing to participate. However, this bias was estimated in one study and was shown to be marginal (3). Furthermore, in hospital-based studies, prevalence of migraine in controls was from 15% to 30% which is within the range expected in the population for this age and sex group. Finally, selection bias for cases or controls is much less of a concern in the two prospective studies, provided that migraine does not alter the threshold for being identified as a case.
TIA vs. stroke
An unusual type of selection bias could occur where exposed subjects are recruited erroneously as subjects with a qualifying event. As mentioned above, TIAs were included as qualifying events in two studies (2, 4). This is of particular importance since symptoms of TIAs are sometimes difficult to distinguish from an attack of migraine with aura, especially the 1.2.5 variety of aura without headaches of the IHS classification (7). This misclassification bias would have for end-result an overestimate of the prevalence of migraine in cases and therefore an over-estimation of the risk. This potential bias is of great concern, as in the only study in which separate analyses (stroke vs. TIA) were performed, migraine was strongly associated with TIA (odds ratio 2.5 [1.2–4.9]) but not with stroke (odds ratio 1.3 [0.7–2.4]) (4). However, in this study special efforts were made to avoid TIA cases who were known to be ligitious (4). Furthermore, the association between migraine and stroke was also found in studies in which TIAs were excluded (1, 3, 5).
A similar problem is possible even when the end-point is stroke. Migraine may have a prolonged neurological aura that lasts longer than 24 h and mimics stroke yet resolves completely. For this reason, it is essential that all studies considering the relationship of migraine to stroke include only cases with definite neuroimaging evidence for stroke (3) not merely cases where the neuroimaging is consistent with stroke (e.g. no evidence for tumour, demyelinating disease, etc.). It is worthy of note that this potential problem is not obviated by a prospective study design.
Diagnosis of migraine
One of the main issues in the study of the association between migraine and stroke is the difficulty in assessing the diagnosis of migraine. This diagnosis is purely clinical and relies on the answers to a series of question that could be interpreted with some variability by the patients as well as the interviewer. This probably partly explains the important differences of the prevalence of migraine observed in the studies on migraine and ischaemic stroke in young women, which ranges from 13.7% to 60% in cases and from 4.8% to 30% in controls.
Issues of both the validity and bias in the diagnosis of migraine need to be considered in epidemiological, and particularly case-control, studies. Accurate diagnosis of migraine is important to avoid non-differential misclassification of exposure, which will bias the risk estimate towards showing no association. If it is not possible to have an expert interviewing all the subjects, a validation study is useful for this purpose. In the validation study, an expert in headaches will replicate the interview on a subset of subjects. In three of the case-control studies, migraine was assessed by an expert through a direct interview with the patients (Table 1) (1, 3). In the other two case-control studies, the diagnosis was assessed by other methods but no validation study was performed.
It is likely that the methods of assessing the diagnosis of migraine of these last two studies are associated with a higher percentage of misclassification. In the WHO study, the diagnosis was obtained with a written questionnaire that reproduced, in part, the IHS criteria (5). In the Danish study, the diagnosis was made through a simple question about migraine and the frequency of attacks without any reference to the IHS criteria (2). The low prevalence of migraine in cases and controls observed in this study could be the consequence of geographical differences, but it is more likely related to the assessment of migraine. Migraine was self-diagnosed and it is possible that, in the population, a significant proportion of women with migraine are not aware of the exact diagnosis of their headaches.
Misclassification, if non-differential between cases and controls, is a less important threat to internal study validity than bias. Because they are retrospective, i.e. conducted after the event of interest, case-control studies are particularly susceptible to bias in the ascertainment of exposure. This bias may take several forms. There may be interviewer bias and recall bias by participants.
Interviewer bias may occur when the interviewer is aware of both the hypothesis tested and the status of the patient interviewed. The interviewer could, involuntarily, over-estimate the frequency of migraine in cases by questioning them more thoroughly than controls. A solution would be to have the interviewer blinded to the clinical status of the patient. This is, however, difficult to set up and was not done in any of the studies published. Other measures to control the potential diagnostic bias due to the interviewer include the use of definite criteria for the diagnosis of migraine in a structured questionnaire (1, 3, 4).
The inter-observer agreement of the diagnosis of migraine with an IHS criteria-based structured questionnaire was tested in one of the studies on migraine and stroke in young women (3). In this study, 60 women (30 cases and 30 controls) had a replicate telephone interview by another neurologist, 5–9 months after the first interview. They were chosen randomly from the lists of cases and controls, and asked by telephone to participate in a complementary study. At first call, case patients were instructed not to talk about their stroke during the interview. Then each call was transferred to a neurologist unaware of the status – case or control – of the patient. In the second interview, only questions regarding headaches were asked in order to keep the observer blinded. The kappa index was 0.79 in cases and 0.83 in controls (3), which confirms that the use of a structured questionnaire based on the IHS criteria is an efficient means for a high reproducibility of the diagnosis of migraine in this setting.
Recall bias
Recall bias could be responsible for the association observed if cases were more prone than controls to accurately remember their headaches. Patients with stroke and their relatives often try to find some explanation to this dramatic event and they are more likely to scrutinize their past medical history than are controls. This would lead to a higher prevalence of migraine in cases compared with controls. However, it is unlikely that this potential bias explains the observed association, since cases were generally unaware of the hypothesis under study. There is no reason why they should report migraine symptoms more completely than controls. Furthermore, migraine is a recurrent and painful condition and it appears unlikely that major differences of recall would exist between cases and controls. However, this kind of bias cannot be estimated directly and its importance might be greater than usually considered. Two studies attempted to have patients make an indirect estimate of a potential recall bias by asking lure questions, that is, questions about head trauma (1) or other neurological diseases (4) that are not related to stroke. In these two studies no over-reporting was found, which is quite reassuring but does not definitively answer the question of a potential recall bias in cases.
Publication bias
The issues of bias and confounding discussed above pertain to the issue of the internal validity of each individual study. External validity or generalizations from published data may also be threatened if there is a selection of which research findings are submitted or accepted for publication. Authors are more prone to submit a study showing a positive association than a study with no association. Although a study specifically and adequately designed to evaluate the association between migraine and ischaemic stroke and showing no association would probably be welcomed by editors, it is possible that authors would be reluctant to submit it in a context of several positive studies already published. Although a publication bias cannot be ruled out, the consistency of the association observed in the published studies does not seem to be related to such a bias.
In conclusion, all the biases we have described could have some influence, although their potential effect should vary across studies according to the care given to limit their importance in the design or the analysis. It is interesting to note that the two largest studies (2, 5) were not dedicated to the particular issue of the association between migraine and stroke, which explains why they have the less stringent assessment and diagnostic criteria and are therefore more exposed to misclassification biases. However, the highly consistent results in several settings and in various countries suggests that this association is neither purely fortuitous nor the result of simple misclassification or confounding.
This does not mean, however, that we can be absolutely confident of the existence of an association between migraine and ischaemic stroke. Some of the biases mentioned above are unavoidable or not fully controllable. These biases could have had some influence in all the studies and produced, when combined, similarly biased estimates. They are inherent to case-control studies, which is the design of all these studies. For this reason, definite confirmation or refutation of the association between migraine and ischaemic stroke in young women will not arise from larger case-control studies but from follow-up studies in which migraine is diagnosed before the occurrence of a stroke. Two follow-up studies have been published in which the association between migraine and stroke has been studied: one in male physicians (9) and the other on a large sample of the population (10). Both showed an association between migraine and stroke. Although limited by the lack of precise diagnostic criteria for migraine and by validation procedures for stroke (10), these findings lend support to the case-control data.
Variables influencing the association
Type of stroke
The risk of stroke in young women with migraine seems to apply not only to TIA and ischaemic stroke, but also to haemorrhagic stroke, although haemorrhagic stroke has been less completely studied and the results are less consistent. A lack of specificity of the association to ischaemic strokes would raise concern regarding the possibility of bias or confounding. Within the group of ischaemic strokes, the risk does not seem to vary according to the subtype of ischaemic stroke but again this was analysed in only one study and the power to detect any difference was probably low (3). It would be of interest to know if there are differences between the subtypes of ischaemic stroke as it could suggest pathophysiological processes. In a previous case-control study, it was shown that migraine diagnosed according to the IHS criteria was more frequent in cases with cervical artery dissection than in non-stroke controls (11). Limiting the association between migraine and ischaemic stroke to some sub-types of stroke, such as cervical artery dissection, would help to focus future investigations concerning the pathophysiological processes. Future studies in this area should try to categorize ischaemic strokes.
Migraine with and without aura
In the three studies that have estimated the risk of ischaemic stroke according to the type of migraine, the risk was highest in migraine with aura, with OR ranging from 3.8 to 6.2 (Table 1) (3–5). In one of these studies, the prevalence of migraine with aura was astonishingly high in cases and in controls, as it represented 70% of patients with migraine (69/96 in controls) (5). These figures are well above all published data, which raises questions about the assessment of migraine with aura in this study. Migraine with aura was diagnosed on the basis of at least one attack with aura at any time, which is at variance with the IHS definition in which at least two attacks with aura are required. However, even with a diagnosis of migraine with aura based on a single attack, a frequency of 70% is far beyond what is expected in European populations. This raises concern not only as regards the diagnosis of migraine with aura but also as regards the assessment of the aura itself.
The odds ratios for migraine without aura were roughly half those found for migraine with aura, ranging from 1.5 to 3.0 (3, 5) and the risk was significantly higher than one in only one of the three studies (3).
It is unclear if migraine with aura and migraine without aura are different diseases or share some common mechanisms. There is little doubt that migraine with aura is consistently associated with ischaemic stroke, but it is too early to confidently state that migraine without aura carries no risk. More studies are needed to explore this important point.
Prophylactic migraine treatment was analysed in two studies and it was concluded that it did not play a significant role in the association between migraine and ischaemic stroke (3, 4).
Oral contraceptive use and tobacco smoking
More than 20 years ago Bickerstaff suggested that there might be an interaction between migraine and oral contraceptives as regards the risk of stroke (12). In two studies the risk of ischaemic stroke was substantially increased in migrainous women who were using oral contraceptives compared to non-oral contraceptives users (OR = 13.9 vs. 3.7 in one study (3) and 16.9 vs. 2.27 in another (5)). However, the numbers were small in each category and statistical interaction was not significant. In both studies the risk of ischaemic stroke was higher in women using high oestrogen content oral contraceptives but migraine could not be studied simultaneously due to the small sample sizes (3, 5).
Since it is now well established that, by contrast to high oestrogen content oral contraceptives (50 μg), low dose oral contraceptives carry very little or no risk in the absence of other risk factors, it is likely that the odds ratio for the risk of ischaemic stroke with the association of low dose oral contraceptives and migraine would be lower than those observed in the two above-mentioned studies. The single study addressing this issue was the pooled analysis of two USA studies (13), which found that patients with migraine were the only subset in which low dose oral contraceptives were associated with an increased risk of ischaemic stroke (OR = 2.08, 95% CI 1.19–3.65). Again, this would be a crucial point for further investigations.
A relatively similar pattern was observed for tobacco smoking, with a higher risk in migrainous women who smoked tobacco compared with non-smokers (OR = 10.2 vs. 5.8 for smokers (20 cigarettes/day in one study (3)). Cigarette smoking has been consistently shown, first, to increase the risk of ischaemic stroke, with a relative risk of around 2, and second to greatly enhance the risk associated with oral contraceptives, with a relative risk of around 6 in most, but not all, studies (13). Therefore, it is not surprising that the combination of the three factors – migraine, oral contraceptive use and tobacco smoking – carried a very high relative risk, 34.4 [3.27–361] in the WHO study. However, it should be emphasized that this dramatic increase was based on only nine cases and two controls (5).
From these results it can be concluded that the risk of ischaemic stroke in young women with migraine is greatly increased if they use oral contraceptives or if they smoke tobacco or both. Whether there is an interaction between migraine and these exposures remains to be clarified. Furthermore, we are aware of no data concurrently stratifying both by the presence vs. the absence of aura and by the use of low oestrogen oral contraceptives vs. non-use of oral contraceptives.
Potential mechanisms
An important argument for the causal nature of an observed association between an exposure and a disease is its biological plausibility, that is to have strong assumptions about an underlying mechanism. Numerous hypotheses have been raised to explain the association between migraine and ischaemic events: vasospasm, endothelial dysfunction, congenital thrombophilia, platelet hyperaggregability, association with cardiac abnormalities (patent foramen ovale or mitral valve prolapse) predisposing to ischaemic stroke, etc. However, to date, no fully convincing evidence has been produced.
Considerations have to take into account that migraine is a frequent condition and ischaemic stroke a rare disease in young women. Considering this constraint, the various hypotheses could be summarized as follows.
Very few women with migraine are at high risk for stroke
Under this hypothesis, a few cases of stroke associated with migraine explain the overall association observed. Two mechanisms could be distinguished.
Migraine as causal factor
This hypothesis states that a stroke is directly and causally related to an acute migraine attack; it is often mistaken as the only hypothesis possible to explain the association. The underlying hypothesis is that the mechanisms involved in migraine could, in some circumstances, directly lead to a stroke. This is the general concept of migrainous stroke or migraine-induced stroke (14). In the past, migrainous stroke has been largely over-diagnosed; it was redefined in a more restrictive way in the IHS classification (7) as a stroke occurring during a migraine attack in a patient with a migraine with aura, the symptoms and signs of the stroke being identical to those of usual migraine attacks, and other causes of stroke being excluded. With this definition, migrainous stroke has become a rare disease and could not explain the increased risk of ischaemic stroke in migraineurs (1, 3, 4).
It could be argued that the present IHS definition is too restrictive and that migraine-induced stroke might exist even without all the features of the IHS classification. In the WHO study, migrainous stroke was defined as the existence of headaches in the 3 days before the stroke and it was concluded that up to 40% of the strokes in migrainous women are migrainous strokes (5). If the observed statistical association between migraine and ischaemic stroke was to be explained mainly by migrainous strokes, some features are expected. First, migraine should be diagnosed more frequently in cryptogenic strokes, that is, strokes with no other causes than migraine, which does not seem to be the case, although the evidence is sparse (3, 11). Second, oral contraceptive use and tobacco smoking are not expected to play a major role in the occurrence of a migrainous stroke and there should not be any interaction between them and migraine. Future studies should, by focusing on these points, bring evidence for or against this hypothesis of a direct causal relationship between migraine and stroke which, at the present time, does not appear to be the most likely.
Migraine as an innocent bystander
Diseases leading by themselves to a stroke may harbour migraine-like headaches as an associated symptom. In this situation, the apparent association between migraine and stroke is misleading and is, in fact, due to the relationship between the causal affection and both migraine and stroke. For example, migraine has been observed in diseases which are known causes of ischaemic stroke like MELAS (15, 16), antiphospholipid antibodies syndrome (17) and recently CADASIL (18). It is possible that, in these diseases, migraine-like headaches are an unspecific marker of the disease and carry no additional risk for ischaemic stroke. The association between migraine and cervical artery dissection described in a previous case-control study (11) also enters this category. Cervical artery dissection is a common cause of stroke in young subjects and if migraine were particularly frequent in this disease, it might help to give an overall apparent association between migraine and ischaemic strokes. These last results need confirmation in future studies.
All women with migraine are at increased risk of stroke
If all women with migraine indeed had a higher risk of stroke, then this risk has to be very low with regard to the high prevalence of migraine and the very low prevalence of stroke in young women. Under this hypothesis, there would be a yet unknown weak prothrombotic diathesis associated with migraine, operating independently or, more likely, synergistically with other factors. One example is the role of migraine as a risk modifier in the relationship between oral contraceptives and ischaemic stroke. Indeed, available data suggest that the risk associated with migraine in conjunction with oral contraceptives is far greater than the simple addition of the risks for each factor taken alone. This interaction is, however, only suggested and more studies are needed to elucidate this issue. It has also been suggested that some conditions predisposing for stroke, including minor cardiac abnormalities like patent foramen ovale (19) and mitral valve prolapse (20) might cause a stroke when combined with migraine but this has not been firmly established.
Conclusion
There are still a lot of unanswered questions on the issue of the association between migraine and ischaemic stroke in young women.
There is still some uncertainty about the existence of an association between migraine and ischaemic stroke in young women. Although all the studies published show an association between migraine and ischaemic stroke, they are all exposed to the same biases that could produce a false association. This cautious attitude is not purely theoretical. In the absence of a clinical or biological marker of migraine, its diagnosis relies only on answers to a series of questions and is therefore exposed to the potentially strong misclassification biases we have described. The confirmation of the association could only be provided by properly designed longitudinal studies. However, considering the very low prevalence of ischaemic stroke in young women, these longitudinal studies would need to include thousands of young women followed-up for many years with strict diagnostic criteria for both migraine, with a face-to-face interview with a structured questionnaire based on the IHS criteria, and stroke, with an adequate neuroimaging evaluation.
Indeed, if there is an association between migraine and ischaemic stroke (which is the most likely in view of the existing literature), what would be its explanation? Is there a common mechanism or are there several different possibilities? Is the mechanism direct or indirect? Does it concern potentially all women with migraine or just a subset with a very particular form of migraine? Is it limited to young women or are there implications for older women and men? Here again there are no convincing answers or even hypotheses. However, we think that this a highly challenging area: the answer to these questions could give us some precious insights into not only the mechanism of migraine but also into the mechanism of ischaemic stroke, which so frequently remains unexplained in young patients. For this explanatory part, case-control studies are still useful provided they use strict criteria for both migraine and stroke and include sophisticated investigations selected on the hypotheses tested.
Finally, what are the individual and public health consequences of this association? Fortunately, the absolute risk of stroke among young women is low: it has been estimated around 5–20 per 100 000 woman years (21–23). In women with migraine, this risk has been estimated to 17–19 per 100 000 woman years (3, 4) which means that women with migraine are still at a low risk for stroke and that there is no preventive treatment to undertake. However, until the situation has been clarified and subgroups at risk clearly identified, physicians should firmly advise young migrainous women not to smoke and, if they use oral contraceptives, to choose low oestrogen content pills or progestogen only pills (although there are no data about the stroke risk of oral contraceptives lacking oestrogen).