Migraine and Hemorrhagic Stroke
Background and Purpose—Several studies have assessed the possible increased risk of hemorrhagic stroke in migraineurs, drawing differing conclusions. No meta-analysis on the topic has been published to date.
Methods—Multiple electronic databases (MEDLINE, EMBASE, Science Citation Index, and the Cochrane Library) were systematically searched up to March 2013 for studies dealing with migraine and hemorrhagic stroke. We selected case–control and cohort studies with a clear definition of the diagnostic criteria for migraine and hemorrhagic stroke, using an adjusted model or a matching procedure that could control for potential confounders, and reporting effect estimates with 95% confidence intervals (CIs) or enough data to allow calculation of those numbers. Adjusted odds ratios and hazard ratios were used to estimate effect size.
Results—Of 11 264 records, we identified 8 studies (4 case–control and 4 cohort studies) involving a total of 1600 hemorrhagic strokes, which were included in the meta-analysis. The overall pooled adjusted effect estimate of hemorrhagic stroke in subjects with any migraine versus control subjects was 1.48 (95% CI, 1.16–1.88; P=0.002), with moderate statistical heterogeneity (I2=54.7%; P value for Q test=0.031). The risk of hemorrhagic stroke in subjects with migraine with aura (1.62; 95% CI, 0.87–3.03; P=0.129) was not significant. Compared with control subjects, the risk of hemorrhagic stroke was greater in females with any migraine (1.55; 95% CI, 1.16–2.07; P=0.003) and in female migraineurs aged less than 45 years (1.57; 95% CI, 1.10–2.24; P=0.012).
Conclusions—Available studies suggest that subjects with migraine have an increased risk of hemorrhagic stroke. Further studies are needed to address the hemorrhagic stroke risk according to migraine type, age, sex, and hemorrhagic stroke type.
See related article, p 2987.
Several observational studies have indicated an association between migraine and vascular diseases and, in particular, between migraine and ischemic stroke, subclinical brain lesions, cardiac events, and vascular mortality. Results of those studies have been included in various meta-analyses1–4 and in a recent review.5
The association between migraine and ischemic stroke is stronger in subjects experiencing migraine with aura (MwA), although it is either less evident or not significant in subjects with migraine without aura (MwoA).6–8 Subjects with MwA have a 2-fold increased risk of ischemic stroke when compared with nonmigraineurs.1,3 However, data on the possible association between migraine and hemorrhagic stroke, including intracerebral hemorrhage (ICH) and subarachnoid hemorrhage (SAH), are inconsistent.5
To evaluate the association between migraine and the risk of hemorrhagic stroke, we performed a meta-analysis of the available studies.
The present meta-analysis was performed according to the criteria reported by the MOOSE group.9
Two investigators (P.R. and R.O.) independently searched published papers indexed in MEDLINE, EMBASE, Science Citation Index, and the Cochrane Library from inception to March 2013, using the terms “migraine,” “headache,” and “migraine disorders” combined with the terms “cerebrovascular accidents,” “cerebrovascular disorders,” “stroke,” “cerebral infarction,” “brain infarction,” “brain hemorrhage,” “cerebral hemorrhage,” “intracerebral hemorrhage,” “intraparenchymal hemorrhage,” or “subarachnoid hemorrhage.” Results were filtered to include only studies written in English. In MEDLINE and EMBASE, the search was restricted to studies performed on humans. The “explode” feature available in EMBASE was used. The search input in MEDLINE included both text words and Medical Subject Heading (MeSH) terms. A manual search among references of selected articles and reviews was also performed.
Studies were included in the analysis if they met the following predetermined criteria: (1) a case–control or cohort design; (2) a clear definition of the diagnostic criteria for migraine, ICH, and SAH; (3) the use of an adjusted model or of a matching procedure that could control for potential confounders; and (4) the report of effect estimates with 95% confidence intervals (CIs) or of enough data to allow calculation of those numbers. In the case of overlapping data among studies, we included the report with the longest follow-up or the largest number of patients if all studies had a cohort design, or with a cohort rather than a case–control design if different designs were used. Moreover, the following exclusion criteria were applied: (1) studies with a design other than case–control or cohort (eg, case reports, letters, interventional studies, reviews); (2) studies not investigating migraine, ICH, or SAH; and (3) studies referring to transient stroke-like syndromes, migrainous stroke, mixed ischemic and hemorrhagic stroke, rare genetic syndromes characterized by both migraine and stroke (eg, cerebral autosomal dominant arteriopathy with subcortical infarcts and leukoaraiosis), or studies carried out on women during pregnancy or puerperium.
We used a 2-step procedure for the selection of eligible studies. In the first step, 2 investigators (P.R. and R.O.) independently screened titles and abstracts to identify and exclude all studies that did not meet the predetermined criteria. In the second step, the same researchers evaluated the full-text version of the remaining studies. All disagreements were resolved by consensus among all the study participants.
A standardized data collection form was used to extract the following information: last name of the first author, title of the study, publication year, country where the study was performed, years of subjects’ inclusion, study design, study size, source population, criteria used for migraine diagnosis, migraine type where available, fully adjusted effect estimates with 95% CIs and confounders, duration of follow-up, and type of hemorrhage. All disagreements were resolved by consensus.
Odds ratios (ORs) and hazard ratios (HRs) were used to estimate effect size. We chose to pool the adjusted effect measures rather than the crude ones because of the role of confounding factors in affecting the validity of observational studies. When different adjusted models were available, we chose the model including the largest number of factors. We included all studies regardless of age and sex distribution of participants. We performed an overall analysis of the association between any migraine and ICH or SAH. We also performed subgroup analyses for MwA and MwoA, females, and females aged <45 years. To obtain the pooled relative risk estimates, the natural logarithm of OR and HR was weighted by the inverse of their variance. We ran a random effects model rather than a fixed effects one because of the high likelihood of between-study variance. We performed a separate analysis according to the study design (case–control or cohort studies). Subgroup analyses were also used to investigate heterogeneity, where present.
In accordance with the Cochrane Collaboration Guidelines for systematic reviews,10 we assessed the clinical, methodological, and statistical heterogeneity of the included studies. Clinical heterogeneity was assessed by evaluating differences in the study populations, exposures, and outcomes; methodological heterogeneity was assessed by comparing the different confounders assessed in the adjusted models; statistical heterogeneity was assessed using the I2 statistic to quantify the proportion of variability in effect estimates attributable to heterogeneity between studies versus sampling error within studies. A value of I2 of 0% to 40% was considered not important, 30% to 60% represented moderate heterogeneity, 50% to 90% substantial heterogeneity, and 75% to 100% considerable heterogeneity. The I2 statistic is a derivative of Cochran’s Q, a test used to assess between-study variance in DerSimonian and Laird’s random effects model.11 We used the Galbraith plot12 to visually examine the impact of individual studies on the overall homogeneity test statistics and performed a sensitivity analysis to quantify the effect on the overall results of each of the included studies. We evaluated the potential publication bias by visually examining for possible skewness in a funnel plot and statistically using Egger’s method.13 Analyses were carried out with R software, using the meta14 and metafor15 packages.
Our search identified 11 264 records (Figure 1). After reviewing titles and abstracts and excluding records not meeting our inclusion criteria, we obtained 39 useful studies. Manual searching of references retrieved 3 additional studies. In the References in the online-only Data Supplement, we listed the 34 studies that were excluded after full-text revision and the reasons for their exclusion from the analysis. We were finally left with 8 studies involving a total of 1600 hemorrhagic strokes.16–23
Study characteristics are summarized in Tables 1 and 2. Four of the included studies had a case–control design,17–19,23 and 4 had a cohort design.16,20–22 The Collaborative Group study had a hospital control group and a neighborhood control group and stratified data according to oral contraceptive use.19 In the primary analysis, we included the data referring to oral contraceptive nonusers and comparison with the neighborhood control group; we also performed an additional analysis considering oral contraceptive nonusers and comparison with the hospital control group. All studies reported data on any migraine, whereas only 3 of them reported data regarding MwA and MwoA.18,21,22 Migraine was diagnosed according to the International Classification of Headache Disorders (ICHD) criteria in 2 studies,17,18 whereas for the Women’s Health Study, the agreement study of self-reported migraine with the ICHD criteria was very good.22 All but 1 of the studies investigated both ICH and SAH; 1 study reported data for ICH and SAH,22 whereas another study reported data on SAH only.17 Four studies investigated a female population,18,19,22,23 1 a male population,16 and the remaining 3 male and female populations.17,20,21 Two studies reported results for males and females separately.20,21 All but 1 of the studies included subjects according to predefined age groups20; 3 studies investigated female subjects aged <45 years.18,19,23 Arterial hypertension, cigarette smoking, age, body mass index, and lipid status were the most frequent potential confounders in the adjusted models.
Migraine and Hemorrhagic Stroke
The overall pooled adjusted effect estimate of hemorrhagic stroke in subjects with any migraine versus control subjects was 1.48 (95% CI, 1.16–1.88; P=0.002) with moderate statistical heterogeneity (I2=54.7%; P value for Q test=0.031). The 4 case–control studies had a pooled OR of hemorrhagic stroke of 1.41 (95% CI, 1.09–1.82; P=0.009) and low statistical heterogeneity (I2=18.6%; P value for Q test=0.298), whereas the 4 cohort studies had a pooled HR of 1.47 (95% CI, 0.97–2.24; P=0.068) with substantial statistical heterogeneity (I2=67.2%; P value for Q test=0.027). The comparison between case–control and cohort studies revealed no significant subgroup difference (P value for Q test=0.855), thus indicating low methodological heterogeneity. Figure 2 graphically reports our findings.
The overall pooled adjusted effect estimate of hemorrhagic stroke in subjects with any migraine versus no migraine, considering for the Collaborative Group study19 data referring to oral contraceptive nonusers and the hospital control group, was 1.38 (95% CI, 1.08–1.78; P=0.011) with moderate statistical heterogeneity (I2=58.0%; P value for Q test=0.020; Figure I in the online-only Data Supplement).
As shown in Figure 3, the pooled adjusted effect estimate of hemorrhagic stroke in patients with MwA versus control subjects was 1.62 (95% CI, 0.87–3.03; P=0.129); the same adjusted effect estimate in patients with MwoA was 1.39 (95% CI, 0.74–2.62; P=0.303). Data referring to migraine type were contributed by 3 studies (I2=62.1%; P=0.072 for MwA and I2=45.3%; P=0.161 for MwoA).18,21,22
The overall pooled adjusted effect estimate of hemorrhagic stroke in female migraineurs of any age versus control subjects was 1.55 (95% CI, 1.16–2.07; P=0.003; statistical heterogeneity I2=42.5%; P=0.138). When considering female migraineurs aged less than 45 years the estimate was 1.57 (95% CI, 1.10–2.24; P=0.012; statistical heterogeneity 16.2%; P=0.303; Figure 4).
One of the included studies did not fulfill 3 of our predefined criteria (Tables I and II in the online-only Data Supplement).16 According to sensitivity analysis, the exclusion of that study from the analyses had a minimal influence on the overall pooled adjusted effect estimate (1.48; 95% CI, 1.15–1.90; P=0.002; I2=60.9%; P value for Q test=0.018). All included studies fell within 2 standard deviations of the z score in the Galbraith plot. We repeated the analyses excluding all the studies, one by one, and we did not find any change in the results (data not shown).
The funnel plot (Figure II in the online-only Data Supplement) of the 8 studies included in the meta-analysis revealed no significant asymmetry. Egger’s test for funnel plot asymmetry returned a nonsignificant result (P=0.512).
The meta-analysis of the 8 available studies on migraine and hemorrhagic stroke (4 case–control and 4 cohort studies), involving a total of 1600 hemorrhagic strokes, found an association between the 2 conditions. The risk of hemorrhagic stroke increased by 50% in subjects with any migraine compared with nonmigraineurs. The overall risk of hemorrhagic stroke was lower than that reported for ischemic stroke in other meta-analyses.1,3,24 The risk of hemorrhagic stroke was also increased when female migraineurs of any age and female migraineurs aged <45 years were compared with control subjects. However, this finding should be considered with caution because data of interest referring to males were reported only in 2 studies,16,21 and a direct comparison of the risk in females with the risk in males was not possible.
This meta-analysis could not prove the association of either MwA or MwoA with an increased risk of hemorrhagic stroke because only 2 cohort studies and 1 case–control study collected data on the risk of hemorrhagic stroke according to migraine type.18,21,22 Both cohort studies21,22 showed an association between MwA and hemorrhagic stroke, whereas only 1 of them showed an association between MwoA and hemorrhagic stroke.21 However, because MwA is the smaller subgroup and the effect size estimate was higher than that for MwoA, we cannot exclude the possibility that the meta-analysis had insufficient power to detect a possible association.
We did not perform any analysis by hemorrhagic stroke type (ICH, SAH) because the available studies referred either to SAH only17 or to ICH and SAH separately only in patients with MwA.22 In the former study,17 no association was found between migraine and SAH, whereas in the latter,22 a positive association was found between MwA and ICH but not between MwA and SAH. Kuo et al21 reported a higher proportion of SAH in the migraine group.
The mechanisms that might explain the association between migraine and hemorrhagic stroke are still unclear. Moreover, thorough investigations are lacking because only 2 of the included studies specifically addressed the topic of hemorrhagic stroke in migraineurs.21,22 There are several mechanisms that may be responsible. First, alterations of the vessel wall, supported by the presence of endothelial dysfunction in migraineurs, may favor hemorrhagic stroke.25,26 Second, comorbid vascular risk factors such as arterial hypertension or platelet dysfunction could link migraine to hemorrhagic stroke as to ischemic stroke.27,28 Third, the chronic use of nonsteroidal anti-inflammatory drugs (NSAIDs), which also have an antiplatelet action, may exert a confounding effect on the association between migraine and hemorrhagic stroke, although this possibility is unlikely because some studies adjusted the model for the confounding effect of NSAID use,16,22 and the possible increased hemorrhagic risk associated with NSAID use has recently been challenged.29,30 Finally, structural brain lesions such as arteriovenous malformations that can mimic migraine may represent another factor in the association between migraine and hemorrhagic stroke.31,32
The 8 studies included in this meta-analysis were of high quality and were at low risk of publication bias. However, analysis based on only 8 studies may limit the conclusiveness of the results. In addition, although the overall number of participants was large, the number of outcome events in subgroups was sometimes small, inviting caution in the interpretation of the data. In particular, the available results did not allow any conclusions to be drawn concerning migraine type or hemorrhagic stroke type. A further limitation is the lack of data stratified by age and sex, the availability of which would have allowed an exhaustive pooled analysis. The retrieved studies did not report essential data for conclusions to be drawn about the relationship between either duration of migraine or active versus past-history of migraine and possible hemorrhagic stroke occurrence. The striking differences in the duration of the follow-up (from 2 to 13.6 years) in the cohort studies may represent a further point of weakness in the search for unbiased conclusions. In fact, as shown by the Women’s Health Study, the association between migraine and hemorrhagic stroke emerged only with the longer follow-up.22 Because the number of patients with migraine who had a hemorrhagic stroke was low in most of the studies, no details were reported regarding size and site of the brain hemorrhage and event-related disability. All but 1 of the studies were carried out during the 1990s or earlier, and only 2 studies reported data on mortality.21,22 Another limitation is the inclusion only of studies in the English language.
In conclusion, although the results of the present meta-analysis show an overall increased risk of hemorrhagic stroke in subjects suffering from any migraine, it should be noted that the number of expected hemorrhagic strokes among patients with migraine is considerably small and that other factors may play a stronger role; in addition, the mechanisms underlying the association are only hypothetical, as are those linking migraine to ischemic stroke.31,32 Consequently, no alert should be given to migraineurs because no changes to their current standard treatments are needed. Indeed, to date, the best recommendation for physicians treating subjects with migraine is to continue to focus carefully on those factors that could increase their risk of vascular events.
We are grateful to Professor Giancarlo Logroscino for his valuable suggestions.
Sources of Funding
This study was funded by the “ex 60%” 2012 grant from the Italian Ministero dell’Istruzione, dell’Università e della Ricerca.
The online-only Data Supplement is available with this article at http://stroke.ahajournals.org/lookup/suppl/doi:10.1161/STROKEAHA.113.002465/-/DC1.
- Received June 11, 2013.
- Accepted July 25, 2013.
- © 2013 American Heart Association, Inc.
- Schürks M,
- Rist PM,
- Bigal ME,
- Buring JE,
- Lipton RB,
- Kurth T
- Schürks M,
- Rist PM,
- Shapiro RE,
- Kurth T
- Sacco S,
- Ricci S,
- Carolei A
- Tzourio C,
- Tehindrazanarivelo A,
- Iglésias S,
- Alpérovitch A,
- Chedru F,
- d’Anglejan-Chatillon J,
- et al
- Higgins JPT,
- Green S
- Egger M,
- Davey Smith G,
- Schneider M,
- Minder C
- Schwartzer G
- Viechtbauer W
- Chang CL,
- Donaghy M,
- Poulter N
- Kurth T,
- Kase CS,
- Schürks M,
- Tzourio C,
- Buring JE
- Schwartz SM,
- Petitti DB,
- Siscovick DS,
- Longstreth WT Jr.,
- Sidney S,
- Raghunathan TE,
- et al
- Etminan M,
- Takkouche B,
- Isorna FC,
- Samii A
- Bak S,
- Andersen M,
- Tsiropoulos I,
- García Rodríguez LA,
- Hallas J,
- Christensen K,
- et al
- Johnsen SP,
- Pedersen L,
- Friis S,
- Blot WJ,
- McLaughlin JK,
- Olsen JH,
- et al