Introduction

Due to the advancing age of the population and disease-promoting lifestyle factors, the incidence of stroke, the second most common cause of death globally surpassed only by coronary heart disease, is increasing. Except for several well-established risk factors, obesity, smoking, and physical inactivity [1], are also thought to be associated with the risk of stroke. Migraine, one of the most common and disabling primary headache disorders, affects 10–20% of the population during the most productive periods of their working lives [2]. Recently, some original epidemiological studies, especially case–control studies and cohort studies, have indicated a possible association between migraine and stroke risk [3], but the direction and magnitude of the findings are inconsistent.

In 2010, a meta-analysis of epidemiological studies [4] was conducted to evaluate the association between migraine and ischemic stroke, but it did not, respectively, provide pooled estimates of association of interest from case–control studies and cohort studies. Furthermore, two of the included eight cohort studies only provided data about total stroke but not ischemic stroke, and the authors regarded mistakenly these as data about ischemic stroke [5, 6]. In 2013, another systematic review identified the association between migraine and hemorrhagic stroke [7], but it only included four case–control studies and four cohort studies. Up to now, although some available data existed [5, 6, 8–13], the relationship between migraine and risk of total stroke has not been summarized quantitatively. Given that migraine is a primary headache disorder and stroke can cause substantial burden of disease, an improved understanding of their relationship should have important public health and clinical implications for stroke.

Taking into consideration of the inconsistent conclusions of existing epidemiological studies, the higher level of evidence from prospective cohort studies, and defect of previous systematic reviews, we conducted an updated meta-analysis of ten prospective cohort studies to evaluate the association between migraine and the risk of ischemic, hemorrhagic, and total stroke.

Methods

Ethical approval is not required for this review.

Search Strategy

We followed the meta-analysis of observational studies in epidemiology guidelines (MOOSE) [14] and the preferred reporting items for systematic review and meta-analyses (PRISMA) [15] to report our review. A systematic literature search of PubMed and Embase was conducted from these databases’ inception through July 2016 using the following search terms with no restrictions: “migraine” and “headache” in combination with “cerebrovascular”, “stroke”, and “cerebral infarction”, “brain infarction”, and “hemorrhage”. Reference lists of the retrieved articles were also reviewed.

Study selection

We included studies which met the following criteria: (1) it had a prospective cohort study design; (2) the exposure of interest, migraine, was reported with a clear definition of the diagnostic criteria; (3) the endpoint of interest was incidence of stroke, whether it was ischemic, hemorrhagic, or both; and (4) the relative risk (RR) or Hazard Ratio (HR) and the corresponding 95% confidence interval (CI) of total or specific stroke relating to the experience of migraine were reported or could be calculated from the data provided. Studies which were retrospective cohort were excluded. We also excluded studies in which the exposure of interest was not migraine but other type of headache, or the endpoint of interest was the death of stroke. If studies investigating the relationship between migraine and the risk of stroke did not reported required data and the required data could not be calculated from the data provided. If two or more published papers reported the results from one study cohort, we included only the one with the most detailed information or with the longest follow-up period.

Data extraction

Data extraction was carried out independently by two investigators (YZ and HZ). The following information was extracted from the included studies: first author, publication year, country, study period, number of cases, size of cohort, RR or adjusted RR (if possible) with 95% CI, subtype of stroke, and adjusted factors. Discrepancies were resolved by discussion with a third investigator (CP).

Quality assessment

The quality assessment of all included studies was performed according to the Newcastle-Ottawa quality assessment scale [16], which is a validated scale for non-randomized studies in meta-analyses. This scale awards a maximum of nine points to each study: four for selection of participants and measurement of exposure, two for comparability of cohorts on the basis of the design or analysis, and three for assessment of outcomes and adequacy of follow-up. Scores of 0–3, 3.5–6, and 6.5–9 were, respectively, regarded as low, moderate, and high quality of studies.

Statistical analyses

With regard to the prospective cohort design, RR was used to measure the association of interest. Since the incidence of stroke is relatively rare in general populations, we ignored the distinction between RR and HR, and equally regarded them as RR [17]. In any included study, when RRs were reported separately for subgroups by the different types of migraine (with aura versus without aura), we combined the results of the subgroups and calculated a common RR for the main analysis using a fixed-effects model. We calculated an overall pooled RR using a random-effects model for the main analysis [18]. Heterogeneity was tested by Q statistic with a significance level at p < 0.10 and I 2 statistic [19]. The I 2 statistic measures the percentage of total variation across studies due to heterogeneity rather than chance. The negative value of I 2 is set at zero, and I 2 varies from 0% (no observed heterogeneity) to 100% (maximal heterogeneity). An I 2 value of ≥50% is considered to represent substantial heterogeneity. All analyses were conducted using STATA statistical software (version 12.0; College Station, Texas, United States). p < 0.05 was considered statistically significant, except where otherwise specified. All statistical tests were two-sided.

Results

Literature search

The process of study identification and inclusion was shown in Fig. 1. We initially retrieved 9124 articles from PubMed and 10,046 articles from Embase. After 5593 duplicates were excluded, 13,577 citations were screened through titles and abstracts, of which 13,516 were excluded mainly because they were not original observational studies or not relevant to migraine and stroke. After full-text review of the remaining 61 articles, 50 articles were excluded for various reasons. Finally, eleven studies [5, 6, 8–13, 20–22] were included.

Fig. 1
figure 1

Flow chart of identification of relevant prospective cohort studies of migraine in relation to stroke

Full size image

Characteristics and quality of the included studies

We summarized the main characteristics of the ten included studies in Table 1. These studies were published between 1995 and 2016. Of them, ten were conducted in United States and one in United Kingdom. The size of the cohorts ranged from 3298 to 1,566,952 (total 2,221,888) and totally involved 3371 patients with stroke. Except three studies which did not reported relevant data, follow-up duration ranged from 17 to 188 months. There were two studies in which only age and gender were adjusted for statistical analyses, and eight studies controlled for a group of conventional risk factors for stroke, including smoking, hypertension, body mass index, oral contraceptive use, and so on. Another study did not report the adjusted variables. Study quality was moderate in four studies and high in other seven studies.

Table 1 Main characteristics of the included cohort studies
Full size table

Main pooled results of meta-analysis

Migraine and risk of total stroke

Figure 2 showed the results from the random-effects model combining the risk estimates of total stroke related to migraine. Of the eight studies reporting the association between migraine and risk of total stroke, five showed a significantly positive association of interest, and three suggested no statistically significant relationship. Overall, compared with individuals who did not experience migraine, the pooled RR of total stroke was 1.55 (95% CI 1.38–1.75) for migraineurs. No significant heterogeneity was observed (p = 0.206, I 2 = 27.8%).

Fig. 2
figure 2

Association between migraine and the risk of total stroke

Full size image

Migraine and risk of hemorrhagic stroke

Of the eleven included studies, four reported the risk estimates of hemorrhagic stroke related to migraine. Combining these four results, we found that migraine was not associated with an increasing risk of hemorrhagic stroke (RR 1.15; 95% CI 0.85–1.56). No heterogeneity was observed (p = 0.547, I 2 < 0.1%) (see Fig. 3).

Fig. 3
figure 3

Association between migraine and the risk of hemorrhagic stroke

Full size image

Migraine and risk of ischemic stroke

Two of the six studies reporting the relationship between migraine and risk of ischemic stroke reported a significantly positive association of interest, and the other four suggested no statistically significant relationship. The pooled RR of ischemic stroke was 1.64 (95% CI 1.22–2.20) for migraineurs, compared with individuals who did not experience migraine, with observable moderate heterogeneity (p = 0.048, I 2 = 55.3%) (see Fig. 4).

Fig. 4
figure 4

Association between migraine and the risk of ischemic stroke

Full size image

Migraine with aura or without aura and stroke

Of the eleven included studies, two reported the association between migraine with or without aura and ischemic stroke [10, 20]. Synthesizing these results, we found that migraine with aura was associated with ischemic stroke (RR 2.14; 95% CI 1.33–3.43) but migraine without aura was not related to ischemic stroke (RR 1.02; 95% CI 0.68–1.51). Only one study reported the association between migraine with or without aura and hemorrhagic stroke [22]. Similarly, migraine with aura was also found to be positively related to the risk of hemorrhagic stroke but migraine without aura was not associated with the increase of the risk of hemorrhagic stroke. The pooled RRs were 2.25 (95% CI 1.11–4.54) and 0.49 (95% CI 0.14–1.70), respectively.

Discussion

Recently, the association between migraine and the risk of cardiovascular cerebrovascular diseases has been of increasing interest to cardiologists and neurological physicians [23, 24], and some studies were published to investigate the potential mechanisms, although the findings were inconclusive. It was suggested that migraine might increase the risk of ischemic stroke by platelet activation [25], vasospasm-induced cerebrovascular hypoperfusion [26], and increased platelet aggregation [27], and so on. Several potential mechanisms explaining the association of migraine with the risk of hemorrhagic stroke were following. First, alterations of the vessel wall in migraineurs may facilitate hemorrhagic stroke [28]. Second, vascular risk factors such as hypertension and platelet dysfunction might link migraine to hemorrhagic stroke [29]. Third, some studies suggested that endothelial dysfunction may also be a possible link between migraine and stroke [30, 31]. In addition, structural brain lesions that can mimic migraine could represent another factor in the relationship between migraine and the risk of hemorrhagic stroke [32].

Though the mechanisms that how migraine impacts stroke risk have not been studied thoroughly, some epidemiological studies, including large prospective cohort studies investigated their association. Our meta-analysis of eleven cohort studies involving 2,221,888 participants showed that, compared with individuals who never experience migraine, migraineurs had significantly increased risk of more than 50% for total stroke and ischemic stroke. But there was no statistically significant relationship between migraine and the risk of hemorrhagic stroke. We also analyzed the association between migraine with or without aura and stroke. Due to the number of studies in which related outcomes were reported was small, we suggested the result was not conclusive. Our result was partly different from the result of a previous meta-analyses [4, 7]. The previous meta-analysis including case–control and cohort studies suggested the risk of ischemic stroke (OR 2.30; 95% CI 1.91–2.76) for migraineur was obviously higher than the risk of hemorrhagic stroke (OR 1.48; 95% CI 1.16–1.88). The different results between our meta-analysis and the previous meta-analysis may be due to the different types and numbers of original studies included. A cohort study gives stronger evidence than a case–control study, and a retrospective cohort design may suffer more confounding factors and biases than a prospective one.

There are several strengths in our study. First, we adequately utilized the reported data of the included studies and provided the relationship between migraine and the risk of total stroke, hemorrhagic stroke, and ischemic stroke. When two RRs were reported separately in terms of the different type of stroke for a single study, we combined the results of them and calculated a common RR by a fixed-effects model for total stroke. Second, big sample sizes of included studies favor the representativeness of our results. Third, included prospective cohort studies with high quality indicated our findings were reliable and robust.

Some limitations in the present meta-analysis should be of concern. First, adjusted confounders varied among the ten included studies and there were two studies only controlling for age and gender. Some possibly important residual confounders such as obesity, smoking, and diabetes were not adjusted in some studies. These different confounders used in the adjusted models of the included studies should be a source of methodological heterogeneity. Second, all the included studies were only from two countries (United States, United Kingdom), although we conducted the literature search systematically and comprehensively as far as possible. Therefore, it should be cautious to extend our results to other countries and relevant studies conducted in other countries were required in future research. Third, taking into consideration the relatively smaller number of studies reporting estimates of association between migraine and risk of total stroke, hemorrhagic stroke, and ischemic stroke, respectively, we did not conduct subgroup analysis and detect publication bias.

In summary, our meta-analysis of prospective cohort studies with the most up-to-date evidence suggests that migraine is associated with a significantly increased risk of total stroke and ischemic stroke. Hemorrhagic stroke was not found to be related to migraine.