Two investigators attained a consensus on the search strategy and inclusion criteria and independently searched the Pubmed database for relevant articles published between January 2000 and July 2015 using the following search terms: “patent foramen ovale” AND “migraine disorders” OR “migraine” AND “clinical trials” (as topic) AND “humans” (not animals). We only included articles published in English. We also undertook an additional manual search of secondary sources. The Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) checklist was followed in the current study (Fig. 1). Moreover, a list of Additional file 1 for the PRISMA flow diagram is shown in Table 1.

Citation abstracts were first screened by two independent reviewers, and complete manuscripts were retrieved if deemed potentially pertinent. The two reviewers independently appraised the identified articles according to the above-mentioned selection criteria, with consensus resolution in cases of disagreement. The inclusion criteria were as follows: (i) observational studies that examined the effect of PFO closure on migraine; (ii) distinction of MA and MwoA as defined by the criteria of the International Headache Society; (iii) PFO detected either by transthoracic echocardiography with peripheral injection (cTTE), transoesophageal echocardiography with peripheral injection (cTEE), or transcranial Doppler ultrasonography with injection (cTCD); (iv) participants were 18 to 60 years of age; (v) a minimum of 10 patients in each group; and (iv) mean duration of follow-up of at least 6 months. The main exclusion criteria were as follows: (i) headache plausibly caused by conditions other than PFO, and (ii) conference abstracts or published data uninformative about patient outcomes. For outcome scoring, the primary efficacy end-point was the cure of migraine or at least 50% improvement in the severity of migraine symptoms. The secondary efficacy end-point represented any difference between basal and final scores in tests including the Migraine Disability Assessment Test (MIDAS) or the Headache Impact Test-6 (HIT-6).

The two independent reviewers judged the quality of each included study using the Grades of Recommendation Assessment Development and Evaluation (GRADE) assessment system [14]. The two independent reviewers prepared a formal and un-blinded abstract of each study on pre-specified forms, and resolved any important differences of opinion by consensus agreement.

We selected the odds ratio (OR) as the parameter for summarizing each study. Heterogeneity was evaluated using the X² test and I² statistics. Briefly, fixed-effect methods were used when P > 0.1 and I² ≤ 50%. When P < 0.1 and I² > 50%, we first identified the origin of heterogeneity, and then carried out subgroup analysis focusing on the attribution of particular factors causing the inconsistency. If there were statistical inconsistencies in the absence of clinical inconsistency, we used a random-effects model. Finally, we calculated the summary ORs and 95% confidence intervals (CIs). In addition, we prepared a funnel plot to depict the possibility of publication bias. We reported two-tailed P values throughout, using a 0.05 threshold for hypothesis testing, when applicable.

The search strategy initially yielded 315 articles. After screening the titles and abstracts, 21 articles were further investigated [6, 14‐32], of which, eight different articles comprising 546 patients were thoroughly reviewed [18, 21‐27].

Most participants had been referred for secondary preventive surgery after presumed paradoxical embolism attributed to PFO. However, the participants included in Rigatelli et al. [18] and Azarbal et al. [21] had no previous history of cryptogenic stroke or transient ischemic attack. PFO was evaluated by cTEE in four of the selected studies [22, 25‐27], by cTCD in one [21], and by cTCD or cTEE in the remaining three studies [30, 33, 34]. In three studies [18, 21, 26], the frequency and severity of migraine were assessed by the MIDAS questionnaire and score. In the remaining studies [22‐25, 27], the authors created their own questionnaires to record the severity and incidence of headache.

The mean age of the included patients ranged from 39 ± 6 years to 53 ± 11 years. Sixty-nine percent of the participants suffered from MA, while the remainder had MwoA. Seven of the eight studies were retrospective, while the remaining study [18] was prospective. The detailed demographic characteristics of the included studies are reported in Table 1.

The effect of PFO closure on migraine was studied in 546 patients (379 MA and 167 MwoA). At follow-up, migraine improved following PFO closure in 306 (81%) patients with MA and in 105 (63%) patients with MwoA. The estimated effect of PFO closure was reflected by a summary OR of 2.5 [95% CI, 1.09–5.73]. In accordance with the evident statistical heterogeneity [I² = 67%, P = 0.003], we carried out subgroup analysis after excluding three ambiguous studies because of their measuring methods, occurrence of different conditions at baseline, inconsistency in the reported length of treatment, and other factors [21, 22, 26]. In the absence of clinical inconsistency, we used a random-effects meta-analytical approach to combine the results of the individual studies. The overall difference in therapeutic efficacy between the MA and MwoA groups was statistically significant [Z = 2.16, P = 0.03].

Figure 2 depicts a funnel plot for the eight studies of percutaneous PFO closure used to treat migraine. The funnel plot demonstrates asymmetry, which suggests possible publication bias.

Several limitations in the present meta-analysis are worth mentioning when drawing conclusions regarding the benefits of PFO in MA patients. First, most of the included studies were retrospective, suggesting that a recall bias cannot be excluded. Second, the post-surgical therapy and protocol for assessing the outcomes differed among studies. Third, as noted above, the surgical procedures employed several different devices. Finally, based on the contact with the corresponding authors, the baseline data on sex and age were not recorded in three of the included studies [21, 22, 26].