Patent foramen ovale closure versus medical therapy for stroke prevention: A systematic review and meta-analysis of randomized controlled trials

Search strategy, inclusion and exclusion criteria

The meta-analysis was performed according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses statement (PRISMA; a completed checklist can be found in Supplementary File 1). PubMed and Cochrane Library were searched for randomized trials that compared the efficacy in stroke prevention of patent foramen ovale (PFO) closure with that of medical therapy. The following search terms were used for PubMed: "patent foramen ovale" AND (stroke OR embolism) and "randomized" AND "Trial". For Cochrane Library, the following terms were used: "patent foramen ovale" AND "closure" AND (stroke OR embolism). The search period was from the beginning of the databases through to 16^th September 2017, with no language restrictions.

The following inclusion criteria were applied: i) the design was a randomized trial in humans, ii) the study compared stroke outcomes for PFO closure versus medical therapy. Quality assessment of randomized controlled trials was performed using the Cochrane Risk Assessment Tool (Supplementary Figure 1 and Supplementary Figure 2).

Data extraction and statistical analysis

Data from the different studies were entered in Microsoft Excel (2016 Version). All publications extracted from the search strategy were assessed for compliance with the inclusion criteria. In this meta-analysis, the extracted data elements consisted of: i) trial name; ii) follow-up duration; iii) quality score; and iv) characteristics of the population, including sample size, sex and age. Two reviewers (GT and MG) independently reviewed each included study and disagreements were resolved by adjudication with input from a third reviewer (TL).

The number of events for: i) primary endpoint, ii) stroke, iii) transient ischemic attack, iv) all-cause bleeding complications, v) gastrointestinal complications [bleeding, ulceration, ulcer perforation], vi) short-term atrial fibrillation or flutter, viii) long-term atrial fibrillation or flutter, were identified and extracted independently by each reviewer from each trial. The Comprehensive Meta-Analysis Software (Version 2) was used for subsequent meta-analyses and statistical analyses. The event rates (events per patient-year) were used to calculate rate ratios for each study, which were pooled in subsequent meta-analyses.

Heterogeneity across studies was assessed using the I² statistic from the standard chi-square test, which describes the percentage of the variability in the effect estimates resulting from heterogeneity. I² > 50% was considered to reflect significant statistical heterogeneity and in such cases the random-effects model was used. To identify the origin of the heterogeneity, sensitivity analysis excluding one study at a time. Funnel plots showing standard errors against the logarithms of the odds ratio were constructed. Begg and Mazumdar rank correlation test was used to test for publication bias and the Egger’s test were used to detect publication bias.

PFO closure and primary endpoint(s), stroke and transient ischemic attack events

All five trials compared the primary endpoints in PFO closure versus medical therapy. The different trials used slightly different primary endpoints (Supplementary Table 2), as follows: 1) CLOSURE I trial: stroke, transient ischemic accident (TIA), 30-day mortality, neurology-related death, 2) PC trial: stroke, TIA, death, peripheral embolism, 3) CLOSE trial: fatal or non-fatal stroke, 4) RESPECT trial: nonfatal ischemic stroke, fatal ischemic stroke, or early death after randomization and 5) REDUCE trial: co-primary endpoints of i) ischemic stroke, and ii) new ischemic stroke or silent brain infarction. Among the 1829 subjects who underwent PFO closure, 70 (3.8%) met the primary endpoint (Supplementary Table 3). By contrast, 112 of the 1972 subjects receiving medical therapy (5.7%) met the primary endpoint. Our meta-analysis shows that PFO closure significantly reduced primary endpoint events when compared to medical therapy with a rate ratio [RR] of 0.60 (95% CI: 0.44-0.83, P < 0.0001; I²: 15%) (Figure 2, top panel). Using hazard ratios [HR] from the trials produced negligible differences from our event rate analyses (HR: 0.61; 95% CI: 0.45-0.83, P < 0.01; I²: 0%).

Figure 2. Rate ratios for primary endpoint(s) (top), strokes (middle) and transient ischemic attacks (TIAs; bottom) in patent foramen ovale (PFO) closure versus medical therapy.

Subgroup analyses were performed for the primary endpoint based on atrial septal aneurysm and shunt size by pooling hazard ratios from the subgroup analyses of the included studies. PFO closure was not significantly better than medical therapy for patients with an atrial septal aneurysm (HR: 0.46; 95% CI: 0.14-1.60, P = 0.22; I²: 62%, Supplementary Figure 15) or without an atrial septal aneurysm (HR: 0.74; 95% CI: 0.47-1.17, P = 0.19; I²: 0%, Supplementary Figure 16). By contrast, PFO closure significantly reduced primary endpoint events in patients with large shunt size (HR: 0.27, 95% CI: 0.14-0.54, P < 0.0001; I²: 0%, Supplementary Figure 17) but not in those with small shunt size (HR: 0.80, 95% CI: 0.49-1.31; P = 0.38; I²: 0%, Supplementary Figure 18)

All five trials compared the stroke events in PFO closure versus medical therapy. The different trials used slightly different primary endpoints, as follows: 1) CLOSURE I trial: acute focal neurological event that is MR imaging positive, regardless of duration of clinical symptoms, or if imaging cannot be performed for confirmation, it was defined as a persistent focal neurological deficit lasting longer than 24 hours; 2) PC trial: any neurologic deficit lasting for >24 hours typically with documentation in magnet resonance imaging (MRI) or computer tomography (CT); 3) CLOSE trial: sudden onset of focal neurological symptoms with the presence of cerebral infarction in the appropriate territory on brain imaging (CT or MRI), regardless of the duration of the symptoms (less than or greater than 24 hours); 4) RESPECT trial: ischemic stroke was defined as an acute focal neurologic deficit, which was presumed to be due to focal ischemia, and either symptoms that persisted for 24 hours or longer or symptoms that persisted for less than 24 hours but were associated with findings of a new, neuroanatomically relevant, cerebral infarct on MRI or CT; and 5) REDUCE trial: an acute focal neurologic deficit, presumably due to ischemia, that either resulted in clinical symptoms lasting 24 hours or more or was associated with evidence of relevant infarction on MRI or, if MRI could not be performed, CT of the brain. Taken together data from all five trials, stroke occurred in 45 patients (2.5%) in the PFO closure group, but in 102 patients (5.2%) in the medical therapy group. This gave a RR of 0.50 that was statistically significant (95% CI: 0.35-0.73, P < 0.0001; I²: 32%) (Figure 2, middle panel). Using HR from the trials produced negligible differences from our event rate analyses (HR: 0.49, 95% CI: 0.34-0.71, P < 0.01; I²: 30%).

TIAs were assessed in all five trials and the various definitions are shown in Supplementary Table 2. These occurred in 44 patients (2.4%) in the PFO closure group and in 67 patients (3.4%) of the medical therapy group (Supplementary Table 3). There was no statistically significant difference in the RR (0.75; 95% CI: 0.51-1.10, P = 0.14; I²: 0%) (Figure 2, bottom panel). Hazard ratios were available from four trials on TIAs, with a pooled HR of 0.73 (95% CI: 0.49-1.09, P = 0.13; I²: 0%).

Adverse events in PFO closure group compared to medical therapy group

Atrial fibrillation or flutter was detected in 76 patients in the PFO closure group (4.2%) and 37 patients (1.9%) in the medical therapy group (Supplementary Table 4). These equated to a significant increase in the risk when PFO closure was used (RR: 1.90, 95% CI: 1.23-2.93, P < 0.01; I²: 43%) (Figure 3, top panel). Subgroup analysis was performed for the type of atrial fibrillation or flutter by dividing the episodes into i) paroxysmal or minor, and ii) permanent or major [as defined by the individual trials]. This revealed that most of the episodes were only paroxysmal or minor for the PFO group (3.0%) when compared to the medical therapy group (0.6%) (RR: 7.70, 95% CI: 2.30-19.77; P < 0.0001; I²: 32%) (Figure 3, middle panel). Permanent or serious atrial fibrillation or flutter occurred in 1.3% in the PFO closure group compared to 0.4% in the medical therapy group without significance between these groups (RR: 2.19, 95% CI: 0.94-5.01; P = 0.07; I²: 0%) (Figure 3, bottom panel).

Figure 3. Rate ratios for atrial fibrillation or flutter (all types: top; paroxysmal or minor: middle; permanent or major: bottom) in patent foramen ovale (PFO) closure versus medical therapy.

All bleeding complications were counted from the included studies (Supplementary Table 5). These were comparable between the groups, occurring in 39 (2.1%) of the PFO closure group and 47 (2.4%) in the medical therapy group, with no significant difference between them (RR: 0.89; 95% CI: 0.44-1.78, P = 0.74; I²: 51%) (Figure 4, top panel). Three trials reported gastrointestinal complications of hemorrhage, ulceration or ulcer perforation (Supplementary Table 5), which occurred in 7 and 8 patients in the PFO closure and medical therapy groups (0.4% for both arms). Therefore, the risk of gastrointestinal complications was not reduced by PFO closure (RR: 0.92; 95% CI: 0.32-2.70, P = 0.88; I²: 0%) (Figure 4, bottom panel).

Figure 4. Rate ratios for bleeding (top) and gastrointestinal complications (bottom) in patent foramen ovale (PFO) closure versus medical therapy.

Strengths and limitations of the included trials and this meta-analysis

However, there are some limitations inherent to the trials themselves. Firstly, the crossover rate is substantial. For example, in REDUCE 6.3% of study subjects crossed over from PFO-closure to medical treatment and 6.3% did the opposite. Secondly, actual mechanical and anatomical closure of the PFO is of paramount importance to prevent recurrent paradoxical embolization. The different trials had different degrees of successful closure (e.g. REDUCE trial: 75%; CLOSURE trial: 86%; CLOSE trial: 93%) but these were not related to outcomes. Secondly, the definitions of TIA were different between the trials, which could have contributed to the between-study heterogeneity. Finally, the trials had large proportions of study subjects who were lost to follow-up, withdrew consent and crossed over to the other study arm, leading to uncertainty in the reported event rates.

There are many strengths of this meta-analysis study. It is the largest meta-analysis of randomized trials to date, including more than 3800 participants from five trials. Moreover, the follow-up duration was sufficiently long for events to be detected. Our quality analysis indicated that the studies had a low risk of bias. Low levels of heterogeneity were observed for most of our analyses, including primary endpoints, stroke, transient ischemic attacks and gastrointestinal complications. These indicate that it was appropriate to pool these studies together. However, several limitations inherent in the present meta-analysis should be noted. Firstly, the meta-analysis of all-cause bleeding across the trials showed a high level of heterogeneity, which may be clinical in nature, especially when different types of bleeding were measured. Secondly, there is an imbalance between events and lost-to-follow-up/withdrawals²⁸. In most trials, the ratio between the two is often in the region of 5 to 1, but is around 0.3 to 1 in the trials included in this meta-analysis. Moreover, publication bias results should be interpreted with caution as publication bias with less than ten studies is not recommended, but the results are presented here for the sake of completeness. Furthermore, there was a moderate degree of heterogeneity for the atrial septal aneurysm vs. no aneurysm comparison. Since our aim was to compare the effectiveness of PFO closure to medical treatment, we did not perform additional analysis by medical group assignment, such as comparing antiplatelet or anticoagulant therapy to PFO closure. The majority of medical management patients were treated with antiplatelet medications, with a minority treated with oral anticoagulation. Since the presumed mechanism of stroke attributable to PFO is paradoxical embolus from venous thrombi, or in situ thrombus formation, oral anticoagulants may be more effective for those causes. Where reported in individual trials, stroke outcome events were numerically less when medical management was with oral anticoagulation. Further analyses are needed to compare PFO closure with anticoagulation treatment alone as this was beyond the scope of the current study.