The Short Version

For about twenty years, the story of migraine and the patent foramen ovale (PFO) has been a frustrating one. Observational data kept showing that people with migraine, especially migraine with aura, were more likely to have a PFO. Then every randomized trial of closing the PFO with a device failed to hit its primary endpoint. The field mostly moved on, and headache specialists learned to say "no, closing it probably will not fix your attacks."

A trial published in The BMJ on 29 July 2026 reopens the question from a different angle. Instead of plugging the hole, investigators across 39 hospitals in China randomized 1,000 adults who had both migraine and a confirmed PFO to one of four daily medications for 12 weeks: aspirin, clopidogrel, rivaroxaban, or metoprolol. The three antithrombotic drugs were all non-inferior to metoprolol, a standard beta-blocker preventive. Rivaroxaban went further and beat it, with a responder rate of 78.4% versus 61.8%, an absolute difference of 16.2% (98.33% CI 6.0 to 26.4, P<0.001). No major bleeding events occurred.

What a PFO Actually Is

Before birth, blood does not need to pass through the lungs to pick up oxygen, because the placenta handles that. So the fetal heart has a flap-valve opening between the right and left atria called the foramen ovale, which lets blood skip the pulmonary circuit entirely. At birth, pressures shift when the lungs inflate and the flap normally seals shut within the first year of life.

In roughly one in four adults, it never fully seals. That leftover flap is a patent foramen ovale. Most of the time it stays functionally closed because left atrial pressure exceeds right atrial pressure. But during any maneuver that raises right-sided pressure, such as coughing, straining, bending over, or the release phase of a Valsalva, the flap can pop open briefly and let a small amount of venous blood cross directly into the arterial circulation. That is a right-to-left shunt.

Why a Hole in the Heart Might Trigger Attacks

There are two competing mechanistic stories, and this trial is interesting partly because it discriminates between them a little.

It is worth naming the alternative explanation honestly: aspirin and rivaroxaban also have effects that have nothing to do with shunts. Aspirin has been used as a migraine treatment on its own merits. Any of these agents might reduce migraine burden in people without a PFO too, and this trial cannot tell us that because everyone enrolled had one.

How the Trial Was Run

This was an investigator-initiated, multicentre, prospective, randomized, active-controlled, open-label trial with blinded outcome assessment and hierarchical hypothesis testing, run across 39 secondary and tertiary care hospitals in China.

Two design choices deserve credit. First, the 12-week prospective screening period before randomization means baseline migraine days were recorded in real time rather than reconstructed from memory, which reduces the recall inflation that plagues headache trials. Second, participants with previous stroke, transient ischemic attack, intracranial hemorrhage, a non-PFO right-to-left shunt, or contraindications to the study drugs were excluded, so this is a reasonably clean population in which the shunt is the only obvious cardiac culprit.

The primary outcome was the proportion of participants achieving at least a 50% reduction in monthly migraine days or attacks from baseline to weeks 9 through 12. That is a conventional and clinically meaningful responder definition in preventive migraine research.

What It Found

All three antithrombotic agents were non-inferior to metoprolol on responder rate. Rivaroxaban then cleared the higher bar of statistical superiority.

Secondary endpoints moved in the same direction. Compared with metoprolol, rivaroxaban produced greater reductions in migraine days and attacks, higher rates of complete migraine cessation, and greater improvement in migraine-specific quality-of-life scores. When the primary and the secondaries all point the same way, that internal consistency is reassuring, though it does not correct for a shared source of bias like unblinding.

On safety, the trial reported no major bleeding events across any arm over 12 weeks. That is a genuinely good result, and it is also exactly what you would expect from 12 weeks of exposure in a young-to-middle-aged population with no prior stroke or hemorrhage. Twelve weeks is not long enough to characterize the bleeding risk of an anticoagulant.

Regimens and Evidence Grades

What Was Tested, and How Strong the Evidence Is

The four regimens from this trial alongside the established preventive options they would compete with. Doses shown for the trial arms are exactly what was used in the protocol. Nothing here is a prescription, and the anticoagulant rows in particular require an individualized bleeding-risk conversation with a physician.

Why the Closure Trials Failed, and Why That Matters Here

To read this new trial fairly, you need the history it is arguing with. Three randomized trials tested closing the PFO with a device in people with migraine, and all three missed.

So the closure literature is not a flat null. It is a pattern of missed primary endpoints with scattered positive secondaries, which is the classic signature of either a small real effect in a subgroup or a set of underpowered trials chasing noise. The honest read is that device closure does not reliably help migraine, and no guideline recommends it for that indication.

What the BMJ trial changes is the target. If the mechanism is microemboli crossing the shunt rather than the shunt itself, then a drug that reduces clot formation could work where a mechanical plug did not, particularly since a device takes months to fully endothelialize and can itself be a nidus for thrombus in the meantime. That is a coherent argument. It is not proof.

"Three randomized trials told us that plugging the hole does not fix migraine. This trial asks a smarter question: what if the hole was never the problem, and what crosses through it is? The answer looks promising, but a 12-week open-label trial is where that conversation starts, not where it ends." - Cerebral Torque

Who This Applies To

The enrollment criteria define the ceiling of what can be claimed, and they are narrower than the headline suggests.

If you have migraine and have never had an echocardiogram, this study says nothing about you. Roughly three quarters of adults do not have a PFO, and there is no reason from this trial to think an anticoagulant would help them. It also says nothing about people over 64, where bleeding risk climbs steeply, or about anyone with a prior stroke or bleed, because they were deliberately kept out.

Safety and Contraindications

The trial reported no major bleeding events, which is genuinely good news, but the drugs involved carry real risk profiles that a 12-week trial in a selected population will not surface. These are worth stating plainly.

Limitations Worth Taking Seriously

Every trial has limitations sections that nobody reads. This one has limitations that change how you should interpret the result, so they are worth spelling out.

The open-label design deserves the most weight. Participants knew whether they were taking a beta-blocker or a novel anticoagulant, and headache frequency is recorded by the participant in a diary. Outcome assessment was blinded, which helps at the adjudication stage, but it cannot fix the fact that the raw data came from people who knew their assignment. Migraine trials have famously large placebo responses, often 20 to 40% responder rates, and knowing you are on the interesting new drug reliably inflates reported benefit. A 16-point difference is large enough that it probably is not entirely expectation, but it is not possible to say how much of it survives blinding.

The absence of a placebo arm compounds this. All four responder rates were high, between 61.7% and 78.4%, which is well above what placebo arms typically produce in preventive migraine trials. That could mean all four drugs work. It could also mean this population, selected for a 12-week prospective run-in and randomized to four active treatments with no placebo, regressed toward the mean and reported broad improvement. There is no arm in the trial that can distinguish those.

Twelve weeks is short for a preventive migraine trial and very short for an anticoagulant. It answers whether the effect appears; it does not answer whether it persists, whether it wanes, or what happens to bleeding rates at one year, five years, or in someone who stays on rivaroxaban indefinitely. Zero major bleeds in 12 weeks in adults under 65 with no prior stroke is close to the expected result even for a drug that will eventually cause bleeding at a measurable rate.

Finally, the metoprolol comparator at 25 mg twice daily is at the low end of therapeutic. A trial that beat a fully titrated beta-blocker would be more convincing than one that beat a modest fixed dose.

Practical Guidance

Here is how this reasonably fits into a real treatment conversation right now.

For clinicians, the pragmatic reading is this: nothing here changes first-line prescribing. What it does is make antithrombotic therapy a defensible option to discuss in the specific subgroup of PFO-positive patients with frequent attacks who have exhausted conventional preventives, which is precisely the group that used to be sent for closure and sent home disappointed. It also makes a strong case for someone to run the confirmatory trial: double-blind, placebo-controlled, longer follow-up, with a fully titrated comparator.

Bottom Line

A 1,000-person randomized trial found that aspirin, clopidogrel and rivaroxaban were all non-inferior to metoprolol for preventing migraine attacks in adults with a patent foramen ovale, and that rivaroxaban was superior, with a 78.4% responder rate versus 61.8% and no major bleeding over 12 weeks.

The PFO and migraine question has generated more disappointment than progress over two decades. This trial is the most substantial positive signal the field has produced, and it is positive in a direction nobody had properly tested. That is worth paying attention to. It is also open label, uncontrolled against placebo, and three months long, which is exactly the profile of a result that sometimes shrinks when someone runs it properly. So, just be weary.