PFO Closure Selection
Definition
PFO closure selection is the process of deciding, in a patient with an ischaemic stroke and a patent foramen ovale, whether the PFO caused the stroke and therefore whether closing it will prevent the next one. It is a Bayesian inference, not a diagnostic finding: the same structure is pathogenic in one patient and incidental in the next, and the instruments that separate them — RoPE and PASCAL — convert clinical and anatomical variables into a probability of causality that predicts the size of the treatment effect.
Key Concepts
The central arithmetic
- The pooled treatment effect comes from an individual-patient-data meta-analysis of all six trials: 3,740 patients, 121 events, median follow-up 57 months. Annualised stroke 0.47% with closure versus 1.09% with medical therapy, adjusted HR 0.41 (0.28–0.60), 2-year ARR 1.72%. (sources/pfo-jama-2021, rating: very high)
- A 59% relative reduction on a ~1% annual risk is the number that determines everything about how this procedure should be offered. Closure is preventive, expensive, durable, and prevents about half a stroke per 100 patient-years.
- The same analysis separates two outcomes that are usually conflated. Closure reduced PFO-associated recurrent stroke — adjudicated as not attributable to another mechanism — by 76% (0.24 vs 0.90 per 100 person-years, HR 0.24, 0.14–0.43, ARR 2.21%), but all ischaemic recurrence by only 59%. The gap is the proportion of recurrences that closing a hole could never have prevented. (sources/pfo-jama-2021)
- The authors calibrate their own result: an NNT of about 50 at two years is "comparable to the 3 mainstays of medical therapy for secondary stroke prevention" — antihypertensives, antiplatelets and statins. Closure is an ordinary-size preventive intervention delivered by an extraordinary route.
RoPE — the causal probability
| Characteristic |
Points |
| No hypertension |
1 |
| No diabetes |
1 |
| No prior stroke or TIA |
1 |
| Non-smoker |
1 |
| Cortical infarct on imaging |
1 |
| Age 18–29 / 30–39 / 40–49 / 50–59 / 60–69 / ≥70 |
5 / 4 / 3 / 2 / 1 / 0 |
- Derived from 8 databases and 3,023 patients. It estimates the probability that a PFO is pathogenic rather than incidental; it is not a recurrence-risk score, and treating it as one is the commonest error made with it (sources/pfo-jama-2025, rating: high).
- PFO prevalence by strata: RoPE <3 → 23% (the general-population rate; attributable fraction ≈ 0); RoPE 9–10 → 77% (attributable fraction ≈ 90%). Validated in ASTRAL: 23% at RoPE ≤3 rising to 72% at RoPE 9–10.
- RoPE does stratify the treatment effect, but not usefully: HR 0.61 (0.37–1.00) below 7 versus 0.21 (0.11–0.42) at 7 or above, interaction P=.02. Both strata benefit — 2-year ARR 1.1% and 2.1%, NNT 91 and 49 — which is why the authors regard the interaction as mechanistically informative and clinically unhelpful. A test can be significant on the relative scale and useless on the absolute one. (sources/pfo-jama-2021, rating: very high)
- RoPE omits anatomy entirely — no shunt size, no aneurysm. That gap is why PASCAL exists.
PASCAL — adding anatomy to causality
| High RoPE (≥7) |
High-risk anatomy (large shunt and/or ASA) |
Category |
HR (95% CI) |
2-year ARR |
NNT |
| No |
No |
Unlikely |
1.14 (0.53–2.46) |
−0.7% (−4.0 to 2.6) |
— |
| No |
Yes |
Possible |
0.38 (0.22–0.65) |
2.1% (0.6–3.6) |
47 |
| Yes |
No |
Possible |
0.38 (0.22–0.65) |
2.1% (0.6–3.6) |
47 |
| Yes |
Yes |
Probable |
0.10 (0.03–0.35) |
2.1% (0.9–3.4) |
47 |
- Large shunt = >20 bubbles in the left atrium; aneurysm = ≥10 mm excursion from midline. The definitions were not identical across trials — the threshold was >25 bubbles in CLOSURE and >30 in CLOSE — so the two anatomical variables that drive PASCAL were harmonised by judgement. (sources/pfo-jama-2021, rating: very high)
- "Unlikely" is about 15% of the trial population, and it is the only stratum where the point estimate crosses 1 and the absolute difference is negative: 2-year stroke risk 4.1% with the device against 3.4% with medical therapy, ARR −0.7% (−4.0 to 2.6). That null — not the gradient between "possible" and "probable" — is what makes PASCAL clinically useful where RoPE alone was not.
- The "possible" and "probable" categories are indistinguishable on the absolute scale — 2.1% ARR and NNT 47 in both. The system functions in practice as a two-level decision: unlikely, or worth closing.
- Both features together, disregarding RoPE: 5.9% recurrence on medical therapy versus 0.4% with closure at 2 years, ARR 5.5% (2.7–8.3), rising to 7.1% at 5 years, NNT 14 (9–27). (sources/pfo-jama-2021)
- PASCAL was derived in this dataset and has never been validated outside it. ESO 2024 rates the evidence down by one level for exactly that reason (sources/pfo-eso-2024).
- The extended PASCAL classification — the part covering thrombus straddling the PFO and concomitant DVT — could not be evaluated at all. Those patients "were not identified in the study data." This matters, because those are the categories used to justify closing an "unlikely" patient (see Contradictions).
What the trials actually showed
| Trial |
n |
Enrolment |
Comparator |
Result |
| CLOSURE I (2012) |
909 |
18–60, TEE-verified PFO |
APT or OAC |
Negative (STARFlex device, later withdrawn) |
| PC Trial (2013) |
414 |
<60 |
APT or OAC |
Negative |
| RESPECT (2013, extended 2017) |
980 |
18–60 |
APT or OAC |
Neutral at 2.1 y; positive at 5.9 y (3.6% vs 5.8%, HR 0.55, P=.046) |
| CLOSE (2017) |
663 |
16–60, large shunt or ASA |
APT or OAC |
Positive (HR 0.03 vs APT) |
| Gore REDUCE (2017) |
664 |
18–59 |
APT only |
Positive (HR 0.23) |
| DEFENSE-PFO (2018) |
120 |
18–80, high-risk anatomy |
APT or OAC |
Positive (0 vs 6 events, NNT 10) |
- The three negative trials and the four positive ones differ in design, not in biology: follow-up duration (RESPECT), comparator (REDUCE used antiplatelet only), anatomical selection (CLOSE, DEFENSE-PFO), and device (CLOSURE I used a device withdrawn from the market).
- Restricting to the two trials with antiplatelet-only comparators (REDUCE, CLOSE), closure gave an HR of 0.18 (0.05–0.63) — but most of those patients had high-risk anatomy (sources/pfo-eso-2024).
The two guidelines, side by side
| Question |
SCAI 2022 |
ESO 2024 |
| 18–60 with PFO-associated stroke |
Strong for closure vs APT alone, moderate certainty |
Strong for in selected patients, high quality |
| Does anatomy gate it? |
No — recommendation independent of anatomy |
Yes — weak against in PASCAL "unlikely"; large shunt + ASA benefit most |
| ≥60 |
Conditional for (very low) |
No evidence-based recommendation |
| <18 |
Out of scope; discouraged |
No evidence-based recommendation; consensus for 13–17 by anatomy |
| Closure vs anticoagulation |
Conditional for closure + APT (low) |
Weak for closure + APT (low) |
| Timing |
No recommendation |
Consensus: within 6 months, as soon as feasible |
(sources/pfo-jscai-2022, rating: very high; sources/pfo-eso-2024)
The harm side
| Adverse outcome |
Device vs no device (as-treated, n=1,762 vs 1,956) |
Relative risk |
| Atrial fibrillation, all |
5.0% vs 1.1% |
4.54 (2.78–7.39) |
| AF beyond 45 days |
2.4% vs 0.8% |
2.60 (1.44–4.70) |
| Any serious adverse event |
28.7% vs 26.4% |
1.07 (0.97–1.19) — no significant excess |
| Major bleeding |
1.4% vs 1.7% |
0.80 (0.47–1.40) |
| Venous thromboembolism |
1.4% vs 0.5% |
2.59 (1.26–5.36) |
(sources/pfo-jama-2021, rating: very high)
- 46% of all atrial fibrillation was transient, confined to the first 45 days. But the excess beyond day 45 remains significant (RR 2.60) — a small permanent liability from a lifelong implant in a population whose median age is 46.
- Venous thromboembolism rose with the device (RR 2.59, P=.007) and the paper offers no mechanism. It is the least-explained finding in the meta-analysis and is not mentioned in either guideline.
- Harm tracks the causal-likelihood category, opposite to benefit: post-45-day AF risk difference 4.41% (1.02–7.80) in PASCAL "unlikely", 1.53% in "possible", 0.65% (−0.41 to 1.71) in "probable".
- Procedural complication rates from device registries — vascular injury requiring surgery 3.0%, tamponade or perforation 0.5%, death 0.3% — come from claims data rather than the trials (sources/pfo-jama-2025).
- Device-specific complication rates may differ — one meta-analysis found AF lowest with the Amplatzer and Gore occluders, and ESO notes left-disc diameter of 25 mm independently raised AF risk (aOR 2.67) — but no trial has compared FDA-approved devices head to head.
After the device
- Stroke risk remains elevated: 2.5% (1.5–4.0%) at 4 years after closure versus 0.4% (0.3–0.6%) in the general population, HR 6.3 (3.1–12.6) (sources/pfo-jama-2025).
- Residual risk is predicted by age, hypertension, diabetes, dyslipidaemia and smoking — not by PFO characteristics.
- Antithrombotic regimen after closure: dual antiplatelet therapy for 1–6 months, then single antiplatelet therapy for at least 5 years (ESO consensus, 9/9). No trial has compared DAPT with SAPT after closure, and SCAI makes no recommendation beyond one month.
Contradictions / Open Questions
- The outcome patients care about most did not reach significance. Disability-worsening recurrent stroke — a new stroke with any increase in the modified Rankin Scale at 30 days — had HR 0.59 (0.37–1.22), P=.14, the only efficacy endpoint in the meta-analysis that failed. The authors attribute it to missing functional-outcome data, which is probably true and does not change the fact that the strongest evidence in the field has not shown that closure prevents disabling stroke. (sources/pfo-jama-2021, rating: very high)
- The categories used to justify closing an "unlikely" patient are the ones the data cannot reach. ESO 2024's consensus statement licenses closure in PASCAL "unlikely" patients with DVT or PE near the index stroke, straddling thrombus, hypercoagulability, or platypnea-orthodeoxia. SCOPE states that the extended PASCAL categories covering thrombus straddling the PFO and concomitant DVT "could not be evaluated because patients in the uncommon extremely high-risk categories were not identified in the study." The escape clause is not merely unevidenced — it is explicitly outside the largest dataset in the field. (sources/pfo-jama-2021; sources/pfo-eso-2024)
- Causal probability and absolute benefit are inversely related. RoPE 9–10 predicts a ~90% attributable fraction and yet carries a 2-year recurrence of only 2% (0–4%) — so the patients for whom the PFO is most certainly the cause have the least absolute risk to remove. Any policy built on attributable fraction alone will over-treat the young and under-treat the older patient with a large shunt. (sources/pfo-jama-2025)
- "Possible" and "probable" are not distinguishable on the outcome that matters. Both give 2.1% absolute risk reduction at 2 years and both give an NNT of 47. The classification's useful discrimination is between "unlikely" and everything else; presenting it as a three-tier gradient implies a precision the absolute numbers do not support. (sources/pfo-jama-2021, rating: very high)
- The two guidelines cannot both be right about anatomy. SCAI makes a strong, anatomy-independent recommendation because subgroup data are too thin. ESO withholds closure specifically in "unlikely" patients and identifies large shunt plus aneurysm as the group with most to gain. The trial evidence supports ESO's reading; the trial populations support SCAI's caution.
- No trial has compared closure with a DOAC. CLOSE's anticoagulation arm was 92.8% warfarin, and the closure-versus-anticoagulation comparison was post-hoc and not prespecified (HR 0.14, 0.00–1.45, P=0.08). Contemporary medical therapy for a young stroke patient with a PFO is a DOAC, and that comparison has never been made. (sources/pfo-eso-2024)
- PASCAL was derived and tested in the same dataset, by the investigators who ran the trials it was tested on. Kent, Saver, Kasner, Carroll, Meier, Mattle, Thaler and Furlan are principal or steering-committee investigators of RESPECT, REDUCE, CLOSURE, PC or CLOSE, with disclosed funding or fees from Abbott and WL Gore; two manufacturers facilitated data access. The methods are prespecified, registered and auditable, which is why this is not an argument against the result — but a selection instrument for a device, derived and validated by the people who ran the device trials, is not independent evidence that the instrument works. (sources/pfo-jama-2021, rating: very high)
- The age cut-off is arbitrary and the evidence outside it is essentially absent. DEFENSE-PFO randomised 34 patients over 60 and CLOSE one under 18. Two guidelines respond by recommending against evidence-based recommendations, leaving the group with the highest stroke incidence — and the highest competing risk — managed by consensus.
- Timing has no evidence at all. Every trial enrolled within months of the index stroke and none was powered to test early closure. ESO's consensus recommends within 6 months and as soon as feasible, which is a clinical instinct rather than a trial result.
- Residual risk after closure is attributed to vascular factors, which implies the procedure was never the whole treatment. A 6.3-fold excess over the population at 4 years suggests that closing the shunt addresses one mechanism in a patient who usually has others, and that the referral pathway to a cardiologist should include risk-factor management as standard.
Connections
Sources