Heterogeneity of Treatment Effects in an Analysis of Pooled Individual Patient Data From Randomized Trials of Device Closure of Patent Foramen Ovale After Stroke

Authors, Journal, Affiliations, Type, DOI

• Authors: David M. Kent (Tufts PACE Center, corresponding author), Jeffrey L. Saver (UCLA), Scott E. Kasner (Penn), Jason Nelson and Benjamin Koethe (Tufts), John D. Carroll (Colorado), Gilles Chatellier (HEGP Paris), Geneviève Derumeaux (Henri Mondor), Anthony J. Furlan (Case Western), Howard C. Herrmann and Benedicte Lefebvre (Penn), Peter Jüni (Toronto), Jong S. Kim, Pil Hyung Lee and Jae-Kwan Song (Asan Medical Center, Seoul), Heinrich P. Mattle and Bernhard Meier (Bern), Mark Reisman (Washington), Richard W. Smalling (UTHealth Houston), Lars Søndergaard (Rigshospitalet), Jean-Louis Mas (GHU Paris), David E. Thaler (Tufts) — the principal investigators of the six trials being pooled
• Journal: JAMA. 2021;326(22):2277–2286 (December 14, 2021); accepted 3 November 2021; CME available
• Type: Individual participant data (IPD) meta-analysis of all 6 randomised phase 3 trials published by September 2021; PRISMA-IPD; PROSPERO CRD42020186537; Risk of Bias 2.0
• DOI: https://doi.org/10.1001/jama.2021.20956
• Funding: Patient-Centered Outcomes Research Institute (PCORI). Abbott and WL Gore facilitated data access without compensation

Overview

The SCOPE consortium analysis, and the paper from which everything about PFO closure selection in this wiki ultimately derives. Six trials, 3,740 patients, individual-level data harmonised across 20 years and six countries — 121 primary events over a median 57 months. It is the strongest evidence in the field and the source of PASCAL.

Its central finding is that the average treatment effect is a fiction. Closure reduced recurrent ischaemic stroke by 59% (adjusted HR 0.41) — an honest and reproducible average — but that average conceals a stratum in which closure does nothing at all and adds atrial fibrillation (PASCAL "unlikely", HR 1.14), and a stratum in which it reduces recurrence by 90% (probable, HR 0.10). The absolute risk reduction is 2.1% at two years in both the "possible" and "probable" categories — an NNT of about 50 — which the authors point out is comparable to antihypertensive, antiplatelet and statin therapy in secondary stroke prevention.

Two results are more interesting than the headline and less often quoted. First, closure reduced PFO-associated recurrent stroke by 76% (HR 0.24) but all-cause ischaemic stroke by only 59% — the difference is the proportion of recurrences that were never going to be prevented by closing a hole. Second, the outcome patients care about most did not reach significance: disability-worsening recurrent stroke, HR 0.59 (0.37–1.22), P=.14, limited by missing functional data.

Keywords

Patent foramen ovale · PFO closure · heterogeneity of treatment effect · RoPE score · PASCAL classification · individual participant data meta-analysis · recurrent ischaemic stroke

Key Takeaways

Design and population

• 6 trials, 3,740 patients (1,889 device, 1,851 medical therapy), enrolled 2000–2017, median follow-up 57 months (IQR 24–64).
• CLOSE was treated as two trials — one antiplatelet versus device, one anticoagulant versus device — because its randomisation had two medical arms.
• Median age 46 years; 54% male. Baseline: hypertension 27% vs 25%, diabetes 5.6% vs 5.7%, prior stroke or TIA 16% vs 15%, index event a TIA in 6.5% vs 7.1%, superficial infarct 71% vs 68%, large shunt 43% vs 47%, atrial septal aneurysm 33% vs 33%.
• All six trials had "some concerns" for risk of bias (mostly the PROBE design and missing outcome data); none was rated high risk.
• Primary analysis adjusted for age, sex, prior MI, diabetes, hypertension, hyperlipidaemia, prior stroke/TIA, smoking, index event, atrial septal aneurysm, shunt size and superficial infarction.

Primary and secondary efficacy outcomes

• The two ischaemic-stroke outcomes differ, and the difference is informative. Closure reduced PFO-associated recurrence by 76% and all ischaemic recurrence by 59%. The gap is the fraction of recurrences attributable to something else — an occult second mechanism that closing a hole cannot prevent.
• Disability-worsening stroke is the only efficacy outcome that failed, and it is the one defined by what the patient experiences rather than by imaging. The adjudicated PFO-associated endpoint was available in only 1,238 and 1,179 patients because the PC and CLOSURE trials did not adjudicate stroke mechanism.
• Median time to the primary event was 13.7 months (IQR 4.8–29.7) — the benefit accrues early.

Robustness

• Tipping-point analysis: imputed outcomes for patients lost to follow-up would have to be approximately twice as frequent in the device group to nullify the primary result. The finding is not fragile to missing data.
• Excluding CLOSURE — the trial of the withdrawn clamshell device — strengthens the effect, which is the expected direction if the device class mattered.

Heterogeneity — RoPE

• Interaction significant as a continuous variable (P<.01) and dichotomised (P=.02), stable in a 9-point RoPE without the imaging item and in a 5-trial analysis excluding the PC trial.
• RoPE halves the reference class but both halves still benefit. The authors make this point explicitly: a statistically significant interaction on the relative scale is clinically uninformative if every stratum benefits.

Heterogeneity — PASCAL

• P for interaction .003. Results stable in a 9-point RoPE analysis, in a 5-trial analysis excluding PC, and with the lead secondary outcome.
• "Unlikely" is a null on both scales and the only stratum where the point estimate crosses 1 — which is what makes PASCAL clinically useful where RoPE alone was not. The classification's value is not the gradient between possible and probable (identical ARR, identical NNT) but the existence of a stratum where the decision changes.
• The authors compare the NNT of ~50 to secondary-prevention antihypertensives, antiplatelets and statins — an important calibration, because it frames closure as comparable to, not better than, ordinary preventive therapy.

Safety (as-treated population: 1,762 device, 1,956 no device)

• 46% of all atrial fibrillation (50 of 109 events) was transient, confined to the first 45 days and attributable to periprocedural irritation rather than to the device as a permanent substrate.
• Beyond day 45 the excess AF remains real and significant (RR 2.60) — a small permanent liability from a lifelong implant in a population with a median age of 46.
• Venous thromboembolism rose with the device (RR 2.59, P=.007) — an unexpected signal that the paper reports without explanation. It is the one safety finding here with no mechanistic account.
• 105 patients assigned to device but never implanted were analysed in the medical group, which is the correct as-treated logic but dilutes the device arm.

Safety tracks the causal category, in the opposite direction to benefit

• Absolute risk increase in AF beyond 45 days with a device: 4.41% (1.02–7.80) in PASCAL "unlikely", 1.53% (0.33–2.72) in "possible", 0.65% (−0.41 to 1.71) in "probable".
• The authors connect this to the known association between low RoPE scores and incident AF: the same vascular risk factors that make a PFO an unlikely cause make the atrium more likely to fibrillate around an occluder.

Limitations of the document

• The benefit for disabling stroke is unproven. The authors list this first: substantial functional-outcome data were missing for recurrent strokes, and the disability-worsening endpoint was null (HR 0.59, P=.14). A therapy that prevents imaging-defined strokes without demonstrably preventing disabling ones has not shown the outcome that justifies it.
• Key variable definitions were not identical across trials. "Large shunt" was >20 bubbles in most trials but >25 in CLOSURE and >30 in CLOSE; atrial septal aneurysm definitions also varied. The two anatomical variables that drive PASCAL were therefore harmonised by judgement.
• The extended PASCAL classification could not be evaluated. The original system includes extremely high-risk categories — thrombus straddling the PFO, concomitant DVT — and the authors state that patients in those categories were not identified in the trial data. The categories that most strongly justify closure in the "unlikely" stratum are precisely the ones this dataset cannot test.
• The authors are the trials' principal investigators, and the conflict-of-interest list is extensive. Kent received grants from WL Gore; Kasner was REDUCE's principal investigator (Gore); Carroll sat on the RESPECT steering committee (Abbott) and was a REDUCE site investigator (Gore); Meier and Mattle received Abbott funding for PC and PRIMA; Thaler and Smalling had Abbott relationships; Furlan was CLOSURE's PI. PCORI funded the analysis and Abbott and Gore facilitated data access. This does not make the analysis wrong — the methods are prespecified, registered and independently auditable — but PASCAL is a selection instrument for a device, derived by the people who ran the device trials.
• No adjustment for multiple testing was applied to hypothesis-testing analyses (the authors state this explicitly), so the exploratory single-variable subgroup analyses should not be interpreted as confirmatory.
• Antithrombotic choice, newer devices and patients over 60 were not addressed — three of the questions a clinician actually faces, acknowledged in the limitations.
• The "possible" and "probable" categories produce the same absolute benefit and the same NNT, so the three-level system functions in practice as a two-level one. The paper presents the gradients without confronting what that means for the value of the distinction.

Key Concepts Mentioned

concepts/Patent-Foramen-Ovale · concepts/PFO-Closure-Selection · concepts/PFO-Detection-and-Anatomy · concepts/Poststroke-AF-Monitoring

Key Entities Mentioned

entities/Ischemic-Stroke · entities/Atrial-Fibrillation

Named but unlinked: RoPE · PASCAL · SCOPE · RESPECT · REDUCE · CLOSE · CLOSURE I · PC Trial · DEFENSE-PFO · Amplatzer PFO Occluder · Gore Cardioform

Wiki Pages Updated

• sources/pfo-jama-2021 — created
• concepts/PFO-Closure-Selection — primary citation for the pooled effect, RoPE and PASCAL strata, and the safety table
• concepts/Patent-Foramen-Ovale — headline figures re-pointed to the primary analysis
• concepts/PFO-Non-Stroke-Indications — AF risk stratified by PASCAL category
• concepts/Poststroke-AF-Monitoring — AF after closure, primary estimates
• wiki/wikiindex · wiki/sourceindex