PFO Detection and Anatomy
Definition
PFO detection is the identification of a right-to-left shunt across a persistent interatrial tunnel, and the anatomical characterisation that follows it. It is a problem without a gold standard: the conventional reference — contrast transoesophageal echocardiography — is itself unvalidated, and its sensitivity against autopsy, cardiac surgery or catheterisation is only 89% (sources/pfo-eso-2024, rating: very high). Every diagnostic accuracy figure in this field is therefore relative to an imperfect comparator, and the European Stroke Organisation declined to issue any evidence-based recommendation on diagnosis at all because of it.
Key Concepts
Three tests, three different jobs
| Test |
Pooled sensitivity |
Pooled specificity |
LR+ |
LR− |
What it can and cannot do |
| c-TCD |
0.96 (0.93–0.98) |
0.90 (0.83–0.95) |
9.74 |
0.04 |
Best rule-out. Comfortable, reproducible Valsalva. Cannot localise the shunt or see the septum |
| c-TTE |
0.71 (0.50–0.86) |
0.99 (0.93–1.00) |
60.77 |
0.29 |
Widely available, very specific, insensitive — a poor rule-out. Harmonic imaging raises sensitivity to ~91% |
| c-TOE |
0.89 vs invasive/post-mortem truth |
0.91 vs the same reference in a 4-study subset |
— |
— |
The only test that measures anatomy. Valsalva is difficult under sedation, which is why it is imperfect |
- Area under the summary ROC: c-TCD 0.96, c-TTE 0.71 (sources/pfo-eso-2024).
- ESO's consensus, in place of a recommendation: use a locally agreed algorithm across whatever is available (8/9); c-TCD when c-TTE and/or c-TOE is inconclusive (8/9); TCD has superior sensitivity to TTE (6/9); do TOE after a positive screen to define anatomy (9/9). The vote splits are small, and they are votes, not evidence.
- c-TCD cannot distinguish a cardiac from an extracardiac shunt — a pulmonary arteriovenous malformation produces the same high-intensity transient signals. TOE or CT is needed when the distinction matters.
- Neither c-TCD nor c-TTE was ever compared directly with the other against c-TOE. ESO's third diagnostic PICO question could not be answered: no study exists.
What counts as a large shunt — and the definitions do not match
| Modality |
Large shunt |
| c-TTE or c-TOE |
>20 microbubbles in the left atrium within 3 cardiac cycles of right atrial opacification |
| c-TCD |
>10 high-intensity transient signals |
| Older TCD criteria |
1–5 or >5 HITS, depending on the study and its era |
- The inconsistency is partly historical: twelve of the 34 studies in ESO's meta-analysis used a 1–5 HIT threshold and predated the consensus recommendations on shunt grading.
- Adding blood to the agitated saline increases sensitivity, up to 100% in one comparison (sources/pfo-eso-2024). The optimal blood volume is unsettled — one study found 1 mL sufficient, another a single drop.
Atrial septal aneurysm and other high-risk anatomy
| Feature |
Definition |
| Atrial septal aneurysm |
Septum primum excursion ≥10 mm from the plane of the atrial septum, into the right and/or left atrium |
| Septal hypermobility (DEFENSE-PFO) |
Phasic septal excursion into either atrium ≥10 mm |
| Aneurysm (DEFENSE-PFO, stricter) |
Protrusion of the dilated septal segment ≥15 mm beyond the level surface of the septum |
| PFO size (DEFENSE-PFO) |
Maximum separation of primum from secundum ≥2 mm during Valsalva |
| Tunnel length |
Longer tunnels close less completely and are associated with residual shunt |
| Eustachian valve |
Embryological remnant directing IVC flow toward the septum — increases shunt and paradoxical embolism risk |
| Chiari network |
Right atrial remnant; associated with paradoxical embolism in observational data |
- Aneurysm and large shunt are multiplicative, not additive. In pooled trial data, patients with both had 5.9% recurrence on medical therapy and 0.4% with closure at 2 years — absolute risk reduction 5.5% (2.7–8.3), against 1.0% for one or neither feature (sources/pfo-jama-2025, rating: high).
- Almost four in ten cryptogenic-stroke patients with a PFO have high-risk morphology. In DEFENSE-PFO, 1,715 patients had TOE, 450 had a PFO, and 175 (38.9%) met the high-risk anatomical definition (sources/pfo-jacc-2018, rating: high). This is the practical argument for doing TOE rather than TTE in this population: the anatomy changes the decision in a large minority.
Alternatives, and why they are worse
| Modality |
Performance |
| Cardiac MRI |
Detected a shunt in 29.3% where TOE found 53.7% (P<0.001) — significantly inferior |
| Single-phase cardiac CT |
Sensitivity 85.3%, specificity 33.3%, negative predictive value 33% |
| Full-cycle cardiac CT |
Sensitivity 89.4%, specificity 92.3% — better, but single-phase sensitivity fell to 72.7% (P=0.002) |
(All from sources/pfo-eso-2024.)
Residual shunt after closure
- No standardised protocol exists for confirming closure, and no gold standard to confirm it against.
- Reported residual shunt 0–19%, falling with time: 19.5% at a mean of 4 months and 8.4% at 11 months in one series; 87% complete closure at 2 years in another, influenced by device type, tunnel length and aneurysm extent.
- Whether residual shunt matters is unresolved. One cohort found no increase in thromboembolic events; a large prospective cohort found stroke or TIA recurrence at 2.32 versus 0.75 per 100 patient-years in patients with residual shunt versus complete closure (sources/pfo-eso-2024).
- Some apparent residual shunts turn out to be other right-to-left connections — pulmonary arteriovenous malformations were identified in one series.
Contradictions / Open Questions
- The reference standard is not a standard. Every sensitivity and specificity in this field is measured against c-TOE, which achieves only 89% sensitivity against autopsy, surgery or catheterisation. The uncertainty is therefore not a property of the tests being evaluated but of the entire framework, and a Bayesian analysis of diagnostic accuracy — which ESO considered and excluded as beyond scope — would give different answers.
- The modality-specific definitions of "large shunt" are not interchangeable, and none of them is calibrated against outcome. >20 bubbles at TOE, >10 HITS at TCD, and 1–5 HITS in the older literature all get called "large" somewhere, yet shunt size is an entry criterion for the strongest treatment effect in the trial data. (sources/pfo-eso-2024)
- c-TCD is the best screen and the least able to answer the question. It has 96% sensitivity and cannot identify where the shunt is, cannot exclude a pulmonary AVM, and cannot measure or describe the septum. A patient screened by TCD alone is a patient whose treatment decision rests on a test that cannot see the structure being treated.
- The PFO size threshold is stated inconsistently within DEFENSE-PFO itself — ≥2 mm in the Methods and Abstract, ">3 mm" in the Figure 1 legend — in a trial whose entire design depends on that threshold. Baseline mean size was 3.2 ± 1.5 mm, so a substantial minority would fail a 3 mm cut. (sources/pfo-jacc-2018)
- Residual shunt might not matter, and might matter a great deal. 2.32 versus 0.75 events per 100 patient-years is a threefold difference in one cohort and zero in another. No trial has randomised on the basis of residual shunt, and no standard exists for when to look for it.
- Eustachian valve and Chiari network are in the risk conversation without being in any score, and without quantitative outcome data. They appear in expert consensus lists of features that may raise causal probability — which is a category of evidence that cannot be checked. (sources/pfo-jama-2025)
- Whether Valsalva should precede or follow contrast injection is not standardised, and it changes the measured shunt. Studies in ESO's analysis did it before injection, immediately after, five seconds after, at right atrial opacification, or did not report — with no evidence that any is correct.
Connections
Sources