ECG-AI Pulmonary Hypertension (PH) 12-Lead algorithm (1020)

Performance Metrics

Clinical Evidence

The evidence base is young and, so far, entirely manufacturer-authored and retrospective. The pivotal publication (DuBrock HM, Wagner TE, Carlson K, et al., European Respiratory Journal 2024;64(1):2400192) describes the PH Early Detection Algorithm (PH-EDA), a convolutional neural network trained on ECG voltage-time data from Mayo Clinic: 39,823 PH-likely patients and 219,404 controls, labelled from right heart catheterisation or echocardiography and split 48/12/40 into training, validation and test sets. On diagnostic ECGs the algorithm reached an AUC of 0.92 in the Mayo test set and 0.88 in external validation at Vanderbilt University Medical Center; on pre-emptive ECGs — taken six to eighteen months before clinical diagnosis — AUCs were 0.86 and 0.81 respectively. That pre-diagnostic discrimination is the device's central clinical claim: pulmonary hypertension is routinely diagnosed years after symptom onset, and an algorithm that flags risk from ECGs already on file could compress that interval. The clearance itself rests on a separate retrospective validation of 21,066 adult patients presenting with dyspnea across five US health systems, reporting 73 percent sensitivity and 74.4 percent specificity. As of the review date this study is described in the manufacturer's clearance announcement and FDA submission but has not been located in a peer-reviewed venue. The gap between the development AUCs and the cleared operating point is worth understanding in procurement review: an AUC of 0.92 describes ranking ability across all thresholds, while the cleared configuration is a single threshold tuned for a symptomatic population. The manufacturer has additionally reported a real-world analysis in which the algorithm identified more than 85 percent of pulmonary arterial hypertension patients and 78 percent of chronic thromboembolic pulmonary hypertension patients from ECGs taken between symptom onset and diagnosis; this, too, is company-reported rather than peer-reviewed. No randomised or prospective interventional evidence — whether earlier notification actually shortens time to right heart catheterisation or improves outcomes — has been published as of the review date.

Inside the Auris+ Listing

Regulatory Approvals

Safety Record

No recalls, FDA safety communications or published adverse-event reports involving the device were identified in publicly available sources as of August 2026; the MAUDE database could not be queried directly from this environment, so this should be read as "none found in public reporting" rather than an exhaustive audit. The device was cleared in March 2026, so the post-market record is necessarily short. As a notification-only aid the harm surface is bounded but real in both directions: a false notification can trigger unnecessary echocardiography or invasive right heart catheterisation, while at 73 percent cleared sensitivity roughly a quarter of true cases in the validation population produced no notification — the absence of a flag must not defer a clinically indicated work-up.

Intended Use & Indications

The ECG-AI PH algorithm is a cardiovascular machine learning-based notification software device cleared in the United States under product code SAT (21 CFR 870.2380). It analyses a routine 12-lead ECG already in the health system's records and raises a notification when it detects patterns associated with pulmonary hypertension, so that clinicians can consider earlier work-up — typically echocardiography and, definitively, right heart catheterisation. It does not output a diagnosis, a haemodynamic estimate, or a treatment recommendation. The clinical rationale is the well-documented diagnostic delay in pulmonary hypertension: symptoms are non-specific and the mean interval from symptom onset to diagnosis is measured in years, by which point disease is often advanced. Because the input is the standard 12-lead ECG rather than a dedicated acquisition, the algorithm can be applied wherever ECGs already accumulate. Per the manufacturer, the software integrates with EHR and ECG management systems and runs within the health-system environment without transferring patient data; those deployment claims are the manufacturer's own and commercial deployments are not yet independently documented. Procurement reviewers should note the regulatory footprint is US-only as of the review date, the cleared performance profile (73 percent sensitivity, 74.4 percent specificity in adults presenting with dyspnea) positions the device as a case-finding aid rather than a rule-out test, and the definitive diagnostic pathway for pulmonary hypertension — right heart catheterisation — is unchanged by a notification either way.