Article
Nov 9, 2026
What AHA 2026 says about AI in heart care
Two AI trials at AHA 2026 tested plaque-guided chest pain care and a voice app for heart failure. Here is what they found and what is still unproven.
The American Heart Association's Scientific Sessions 2026 ran November 6 to 9 in Chicago, and artificial intelligence had its own late-breaking session. Two results drew attention this weekend: an AI-guided approach to chest pain workups, and a smartphone app that listens to a patient's voice to spot heart failure trouble early. Both look promising. Both still have a lot to prove.
What happened
Two companies reported trial results at or around the meeting. Cleerly announced that PARAMOUNT, which it describes as the first randomized trial of AI-enabled plaque-based care, met its primary endpoint. Cordio Medical reported that its HearO voice app predicted most heart failure events weeks before they happened in a pivotal study called DETECT-HF.
Clinicians should read both with care. The figures below come from company announcements and trade press coverage of the presentations. Neither result has been published in a peer-reviewed journal yet, and Cleerly says its own numbers are preliminary.
PARAMOUNT: AI plaque analysis for stable chest pain
PARAMOUNT enrolled symptomatic outpatients with suspected coronary artery disease, an LDL cholesterol of at least 100 mg/dL, and no history of coronary disease. According to Cleerly, 530 patients were randomized at 21 sites between January 2025 and February 2026. One group had care guided by AI analysis of coronary plaque on CT angiography. The other had usual care guided by functional stress testing.
The primary endpoint was a hierarchical composite at 180 days, analyzed with a win ratio. In a preliminary analysis of 523 patients, Cleerly reported a win ratio of 2.00. The biggest driver was diagnostic certainty: knowing whether the patient's chest pain was caused by coronary disease or not.
What changed for patients
More preventive treatment. Use of lipid-lowering therapy rose from 31% to 75% in the AI arm, compared with 32% to 47% in the usual care arm.
Fewer unnecessary catheterizations. Among patients sent for invasive angiography, 38% in the AI arm ended up with no stent or bypass, compared with 60% in usual care.
More women studied. Women made up 300 of the 523 patients analyzed, more than half the trial.
Co-principal investigator Todd Villines of the University of Virginia said plaque-based care meant significantly fewer patients without obstructive disease were referred to the cath lab, without raising overall catheterization or revascularization rates.
The caveats
The trial is open label, and Cleerly sponsored it. The company's own release says the results are preliminary, not yet peer reviewed, and that full results are expected in early 2027. The trial registry on ClinicalTrials.gov lists PARAMOUNT as active and not recruiting, with the primary endpoint measured through study completion. The 180-day numbers are an early look, not the final word.
DETECT-HF: a voice check for heart failure
Cordio's HearO app asks patients with heart failure to record short phrases on a smartphone every day. It builds a voice model for each patient and flags changes that may signal fluid buildup before symptoms bring the patient to hospital.
According to Medical Device Network, the blinded DETECT-HF study followed 563 patients and HearO predicted over 70% of heart failure events, on average 24.6 days before they occurred. Lead author William Abraham of Ohio State University Wexner Medical Center said the study showed an AI-assisted voice biomarker can identify deterioration weeks before an event.
The caveats
DETECT-HF is registered as an observational, single-arm study. Patients recorded their voices and the recordings were analyzed afterwards. That tells us how well the algorithm detects trouble. It does not yet show that acting on its alerts keeps patients out of hospital. The coverage also did not report how often the app raised alerts that turned out to be nothing, which is the number that decides whether a clinic can live with it. Cordio says it plans to seek FDA clearance, with a US launch anticipated in 2027.
The rest of the AI session
The meeting's dedicated session on AI in cardiovascular diagnosis also included a consumer smartwatch ECG study for weak heart pumping, a blinded randomized trial of AI-assisted echocardiogram interpretation, and an analysis of AI plaque assessment in more than 269,000 patients, according to Radcliffe Cardiology's preview of the program. The direction is clear. AI in cardiology is moving from reading images faster to changing who gets tested, treated and sent to the cath lab.
What this means for clinics
For family physicians and cardiology practices in Canada and the US, nothing here changes practice today. Neither tool is part of guidelines, and HearO is not yet cleared in the US. But three lessons are worth keeping.
Ask what the AI changes, not just what it detects. PARAMOUNT is interesting because it measured decisions: statin use and needless catheterizations. Detection accuracy alone does not tell you whether patients do better.
Look for the false alarm rate. Any monitoring tool that sends alerts needs a clear answer on how many are wrong, and who on your team will respond to them.
Wait for the paper. Sponsor announcements at a conference are a starting point. Peer-reviewed results, with full methods and limits, are what should drive purchasing or referral decisions.
Voice is becoming a serious clinical input. At Athel we use it for a different job: our AI co-clinician interviews patients before the visit by voice, web chat or video, and the physician reviews a structured history before walking into the room. The same rule applies to us as to any AI in cardiology: the clinician stays in charge of every decision.