Interview for Heart Failure Academy

Interviewer: Mirjam Boros, Heart Failure Academy

Speaker: Dr Lin Yee Chen, University of Minnesota Medical School, Minneapolis, Minnesota, US

Mirjam Boros: In 2026, where is ECG-based AI already changing heart failure screening or risk stratification?

Dr Lin Yee Chen: AI-enabled ECG models are already improving heart failure management in two important ways. First, they can detect left ventricular systolic dysfunction before symptoms appear. Some medical centres are already using these models to identify people with asymptomatic ventricular dysfunction, allowing clinicians to start appropriate medical therapies and preventive interventions earlier, reducing the risk of progression to heart failure.

Second, AI ECG can help predict which patients are at risk of developing atrial fibrillation. As a cardiac electrophysiologist, this is an area of particular interest to me. Atrial fibrillation is a major contributor to heart failure, so identifying high-risk individuals creates an opportunity for earlier intervention and prevention.

Mirjam Boros: Which AI-derived rhythm or atrial signals do you currently trust most for flagging heart failure risk?

Dr Lin Yee Chen: The ECG features I focus on are those that reflect atrial dysfunction or enlargement, particularly P-wave parameters. These include P-wave duration, P-wave axis, P-wave terminal force in lead V1 and markers of interatrial block. Together, these signals provide valuable insight into underlying atrial remodelling and dysfunction, both of which are associated with increased cardiovascular risk.

Mirjam Boros: For a new heart failure AI tool, what checks around population fit, calibration, bias and interpretability must be in place before you rely on it?

Dr Lin Yee Chen: All AI models, including AI ECG models, must be validated across multiple clinical settings and patient populations to demonstrate that they are generalisable and perform consistently outside the original development environment.

Calibration is also essential. A model may demonstrate excellent discrimination, but if the predicted probabilities do not align with actual observed event rates, its clinical usefulness is limited. In other words, both calibration and discrimination are important.

Bias must also be carefully assessed. Models should perform equally well across different demographic groups, including younger and older patients, men and women, and other relevant populations.

Finally, explainability is critical. Clinicians need to understand what factors within the model are driving its predictions. Explainable AI is key to building physician confidence and supporting widespread adoption.

Mirjam Boros: In the next two to three years, which AI applications do you expect to enter routine heart failure care, and which will likely remain experimental?

Dr Lin Yee Chen: I expect AI ECG models that detect asymptomatic left ventricular systolic dysfunction to become increasingly common in routine clinical practice. Similarly, AI tools that predict future atrial fibrillation before symptoms develop are likely to see broader adoption.

Applications that will probably take longer to enter routine care include AI models designed to predict which patients will respond to specific heart failure therapies and which will not. These approaches are promising, but further validation is needed before they become standard practice.

Mirjam Boros: For a busy heart failure clinic today, which one or two AI tools clearly add value without increasing workflow burden?

Dr Lin Yee Chen: The AI tool that I believe offers the most immediate value is large language model-assisted clinical documentation. Ambient AI systems can generate clinical notes in real time during patient consultations, creating a draft note automatically as the conversation takes place.

This can significantly reduce administrative workload, improve efficiency and allow clinicians to spend more time focusing on patient care. By reducing documentation burden, heart failure specialists can potentially see more patients while maintaining high-quality care.

Mirjam Boros: If every clinic could adopt one AI tool for heart failure, which would you choose, and what is the biggest risk of using it uncritically?

Dr Lin Yee Chen: I would choose large language model-assisted note-taking and clinical documentation. The benefits for both clinicians and patients are substantial because it streamlines workflow and reduces administrative burden.

However, the greatest risk is over-reliance. Physicians must carefully review and verify AI-generated notes to ensure they are accurate and complete. If clinicians become overly dependent on these systems without appropriate oversight, documentation errors could be introduced into the medical record.