AI for Heart Failure Care Is Evolving Rapidly, THT 2026 Makes Clear

BOSTON, MA—Several sessions last week at THT 2026 made clear that the integration of artificial intelligence (AI) into the management of patients with heart failure (HF) is occurring quickly, touching on everything from screening and diagnosis to treatment.

Explorations into use of AI to detect patients with undiagnosed HF and direct them to echocardiography, the potential for AI-enabled HF clinics, and how the technology can help streamline clinical trials were part of discussions. There was also a talk on the future of human-AI interactions and another focused on issues around ethics, bias, and regulation.

One common theme included the hope that AI could handle some of the more tedious aspects of care so that physicians could get back to having more personal interactions with their patients.

Tariq Ahmad, MD (Yale School of Medicine, New Haven, CT), talked about that “sacred interaction” during his presentation discussing how AI can be used to enhance clinic visits.

“We went into medicine because of a need to help patients,” he said. “We did not go to do all the things that we’re doing in clinics right now, and I think if technology and AI can help us get back to that, then we would have saved medicine from where it’s heading.”

Anu Lala, MD (Icahn School of Medicine at Mount Sinai, New York, NY), who spoke on AI’s role in HF diagnostics and clinical decision-making, said, “Maybe it’s a relief of some of that burden that we’re feeling that will facilitate our openness, our receptivity, to go back to the bedside, [to] be more patient.”

“It’s not us or AI, it’s and. So how can we use it to allow for us to be the best versions of ourselves as physicians, as healers, as clinicians?” Lala said.

Earlier and Better Decisions for Patients

At the beginning of her talk, Lala took an unofficial poll, showing that while some attendees had turned to AI for clinical use recently, not all documented that they had done so.

“The question really is not is AI to be used in helping clinical decision-making. It’s already a part of clinical decision-making in ways that we perhaps are not as aware of or that we are admitting,” she said, adding that the guardrails around clinical use of AI haven’t quite been defined yet.

Currently, how clinicians manage patients can be very reactive in terms of waiting for symptoms and then hospitalizing patients and escalating care, and AI can help change that, Lala said.

“With proper integration and leverage of AI, we can anticipate,” she said. “We can personalize and ultimately make better decisions for the right patients. Because the future of heart failure . . . care is not more data. It’s earlier and better decisions.”

The future of heart failure . . . care is not more data. It’s earlier and better decisions. Anu Lala

In another presentation, Heidi Hartman, MD (NewYork-Presbyterian/Columbia University Irving Medical Center, New York, NY), highlighted how AI can play a role very early in management—by identifying patients who don’t know they have a disease. Though an echocardiogram can find most forms of structural heart disease, it comes with a higher cost and the need for expertise to interpret it, limiting accessibility. ECG, on the other hand, is more widely available.

Hartman described EchoNext, a deep learning model trained on echocardiograms and ECGs from eight sites affiliated with NewYork-Presbyterian that detects structural heart disease from 12-lead ECGs and then steers patients toward getting evaluated by echo. In a silent deployment cohort, the researchers determined that nearly half of patients at high risk for structural heart disease (45%) would not have received an echo with routine clinical care.

They then performed the CACTUS pilot study of the EchoNext model, which showed that AI-ECG can serve as a “safety net for patients with undiagnosed heart failure,” Hartman said. In one notable case, a 45-year-old man with worsening exertional dyspnea and cough who was initially sent home from the emergency department with bronchodilators and steroids for presumed reactive airway disease eventually underwent a heart transplant after being flagged by EchoNext and sent for an echo. The AI model ordered more echos than any physician at the hospital and also had the highest positivity rate.

“This is the beginning of the next era of medical diagnostics where AI-based technologies are facilitating detection of diseases that would’ve otherwise gone undiagnosed,” Hartman said. “And in this way, AI-based opportunistic screening is going to fundamentally change the way we’re practicing medicine.”

Improving HF Visits

In his talk discussing AI-enabled clinic visits, Ahmad noted that HF clinics operate not all that differently from how they ran 25 years ago, with fixed times for all patients, checks of blood pressure and the ECG, a 10-minute chat with the patient to review data and come up with a plan, and 20 minutes of documenting, prescribing, and billing. What has changed, however, is that costs have risen, physician satisfaction has dropped, and patient outcomes have stagnated or possibly worsened.

“What we can do with AI is we can make the visit in accordance with the actual need of the patient,” Ahmad said.

AI can help by synthesizing data from patients’ wearables and providing a summary ahead of time, automating documentation and billing, and allowing more time to discuss issues like prognosis, which often gets short-changed. This may ultimately reduce costs within healthcare systems by limiting the number of personnel needed for administrative tasks.

“We need to reduce the number of human beings involved and use technology to maximize the physician-patient interaction,” Ahmad said.

What we can do with AI is we can make the visit in accordance with the actual need of the patient. Tariq Ahmad

Afnan Tariq, MD, JD (North Coast Cardiology, Encinitas, CA), presented a study detailing a novel AI-based platform (Alyf) that integrates physiologic information from more than 300 consumer wearables, with the aim of providing physicians with information on how patients are faring between clinic visits based on a variety of measures, including heart rate, respiratory rate, activity patterns, circadian variability, and sleep.

“It’s that technical bridge that allows clinicians to be clinicians again and really see that data there. And then clinicians can maximize their impact,” Tariq, who is the co-founder and CEO of Alyf, told TCTMD.

In a feasibility study, the Alyf platform was used by 71 patients (mean age 75 years; 54% women) with HF or HF equivalents who were then followed for a mean of 18.2 months. During that time, the platform processed about 75 million physiologic measurements, or about 2,000 per patient per day.

This lays the foundation for the “infrastructure layer for between-visit care,” Tariq said. “Now, with appropriate tools, you can be armed at the point of care with a full summary of what happened between visits. If there is something that the care team thinks that you need to address between visits, they’re now armed with the information to do that. I think it returns the conversation to being more human.”

Ami Bhatt, MD, chief innovation officer of the American College of Cardiology, in a talk about ethics, bias, and regulation, framed the use of AI as an ethical issue: “We need to get that data, and we need it to be presented to us so we can use our clinical acumen. It is unethical for our field to not do that moving forward.”

More Work Ahead

Hartman said that as AI tools grow in number, “there’s an urgent need to improve the implementation science behind AI models and develop the infrastructure needed to support [them].” Whether they’re cost-effective and how to pay for them are other key issues, she said.

Asked how the healthcare system will handle, for example, an increasing number of patients being sent for echocardiograms, Hartman acknowledged “that we haven’t solved yet.” It also hasn’t been resolved how all the newly diagnosed patients will be connected to care or if there are enough cardiologists to manage them.

Harlan Krumholz, MD (Yale School of Medicine), editor-in-chief of JACC, argued that “if we have the ability to find people who [have] left ventricular systolic dysfunction for pennies a day [and] we can prevent progression to heart failure and we are failing to do it, then I believe that we have violated our moral obligation to our patients. If we have the ability to do this, we need to be able to figure out how to do it operationally. It is unacceptable to say we are going to be overloaded.”

We do not get to stop this revolution. Ilan Richter

In a forward-looking talk on human-AI interactions, Ilan Richter, MD (NewYork-Presbyterian/Columbia University Irving Medical Center), highlighted three main AI-related tensions likely to be in play over the next decade.

“The first is automation versus authority,” he said, referring to the choice that comes with the calibration of AI interfaces designed to “maximize their beneficial effect or for us to categorize our control.”

The next is accountability versus autonomy. “How concerned should we be scientifically, ethically, and even legally about being accountable for medical decisions or outcomes that are the result of AI algorithms?” Richter said. “And how will the autonomy of these models actually come into reality?”

And the last involves explainability versus performance, he said, alluding to questions about whether physicians need to know how AI works if patient outcomes are improving. “It would test the question of what really is our ultimate role as heart failure physicians? Is it to facilitate the appropriate clinical decision or to understand how we got there?”

The field of AI in cardiovascular disease is likely to move quickly in the coming years. The US Advanced Research Projects Agency for Health (ARPA-H) has initiated a cardiovascular disease-focused program called ADVOCATE that is soliciting proposals for agentic AI systems that can provide around-the-clock holistic clinical care by making changes to appointments, medications, diet, and exercise on their own. “The program has three technical areas: development of a patient-facing clinical AI agent, a supervisory agent that ensures clinical AI agents’ consistent safety and effectiveness, and a scalable plan for integration into clinical workflows in healthcare organizations,” according to the ADVOCATE site.

“We do not get to stop this revolution,” Richter said. “We either get to drive it to the place where our patients benefit the most and we let go of some of our old habits to move our profession onwards or we’re left behind. I don’t believe there’s another place in between.”

Krumholz predicted that there would be much change around AI and HF care, and soon. “Just imagine now what it’s going to be like at THT a year from now when we talk about AI,” he said. “The only thing I can guarantee you, it’s going to be amazingly different than it is today.”