AI in HF: How to Close the Gap Between Knowledge and Practice

Several years ago, I took care of a patient in her 50s who had just returned from overseas. During that trip, she lost her partner in a hiking accident. When she came to the hospital, she was grieving and short of breath, and had an abnormal electrocardiogram. Her echocardiogram showed an ejection fraction of 18%.

At first, I hoped she had stress cardiomyopathy. The story seemed to fit. But her ventricular function didn’t rapidly recover. Only later did we learn that she carried a pathogenic TTN variant, suggesting that her presentation was the first sign of an underlying genetic cardiomyopathy.

During her hospitalization, I initiated two low-dose heart failure medications. I saw her again 1 week after discharge and added the remaining therapies. Over the next 3 months, through multiple visits, phone calls, lab checks, and medication adjustments, we eventually got her to maximally tolerated guideline-directed medical therapy (GDMT).

The care was successful, but the process felt inefficient. Did she really need four in-person visits to receive lifesaving medications? Was there a better way to monitor her blood pressure, heart rate, kidney function, potassium, symptoms, and medication tolerance between visits? Could we use the data already being generated outside the clinic to make heart failure care more continuous, proactive, and scalable?

The question was not whether she needed guideline-directed medical therapy. The question was how to deliver it safely, efficiently, and repeatedly after she left the hospital.

That patient’s story stayed with me because it embodies one of the central challenges in heart failure care: ensuring that patients receive lifesaving treatment amid increasing demand and constrained resources. The challenge is especially evident in the persistent gap between what clinical trials demonstrate and what traditional care models actually deliver.

Randomized trials have established the benefits of guideline-directed medical therapy, but applying these therapies in practice is complex and highly individualized. Medication decisions depend on blood pressure, heart rate, renal function, potassium, frailty, prior intolerance, patient preference, and clinical judgment. Yet heart failure is still largely managed through fragmented, visit-based models of care, leaving clinicians with limited visibility into what happens between appointments and few resources to safely adjust therapy before patients decompensate.

Closing this gap will not come from adding another quality metric or asking cardiologists to do more with the same fragmented tools that already contribute to burnout. Quality metrics can measure undertreatment, but they do not give clinicians the time, electronic medical record (EMR)-integrated data, or care teams needed to manage patients between visits.

Improving outcomes requires systems that make longitudinal heart failure care easier to deliver, with remote monitoring, team-based workflows, and actionable data that empower cardiologists and support individualized clinical judgment. That includes the ability to titrate therapy safely in frail patients when it is appropriate and avoid escalation when it is not. Managing heart failure is hard work, and without the right systems and support, population health initiatives risk adding administrative burden without meaningfully improving outcomes.

The problem is not a lack of evidence, and increasingly, it is not a lack of data. Remote patient monitoring, home vital sign collection, and implanted cardiac devices have been generating clinically relevant information for years, yet these data streams have not been systematically incorporated into heart failure care. Historically, they have remained fragmented across manufacturers and poorly integrated into workflows and the EMR, limiting their ability to drive timely, longitudinal clinical action.

At Cedars-Sinai, drawing in part from the model established in electrophysiology—where continuous device data informs longitudinal management—we saw an opportunity in chronic disease management at the intersection of two capabilities that have historically remained separate: platforms that aggregate ICD and pacemaker data from multiple manufacturers and reimbursement-enabled infrastructure for remote, asynchronous chronic care management.

Cardiac device monitoring companies are well positioned for this next phase because managing remote data, building workflows, and increasingly supporting care teams are core to their business. Our approach has been to combine these capabilities—integrating continuous cardiac data infrastructure with structured, reimbursable care delivery and aligned physician incentives—to translate remote monitoring from a tool for observation into a foundation for actionable, longitudinal care. Together, this strategy makes asynchronous chronic disease management both operationally and financially viable.

This approach differs from remote chronic care models that rely primarily on external clinical teams to manage patients between visits. Rather than outsourcing heart failure management, our approach has been to strengthen the capabilities of existing cardiologists, advanced practice providers, and health-system staff by providing the remote patient monitoring infrastructure, clinical decision support, and reimbursement-aligned workflows needed to manage patients longitudinally between visits. The goal is not only better care, but a model in which the clinical and economic value created by longitudinal care accrues to the clinicians and health systems already responsible for these patients, rather than being shifted outside the health system.

Specifically, to implement this vision, we’ve partnered with RhythmScience, a cardiac data platform that integrates implanted device data, remote monitoring inputs, and vital signs into longitudinal care workflows. Together, we’ve integrated this cardiac data platform with HF-AI, an artificial intelligence-driven GDMT clinical decision support system developed by Cedars-Sinai cardiologists in conjunction with Cedars-Sinai Technology Ventures and validated in the prospective TURING-HF clinical trial. In TURING-HF, HF-AI with home vital signs was associated with more than doubling GDMT utilization and enabled approximately 80% of medication titrations to be performed remotely, with significant reductions in heart failure hospitalizations from before to after trial enrollment.

Collaborating with our partner company, our team has also developed reimbursement and care models that account for the time spent by clinical staff, including medical assistants, registered nurses, pharmacists, and cardiologists, allowing health systems to expand longitudinal care capacity through aligned reimbursement.

More broadly, chronic disease management improves when health systems build the capabilities to turn remote monitoring into actionable longitudinal care at scale. That means expanding clinical capacity, aligning clinicians and health systems around the economic value created by that care, and building workflows where future AI tools can support patients and care teams between visits. For AI to matter in chronic disease management,  it will need to be  embedded within real clinical workflows and operate within sustainable, reimbursable population health programs.

As we gain experience applying this model at Cedars-Sinai, we are building the infrastructure to integrate data not only from implantable cardiac devices and home blood pressure cuffs, but also from medical-grade and consumer wearables, into a longitudinal digital cardiology practice. Our goal is to lay the foundation for a virtual, asynchronous model of cardiac care that improves outcomes, increases clinical capacity, and makes high-quality cardiovascular care more continuous, proactive, and scalable.

Off Script is a first-person blog written by leading voices in the field of cardiology. It does not reflect the editorial position of TCTMD.