The Advanced Research Projects Agency for Health (ARPA-H), a federal agency within the U.S. Department of Health and Human Services (HHS), has officially launched a multi-institutional initiative to transform the management of chronic heart failure through the development of "agentic" artificial intelligence. By partnering with a consortium of technology firms and academic medical centers, including Tempus AI, Updoc, Stanford University, Duke University, and Kaiser Permanente, the agency aims to bridge the critical gap in care that exists between traditional outpatient physician visits. The program, titled ADVOCATE (Autonomous Digital Virtual Over-watch for Care-Agile Teams), represents one of the most ambitious federal forays into autonomous clinical technology to date, seeking to provide continuous, high-level monitoring for a condition that affects millions of Americans and places an immense strain on the national healthcare infrastructure.
The Crisis of Cardiovascular Care and Rural Access
The impetus for the ADVOCATE program is rooted in a worsening public health crisis. According to data from the American Heart Association (AHA), approximately 6.7 million American adults are living with heart failure, a number projected to rise to over 8 million by 2030. Heart failure is not only a leading cause of hospitalization for adults over the age of 65 but also carries a staggering economic burden, with annual costs exceeding $30 billion in the United States alone. These costs are driven largely by frequent emergency room visits and hospital readmissions that occur when a patient’s condition destabilizes in the weeks or months between scheduled appointments.
Compounding the clinical challenge is a severe shortage of specialized medical professionals. ARPA-H reports that nearly half of all U.S. counties lack a practicing cardiologist. For patients in rural or underserved urban communities, the "care gap" is not merely a matter of convenience but a life-threatening barrier. Without regular access to specialists who can adjust medications—such as beta-blockers, ACE inhibitors, and diuretics—patients often experience a gradual decline in heart function that goes unnoticed until it reaches a point of acute crisis. The ADVOCATE program is designed to function as a "digital member of the care team," providing the continuous oversight necessary to prevent these escalations.
Defining the Shift to Agentic AI
At the core of this initiative is a transition from traditional generative AI to "agentic AI." While standard AI models are typically designed to respond to prompts or identify patterns in data, agentic AI is characterized by its ability to perceive its environment, reason through complex scenarios, and take autonomous actions toward a specific goal. In a clinical context, an agentic system does not just summarize a patient’s vitals; it interprets those vitals against the patient’s medical history, recognizes signs of fluid retention or hemodynamic instability, and initiates a protocol-driven response, such as recommending a dosage adjustment or scheduling an urgent telehealth consultation.
Haider Warriach, the ADVOCATE program manager and a practicing cardiologist, emphasized that the goal is to create a system that can provide care autonomously while knowing exactly when to engage a human clinician. This "human-in-the-loop" model ensures that the AI acts as an extension of the physician’s expertise rather than a replacement, allowing doctors to focus their limited time on the most complex cases while the AI manages the routine, high-frequency monitoring of stable patients.
A Three-Pillar Strategy for Development and Deployment
The ADVOCATE program is organized into three distinct technical areas, each tasked with a specific component of the system’s lifecycle, from initial coding to real-world integration.
Pillar One: The Patient-Facing Agent
The first team is responsible for building the core agentic AI that interacts directly with patients. This group includes Tempus AI, a leader in data-driven precision medicine, and Updoc, a platform specializing in automated clinical workflows. They are joined by Atman Health. This team’s primary objective is to develop a system capable of supporting heart failure patients in their homes. This involves processing data from wearable devices, remote monitoring tools, and patient-reported symptoms. A critical requirement for this group is the submission of a regulatory package to the U.S. Food and Drug Administration (FDA) for authorization within a strict two-year window, reflecting the urgency of the mission.
Pillar Two: The Supervisory "Guardrail" System
Recognizing the safety risks inherent in autonomous medical software, the second pillar is led by Stanford University. This team is tasked with building a supervisory AI system—an "AI for the AI"—that monitors the patient-facing agents. This system is designed to be disease-agnostic and will focus on detecting "out-of-distribution" (OOD) behavior. OOD refers to instances where the AI receives data that is significantly different from its training set, which can lead to "hallucinations" or unsafe medical recommendations. By providing an independent layer of oversight, Stanford’s system aims to ensure that the ADVOCATE agents remain within safe clinical boundaries at all times.
Pillar Three: Real-World Implementation and Health Equity
The third group focuses on the practical application of these tools in diverse clinical settings, with a heavy emphasis on rural and underserved populations. Duke University and Kaiser Permanente lead this effort. Duke will test the technology across five different health systems and various rural sites, ensuring the AI is compatible with the two dominant electronic health record (EHR) platforms: Epic and Oracle Cerner. Meanwhile, Kaiser Permanente will deploy the agents across 21 medical centers and more than 260 clinics, embedding the ADVOCATE system directly into the workflows of its integrated care delivery model.
Regulatory Landscape and Safety Oversight
The development of the ADVOCATE program coincides with a pivotal moment for federal regulation of artificial intelligence. The FDA has recently increased its scrutiny of medical devices that incorporate generative AI, seeking public and industry feedback on how to ensure these "black box" systems remain transparent and reliable.
Rick Abramson, the director of the FDA’s Digital Health Center of Excellence, noted that the ADVOCATE initiative provides a vital testing ground for new regulatory frameworks. "Efforts like these will be critical as we think about continuous ongoing monitoring of autonomous clinical AI for both safety and effectiveness," Abramson stated. The agency is particularly interested in how software can be monitored after it has been deployed, as traditional medical device regulations are often ill-suited for software that evolves or learns over time.
To provide an additional layer of objective validation, Johns Hopkins University’s Applied Physics Laboratory (APL) has been recruited to provide independent oversight. The APL will assess the technical performance of the AI agents and, more importantly, evaluate the clinical outcomes to determine if the technology actually reduces hospitalizations and improves patient longevity.
Implications for the Future of Healthcare
The successful deployment of the ADVOCATE program could signal a paradigm shift in how chronic diseases are managed globally. Beyond heart failure, the framework established by ARPA-H—combining agentic AI with a supervisory safety layer and deep EHR integration—could be applied to diabetes, chronic obstructive pulmonary disease (COPD), and chronic kidney disease.
From an economic perspective, the program aims to establish clear reimbursement pathways. For AI-driven care to be sustainable, health insurance providers and the Centers for Medicare & Medicaid Services (CMS) must determine how to value autonomous clinical work. By involving major health systems like Kaiser Permanente and Duke, ARPA-H is positioning the ADVOCATE program to provide the data necessary to justify these new payment models.
Furthermore, the initiative addresses the "burnout" crisis currently affecting the medical workforce. By automating the routine aspects of chronic disease management, such as tracking daily weights or adjusting diuretic doses, the AI can alleviate the administrative and cognitive load on cardiologists and primary care physicians. This allows the human workforce to operate at the "top of their license," focusing on diagnostic dilemmas and patient counseling while the AI handles the "over-watch."
As the two-year deadline for FDA submission approaches, the medical and technology communities will be watching closely. The ADVOCATE program is not merely a test of whether AI can understand medicine, but a test of whether a federal agency can coordinate a complex ecosystem of private and public partners to solve a systemic failure in the American healthcare delivery model. If successful, the agentic AI developed by Tempus AI, Updoc, and their partners may become the new standard for "care-agile" teams, ensuring that no patient is ever truly "between visits."
