ARPA-H ADVOCATE Program: Building the World's First FDA-Authorized Heart Failure AI Agent
Introduction: A Regulatory Breakthrough for Healthcare AI Agents
On September 9, 2026, the Advanced Research Projects Agency for Health (ARPA-H) announced the launch of the ADVOCATE program (Agentic AI-EnableD CardioVascular CAre TransfOrmation), a four-year initiative with a total investment of $62.7 million. Its core objective is to develop and obtain FDA authorization for the world's first patient-facing proactive AI agent specifically designed for 24/7 heart failure care.
The significance of this program extends far beyond the technical realm—it represents the formal endorsement by the US federal government of a regulatory framework transition for healthcare AI agents from "assistive tools" to "autonomous clinical systems."
Program Background: Systemic Gaps in Heart Failure Care
Medical Resource Deserts in Rural Areas
The core driver of the ADVOCATE program is the severe shortage of cardiology resources in rural America. According to ARPA-H's assessment:
- Numerous rural US counties lack resident cardiologists
- Heart failure patients face 30-day readmission rates of 20-25% after discharge
- Rural patients often must drive hours to access specialist care
- Existing telemedicine solutions cannot provide continuous personalized monitoring
The "always-on" nature of AI agents makes them an ideal solution for filling this gap.
Regulatory Environment Maturation
In October 2026, the FDA completed an important regulatory milestone: approving the first AI device capable of providing diagnostic recommendations without requiring clinician concurrence (SKOUT Autonomous). The establishment of this "autonomy carve-out" template laid the regulatory foundation for ADVOCATE's FDA authorization application.
Three-Tier Architecture: From Patient Interaction to Large-Scale Deployment
The ADVOCATE program employs a carefully designed three-tier architecture to ensure clinical safety and scalability:
Tier 1 (TA1): Patient-Facing Clinical AI
Executing institutions: Atman Health, Tempus AI, Updoc
This tier is the interface for direct patient interaction, responsible for:
- Symptom monitoring: Continuously tracking heart failure symptoms through wearable devices and patient-reported data
- Medication titration support: Automatically adjusting medication dosage recommendations based on patient status (requiring final physician confirmation)
- Patient education: Providing personalized disease management guidance and lifestyle recommendations
- Emergency escalation: Automatically referring patients to human medical teams when high-risk signals are detected
Tempus AI has previously accumulated extensive oncology AI data; its expansion into cardiology will bring important multimodal data integration capabilities.
Tier 2 (TA2): Supervisory AI Agent
Executing institution: Stanford University
Stanford University is responsible for developing a disease-agnostic supervisory layer, which is the core safety guarantee of the entire system:
- Real-time performance monitoring: Continuously evaluating the quality and safety of TA1 agent recommendations
- Compute escalation pipeline: Automatically triggering deeper computational resources for secondary verification when potentially unsafe recommendations are detected
- Anomaly detection: Identifying systematic biases or error patterns that may emerge in TA1 agents
- Explainability reporting: Providing clinicians with transparent explanations of AI decisions
This "AI supervising AI" architecture is an important innovation for the safe deployment of healthcare AI agents.
Tier 3 (TA3): Large-Scale Clinical Implementation
Executing institutions: Duke University, Kaiser Permanente
This tier is responsible for moving the system from the laboratory to real clinical environments:
- Diverse environment deployment: Simultaneous deployment in rural clinics, community hospitals, and large medical centers
- EHR integration: Deep integration with mainstream electronic health record systems such as Epic and Cerner/Oracle
- Workflow adaptation: Customizing deployment plans according to different healthcare institutions' workflows
- Outcome evaluation: Establishing rigorous clinical outcome tracking frameworks
Kaiser Permanente, as one of America's largest integrated healthcare systems, ensures validation of the system in large-scale real-world environments.
FDA Authorization Pathway: The 24-Month Regulatory Challenge
The ADVOCATE program has set an ambitious goal: submitting an FDA authorization application within 24 months of contract award. This timeline faces multiple challenges:
The "Autonomy Carve-Out" Regulatory Framework
The "de novo autonomy carve-out" template established by the FDA in October 2026 provides a regulatory pathway for the ADVOCATE program. This framework requires:
- Strict indication limitations: Clearly defining the operational scope and boundaries of the AI agent
- Mechanical performance thresholds: Setting specific performance metrics that trigger human intervention
- Mandatory post-market registries: Continuously tracking real-world safety and efficacy data
- Pre-committed "stopping rules": Automatically suspending the system when performance drift exceeds thresholds
Complementary Regulation from HHS and ONC
Beyond the FDA, the Department of Health and Human Services (HHS) and the Office of the National Coordinator for Health Information Technology (ONC) are simultaneously advancing complementary regulatory frameworks:
- HHS: Developing rules regarding clinician liability, scope-of-practice boundaries, and human-in-the-loop escalation triggers
- ONC: Establishing a federal registry of clinical decision support algorithms through the HTI-2 enforcement framework, ensuring all autonomous AI deployments operating within certified EHRs are auditable
Lessons for the Asia-Pacific Region
The ADVOCATE program holds important reference value for healthcare AI development in the Asia-Pacific region. APAC faces similar challenges:
- Unequal rural healthcare resources: Rural areas in China, India, and Southeast Asia face severe shortages of cardiology specialists
- Accelerating population aging: Aging trends in Japan, South Korea, and China are rapidly increasing the number of heart failure patients
- Healthcare system pressure: Limited medical resources must serve large patient populations
Regulatory authorities in APAC countries including China, Japan, and Australia are closely monitoring the FDA's autonomous AI authorization framework, with similar local regulatory pathways expected to emerge in 2027-2028.
Technical Challenges: From Clinical Trials to Real-World Deployment
The core technical challenges facing the ADVOCATE program include:
Data heterogeneity: EHR data formats, quality, and completeness vary enormously across different healthcare institutions, requiring robust data standardization capabilities.
Individual variation adaptation: Heart failure patient conditions are highly individualized; AI agents need to balance population models with individual adaptation.
Edge case handling: Rural environments may present unstable network connectivity and device limitations; the system needs offline degradation capabilities.
Cultural and linguistic adaptation: Deployment in diverse ethnic communities requires consideration of language barriers and cultural differences in patient interaction.
Industry Impact: Regulatory Precedent for Healthcare AI Agents
The most important significance of the ADVOCATE program lies in establishing a replicable regulatory precedent for FDA authorization of healthcare AI agents. If successful, this framework could be applied to:
- Diabetes management AI agents
- Chronic obstructive pulmonary disease (COPD) monitoring systems
- Mental health crisis intervention agents
- Post-surgical recovery monitoring platforms
Conclusion
ARPA-H's ADVOCATE program represents a historic leap for healthcare AI from "diagnostic assistance tools" to "autonomous clinical agents." Its three-tier architecture design, rigorous safety supervision mechanisms, and clear FDA authorization pathway provide an important model for the responsible deployment of healthcare AI agents globally.
For healthcare institutions and regulators in the Asia-Pacific region, closely tracking the progress of the ADVOCATE program will provide valuable reference experience for building local healthcare AI agent regulatory frameworks.
