ARPA-H Awards Contracts for AI That Manages Heart Failure Patients Between Clinic Visits

ARPA-H Awards Contracts for AI That Manages Heart Failure Patients Between Clinic Visits

On September 9, ARPA-H announced contract awards under ADVOCATE, a 39-month program. In the first phase, teams have 24 months to build an AI agent that supports heart failure patients between visits and to submit an FDA medical device authorization application.

Heart failure treatment requires clinicians to adjust medications incrementally, balancing symptoms, blood pressure, kidney function, and drug tolerability. ADVOCATE aims to move part of that recurring work outside of office visits: the agent would communicate with the patient, assist with self-management, and escalate cases to the clinical team. First-phase contracts went to Atman Health, Tempus AI, and UpDoc.

The program simultaneously funds two additional layers of scrutiny. A separate team is building an AI monitor that, once the system is deployed, will search in real time for unsafe recommendations and actions outside known scenarios. The Johns Hopkins University Applied Physics Laboratory is tasked with independently evaluating both the technical outputs and the clinical outcomes.

This scrutiny matters because an algorithm can absorb the habits of a single hospital. A review on medical AI validation explains how patient records carry traces of local protocols, such as the ordering of tests and the makeup of the patient population. A system can then predict diagnoses accurately within one archive yet perform worse at a different institution. UpDoc's plan includes head-to-head comparison of the agent against cardiologists and against usual care.

Atman uses a language model to read medical records and converse with the patient. Treatment decisions are made by a separate system governed by explicit clinical rules; the company states that clinicians will be able to see the conditions and logic behind each recommendation.

UpDoc outlines a 39-month path: independent verification, benchmarking of the agent's decisions against those of cardiologists, then a clinical trial under an FDA investigational device exemption. The team then plans to integrate the system into the electronic health records of partner health systems and compare agent-assisted care with usual care.

The monitor, the independent evaluation, and the comparison of agent-assisted care with usual care form the program's path from contract awards to clinical validation.