ARPA-H just bet $62.7 million that an AI can manage heart failure patients around the clock, prescribing drugs and ordering tests without waiting for a doctor. If it works, it could reshape how we treat the fastest-growing heart condition in America.
Your Cardiologist Might Get a Digital Coworker
Imagine calling your doctor at 2 a.m. because your ankles are swollen and you can't breathe. Now imagine an AI picks up instead, assesses your symptoms, adjusts your medication, and orders a blood test before you even finish describing what's wrong.
That's not science fiction. That's the actual goal of a new $62.7 million government program announced this week.
ARPA-H, the federal agency created to fund moonshot health projects, just launched ADVOCATE (Agentic AI-EnableD CardioVascular CAre TransfOrmation). The program's mission: build the first FDA-authorized AI system that can function as a real member of a heart failure care team, working around the clock.
Not a chatbot. Not a glorified symptom checker. An AI agent that could assess how sick you are, recommend or prescribe drugs, and escalate to a human doctor when things get serious.
Why Heart Failure? Because the Numbers Are Brutal
Heart failure is the kind of disease that sounds manageable until you look at the data. About 6.7 million American adults currently live with it, making it the fastest-growing cardiovascular condition in the country. It showed up on more than 452,000 death certificates in 2023 alone.
The older you get, the worse the odds. Among men over 80, roughly 14% have heart failure. And managing it is expensive.
The core problem isn't that we lack good treatments. It's that heart failure requires constant monitoring and frequent medication adjustments, the kind of attentive, always-on care that our healthcare system simply can't deliver at scale. Think of it like owning a high-maintenance car but only being able to visit the mechanic once every three months.
ARPA-H believes AI could fill that gap. If ADVOCATE works as planned, the agency estimates it could prevent enough hospitalizations to save $28 billion annually across the heart failure population.
Get tomorrow's biotech intelligence before your competitors.
Join thousands of biotech professionals who start their day with our free, daily briefing.
ARPA-H picked five teams across three roles, structured almost like positions on a soccer team.
The forwards (patient-facing AI builders): Atman Health, UpDoc, and Tempus AI. These companies are tasked with creating the actual AI agents that interact with patients, assess symptoms, and make clinical recommendations.
The goalkeeper (supervisory AI): Stanford University is building a watchdog system designed to catch unsafe or inaccurate recommendations before they reach patients. Think of it as an AI that watches the other AI, flagging anything suspicious.
The midfielders (real-world deployment): Duke University will test and validate the technology across actual hospitals and clinics, making sure it works outside the lab.
The funding structure is phased. ARPA-H committed up to $33.7 million in year one of a 39-month program, with remaining funds subject to renegotiation based on progress. Translation: hit your milestones or the money dries up.
The FDA Gauntlet
Building an AI that can practice medicine is hard. Getting the FDA to authorize it is arguably harder.
The regulatory path ADVOCATE teams are expected to follow has three stages packed into a tight timeline. First, an FDA pre-submission meeting in year one to align on expectations. Then an Investigational Device Exemption (basically, permission to test the device on real patients) around the 12-month mark. Finally, a De Novo submission within 24 months.
De Novo is the FDA pathway for novel devices that don't have a clear precedent on the market. It's slower and more scrutinized than the standard 510(k) route, which lets companies piggyback on existing products. But when you're building something the FDA has never seen before (an autonomous AI prescribing heart medications, for instance), De Novo is the only realistic option.
The FDA has already authorized roughly 1,450 AI-enabled medical devices, most of them in radiology, cardiology, and neurology. But nearly all of those are decision-support tools: they help doctors interpret scans or flag abnormalities. ADVOCATE is aiming for something fundamentally different, an AI that doesn't just advise but acts.
That distinction matters enormously. The FDA's January 2025 draft guidance for AI medical devices requires sponsors to describe the AI's intended use, its limitations, how it fits into clinical workflow, and how it will be monitored after launch. An August 2025 update introduced an optional Predetermined Change Control Plan, which is essentially a roadmap for how the AI will be updated over time without needing a new FDA submission each time.
The Bull Case and the Bear Case
The optimistic view is compelling. Heart failure patients need 24/7 monitoring that human healthcare systems can't provide. AI could fill that void, reduce hospitalizations, and save billions. Federal backing gives these teams credibility, funding, and a direct line to regulators.
For companies like Tempus AI (already publicly traded) and Atman Health, an ARPA-H award is a powerful stamp of validation. It won't generate revenue tomorrow, but it positions them at the front of the line if agentic clinical AI becomes a real product category.
The skeptical view has teeth, too. Letting an AI prescribe medications raises massive questions about liability, bias, and patient safety. Who's responsible when the algorithm gets it wrong: the company, the hospital, or the supervising physician who may not have been watching? Privacy concerns loom large when wearable data and patient records feed into cloud-based AI systems.
And then there's the execution risk. Two years to go from concept to FDA authorization is wildly aggressive. Most medical device companies spend longer than that just on clinical trials.
Why This Feels Like a Turning Point
ARPA-H has been steadily building an AI healthcare portfolio. In 2024 alone, the agency committed $48.3 million to Every Cure for AI-powered drug repurposing, up to $98.5 million for rare disease diagnostics through its RAPID program, and up to $125 million for biological modeling through CATALYST.
ADVOCATE fits into that pattern but pushes the boundary further. Previous programs used AI to find treatments or identify diseases. This one wants AI to deliver care directly.
If even one of these teams gets FDA authorization within the next two years, it would be a landmark moment: the first AI system cleared to function autonomously in cardiovascular medicine. It wouldn't replace cardiologists. But it might give 6.7 million heart failure patients something they desperately need: a care team member that never sleeps, never takes a vacation, and never forgets to check in.
Deals & M&A5 min read
Roche Just Paid $1.53 Billion for a Drug That's Never Been in a Human
Roche just locked up exclusive global rights to a preclinical trispecific antibody from Chinese biotech Simcere Zaiming for up to $1.53 billion. The drug has never been tested in a human, but it targets B cells in a way Roche's legacy blockbusters can't.
