Duke Leads $15.5M ARPA-H SYMPHONY Consortium to Test Heart-Failure Agentic AI

Heart failure is the reason a person over 65 gets readmitted to the hospital more often than almost anything else. The tools meant to help are arriving fast, and the people who study them will tell you the evidence is not keeping up. North Carolina is now the place where that gap gets tested at national scale.

Duke University is the prime awardee for the SYMPHONY Consortium under ARPA-H's ADVOCATE program, with an award of up to $15.5 million listed on ARPA-H award page 4704. The award start date is Sept. 1, 2026. ARPA-H announced the ADVOCATE performer awards Sept. 9, and the Duke Clinical Research Institute published its news release Sept. 10.

The most important thing to know about SYMPHONY is what it is not. It is not the builder of a patient-facing clinical agent. It does not ship an FDA-authorized product today, and neither does anyone else in this program.

Where SYMPHONY sits in the program

ADVOCATE is a four-year, $62.7 million program with up to $33.7 million in year one, framed by ARPA-H as a bid-build test of FDA-authorized clinical AI for cardiovascular care. The work splits into technical areas.

SYMPHONY sits in Technical Area 3: scalable implementation, multi-site validation, evaluation, monitoring, and evidence generation. The patient-facing agent itself is built in TA1 by Atman Health, Tempus AI, and Updoc. TA2, Stanford's STEWARD, covers supervisory AI over those agents. Peer TA3 work includes Kaiser Permanente's ORCHESTRATE-HF.

The division matters. Duke's check is up to $15.5 million, not the program envelope. TA1 performers must submit an FDA-authorization package within 24 months of contract award, and no agentic heart-failure tool in this program is FDA-authorized today.

What the consortium actually does

DCRI says it will lead SYMPHONY to rigorously test agentic AI tools for cardiovascular and heart-failure care. The acronym expands to: SYstematic evaluation and Monitoring of scaled dePloyment and Harmonization of agentic AI tOols to contiNuouslY improve national cardiovascular health.

The design calls for testing across five health systems and rural sites, on Epic and Cerner/Oracle EHRs, amplified by American Heart Association assets including the Get With The Guidelines and Outpace CVD registries and the Precision Medicine Platform. The consortium is not a single-product build. It creates processes and infrastructure to evaluate multiple AI-enabled solutions at scale: patient monitoring, care coordination, medication management, and related workflows.

Named leads include Manesh Patel, MD, chief of cardiology at Duke and SYMPHONY principal investigator and technical lead, who is also listed as president of the American Heart Association; Mariell Jessup, MD, the AHA's chief science and medical officer, consortium co-PI; Raj Ratwani, director of MedStar's National Center for Human Factors in Healthcare, co-PI; and Adam DeVore, MD, Sreekanth Vemulapalli, MD, and Rishikesan Kamaleswaran, PhD, as program investigators.

Patel said in the DCRI release:

"AI is already reshaping medicine with great promise but little proof. The potential is to reach people where they are, but we need evidence that these tools, when used in care, will improve health. Heart patients deserve tools that have been tested with the same rigor we demand for any drug or device. SYMPHONY is building that evidence and the blueprint that lets any health system, urban or rural, deploy these agents responsibly."

DeVore: "I am excited that we are evaluating these tools in patients with heart failure as we know there are gaps in care, particularly around early access to medical therapy for heart failure, that may be addressed with different care models."

Jessup: "Through SYMPHONY, the Heart Association is helping to ensure that these tools are tested against real-world data, evaluated in a variety of care settings and designed to benefit patients in every community. Too often, the patients who could benefit most from new technologies are the last to see them."

The North Carolina piece

Partners named on the DCRI release include Duke Health, Ohio State University Health, Baylor Scott & White Health, Ochsner Health, MedStar-Georgetown, MedStar's human-factors center, rural health partners in multiple states (unnamed in the release), Amazon Web Services, the Health AI Partnership, and the North Carolina Department of Health and Human Services. ARPA-H lists the award location as Raleigh-Durham, NC.

Treat the DHHS role carefully. No standalone NC DHHS press release on SYMPHONY was found in the reporting pass, so DHHS is a listed collaborator, not yet a described one. What the state gets out of this, and what it contributes, is one of the open questions worth following.

What this means for the state

North Carolina's DCRI is the prime for a national ARPA-H evaluation platform on heart-failure agentic AI. That is authority territory: evidence, and rural and urban deployment, not a chatbot launch day.

What to watch

• The $15.5 million attribution. The DCRI release omits the figure. Any dollar claim should cite ARPA-H award page 4704, and the "up to" stays in the sentence.
• The first patient-touching evaluation. Shadow mode versus IDE, and the timeline to it, is not locked anywhere public.
• The rural rollout. Which clinics, which states, and how the rural sites plug into the five health systems.
• The dual role disclosure. Patel serving as both AHA president and SYMPHONY PI, with the AHA as a partner, is a relationship worth watching get handled with care.
• NC DHHS's actual role. Data, Medicaid, rural clinics, governance, or all of the above. No MOU has surfaced yet.

Sources: DCRI news release (Sept. 10, 2026) · ARPA-H award card 4704 · ARPA-H ADVOCATE announcement (Sept. 9, 2026) · ARPA-H ADVOCATE program hub · Fierce Healthcare · Healthcare Dive. Award amounts are "up to" ceilings from ARPA-H. This is not medical advice. SYMPHONY is an evaluation and implementation consortium; program goals are not live clinical performance in North Carolina hospitals.