Four Stanford Medicine researchers have been awarded Advanced Research Projects Agency for Health contracts.

ARPA-H is an agency within the U.S. Department of Health and Human Services that advances high-potential, high-impact biomedical and health research that cannot be readily accomplished through traditional research or commercial activity. Increasingly, these projects include the development of artificial intelligence features to enhance health care outcomes or to create the monitoring tools necessary for safe implementation of AI in health care settings.

Roxana Daneshjou, MD, PhD, assistant professor of biomedical data science and of dermatology; Euan Ashley, MB ChB, DPhil, the Arthur L. Bloomfield Professor in Medicine, the Roger and Joelle Burnelle Professor in Genomics and Precision Health, and a professor of biomedical data science; and Akshay Chaudhari, PhD, associate professor of radiology and of biomedical data science, will lead a team that is creating an AI monitoring system to ensure that autonomous AI agents used in health care settings are clinically reliable and operationally feasible. Purvesh Khatri, PhD, professor of medicine, will develop an AI-based digital twin of a patient’s immune response to predict outcomes for that patient in an intensive care unit.

Monitoring agentic AI in clinical settings

While agentic AI systems can enhance access to high-quality care, they require continuous monitoring to ensure that they can achieve safe, trustworthy, and verifiable performance.

A Stanford Medicine team, co-led by Daneshjou, Ashley, and Chaudhari, received up to $14.9 million in ARPA-H funding to create a best-in-class monitoring tool for agentic AI systems being used in cardiovascular care.

The team has proposed a three-stage supervisory system, STEWARD (safety, trust, evaluation, and watchdog for embedded agentic care delivery), which will monitor the AI agents deployed in clinical settings, giving on-site reports to clinical teams. The team hopes that STEWARD will also help create a regulatory paradigm for what safe and effective continuous monitoring systems should look like within health care delivery settings.

Their project is a part of ARPA-H’s Agentic AI-Enabled Cardiovascular Care Transformation program, which is transforming cardiovascular disease management with an agentic “clinician extender” AI system to provide round-the-clock care. The ADVOCATE program is the first attempt to build a U.S. Food and Drug Administration-authorized clinical AI system. The program team brings together academic researchers, product developers, health systems and regulators to work on the same systems.

“We are honored to contribute to ARPA-H’s ADVOCATE program,” Daneshjou said. “Our team will develop AI monitoring tools to ensure that autonomous agents operating in clinical environments remain safe, transparent and accountable.”

The STEWARD system would be the first of its kind, the researchers said, as no systems exist for regulatory-grade monitoring of autonomous health care AI agents, and ensuring that these tools are safely used is paramount.

“We are excited by the potential of AI agentic workflows to improve patient care in cardiology, but the steps toward safe implementation are not yet clear,” Ashley said. “ADVOCATE brings together exactly the right partners to understand what we need to do to safely regulate these powerful tools and deploy them at scale in health care systems.”

Chaudhari added, “We look forward to leading the work on responsible continuous monitoring that marries innovative AI tooling with regulatory pragmatism. Alongside our partners at OpenAI, University of Washington and Tuple Health, we are well equipped to not only make technical breakthroughs, but also to help create new regulatory paradigms.”

Improving critical care by making a digital twin

A team led by Purvesh Khatri, PhD, professor of medicine and a member of the Institute for Immunity, Transplantation and Infection, was awarded up to $41.4 million to direct a part of ARPA-H's Critical Illness Immunological Reprogramming and Control Point Learning Engine (CIRCLE) program, which is aimed at improving the quality of care for patients in intensive care units.

Patients with critical illnesses, such as sepsis or trauma, are treated without timely insights into their rapidly changing immune function and organ-injury trajectories. There are no real-time, multivariate immune readouts available to quantify a patient’s immune status in the ICU, and more than 100 drug trials attempting to treat these patients have failed. The goal is to use digital twin technology to identify specific immunotherapies a patient will respond to and improve outcomes.

The team is developing the Immune Mapping and Perturbation for Real-time Identification and Treatment collaborative, alongside commercial partners Inflammatix Inc. and Planned Systems International.

The IMPRINT team will assay a wide range of immune system and organ-specific markers. These data, as well as electronic medical records, will inform a multimodal mechanistic digital twin that explicitly models the organs affected by critical illness and propose therapeutics approved by the U.S. Food and Drug Administration. “This project will allow us to redefine critical illness based on the underlying immunology, predicting the trajectory a patient is on and identifying appropriate treatment for the patient to improve overall outcomes,” Khatri said. “Through ARPA-H’s CIRCLE program, our research will allow us to bring precision immunology at the point of care.”