ARPA-H funds CIRCLE AI to watch ICU immune swings in real time

ARPA-H named the CIRCLE performer teams that will build AI-powered digital twins and sensors so ICU clinicians can see a patient’s immune system swing from overactive to underactive in real time — and intervene before organ damage sticks.

HEALTH desk — ICU immune chaos moves faster than yesterday’s lab result; CIRCLE is a federal bet that digital twins can turn that blur into control points before organ failure and post-ICU syndrome lock in.

Who the program is for. CIRCLE aims to reimagine care for the 4.6 million Americans admitted to intensive care units each year. An intensive care unit, shortened to ICU, is the hospital ward for people sick enough that a regular floor cannot watch them minute to minute. Among those who leave, the page says many struggle with long-term complications such as post-ICU syndrome, a set of debilitating physical and cognitive symptoms that can last for months or years after discharge. The selected teams will build and test tools that give hospital care teams a real-time window into a patient’s immune system, so they can intervene sooner and try to prevent both immediate death and lasting complications. 4.6 million is the agency’s figure. This desk did not count admissions.

Why yesterday’s lab result is the wrong clock. The page says critical illness — a major injury, an infection, or a serious complication of a chronic disease — is complex and moves quickly. Patients’ immune systems frequently swing from overactive to underactive in the time a traditional lab test is still being processed. Overactive means the immune system is attacking too hard, including the patient’s own organs. Underactive means it has gone quiet when the body still needs a defense. Without seeing those changes as they happen, the page says ICU teams typically treat immune-related organ damage only after it occurs, with broad, infrequent interventions. Broad means a treatment aimed at the whole immune system, not one measured switch. Infrequent means the team cannot adjust as fast as the immune system moves. These lines are the agency’s. This desk did not watch an ICU shift.

The death figure, and the aftermath, and whose numbers they are. Without more precise care, the page says, as many as 30 percent of ICU patients die in the hospital. Thirty percent is about three in ten. Among those who survive, many leave with post-ICU syndrome and other lasting complications tied to a higher chance of dying later, not only in the hospital. The syndrome can look different from one patient to the next. It often combines cognitive problems — confusion, poor concentration, forgetfulness — physical symptoms — fatigue, shortness of breath, and muscle weakness — and emotional responses to trauma. Rehabilitation can help manage symptoms and recover abilities. Many patients, the page says, live with serious impairments for months or years. File the 30 percent and that symptom list as the agency’s. This desk did not review a chart.

The program manager, as the agency’s words, not as a saved life. Yoram Vodovotz, Ph.D., CIRCLE program manager, said major injuries and infections usually happen without warning, so everyone is at risk of critical illness. He said patients arrive with unique biology, and care teams have limited tools to see how dangerous the underlying immune dysregulation is, and few chances to individualize treatment. Immune dysregulation means the immune system is outside its normal range. He said the field is overdue for technology that can show that biological detail and give each patient the best chance at survival and long-term health. Through CIRCLE, he said, ARPA-H is stepping in to build the tools ICU teams need. File the name, the Ph.D., the title, and those sentences as the page’s. A quote is not a measured drop in deaths.

What the tools are supposed to be. CIRCLE aims to give care teams real-time monitoring and response tools that point them toward the most effective immunotherapies for one patient. Immunotherapy, here, means a treatment that changes the immune system. That will include AI-powered digital twins of patients’ immune systems. A digital twin, on this page, is a computer model of one patient’s immune system that can try strategies before a clinician acts on the body. The page says those models can address dangerous immune-driven inflammation and interrupt the cascade toward organ failure before it starts. A cascade is a chain of failures, one organ after another. Each Technical Area team will strive for a program goal of a 25 percent shorter ICU stay. Twenty-five percent shorter means a four-day stay would be about three days, if the goal holds. The path the page names is one integrated system: measure parts of the immune system, model that state with digital twins, find actionable immune control points, and modulate those points to improve health. A control point is a place where a treatment could change the course. Modulate means turn that point up or down, not only watch it. Strive, and if successful, are the page’s words. This is a goal. It is not a result this desk measured.

Three Technical Area teams, in the page’s order. The University of Vermont will monitor multiple immune mediators, immune cell function, and neurophysiological state. A mediator is a chemical immune cells use to talk to each other. Neurophysiological state is how the nervous system is behaving, not only what is in the blood. Those readings, plus the electronic medical record, will feed an enhanced digital twin. The model simulates potential novel therapies the U.S. Food and Drug Administration, shortened to FDA, has already approved, for one patient at the level of immune cells. Stanford University will measure a wide range of transcriptomic immune markers and organ-specific markers. Transcriptomic means which genes are switched on, read from the RNA the cells are making. Those data and the medical record will feed a multi-modal mechanistic digital twin that models the organs critical illness hits and proposes FDA-approved therapeutics. Multi-modal means more than one kind of measurement. Mechanistic means the model tries to follow the biology, not only match a past pattern. The team will check predictions on a multi-organoid assembloid platform. An organoid is a tiny lab-grown stand-in for an organ. An assembloid, in the page’s quotation marks, is several of those grown so they interact. Novelna, Inc. will build a dual-channel, multi-plex sensor for high-frequency immune measurements that can be tied to a tissue. Dual-channel means two readout paths. Multi-plex means many immune signals on one device. High-frequency means often, not once a day. Those readings will feed an ensemble digital twin — several model shapes used together — to identify and rank control points for FDA-approved drugs and for new interventions. File the three teams as the page’s. This desk did not visit a lab.

Three Acceleration Platform teams. The page says they will both help and evaluate the Technical Area performers. Verily Health, Inc. will provide a CIRCLE program database. It will pull in public and private critical-illness datasets, plus data the performers generate, for training and checking the models. The team will also build data products meant to become resources for the wider critical-care community. Sage Bionetworks will provide a sandbox where the performers develop, check, and test digital twins. A sandbox is a shared practice environment, not the bedside. The page says it can run simulated clinical trials and use synthetic data from large computer simulations. Synthetic data means numbers a model generates, not a named patient’s chart. The same piece is meant to help teams get ready for future FDA applications covering digital twins and Software as a Medical Device. Software as a Medical Device, shortened to SaMD, is software the regulator treats as a medical product. Vanderbilt University Medical Center will set up adaptive clinical trials of digital-twin-guided immune treatments through its network of partners and trial sites. Adaptive means the trial can change as evidence comes in, under its own rules. The team will also judge whether the devices and software are acceptable and usable in real ICUs. Will, and facilitate, are the page’s. This filing does not say a trial has opened, and it does not name a hospital ward.

The money, and what the ceiling is. The agency’s commitment is up to $144.9 million over five years. Up to is a ceiling, not a check this desk saw. Awards vary by team. They depend on each team meeting aggressive, accelerated research milestones. Contingent means the later money is not guaranteed. If the program works, the page says, critical-care teams would get real-time control of a patient’s immune system, which could limit organ-damaging inflammation, save lives in the ICU, and reduce post-ICU syndrome after discharge. If successful is the agency’s condition. Do not split $144.9 million into six equal checks. The page does not print a dollar figure for any one team.

Plain English for the rest of the card: ARPA-H = the Advanced Research Projects Agency for Health, inside HHS. CIRCLE = Critical Illness Immunological Reprogramming and Control Point Learning Engine. ICU = intensive care unit. 4.6 million = the agency’s count of Americans admitted to an ICU each year. 30 percent = as many as three in ten ICU patients dying in the hospital, on the page’s “as many as” line, without more precise care. post-ICU syndrome = lasting cognitive, physical, and emotional problems after intensive care. digital twin = a computer model of one patient’s immune system. control point = a place in that system where a treatment could change the course. 25 percent = the program’s goal for a shorter ICU stay, not a result. mediator = a chemical immune cells use to talk. transcriptomic = which genes are switched on. assembloid = lab-grown mini-organs combined so a team can check a model. multi-plex = many immune readings on one sensor. ensemble = several models used together. sandbox = a practice environment for the models. synthetic data = numbers a computer generates. SaMD = Software as a Medical Device. $144.9 million over five years = the agency’s ceiling, contingent on milestones. This filing is the 22 Sep award announcement. It is not a drug approval, and it is not a patient outcome.

PRIMARY here: ARPA-H’s 22 Sep 2026 page, “ARPA-H selects teams to transform critical care with new lifesaving technology,” at arpa-h.gov — Tier A PRIMARY, the agency’s own record. The visible date is September 22, 2026. The time element’s datetime is 2026-09-22T19:43:10+00:00. The CIRCLE expansion, the HHS line, the 4.6 million admissions, the post-ICU syndrome description, the overactive-to-underactive swing, the as-many-as-30-percent in-hospital deaths, the symptom list, the Vodovotz title and quotes, the digital-twin and immunotherapy lines, the 25 percent shorter-stay goal, the Vermont, Stanford, and Novelna, Inc. technical plans, the Verily Health, Inc., Sage Bionetworks, and Vanderbilt University Medical Center platform plans, and the up-to-$144.9-million five-year ceiling with milestone conditions are the page’s. NOT claimed: that this desk awarded a contract, counted the 4.6 million, reviewed a chart, measured a 25 percent shorter stay, split the $144.9 million by team, opened a trial, named a hospital, treated “if successful” as a result, a drug approval, a stock tip, or investment advice. Distinct from the already-filed heidi-340m, ramona-optics-25m, and envisagenics-boehringer.