Cedars-Sinai team targets the hours before sudden cardiac arrest with real-time AI risk modeling
More than 600,000 cardiac arrests occur in the United States each year, and survival remains low both inside and outside the hospital. The new research funded through the American Heart Association’s Cardiac Arrest Research Team network is aimed at a problem cardiology still cannot solve reliably: identifying which patients are entering the danger window before out-of-hospital cardiac arrest happens.
The project, A Personalized and Scalable Early Warning System for Prediction of Imminent Out-of-Hospital Cardiac Arrest, will be led by a Cedars-Sinai Medical Center team in Los Angeles. Sumeet Chugh, M.D., will serve as principal investigator, with Kyndaron Reinier, M.P.H., Ph.D., as co-principal investigator. Their work is being carried out under the newly formed CART Network, a collaboration between the American Heart Association and Heart & Stroke™, the Heart and Stroke Foundation of Canada.
This is not a surveillance tool for abstract risk over years. It is designed to detect the short fuse: the symptoms, physiologic changes and behavioral signals that may appear in the hours or days before collapse.
Multiple data streams may expose a pre-arrest physiologic pattern
The project will combine several distinct data sources into one predictive system, including symptoms such as chest pain or trouble breathing, Fitbit activity tracker data, medical history, blood tests and electrocardiogram data. Artificial intelligence will be used to synthesize those inputs and update risk in real time.
That approach reflects the basic problem in sudden cardiac arrest prevention. The event often arrives without enough warning for conventional clinic-based assessment to intervene. A single office visit, a static risk score or an isolated ECG cannot capture the dynamic shifts that precede arrest in real life. A falling activity level, worsening dyspnea, evolving symptoms and laboratory or electrical abnormalities may together signal instability long before a patient reaches emergency care.
Wearable data may be especially useful because it fills in the gaps between encounters. Daily movement patterns, sleep disruption or abrupt drops in exertion can provide a continuous window into decompensation, even when the patient has not yet presented for medical attention. When paired with symptom reporting and clinical history, those measurements may help distinguish a routine fluctuation from a meaningful deterioration in cardiovascular status.
A real-time model is also attractive because cardiac arrest risk is not fixed. It can change quickly as ischemia, arrhythmia, heart failure stress or another acute trigger develops. An algorithm that continually recalibrates risk may be better suited to this unstable physiology than an approach built on outdated snapshots.
The CART Network is designed to speed translation, not just discovery
The grant is the newest scientific award within the CART Network, which launched in June 2026 to advance prediction, detection, treatment and recovery after cardiac arrest. The network brings together teams in the United States and Canada to coordinate expertise, data resources and research methods, with the goal of translating discoveries into practice more quickly.
In practical terms, the network is trying to bridge a familiar gap in cardiovascular science. Many ideas for earlier detection never make it beyond local datasets or single-institution pilots. A coordinated research structure creates a path for shared methodology, broader validation and eventually a tool that could be used beyond one academic center.
The California team will operate under a newly designated identity as the K. Scott Gordon Memorial Research Team for Cardiac Arrest Prediction and Prevention, supported by a donor gift to the Association. That naming reflects philanthropic backing, but the clinical objective remains straightforward: identify impending arrest before a patient becomes pulseless.
Why an early warning system could change bedside response
Sudden cardiac arrest is a race against time, and the margin for effective treatment is measured in minutes. Manesh R. Patel, M.D., the American Heart Association’s 2026-2027 volunteer president, framed the scientific need in direct terms: seconds matter when lifesaving treatment must arrive at the right time.
A system that can flag rising risk in near real time could influence several points in care. It could prompt urgent reassessment, accelerate emergency evaluation, or support targeted intervention before the patient deteriorates further. In the outpatient setting, it could also help clinicians triage symptoms that might otherwise be dismissed as nonspecific until the event has already occurred.
The biological rationale is sound. Cardiac arrest often sits at the end of a cascade rather than appearing from nowhere. Electrical instability, myocardial ischemia, hemodynamic stress and worsening systemic illness can produce subtle changes before collapse. The challenge has always been recognizing that pattern early enough to act on it.
A model that merges symptoms, wearable activity, prior medical history, blood testing and ECG data may do more than count risk factors. It may detect trajectory. In sudden cardiac arrest prevention, trajectory is the clinically meaningful signal.
Research funding remains central to the Association’s strategy
The American Heart Association says it has now funded more than $6.3 billion in cardiovascular, cerebrovascular and brain health research since 1949, making it the single largest non-profit, non-government supporter of heart and brain health research in the United States. The new grant fits squarely into that long-term research strategy: support projects that can move from conceptual promise to real-world clinical utility.
The Association’s broader claim is not about prestige; it is about the mechanism by which new science reaches patients. Basic discovery, computational modeling, shared infrastructure and donor support all feed the same pipeline. Without that sequence, even strong ideas can stall before they influence emergency response or prevention programs.
The unanswered clinical question is whether a multimodal AI system can identify impending out-of-hospital cardiac arrest with enough accuracy, timeliness and usability to change outcomes in routine care. The project now has funding, a research network and a defined target. The hard work is proving that the signal is strong enough to guide action before the arrest happens.