The U.S. Food and Drug Administration made a significant step in the use of predictive algorithms for sepsis detection care.
The FDA cleared the first AI-based early-warning system designed to detect sepsis in hospitalized patients. The John Hopkins University developed the Targeted Real-Time Early Warning System to use real-time patient data from electronic health records to identify patterns signaling early stages of sepsis.
Once it identifies an at-risk patient, the platform sends automated alerts to notify hospitals and attending physicians to intervene. The researchers behind the tool noted that there’s no other technology that monitors patients for sepsis. They added the field has largely relied on manual suspicion.
“Pre-suspicion screening is what creates lead time, and lead time is what changes outcomes in sepsis. Once a clinician already suspects sepsis, the clock has been running — often for hours or even days,” said lead researcher Suchi Saria, a Johns Hopkins professor and director of the AI & Healthcare Lab. “No other cleared test or device monitors for sepsis prior to clinician suspicion.”
The approval reflects a growing regulatory approval of machine learning and AI tools aimed at reducing preventable deaths and complications. Machine learning–enabled sepsis detection tools could shift sepsis management from reactive recognition to proactive surveillance. For health systems, care teams may anticipate lowered manual surveillance requirements, and more data to support evidence-based clinical decision-making.
AI’s Targeting One Of Healthcare’s Deadliest Conditions
The idea for the Targeted Real-time Early Warning System began with a personal loss. Saria developed the technology after her nephew died from septic shock during a hospital stay.
Sepsis is a challenge that has long frustrated clinicians and health systems. It remains one of the leading causes of hospital deaths. There is no single definitive test to detect it in patients.
For clinicians, it can be difficult to recognize its earliest stages, when treatment is most effective. Yet, early signs are subtle and non specific with most overlapping with other conditions, such a dehydration and stress.
Clinicians are often working with incomplete or delayed information. Patient’s vitals fluctuate and clinicians do not always have lab results or electronic health record data reflecting real-time bedside changes.
To address that problem, Saria and her team developed a system that continuously evaluates electronic health records for early sepsis signs. Their goal was to give clinicians more time to act.
By identifying patients at risk before their condition rapidly deteriorates, the platform aims to enable earlier treatment. It also aims to improve outcomes for a condition that remains one of the leading causes of hospital deaths.
Real-Time Continuous Monitoring To Identify At-Risk Patients For Sepsis
Developers describe the platform as a clinically intelligent electronic health record system that reasons like a clinician. It evaluates patients clinical baselines and identifies patterns that may signal deterioration before symptoms become obvious.
The technology reduced sepsis mortality by 18%, according to researchers. The system also has a built in bias and governance layer. The model monitors its performance across patient populations to identify potential bias and reliability concerns.
The researchers say they designed the technology to support, rather than replace, clinician judgment. When the system flags a patient, it shows the rationale, recommends next steps, and tracks clinician follow-up actions.
The system has expanded to dozens of hospitals across the U.S. The growth reflects increasing interest in AI tools that help identify high-risk patients earlier, before complications become life-threatening.
Federal Research Ecosystem Played A Key Role In Breakthrough Prediction System
Federal and philanthropic funding made development of the AI sepsis detection tool possible.
Johns Hopkins researchers have shared that National Science Foundation support was instrumental in the development of the technology. The NSF provided direct support for the underlying research through its Future of Work at the Human-Technology Frontier initiative.
Bayesian Health later received a NSF Small Business Innovation Research award to help commercialize the technology. The company licensed the Targeted Real-time Early Warning System from Johns Hopkins University. It has since expanded deployment of the system and related clinical prediction tools across hospitals.
Published studies of the Targeted Real-time Early Warning System also acknowledge support from the Gordon and Betty Moore Foundation. The non profit funded the research and evaluation efforts for the system’s development and deployment across health systems.
The team received an Alfred P. Sloan Foundation Research Fellowship.
Saria also held grants from several federal and nonprofit organizations, including the National Institutes of Health, the Defense Advanced Research Projects Agency, the United States Food and Drug Administration, and the American Heart Association.
In 2023, the FDA granted the technology Breakthrough Device Designation. The program aims to accelerate the development and review of devices that address life-threatening conditions. The designation supported broader deployment across health systems, including Cleveland Clinic, MemorialCare, and the University of Rochester Medical Center.
The Tool Offers A Helping Hand To Patients And Clinicians Fighting Sepsis
Sepsis places a high demand on frontline doctors and nurses. The condition requires high-acuity, time-sensitive demands that can strain hospital staffs.
Frontline clinicians frequently make decisions under uncertainty while balancing competing clinical priorities across multiple patients.
Patients with sepsis can deteriorate rapidly. Management requires frequent reassessment, timely diagnostics, and rapid escalation of care when clinical status worsens.
Maintaining continuous situational awareness across high-risk patients is challenging even for experienced teams. The difficulty heightens within crowded emergency departments and inpatient units operating under staffing constraints.
The Targeted Real-time Early Warning System addresses a core challenge in sepsis care and inpatient monitoring. Clinicians often have a narrow window to recognize clinical deterioration and initiate treatment.
Clinicians often have a narrow window to recognize clinical deterioration and initiate treatment.
Proponents describe the technology as a potential shift in standard monitoring practices. Some compare its role in acute care settings to continuous physiologic monitors used in emergency and critical care environments.
Still, questions remain. Integration into clinical workflows raises concerns about alert volume, clinician trust, and how clinicians interpret AI-generated signals alongside existing clinical data streams.
As these systems scale across hospitals, performance across diverse patient populations, workflow fit, and signal-to-noise ratio will be key measures of success.
Ultimately, these tools will need to support high-acuity care environments by reducing cognitive and operational burden without adding additional noise or unnecessary alerts.
