Johns Hopkins University

Sepsis is a life-threatening condition where early detection is critical to survival. Hospitals lacked sufficiently accurate, timely tools to screen patients at risk of sepsis before their condition deteriorated, contributing to preventable mortality and extended hospital stays.

Bayesian Health's Targeted Real-time Early Warning System (TREWS) — an adaptive machine learning platform — was deployed in a clinical setting at Johns Hopkins to provide real-time early warning screening for sepsis. The system was validated across three peer-reviewed studies published in Nature and npj Digital Medicine.

Timely use of the TREWS platform was associated with an 18.2% relative reduction in mortality among sepsis patients. The studies also demonstrated reductions in morbidity and length of stay for hospitalized sepsis patients.

• Peer-reviewed validation across three studies lends strong credibility to real-world clinical AI outcomes.
• Timeliness of clinician response to AI alerts was a key driver of mortality reduction, underscoring that adoption behavior matters as much as model accuracy.
• Adaptive ML can move beyond algorithm-only approaches to meaningfully change patient outcomes when integrated into clinical workflows.