Johns Hopkins Medicine (five-hospital multi-site study)

Sepsis contributes to 1 in 2–3 hospital deaths, and early identification is critical to reducing mortality. Despite their importance, clinical decision support tools typically see adoption rates in the low double-digits, limiting their real-world impact. Existing tools lacked the precision, sensitivity, and workflow integration needed to drive meaningful clinical action at scale.

Bayesian Health deployed the TREWS (Targeted Real-time Early Warning System) across five hospitals including Johns Hopkins, integrating directly into EMR workflows. Over a two-year study period involving 2,000+ providers, the platform was continuously monitored and tuned to account for real-world variations in patient populations and care team behaviors. The system achieved 80%+ sensitivity while maintaining high precision (1 in 3 alerts confirmed by providers).

Provider adoption reached a sustained 89%, far exceeding the industry norm of low double digits. The system enabled 1.85 hours faster life-saving sepsis treatment. Three peer-reviewed studies published in Nature Medicine demonstrated that higher adoption was directly associated with significant reductions in patient mortality, morbidity, and hospital cost.

• High-precision alerting (1 in 3 alerts actionable) is essential to building physician trust and achieving sustained adoption.
• Direct EMR workflow integration was a key driver of the 89% adoption rate.
• Clinically deployed AI can demonstrably reduce sepsis mortality when adoption is high and the model is continuously monitored for drift.