An artificial intelligence-powered learning health system to improve sepsis detection and quality of care: a before-and-after study

Comprehensive Summary

This study developed an AI-powered Sepsis Learning Health system (SLHS) to improve sepsis care by enhancing sepsis identification, monitoring, and quality improvement. The SLHS integrates an AI-driven digital monitoring system with a standardized clinical pathway. The AI algorithm, HERACLES, retrospectively generates probability distributions for sepsis classification, including no sepsis, possible sepsis, and confirmed sepsis. These predictions are used to support targeted interventions and monitor adherence to evidence-based sepsis care. The study found a significant decrease in in-hospital and 90-day mortality in SLHS wards, as well as improved time-to-antibiotics among patients managed according to the sepsis pathway. HERACLES consistently identified ~9-10% of sepsis cases across SLHS wards and 5% within the emergency, medicine, and gastrointestinal surgery wards, suggesting improved sepsis detection. Limitations of the system include the retrospective operation of HERACLES rather than real-time prediction and the lack of implementation in intensive care units.

Outcomes and Implications

Sepsis care remains challenging due to the difficulties in early recognition and consistent implementation of evidence-based management. This study demonstrates that an AI-Powered Sepsis Learning Health System can help standardize sepsis care through continuous monitoring and sepsis probability classification, contributing to reduced sepsis-related mortality.