An artificial intelligence-powered learning health system to improve sepsis detection and quality of care: a before-and-after study
Jérémie Despraz, Raphaël Matusiak, S. Nektarijevic, Valerio Rossetti, François Bastardot, Rachid Akrour, Andreas Konasch, Émeline Gauthiez, Olivier Pignolet, Santino Pepe, Jean-Daniel Chiche, Daniel E. Kaufmann, Thierry Calandra, Jean Louis Raisaro, Sylvain Meylan, Alicia Cancela Costa, Emmanouil Glampedakis, Laurence Rochat Stettler, Sebastien Vingerhoets, Coralie Galland-Decker, Alain Junger, Isabelle Lehn, Matthias Roth-Kleiner
Abstract
Sepsis is a major global health crisis where early recognition and effective management remain significant challenges for healthcare systems. As part of the Lausanne University Hospital sepsis quality of care program, we developed and validated an Artificial Intelligence (AI)-powered Sepsis Learning Health System (SLHS) to enhance sepsis care. The SLHS combines a standardized clinical pathway with HERACLES, an AI algorithm that retrospectively classifies patient data into confirmed, possible, or invalidated sepsis cases every 6 h. Predictions inform dynamic dashboards displaying quality-of-care indicators to guide clinical interventions. Analysis of 97,559 stays in wards using the SLHS and 25,851 stays in control wards showed that in-hospital and 90-day mortality decreased for HERACLES-flagged sepsis in SLHS wards, while control wards did not. Further, sepsis coding increased in SLHS wards but did not change in control wards. This real-world example demonstrates how clinician-integrated AI systems can improve sepsis detection and outcomes.