Abstract Sepsis, a heterogeneous syndrome characterized by whole-body inflammation caused by the body's response to an infection, is the most expensive and deadly condition treated in hospitals, with over 270,000 cases of sepsis-related deaths in the U.S. alone. The cornerstones of optimal sepsis care are early recognition accompanied by appropriate antimicrobial therapy, and use of evidence-based hemodynamic therapies such as fluid resuscitation and vasoactive medications. While data-driven approaches based on machine learning (ML) have shown promise in finding patterns in high-dimensional clinical data to forecast sepsis among hospitalized patients, there are no clinically validated and FDA-approved clinical decision support (CDS) system that can reliably identify patients at risk of developing sepsis. Moreover, existing ML-based solutions are as good as the quality of the data presented to them, and the presence of outliers and missingness can have deleterious effects on their performance. For instance, it has been suggested that such systems are essentially looking over clinician's shoulders-using clinical behavior as the expression of preexisting intuition and suspicion to generate a prediction. As such, there is a critical need for sepsis prediction tools that can effectively use the routinely collected EHR data, assess prediction confidence, and if needed, take necessary steps to gather additional information to reduce prediction uncertainty and improve diagnostic accuracy without significant demand on the end-users. This project aims to assess the clinical utility, safety, and efficacy of a novel uncertainty-aware sepsis prediction system designed and developed in collaboration between UC San Diego Health and Healcisio Inc., a UCSD start-up focused on scalable development and commercialization of advanced analytical systems in critically care settings. The Healcisio system is explicitly designed to improve compliance with the Centers for Medicaid and Medicare Services (CMS) care protocol for sepsis (the SEP1 bundle) and to address the existing delays and variabilities in determining the sepsis onset time, so that life-saving antibiotics and hemodynamic support can be delivered in a timely fashion. To maintain software quality assurance a quality management system (QMS) will be developed to accompany a 510(k) FDA submission package to demonstrate safety and effectiveness. To enhance hospital quality improvement (QI) teams’ ability to measure impact of earlier recognition and SEP-1 bundle compliance, a novel quality measure (SEP1+) and a causal impact analysis tool is introduced. Ultimately, the novel technologies developed and tested under this project will enhance our ability to use advanced analytics to predict adverse events, assess patients’ response to therapy, and optimize and personalize care at the beside through a rapid-cycle ‘learning healthcare system’ framework.

Narrative The proposed project will make use of artificial intelligence (AI) to collect and analyze data from hospitalized patients with sepsis. We plan to conduct a clinical study to quantify the benefit of an AI assisted care protocol to reduce organ failure in patients with sepsis. We intend to seek regulatory approval for our AI system to maintain quality assurance and ensure patient safety. We have a very strong team of doctors and researchers who work closely together, covering all aspects of the proposed project, which we hope will help us improve the care of hospitalized patients with sepsis.