Rule-Based Artificial Intelligence and Workflow to Prompt Early Sepsis Management: A Quality Improvement Project.

Abstract

Introduction: Early identification and management of sepsis improves patient outcomes, yet hospitals struggle to consistently screen patients on arrival and during hospitalization. The Centers for Disease Control published Hospital Sepsis Program Core Elements to guide hospital sepsis management and outcomes improvement efforts and will measure the sepsis core elements with the National Healthcare Safety Network Annual Hospital Survey.

Methods: To further sepsis care management, our community-owned, nonprofit hospital implemented an emergency department quality improvement project, introducing rule-based artificial intelligence (AI) for sepsis identification with a workflow. Objectives were to measure rule-based AI sensitivity, sepsis management compliance, length of stay (LOS), and mortality rate.

Results: A total of 895 cases were included in the final dataset, 370 preimplementation and 525 postimplementation. Postimplementation rule-based AI alerts identified 93.9% (493 of 525) cases for sepsis management interventions. After rule and workflow implementation, combined 3-hour compliance for antibiotic given, blood culture drawn, and lactate measured was 89.5%. Average LOS decreased by 2.3 days (p < .001), and mortality per 100 cases decreased by 22.3% (p = .0998).

Conclusions: Implementing rule-based AI software to identify severe sepsis in conjunction with a sepsis workflow decreased LOS for patients diagnosed with either severe sepsis or septic shock.