SHM Converge Scientific Abstract ePoster Gallery-Enhancing Sepsis Care Using Ai-Driven Care Pathways

Pdf Summary

This proof-of-concept pilot at Tampa General Hospital used electronic health record data, natural language processing, and AI tools to improve sepsis care pathways and patient flow. The team partnered with Palantir Technologies to build a system on the Foundry platform that identified historically benchmarked care milestones beyond the standard CMS sepsis bundle. Early milestones considered included physical therapy, case management, infectious disease consults, antibiotic de-escalation, and timing of imaging and lab work.<br /><br />The pilot focused first on antibiotic de-escalation for sepsis patients who had remained on broad-spectrum antibiotics for more than 72 hours. When this threshold was met, an alert was sent to the antimicrobial stewardship (ASP) team for review. If appropriate, ASP contacted the physician to recommend narrowing, reducing, or stopping antibiotics. The alert system launched in March 2023.<br /><br />Baseline data from 3,651 sepsis patients showed an average length of stay of 27.7 days and an average of 184.9 hours on broad-spectrum antibiotics. After implementation, the alert fired about 30 times per week. Forty percent of alerts resulted in antibiotic de-escalation, and 32% led to duration reduction or discontinuation. In the intervention group, mean length of stay decreased to 19.5 days, a 30% reduction, and mean time on broad-spectrum antibiotics fell to 161.7 hours, a 13% reduction.<br /><br />The project suggests that combining AI with historically benchmarked milestones can standardize sepsis care, improve efficiency, reduce adverse events, support antimicrobial stewardship, and potentially increase bed capacity. Future plans include expanding milestones, automating alert creation, and using large language models to identify additional opportunities for improvement.

Asset Subtitle

Jaimie Weber

Meta Tag

Author List Audra Regan, Chris Conn, Christopher Ponsdomenech, Douglas J. McConnell, Heather C. Petri, Jaimie Weber, Kristen Zeitler, Melissa O'Neal, Nicholas Piccicacco

Category Innovations

Concept Artificial Intelligence

Concept Care Pathway

Concept Sepsis

Concept Electronic Health Record

Concept Natural Language Processing

Distinguished Non-Finalist

Presenter Organization Tampa General Hospital

Presenting Author Jaimie Weber

Track Technology in Hospital Medicine

Keywords

sepsis care

electronic health record

natural language processing

artificial intelligence

antibiotic de-escalation

antimicrobial stewardship

patient flow

Palantir Foundry

length of stay

broad-spectrum antibiotics

Artificial Intelligence

Care Pathway

Sepsis

Electronic Health Record

Natural Language Processing