SHM Converge Scientific Abstract ePoster Gallery-Prototype Development of an Ai-Integrated Patient Assignment Workflow

Pdf Summary

This pilot project describes the development of an AI-assisted workflow to improve overnight hospital medicine patient assignment at UC San Diego Health. Currently, nocturnists manually assign morning admissions to day teams at the end of an overnight shift, a cognitively difficult time when fatigue and circadian disruption increase the risk of errors and contribute to burnout.<br /><br />A baseline survey of 18 of 21 nocturnists (86% response rate) showed the process was highly burdensome. No clinicians reported perfect accuracy. Many reported frequent misassignments, with only a minority saying they were correct more than 75% of the time. The task also caused notable anxiety: nearly 40% reported moderate anxiety, about one-third reported significant anxiety or described it as “the worst part of their shift,” and no one reported no anxiety. The work was time-consuming as well, with none completing it in under 10 minutes and more than half needing over 30 minutes.<br /><br />To address this, the team built a prototype using Microsoft Copilot, a HIPAA-compliant large language model powered by OpenAI, to generate patient-to-team assignments from overnight admission lists. They used constraint-aware logic applied to nightly Epic flat-file exports containing census and admission data, and iteratively refined prompts through trial and error. The AI-generated distributions were compared with post hoc human assignments to evaluate performance.<br /><br />The authors conclude that morning patient distribution is inefficient, stressful, and error-prone. They propose that an LLM-based assignment support tool could reduce misassignments, clerical workload, and burnout. Future work will focus on prospective evaluation and integration with Epic and QGenda APIs for more automated, constraint-driven patient distribution.

Asset Subtitle

Alan Moazzam

Meta Tag

Author List Alan Moazzam, Aneesh Kuruvilla, Brian Kwan, Carlos Lago Hernandez, Jeffrey Pan, John Bell, Keisuke Nakagawa, Reid Sasaki, Vineet Gupta

Category Innovations

Concept Patient Assignment

Concept Shift

Concept Misassignment

Concept Manual Distribution

Concept Order Accuracy

Distinguished Non-Finalist

Presenter Organization University of California San Diego Health Division of Hospital Medicine

Presenting Author Alan Moazzam

Track Technology in Hospital Medicine

Keywords

AI-assisted workflow

hospital medicine

patient assignment

nocturnist

overnight admissions

large language model

Microsoft Copilot

Epic data

burnout

workflow automation

Patient Assignment

Shift

Misassignment

Manual Distribution

Order Accuracy