SHM Converge Scientific Abstract ePoster Gallery-Creation of an Artificial Intelligence Tool to Screen Surgical Patients

Pdf Summary

The document describes the development and deployment of the Surgical Screening Tool (SST), an EHR-integrated, LLM-powered, human-in-the-loop workflow designed to automate screening of surgical patients for surgical co-management (SCM) consultation. SCM pairs hospitalists with surgical teams to manage medically complex perioperative patients, improving outcomes, shortening hospital stays, and reducing costs. However, manual screening is time-consuming and inconsistent across providers.<br /><br />The SST uses clinical criteria related to perioperative morbidity, preoperative documentation, and structured surgical data to classify patients as “Affirmative,” “Negative,” or “Maybe” for SCM consultation. Since deployment, the tool has screened 6,221 surgical cases in real time and recommended consultation for 1,588 cases (25.5%).<br /><br />Against treating physicians’ final decisions as the reference standard, the SST showed high sensitivity (93.6%; 95% CI 90.7%–96.3%) and moderate specificity (73.8%; 95% CI 70.7%–76.9%). The authors note that some apparent false positives may not truly be unnecessary, because many patients initially declined by physicians were later referred to SCM during the same admission.<br /><br />Analysis of discordant cases suggested that many disagreements were due not to LLM failure, but to incomplete screening criteria, missing documentation, provider practice variation, or workflow-related human/structural errors. Some “false positives” may even indicate the tool identified patients clinicians initially overlooked.<br /><br />Overall, the study concludes that an LLM-powered, human-in-the-loop screening system can safely and accurately triage surgical patients for specialty consultation, and may augment or eventually automate labor-intensive clinical workflows. Future work will examine the tool’s impact on patient outcomes, operational efficiency, clinician EHR time, and total cost of care.

Asset Subtitle

Jane Wang

Meta Tag

Author List Abby Pandya, April S. Liang, Jane Wang, Jerry Liu, Jonathan H. Chen, Kevin Schulman, Natasha Steele, Rita Pandya, Stephen P. Ma, Timothy Keyes

Category Innovations

Concept Surgical Screening Tool

Concept Surgical Co-management Consultation

Concept Perioperative Care

Concept Unstructured Clinical Note

Concept Medication List

Distinguished Finalist

Presenter Organization Stanford University School of Medicine

Presenting Author Jane Wang

Track Perioperative

Keywords

Surgical Screening Tool

EHR-integrated

LLM-powered

human-in-the-loop

surgical co-management

perioperative screening

clinical decision support

sensitivity and specificity

hospitalist consultation

workflow automation

Surgical Screening Tool

Surgical Co-management Consultation

Perioperative Care

Unstructured Clinical Note

Medication List