A Novel Low-Code Platform for LLM-Powered Healthcare Informatics Workflows

Presentation Type: Podium Abstract
Presentation Time: 02:00 PM - 02:12 PM

Abstract Keywords: Artificial Intelligence, Usability, Information Extraction, Large Language Models (LLMs), Informatics Implementation, Workflow, Data transformation/ETL, Governance
Programmatic Theme: Clinical Informatics

Large language models (LLMs) have proven useful for a broad range of tasks relating to the processing of clinical data, and their flexibility and capabilities make them attractive tools for use in informatics workflows. However, there exists a gap between a functional LLM prompt and a functional application in an operational healthcare setting. We present a low-code informatics platform that removes this barrier and empowers users to leverage LLMs for chart review-based tasks at scale.

Speaker(s):
Steven Bedrick, PhD
Oregon Health & Science University

Author(s):
Steven Bedrick, PhD - Oregon Health & Science University; Robert Gale, MS - Oregon Health & Science University; Ivana Jankovic, MD - OHSU;

Multi-Agent Adjudication Enhances Vaping Cessation Stage Classification in Reddit Discourse

Presentation Type: Podium Abstract
Presentation Time: 02:12 PM - 02:24 PM

Abstract Keywords: Large Language Models (LLMs), Natural Language Processing, Data Mining
Programmatic Theme: Public Health Informatics

Automated Transtheoretical Model staging of short-form social media text using large language models remains unstable, particularly across early stages. We evaluated 36 configurations (30 prompt-based and 6 multi-agent) across six large language models using a clinician-annotated Reddit dataset (n=1,184). Collapsing early stages and implementing structured multi-agent adjudication significantly improved macro-averaged F1 and No Stage discrimination. The best configuration achieved macro-F1 = 0.862 and κ = 0.818, demonstrating scalable, reliable behavioral staging for digital cessation interventions.

Speaker(s):
Lucas Aust, PhD Student
Hi3 Tech Lab

Author(s):
Lucas Aust, PhD Student - Hi3 Tech Lab; Anthony Fu, High School - Dutch Fork High School; DIAN HU, PhD - University of Maryland School of Medicine; Erin Kasson, MS, MSW - Washington University in St. Louis; Li-Shiun Chen, PhD - Washington University in St. Louis,; Patricia Cavazos-Rehg, PhD - Patricia Cavazos-Rehg4; Hongfang Liu, PhD - University of Texas Health Science Center at Houston; Ming Huang, PhD - UTHealth Houston;

Agentic Triage of Documentation Ambiguity in a Large Language Model-Based Colonoscopy Surveillance Clinical Decision Support System

Presentation Type: Paper - Student
Presentation Time: 02:24 PM - 02:36 PM

Abstract Keywords: Clinical Decision Support, Large Language Models (LLMs), Healthcare Quality, Evaluation, Natural Language Processing, Usability, Workflow
Programmatic Theme: Clinical Informatics

Reliable extraction of surveillance-relevant variables from colonoscopy and pathology reports is essential for colonoscopy surveillance clinical decision support (CDS). In prior work, we developed an LLM-based CDS pipeline and observed that residual extraction failures were primarily associated with ambiguity in source documentation. In this study, we evaluated five large language models to characterize the documentation-driven limits of single-pass LLM extraction and developed an agentic workflow for ambiguity detection and triage. In a bi-institutional cohort of 863 cases, shared errors across all five models indicated persistent documentation-driven failures not resolved by model selection alone. Among 20 non-urgent cases with shared cross-model errors, confidence-based flagging identified 55.0% at an 11.2% flag rate, whereas agentic triage identified 80.0% at a 22.1% flag rate and reduced error escape from 45.0% to 20.0%. These results support ambiguity-aware agentic triage as a practical safety layer for LLM-based CDS.

Speaker(s):
Ashwin Rao, MD
Baylor College of Medicine

Author(s):
Xiaoqian Jiang, PhD - University of Texas Health Science Center at Houston; Aman Bali, MD - Baylor College of Medicine; Vinh Tran, MD - Baylor College of Medicine; Shaleen Vasavada, MD - UTHealth Houston; Jinna Chu, MD - Baylor College of Medicine; Fasiha Kanwal, MD - Baylor College of Medicine; Hashem El-Serag, MD - Baylor College of Medicine; Craig Rusin, PhD - Baylor College of Medicine;

Agentic AI Framework for Longitudinal Rheumatoid Arthritis Ascertainment

Presentation Type: Podium Abstract
Presentation Time: 02:36 PM - 02:48 PM

Abstract Keywords: Large Language Models (LLMs), Natural Language Processing, Information Extraction, Artificial Intelligence
Programmatic Theme: Clinical Research Informatics

Early rheumatoid arthritis (RA) detection is critical to prevent irreversible structural joint damage. This study presents a novel agentic AI framework to chronologically process longitudinal electronic health records. Deploying one orchestrator and nine specialized sub-agents , the system extracts both 1987 and 2010 ACR/EULAR diagnostic criteria.The framework achieved an accuracy of 81.1% to 87.4% and an F1 score of 0.855 to 0.902 , demonstrating significant potential to minimize diagnostic delays compared to manual nurse abstraction.

Speaker(s):
Xingyi Liu, Ph.D.
Mayo Clinic

Author(s):
Xingyi Liu, Ph.D. - Mayo Clinic; Roslin Jose George, M.B.B.S., M.P.H. - Mayo Clinic; Mounika Pusa, Bachelor - Mayo Clinic; Sunghwan Sohn, PhD - Mayo Clinic; Cynthia S. Crowson, Ph.D. - Mayo Clinic;

A Skill-Oriented Agent Framework for Meta-Analysis in Systematic Reviews

Presentation Type: Podium Abstract
Presentation Time: 02:48 PM - 03:00 PM

Abstract Keywords: Information Retrieval, Artificial Intelligence, Large Language Models (LLMs)
Programmatic Theme: Translational Bioinformatics

Systematic reviews and meta-analyses require specialized statistical and programming expertise, limiting accessibility for many researchers. We propose a skill-oriented agent framework in which meta-analysis knowledge, workflows, and reusable code are encapsulated in skill specifications for centralized management and reuse. An orchestrator agent maps user queries to skills and iteratively generates executable code to construct reproducible analyses. A prototype demonstrates the feasibility of agent-assisted evidence synthesis.

Speaker(s):
Syed Arsalan Ahmed Naqvi, MD
Mayo Clinic

Author(s):
Kaneez Zahra Rubab Khakwani, MBBS - University of Arizona; Syed Arsalan Ahmed Naqvi, MD - Mayo Clinic; Muhammad Ali Khan, MD - Mao Clinic; Zaryab Bin Riaz, MD - University of Arizona; Bryan Rumble, MSc - American Society of Clinical Oncology; Thomas K. Oliver, BA - American Society of Clinical Oncology; Jeremy Warner, MD, MS - Brown University; Huan He, Ph.D. - Yale University; Irbaz Riaz, MD, MS, MBI, PhD - Mayo Clinic;

A Self-triage Chatbot Grounded by Medical Flowcharts

Presentation Type: Podium Abstract
Presentation Time: 03:00 PM - 03:12 PM

Abstract Keywords: Public Health, Large Language Models (LLMs), Delivering Health Information and Knowledge to the Public
Programmatic Theme: Consumer Health Informatics

Online health resources and large language models are increasingly used for medical decision support, yet concerns remain regarding reliability and interpretability. We present TriageMD, a conversational self-triage system guided by clinician-validated flowcharts from American Medical Association. Using a multi-agent architecture, the system retrieves and navigates triage flowcharts during conversations. Evaluation with simulated datasets shows 84.1% retrieval accuracy and 99.1% navigation accuracy, demonstrating a promising direction for AI-assisted decision support that combine LLMs with structured protocols.

Speaker(s):
Yujia Liu, MS
University of California, San Diego

Author(s):
Sophia Yu, BS - University of California, San Diego; Hongyue Jin, MS - University of California, San Diego; Jessica Wen, MD - Kaiser Permanente; Alexander Qian, MD - University of California San Francisco; Terrence Lee, MD - University of California San Francisco; Mattheus Ramsis, MD - University of California, San Diego; Gi Won Choi, MD, PhD - Korea University Ansan Hospital; Lianhui Qin, PhD - University of California, San Diego; Xin Liu, PhD - Google Research; Edward Wang, PhD - University of California, San Diego;