- Home
- 2026 Amplify Informatics Conference Program Gallery
- Poster Session 1 and Reception – Sponsored by University of Colorado Anschutz Center for Health AI
Times are displayed in (UTC-06:00) Mountain Time (US & Canada) Change
Custom CSS
double-click to edit, do not edit in source
5/19/2026 |
5:00 PM – 6:30 PM |
Aspen Ballroom
Poster Session 1 and Reception – Sponsored by University of Colorado Anschutz Center for Health AI
Presentation Type: Poster
Description
Join us for an engaging Poster Session where ideas come to life through one-on-one conversations with presenters. Explore a diverse range of topics, learn directly from the researchers behind the work, and dive deeper into the studies that spark your interest. This is your opportunity to connect with others who share your passions, exchange perspectives, and build new professional relationships. Whether you’re looking to gain insights, ask questions, or network with peers, the Poster Session offers a dynamic, interactive environment to expand your knowledge and your professional circle. Sponsored by University of Colorado Anschutz Center for Health AI
Use of AI Technology to Enhance Patient Engagement and Rehabilitation Outcomes
Presentation Type: Poster - Student
Poster Number: 100
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Analytical Artificial Intelligence: ML, Digital Pathology, Imaging AI, Predictive Analytics, Governance, Outcomes Improvement and Equity, Population Health, Digital Therapeutics, Remote Patient Monitoring (RPM), and Digital Engagement, Augmented Reality and Virtual Reality in Care, Clinical Decision Support and Care Pathways, Clinician Well-Being, Human Factors and Usability, Telemedicine, Health at Home, and Virtual Care
Primary Track: Advancing Wellness for Providers and Community with Consideration of Human Factors
An AI-powered, markerless motion-capture platform was evaluated in two hospital rehabilitation clinics to assess its impact on patient engagement and usability. Seventy-nine patients completed over 150 sessions with 214 surveys analyzed. Eighty-eight percent reported increased motivation, 82% greater confidence, and 95% positive usability ratings. Therapist feedback confirmed workflow integration and clinical feasibility. Results demonstrate that accessible, low-cost AI technologies can enhance adherence, patient experience, and data-driven rehabilitation outcomes.
Speaker(s):
Shayen Bhatia, Student
Prime Healthcare
Author(s):
Shayen Bhatia, Student - Prime Healthcare; Luke Riggan, Ph.D - California NanoSystems Institute;
Presentation Type: Poster - Student
Poster Number: 100
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Analytical Artificial Intelligence: ML, Digital Pathology, Imaging AI, Predictive Analytics, Governance, Outcomes Improvement and Equity, Population Health, Digital Therapeutics, Remote Patient Monitoring (RPM), and Digital Engagement, Augmented Reality and Virtual Reality in Care, Clinical Decision Support and Care Pathways, Clinician Well-Being, Human Factors and Usability, Telemedicine, Health at Home, and Virtual Care
Primary Track: Advancing Wellness for Providers and Community with Consideration of Human Factors
An AI-powered, markerless motion-capture platform was evaluated in two hospital rehabilitation clinics to assess its impact on patient engagement and usability. Seventy-nine patients completed over 150 sessions with 214 surveys analyzed. Eighty-eight percent reported increased motivation, 82% greater confidence, and 95% positive usability ratings. Therapist feedback confirmed workflow integration and clinical feasibility. Results demonstrate that accessible, low-cost AI technologies can enhance adherence, patient experience, and data-driven rehabilitation outcomes.
Speaker(s):
Shayen Bhatia, Student
Prime Healthcare
Author(s):
Shayen Bhatia, Student - Prime Healthcare; Luke Riggan, Ph.D - California NanoSystems Institute;
Shayen
Bhatia,
Student - Prime Healthcare
Annotating Patient-Reported Medical Cannabis Treatment Experiences Using Open Source Large-language Models
Presentation Type: Poster - Student
Poster Number: 101
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Analytical Artificial Intelligence: ML, Digital Pathology, Imaging AI, Predictive Analytics, Governance, Health Data Science, Clinical Decision Support and Care Pathways
Primary Track: Advancing Wellness for Providers and Community with Consideration of Human Factors
Medical cannabis use has expanded significantly across the United States, driven by evolving state laws and increasing patient interest in alternative therapies for various health conditions. Despite this growth, there remains a substantial gap in understanding the real-world effectiveness and safety of different cannabis products from the patient’s perspective. In this study, we compare several large language models for their ability to annotate patient-reported medical experiences present in a multi-state survey of over 3500 patients.
Speaker(s):
Richard Boyce, PhD
University of Pittsburgh
Author(s):
Gigi Zheng, Master of Science - Richard David Boyce;
Presentation Type: Poster - Student
Poster Number: 101
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Analytical Artificial Intelligence: ML, Digital Pathology, Imaging AI, Predictive Analytics, Governance, Health Data Science, Clinical Decision Support and Care Pathways
Primary Track: Advancing Wellness for Providers and Community with Consideration of Human Factors
Medical cannabis use has expanded significantly across the United States, driven by evolving state laws and increasing patient interest in alternative therapies for various health conditions. Despite this growth, there remains a substantial gap in understanding the real-world effectiveness and safety of different cannabis products from the patient’s perspective. In this study, we compare several large language models for their ability to annotate patient-reported medical experiences present in a multi-state survey of over 3500 patients.
Speaker(s):
Richard Boyce, PhD
University of Pittsburgh
Author(s):
Gigi Zheng, Master of Science - Richard David Boyce;
Richard
Boyce,
PhD - University of Pittsburgh
Nursing-BERT: Domain-Specific Pretraining on Nursing Notes for Fall-Risk Named Entity Recognition
Presentation Type: Poster Invite - Regular
Poster Number: 102
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Analytical Artificial Intelligence: ML, Digital Pathology, Imaging AI, Predictive Analytics, Governance, Health Data Science, Clinical Decision Support and Care Pathways
Primary Track: Big Data for Health
This study developed and evaluated two domain-adapted transformer-based language models, trained on large-scale nursing documentation, to extract fall-related nursing phenotypes from unstructured notes. Using more than 589,000 nursing notes and an expert-annotated corpus for named entity recognition, the models outperformed general-domain transformer baselines, achieving performance scores of 0.8648 for Korean and 0.9188 for English. These results highlight their utility for identifying actionable nursing risk factors and supporting nursing-focused clinical decision support.
Speaker(s):
INSOOK CHO, PhD
Inha University
Author(s):
Sujee Lee, PhD - Sungkyunkwan University; Hyekeyong Shin, MS - Inha University; Byeong Sun Park, PhD - Inha University; Dawit Cho, BA - Sungkyunkwan University;
Presentation Type: Poster Invite - Regular
Poster Number: 102
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Analytical Artificial Intelligence: ML, Digital Pathology, Imaging AI, Predictive Analytics, Governance, Health Data Science, Clinical Decision Support and Care Pathways
Primary Track: Big Data for Health
This study developed and evaluated two domain-adapted transformer-based language models, trained on large-scale nursing documentation, to extract fall-related nursing phenotypes from unstructured notes. Using more than 589,000 nursing notes and an expert-annotated corpus for named entity recognition, the models outperformed general-domain transformer baselines, achieving performance scores of 0.8648 for Korean and 0.9188 for English. These results highlight their utility for identifying actionable nursing risk factors and supporting nursing-focused clinical decision support.
Speaker(s):
INSOOK CHO, PhD
Inha University
Author(s):
Sujee Lee, PhD - Sungkyunkwan University; Hyekeyong Shin, MS - Inha University; Byeong Sun Park, PhD - Inha University; Dawit Cho, BA - Sungkyunkwan University;
INSOOK
CHO,
PhD - Inha University
Calibration and Reliability of NEDOCS, Unrestricted NEDOCS and XGBoost Models for Predicting Pediatric Emergency Department Overcrowding
Presentation Type: Poster - Regular
Poster Number: 103
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Analytical Artificial Intelligence: ML, Digital Pathology, Imaging AI, Predictive Analytics, Governance, Workforce Automation, Communication, and Workflow Efficiency, Leadership and Strategy
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
NEDOCS predicts ED overcrowding but requires recalibration for optimal performance. We compared NEDOCS and XGBoost machine learning predictive performance for 8-hour left without being seen rates using prospectively collected ED crowding data over the course of one year. The XGBoost had a higher point estimate AUROC than using NEDOCS alone; both demonstrated problems with calibration. This study shows the promise of machine learning to predict ED crowding and the need for routine predictive tool recalibration.
Speaker(s):
Adam Munday, M.D.
Mount Sinai Hospital
Author(s):
Adam Munday, M.D. - Mount Sinai Hospital; Kenneth McKinley, MD - Children's National;
Presentation Type: Poster - Regular
Poster Number: 103
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Analytical Artificial Intelligence: ML, Digital Pathology, Imaging AI, Predictive Analytics, Governance, Workforce Automation, Communication, and Workflow Efficiency, Leadership and Strategy
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
NEDOCS predicts ED overcrowding but requires recalibration for optimal performance. We compared NEDOCS and XGBoost machine learning predictive performance for 8-hour left without being seen rates using prospectively collected ED crowding data over the course of one year. The XGBoost had a higher point estimate AUROC than using NEDOCS alone; both demonstrated problems with calibration. This study shows the promise of machine learning to predict ED crowding and the need for routine predictive tool recalibration.
Speaker(s):
Adam Munday, M.D.
Mount Sinai Hospital
Author(s):
Adam Munday, M.D. - Mount Sinai Hospital; Kenneth McKinley, MD - Children's National;
Adam
Munday,
M.D. - Mount Sinai Hospital
Understanding When Retrieval Helps: Physician Evaluation of GPT-4 and Retrieval-Augmented Generation Approaches for Rare Disease Question Answering
Presentation Type: Poster Invite - Student
Poster Number: 104
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Analytical Artificial Intelligence: ML, Digital Pathology, Imaging AI, Predictive Analytics, Governance, Generative AI in Clinical Workflow: Ambient Listening, Chart Summarization, Automated Response with LLM, Health Data Science
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Retrieval-augmented generation(RAG) is increasingly used to enhance LLMs output in biomedicine, yet its benefit for rare-disease question-answering remains uncertain. Using physician ratings of accuracy, completeness, and relevance for patient questions on Complex Lymphatic Anomalies, we found that GPT-4 alone already produced high-quality answers, and RAG offered only modest gains in relevance without improving accuracy or completeness. These results underscore the need for context-specific retrieval strategies and rigorous evaluation when applying RAG to clinical information delivery.
Speaker(s):
Min Zhao, MS
Washington University in St. Louis, School of Medicine
Author(s):
Min Zhao, MS - Washington University in St. Louis, School of Medicine; Ethan Hillis, MS - Institute for Informatics at Washington University School of Medicine in St. Louis; Inez Oh, PhD - Institute for Informatics at Washington University in St. Louis, School of Medicine; Aditi Gupta - Washington University in St. Louis; Sally Cohen-Cutler, MD - Children’s Hospital of Philadelphia; Kathryn Harmoney, DO - University of New Mexico's Health Sciences Center; Albert Lai, PhD, FACMI, FAMIA - Washington University; Bryan Sisk, MD, MSCI - Washington University in St. Louis, School of Medicine;
Presentation Type: Poster Invite - Student
Poster Number: 104
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Analytical Artificial Intelligence: ML, Digital Pathology, Imaging AI, Predictive Analytics, Governance, Generative AI in Clinical Workflow: Ambient Listening, Chart Summarization, Automated Response with LLM, Health Data Science
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Retrieval-augmented generation(RAG) is increasingly used to enhance LLMs output in biomedicine, yet its benefit for rare-disease question-answering remains uncertain. Using physician ratings of accuracy, completeness, and relevance for patient questions on Complex Lymphatic Anomalies, we found that GPT-4 alone already produced high-quality answers, and RAG offered only modest gains in relevance without improving accuracy or completeness. These results underscore the need for context-specific retrieval strategies and rigorous evaluation when applying RAG to clinical information delivery.
Speaker(s):
Min Zhao, MS
Washington University in St. Louis, School of Medicine
Author(s):
Min Zhao, MS - Washington University in St. Louis, School of Medicine; Ethan Hillis, MS - Institute for Informatics at Washington University School of Medicine in St. Louis; Inez Oh, PhD - Institute for Informatics at Washington University in St. Louis, School of Medicine; Aditi Gupta - Washington University in St. Louis; Sally Cohen-Cutler, MD - Children’s Hospital of Philadelphia; Kathryn Harmoney, DO - University of New Mexico's Health Sciences Center; Albert Lai, PhD, FACMI, FAMIA - Washington University; Bryan Sisk, MD, MSCI - Washington University in St. Louis, School of Medicine;
Min
Zhao,
MS - Washington University in St. Louis, School of Medicine
Early Detection of Clinical Deterioration in a Pediatric Cardiology Acute Care Unit using Machine Learning Techniques
Presentation Type: Poster - Regular
Poster Number: 105
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Analytical Artificial Intelligence: ML, Digital Pathology, Imaging AI, Predictive Analytics, Governance, Health Data Science, Clinical Decision Support and Care Pathways
Primary Track: Big Data for Health
This single-center retrospective study explores the use of machine learning (ML) methods to predict clinical deterioration (CD) in pediatric cardiology patients. By analyzing registries and EPIC Clarity data, sixteen key predictors of CD were identified and used to develop a ML model that predicts CD within 6 hours of deterioration. ML model selection included the use of a model evaluation dashboard, which had varying feature datasets, balancing methods, and model types.
Speaker(s):
Michelle Gleason, MPH
Children's Healthcare of Atlanta
Author(s):
Naveen Muthu, MD - Children's Healthcare of Atlanta; Nikhil Chanani, MD - Childrenís Healthcare of Atlanta; Sherry Smith, RN - Children's Healthcare of Atlanta; David Kulp, MSc - Emory University School of Medicine; Nikolay Braykov, MS - Children's Helathcare of Atlanta; Zachary West, MD - Children's Healthcare of Atlanta; Michelle Gleason, MPH - Children's Healthcare of Atlanta;
Presentation Type: Poster - Regular
Poster Number: 105
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Analytical Artificial Intelligence: ML, Digital Pathology, Imaging AI, Predictive Analytics, Governance, Health Data Science, Clinical Decision Support and Care Pathways
Primary Track: Big Data for Health
This single-center retrospective study explores the use of machine learning (ML) methods to predict clinical deterioration (CD) in pediatric cardiology patients. By analyzing registries and EPIC Clarity data, sixteen key predictors of CD were identified and used to develop a ML model that predicts CD within 6 hours of deterioration. ML model selection included the use of a model evaluation dashboard, which had varying feature datasets, balancing methods, and model types.
Speaker(s):
Michelle Gleason, MPH
Children's Healthcare of Atlanta
Author(s):
Naveen Muthu, MD - Children's Healthcare of Atlanta; Nikhil Chanani, MD - Childrenís Healthcare of Atlanta; Sherry Smith, RN - Children's Healthcare of Atlanta; David Kulp, MSc - Emory University School of Medicine; Nikolay Braykov, MS - Children's Helathcare of Atlanta; Zachary West, MD - Children's Healthcare of Atlanta; Michelle Gleason, MPH - Children's Healthcare of Atlanta;
Michelle
Gleason,
MPH - Children's Healthcare of Atlanta
Automated Replication of Clinical Prediction Models via Agentic Workflows
Presentation Type: Poster - Student
Poster Number: 106
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Analytical Artificial Intelligence: ML, Digital Pathology, Imaging AI, Predictive Analytics, Governance, Clinical Decision Support and Care Pathways, Generative AI in Clinical Workflow: Ambient Listening, Chart Summarization, Automated Response with LLM
Primary Track: Big Data for Health
Reproducing clinical prediction models is notoriously difficult; fewer than 1% of models are independently replicated at the code level. This study evaluates whether an agentic LLM system can autonomously extract methods, reconstruct pipelines, replicate performance, and optimize methodologies using the InSight sepsis model as a case study. The system successfully reproduced core elements (Baseline AUROC 0.828 in MIMIC-III) and autonomously diagnosed low sensitivity, implementing corrections that increased sensitivity to 0.722 while maintaining specificity at 0.845, suggesting that agentic workflows can accelerate clinical ML validation.
Speaker(s):
Braden Eberhard, BS
University of Pennsylvania
Author(s):
Yidi Huang, BS - University of Pennsylvania; Dokyoon Kim, PhD - University of Pennsylvania;
Presentation Type: Poster - Student
Poster Number: 106
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Analytical Artificial Intelligence: ML, Digital Pathology, Imaging AI, Predictive Analytics, Governance, Clinical Decision Support and Care Pathways, Generative AI in Clinical Workflow: Ambient Listening, Chart Summarization, Automated Response with LLM
Primary Track: Big Data for Health
Reproducing clinical prediction models is notoriously difficult; fewer than 1% of models are independently replicated at the code level. This study evaluates whether an agentic LLM system can autonomously extract methods, reconstruct pipelines, replicate performance, and optimize methodologies using the InSight sepsis model as a case study. The system successfully reproduced core elements (Baseline AUROC 0.828 in MIMIC-III) and autonomously diagnosed low sensitivity, implementing corrections that increased sensitivity to 0.722 while maintaining specificity at 0.845, suggesting that agentic workflows can accelerate clinical ML validation.
Speaker(s):
Braden Eberhard, BS
University of Pennsylvania
Author(s):
Yidi Huang, BS - University of Pennsylvania; Dokyoon Kim, PhD - University of Pennsylvania;
Braden
Eberhard,
BS - University of Pennsylvania
Lending a Helping Headset: Virtual Reality Facilitates Epidural Placement Procedures in Laboring Patients
Presentation Type: Poster - Student
Poster Number: 107
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Augmented Reality and Virtual Reality in Care, Human Factors and Usability, Workforce Automation, Communication, and Workflow Efficiency
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Virtual reality (VR) shows promise as a noninvasive therapy for patients experiencing pain and anxiety. However, less is known about its effects in laboring parturients. We present a trial of VR during epidural placement, chosen because successful epidural anesthesia may confer significant medical benefits in addition to pain control. We present lessons learned, challenges, and user satisfaction ratings. We advocate for continued visibility of this patient population in the human-computer interaction domain of clinical informatics.
Speaker(s):
Michael Zhao, MD
University of Chicago
Author(s):
Michael Zhao, MD - University of Chicago; Melanie Zuniga, MBA - University of Chicago; Bhuvaneswari Sandeep Ram, MBBS - University of Chicago; Myra Collins, BA - University of Chicago; Leziga Obiyo, MD - University of Chicago;
Presentation Type: Poster - Student
Poster Number: 107
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Augmented Reality and Virtual Reality in Care, Human Factors and Usability, Workforce Automation, Communication, and Workflow Efficiency
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Virtual reality (VR) shows promise as a noninvasive therapy for patients experiencing pain and anxiety. However, less is known about its effects in laboring parturients. We present a trial of VR during epidural placement, chosen because successful epidural anesthesia may confer significant medical benefits in addition to pain control. We present lessons learned, challenges, and user satisfaction ratings. We advocate for continued visibility of this patient population in the human-computer interaction domain of clinical informatics.
Speaker(s):
Michael Zhao, MD
University of Chicago
Author(s):
Michael Zhao, MD - University of Chicago; Melanie Zuniga, MBA - University of Chicago; Bhuvaneswari Sandeep Ram, MBBS - University of Chicago; Myra Collins, BA - University of Chicago; Leziga Obiyo, MD - University of Chicago;
Michael
Zhao,
MD - University of Chicago
Navigating Complex Build Projects using Visualizations
Presentation Type: Poster - Regular
Poster Number: 108
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Change Management, Human Factors and Usability, Workforce Automation, Communication, and Workflow Efficiency
Primary Track: Driving Change at Scale through Effective Leadership and Governance
Building robust clinical tools within an electronic health record system can result in complicated build architectures with interwoven cascades of EHR elements and dependencies that are challenging to organize, document, and communicate. Visualization techniques such as network diagrams are often used in software development to accomplish these tasks. We present real-world examples of visualizations utilized in the context of complex EHR builds originating from an academic medical center’s clinician builder program.
Speaker(s):
Karl Santiago, MD
Boston Children's Hospital
Author(s):
Karl Santiago, MD - Boston Children's Hospital; Heather O'Donnell - Boston Children's Hospital; Jonathan Hron, MD - Boston Children's Hospital;
Presentation Type: Poster - Regular
Poster Number: 108
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Change Management, Human Factors and Usability, Workforce Automation, Communication, and Workflow Efficiency
Primary Track: Driving Change at Scale through Effective Leadership and Governance
Building robust clinical tools within an electronic health record system can result in complicated build architectures with interwoven cascades of EHR elements and dependencies that are challenging to organize, document, and communicate. Visualization techniques such as network diagrams are often used in software development to accomplish these tasks. We present real-world examples of visualizations utilized in the context of complex EHR builds originating from an academic medical center’s clinician builder program.
Speaker(s):
Karl Santiago, MD
Boston Children's Hospital
Author(s):
Karl Santiago, MD - Boston Children's Hospital; Heather O'Donnell - Boston Children's Hospital; Jonathan Hron, MD - Boston Children's Hospital;
Karl
Santiago,
MD - Boston Children's Hospital
LLM-Assisted Thematic Sentiment Analysis to Understand Delays in IT Request for Change Projects at a Community Hospital System
Presentation Type: Poster - Regular
Poster Number: 109
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Change Management, Analytical Artificial Intelligence: ML, Digital Pathology, Imaging AI, Predictive Analytics, Governance, Generative AI in Clinical Workflow: Ambient Listening, Chart Summarization, Automated Response with LLM
Primary Track: Driving Change at Scale through Effective Leadership and Governance
Rochester Regional Health uses Jira (project management application) to manage Information Technology (IT) related Requests for Change. To understand project delays, a closed source LLM was used to perform a thematic analysis of comments that monitor progress. An unsupervised clustering model returned five major reasons for delays that could be addressed. This demonstrates the potential utility of LLMs to analyze work processes at scale, enabling organizations to identify systemic issues and implement interventions
Speaker(s):
Abdul Mannan, D.O.
Rochester Regional Health
Author(s):
Abdul Mannan, D.O. - Rochester Regional Health; Orlando Nunez, Physician - Rochester Regional Health; Jee Lee, MD - RRH; Venugopal Mudgundi, MD, MPH, MSHI - Rochester Regional Health;
Presentation Type: Poster - Regular
Poster Number: 109
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Change Management, Analytical Artificial Intelligence: ML, Digital Pathology, Imaging AI, Predictive Analytics, Governance, Generative AI in Clinical Workflow: Ambient Listening, Chart Summarization, Automated Response with LLM
Primary Track: Driving Change at Scale through Effective Leadership and Governance
Rochester Regional Health uses Jira (project management application) to manage Information Technology (IT) related Requests for Change. To understand project delays, a closed source LLM was used to perform a thematic analysis of comments that monitor progress. An unsupervised clustering model returned five major reasons for delays that could be addressed. This demonstrates the potential utility of LLMs to analyze work processes at scale, enabling organizations to identify systemic issues and implement interventions
Speaker(s):
Abdul Mannan, D.O.
Rochester Regional Health
Author(s):
Abdul Mannan, D.O. - Rochester Regional Health; Orlando Nunez, Physician - Rochester Regional Health; Jee Lee, MD - RRH; Venugopal Mudgundi, MD, MPH, MSHI - Rochester Regional Health;
Abdul
Mannan,
D.O. - Rochester Regional Health
Interventions to reduce differential ordering with CBCs
Presentation Type: Poster - Regular
Poster Number: 110
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support and Care Pathways, Outcomes Improvement and Equity, Education and Training
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
We employed a strategy of eliminating CBC with differential from Top Search Results, pushing it to the bottom of the order search bar, followed by targeted interventions. Subsequently, educational emails to primary care clinics with indications for ordering differentials and current ordering practices across primary care sites were sent, and a duplicate alert for repeat CBC with diff within 24 hours for inpatient areas was implemented. Together, the three strategies significantly reduced differential ordering with cumulative impact across the health system of ~15%.
Speaker(s):
Rohith Palli, MD, PhD
University of Washington
Author(s):
Rohith Palli, MD, PhD - University of Washington; Regina Kwon, MD, MPH - University of Washington; Nikki Gentile, MD, PhD - University of Washington; Thomas Meloy, MLS - University of Washington; Patrick Mathias, MD, PhD - University of Washington;
Presentation Type: Poster - Regular
Poster Number: 110
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support and Care Pathways, Outcomes Improvement and Equity, Education and Training
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
We employed a strategy of eliminating CBC with differential from Top Search Results, pushing it to the bottom of the order search bar, followed by targeted interventions. Subsequently, educational emails to primary care clinics with indications for ordering differentials and current ordering practices across primary care sites were sent, and a duplicate alert for repeat CBC with diff within 24 hours for inpatient areas was implemented. Together, the three strategies significantly reduced differential ordering with cumulative impact across the health system of ~15%.
Speaker(s):
Rohith Palli, MD, PhD
University of Washington
Author(s):
Rohith Palli, MD, PhD - University of Washington; Regina Kwon, MD, MPH - University of Washington; Nikki Gentile, MD, PhD - University of Washington; Thomas Meloy, MLS - University of Washington; Patrick Mathias, MD, PhD - University of Washington;
Rohith
Palli,
MD, PhD - University of Washington
CAP-tivating Change: Reducing Antibiotic Waste Through Smart EHR Updates
Presentation Type: Poster Invite - Regular
Poster Number: 111
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support and Care Pathways, Outcomes Improvement and Equity, Quality Informatics and Lean
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
This quality improvement project reduces antibiotic waste in a pediatric hospital by optimizing inpatient prescribing for community-acquired pneumonia (CAP). Electronic Health Record (EHR) changes shorten IV ampicillin duration and automate conversion to oral amoxicillin, standardizing selection and duration per clinical guidelines. This promotes earlier IV-to-PO conversion, reducing unnecessary antibiotic use, minimizing resistance risk, and enhancing antimicrobial stewardship. Success is measured through ampicillin waste and prescribing patterns, improving patient care and resource efficiency.
Speaker(s):
Julia Yarahuan, MD, MBI
Children's Healthcare of Atlanta/Emory University
Author(s):
Taylor Ford, MD - Emory University; Julia Yarahuan, MD, MBI - Children's Healthcare of Atlanta/Emory University; Sarah Thompson, MSHIMI, BSN, RN - Children's Healthcare of Atlanta; Swaminathan Kandaswamy, PhD - Emory University School of Medicine; Preeti Jaggi, DM - Emory/Children's Healthcare of Atlanta; Whitney Sherry, MD - Emory/Children's Healthcare of Atlanta; Andrew Smelser, PharmD - Emory/Children's Healthcare of Atlanta; Lucie Fan, MD - Emory/Children's Healthcare of Atlanta; Zayd Ahmad, PharmD - Emory/Children's Healthcare of Atlanta;
Presentation Type: Poster Invite - Regular
Poster Number: 111
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support and Care Pathways, Outcomes Improvement and Equity, Quality Informatics and Lean
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
This quality improvement project reduces antibiotic waste in a pediatric hospital by optimizing inpatient prescribing for community-acquired pneumonia (CAP). Electronic Health Record (EHR) changes shorten IV ampicillin duration and automate conversion to oral amoxicillin, standardizing selection and duration per clinical guidelines. This promotes earlier IV-to-PO conversion, reducing unnecessary antibiotic use, minimizing resistance risk, and enhancing antimicrobial stewardship. Success is measured through ampicillin waste and prescribing patterns, improving patient care and resource efficiency.
Speaker(s):
Julia Yarahuan, MD, MBI
Children's Healthcare of Atlanta/Emory University
Author(s):
Taylor Ford, MD - Emory University; Julia Yarahuan, MD, MBI - Children's Healthcare of Atlanta/Emory University; Sarah Thompson, MSHIMI, BSN, RN - Children's Healthcare of Atlanta; Swaminathan Kandaswamy, PhD - Emory University School of Medicine; Preeti Jaggi, DM - Emory/Children's Healthcare of Atlanta; Whitney Sherry, MD - Emory/Children's Healthcare of Atlanta; Andrew Smelser, PharmD - Emory/Children's Healthcare of Atlanta; Lucie Fan, MD - Emory/Children's Healthcare of Atlanta; Zayd Ahmad, PharmD - Emory/Children's Healthcare of Atlanta;
Julia
Yarahuan,
MD, MBI - Children's Healthcare of Atlanta/Emory University
Resident-led initiative to improve febrile infant documentation efficiency
Presentation Type: Poster - Student
Poster Number: 112
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support and Care Pathways, Education and Training, Human Factors and Usability, Quality Informatics and Lean
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
While there is a clinical pathway at Nationwide Children’s Hospital for treatment and evaluation of well appearing febrile neonates, there is no standardized documentation. We are building a resident-driven standardized history and physical to increase consistency and efficiency in documentation. Pre-survey data of 40 pediatrics residents indicates using an unweighted NASA task load index score showed a score of 49.2 indicating there is an opportunity to improve on the task load of resident documentation.
Speaker(s):
Steven Dick, MD
Nationwide Children's Hospital
Author(s):
Steven Dick, MD - Nationwide Children's Hospital; Emily Sentman, MD - Nationwide Children's Hospital; Gerd McGwire, MD, PhD - Nationwide Children's Hospital, Ohio State Univeristy;
Presentation Type: Poster - Student
Poster Number: 112
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support and Care Pathways, Education and Training, Human Factors and Usability, Quality Informatics and Lean
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
While there is a clinical pathway at Nationwide Children’s Hospital for treatment and evaluation of well appearing febrile neonates, there is no standardized documentation. We are building a resident-driven standardized history and physical to increase consistency and efficiency in documentation. Pre-survey data of 40 pediatrics residents indicates using an unweighted NASA task load index score showed a score of 49.2 indicating there is an opportunity to improve on the task load of resident documentation.
Speaker(s):
Steven Dick, MD
Nationwide Children's Hospital
Author(s):
Steven Dick, MD - Nationwide Children's Hospital; Emily Sentman, MD - Nationwide Children's Hospital; Gerd McGwire, MD, PhD - Nationwide Children's Hospital, Ohio State Univeristy;
Steven
Dick,
MD - Nationwide Children's Hospital
Streamlining ACS Work-Ups with Non-Interruptive Decision Support in the ED
Presentation Type: Poster - Student
Poster Number: 113
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support and Care Pathways, Workforce Automation, Communication, and Workflow Efficiency, Diagnostics, Clinician Well-Being
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
We deployed a silent, non-interruptive clinical decision support designed to enhance the completeness and timeliness of acute coronary syndrome (ACS) workups in the emergency department. The system automatically adds an EKG when a troponin is ordered in the absence of a recent tracing. Preliminary findings demonstrate marked reductions in encounters with missing and delayed EKGs as well as increase in encounters with simultaneous ordering of EKGs and troponins.
Speaker(s):
Vincent Xiao, MD
NYU Langone Health
Author(s):
Vincent Xiao, MD - NYU Langone Health; Eric Epstein, MD - NYU Langone Health; Edwin Pineda, RN, MSN - NYU Langone Health; Nicholas Genes, MD, PhD - NYU Grossman School of Medicine;
Presentation Type: Poster - Student
Poster Number: 113
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support and Care Pathways, Workforce Automation, Communication, and Workflow Efficiency, Diagnostics, Clinician Well-Being
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
We deployed a silent, non-interruptive clinical decision support designed to enhance the completeness and timeliness of acute coronary syndrome (ACS) workups in the emergency department. The system automatically adds an EKG when a troponin is ordered in the absence of a recent tracing. Preliminary findings demonstrate marked reductions in encounters with missing and delayed EKGs as well as increase in encounters with simultaneous ordering of EKGs and troponins.
Speaker(s):
Vincent Xiao, MD
NYU Langone Health
Author(s):
Vincent Xiao, MD - NYU Langone Health; Eric Epstein, MD - NYU Langone Health; Edwin Pineda, RN, MSN - NYU Langone Health; Nicholas Genes, MD, PhD - NYU Grossman School of Medicine;
Vincent
Xiao,
MD - NYU Langone Health
Scaling Clinical Pathways: A Practical Framework for Sustainable CDS
Presentation Type: Poster Invite - Regular
Poster Number: 114
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support and Care Pathways, Leadership and Strategy, Human Factors and Usability
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
We developed a clinical pathways program that scaled to 40+ pathways with 900+ users and 15,000+ annual encounters. Key strategies included process-based governance, ad hoc multidisciplinary workgroups, the “80% rule” for usability, integration of clinical informatics, proactive maintenance, and continuous dissemination. We will share our successful approach demonstrating how intentional governance, structured engagement, and planned maintenance support sustainable, scalable CDS across EHR environments.
Speaker(s):
Laura Macke, MD
University of Colorado Anschutz
Author(s):
Laura Macke, MD - University of Colorado Anschutz; Lisa Schilling, MD, MSPH - U of Colorado Anschutz;
Presentation Type: Poster Invite - Regular
Poster Number: 114
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support and Care Pathways, Leadership and Strategy, Human Factors and Usability
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
We developed a clinical pathways program that scaled to 40+ pathways with 900+ users and 15,000+ annual encounters. Key strategies included process-based governance, ad hoc multidisciplinary workgroups, the “80% rule” for usability, integration of clinical informatics, proactive maintenance, and continuous dissemination. We will share our successful approach demonstrating how intentional governance, structured engagement, and planned maintenance support sustainable, scalable CDS across EHR environments.
Speaker(s):
Laura Macke, MD
University of Colorado Anschutz
Author(s):
Laura Macke, MD - University of Colorado Anschutz; Lisa Schilling, MD, MSPH - U of Colorado Anschutz;
Laura
Macke,
MD - University of Colorado Anschutz
Checkmate on Chemo Errors: How an EHR Based Checklist Can Reduce Chemotherapy Errors
Presentation Type: Poster - Regular
Poster Number: 115
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support and Care Pathways, Outcomes Improvement and Equity, Human Factors and Usability
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Chemotherapy administration is complex and risky. Nursing staff have standard items that must be double checked prior to all chemotherapy administrations. Integrating these items into a checklist within the EHR can reduce chemotherapy errors.
Speaker(s):
Claire Stokes, MD, MPH
Children's Healthcare of Atlanta and Emory University
Author(s):
Tonya Bennett, BSHI; Rachael LeRoux, MSHS, CPPS - Children's Healthcare of Atlanta; Mikaela Gray, BSN, RN, CPHON - Children's Healthcare of Atlanta; Tessa Shevlin, BSN,RN,CPHON - Children's Healthcare of Atlanta; Terese Weart, MSN,RN,CPNP-PC - Children's Healthcare of Atlanta; Ansley Jones, BSN,RN,CPHON - Children's Healthcare of Atlanta; Elyse Lindquist, BSN,RN,CPHON - Children's Healthcare of Atlanta; Brooke Amason, BSN,RN,CPHON - Children's Healthcare of Atlanta; Racheal Ogletree, MSN,RN,CNML - Children's Healthcare of Atlanta; Julia Yarahuan, MD, MBI - Children's Healthcare of Atlanta/Emory University;
Presentation Type: Poster - Regular
Poster Number: 115
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support and Care Pathways, Outcomes Improvement and Equity, Human Factors and Usability
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Chemotherapy administration is complex and risky. Nursing staff have standard items that must be double checked prior to all chemotherapy administrations. Integrating these items into a checklist within the EHR can reduce chemotherapy errors.
Speaker(s):
Claire Stokes, MD, MPH
Children's Healthcare of Atlanta and Emory University
Author(s):
Tonya Bennett, BSHI; Rachael LeRoux, MSHS, CPPS - Children's Healthcare of Atlanta; Mikaela Gray, BSN, RN, CPHON - Children's Healthcare of Atlanta; Tessa Shevlin, BSN,RN,CPHON - Children's Healthcare of Atlanta; Terese Weart, MSN,RN,CPNP-PC - Children's Healthcare of Atlanta; Ansley Jones, BSN,RN,CPHON - Children's Healthcare of Atlanta; Elyse Lindquist, BSN,RN,CPHON - Children's Healthcare of Atlanta; Brooke Amason, BSN,RN,CPHON - Children's Healthcare of Atlanta; Racheal Ogletree, MSN,RN,CNML - Children's Healthcare of Atlanta; Julia Yarahuan, MD, MBI - Children's Healthcare of Atlanta/Emory University;
Claire
Stokes,
MD, MPH - Children's Healthcare of Atlanta and Emory University
PEN-PAL: An EHR Informed AI Assistant for Penicillin Allergy Risk Assessment
Presentation Type: Poster - Student
Poster Number: 116
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support and Care Pathways, Generative AI in Clinical Workflow: Ambient Listening, Chart Summarization, Automated Response with LLM, Innovation Partnerships, Implementation Science, and Learning Health Systems, Standards, Terminology, and Interoperability, TEFCA, FHIR, Workforce Automation, Communication, and Workflow Efficiency
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Penicillin allergy labels (PALs) affect ~15% of patients but are rarely confirmed, leading to suboptimal antibiotic use. PEN-PAL is an EHR-integrated AI assistant conducting conversational risk assessments using validated frameworks. In 12 mock patient interactions, PEN-PAL achieved 100% diagnostic accuracy with the Stone framework in 3.4 minute conversations. This tool addresses de-labeling barriers by efficiently collecting histories and providing accurate risk stratification, potentially improving antibiotic stewardship and outcomes.
Speaker(s):
Megan Wang, Undergraduate Student
Vanderbilt University
Author(s):
Allison McCoy, PhD, ACHIP, FACMI, FAMIA - Vanderbilt University Medical Center; Siru Liu, PhD - Vanderbilt University Medical Center; Adam Wright, PhD - Vanderbilt University Medical Center; Cosby Stone, MD, MPH - Vanderbilt University Medical Center;
Presentation Type: Poster - Student
Poster Number: 116
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support and Care Pathways, Generative AI in Clinical Workflow: Ambient Listening, Chart Summarization, Automated Response with LLM, Innovation Partnerships, Implementation Science, and Learning Health Systems, Standards, Terminology, and Interoperability, TEFCA, FHIR, Workforce Automation, Communication, and Workflow Efficiency
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Penicillin allergy labels (PALs) affect ~15% of patients but are rarely confirmed, leading to suboptimal antibiotic use. PEN-PAL is an EHR-integrated AI assistant conducting conversational risk assessments using validated frameworks. In 12 mock patient interactions, PEN-PAL achieved 100% diagnostic accuracy with the Stone framework in 3.4 minute conversations. This tool addresses de-labeling barriers by efficiently collecting histories and providing accurate risk stratification, potentially improving antibiotic stewardship and outcomes.
Speaker(s):
Megan Wang, Undergraduate Student
Vanderbilt University
Author(s):
Allison McCoy, PhD, ACHIP, FACMI, FAMIA - Vanderbilt University Medical Center; Siru Liu, PhD - Vanderbilt University Medical Center; Adam Wright, PhD - Vanderbilt University Medical Center; Cosby Stone, MD, MPH - Vanderbilt University Medical Center;
Megan
Wang,
Undergraduate Student - Vanderbilt University
Enhancing Cardiology Care Continuity With an EHR-Integrated Clinical Decision Support System
Presentation Type: Poster - Regular
Poster Number: 118
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support and Care Pathways, Outcomes Improvement and Equity, Change Management, Quality Informatics and Lean
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Background: Continuity of cardiology care is essential for patients with heart failure, arrhythmias, and structural heart disease, yet many eligible patients do not receive referrals or are lost to follow-up. Prior studies show that 52% of heart failure patients lack timely cardiology follow-up [1]. This project implemented an EHR-integrated CDS system to identify at-risk cardiac patients at key points of care and improve specialty follow-up.
Methods: A custom CDS algorithm was deployed in November 2024 within a large academic health system. The system reviews clinical data and triggers alerts under three conditions: an inpatient meets criteria for a cardiology consult but has no active consult order, a patient meets referral criteria at their primary care visit, or a qualifying patient is discharged without being evaluated by cardiology. Alerts are delivered as passive flags, in-basket messages, or interruptive notifications and routed to hospitalists, primary care clinicians, or cardiology clinic staff. Referral and alert volumes were monitored over 12 months.
Results: During the 12-months, the CDS system issued 4,249 notifications, which led to 702 cardiology referrals. Heart failure clinic referrals showed the largest increase, rising from an average of 10 per month to more than 25. The greatest gains occurred among hospitalized high-risk heart failure patients who had no cardiology involvement during their admission.
Conclusion: The CDS system improved identification of high-risk cardiac patients and increased successful cardiology follow-up across inpatient and outpatient settings. Automated care-gap detection paired with navigator review enhanced continuity and represents a scalable approach for chronic disease management.
Speaker(s):
Leigh Anne Goodman, DO
University of Arizona College of Medicine - Phoenix, Banner Health
Author(s):
Timothy Shimon, MD - Banner Health; Leigh Anne Goodman, DO - University of Arizona College of Medicine - Phoenix, Banner Health;
Presentation Type: Poster - Regular
Poster Number: 118
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support and Care Pathways, Outcomes Improvement and Equity, Change Management, Quality Informatics and Lean
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Background: Continuity of cardiology care is essential for patients with heart failure, arrhythmias, and structural heart disease, yet many eligible patients do not receive referrals or are lost to follow-up. Prior studies show that 52% of heart failure patients lack timely cardiology follow-up [1]. This project implemented an EHR-integrated CDS system to identify at-risk cardiac patients at key points of care and improve specialty follow-up.
Methods: A custom CDS algorithm was deployed in November 2024 within a large academic health system. The system reviews clinical data and triggers alerts under three conditions: an inpatient meets criteria for a cardiology consult but has no active consult order, a patient meets referral criteria at their primary care visit, or a qualifying patient is discharged without being evaluated by cardiology. Alerts are delivered as passive flags, in-basket messages, or interruptive notifications and routed to hospitalists, primary care clinicians, or cardiology clinic staff. Referral and alert volumes were monitored over 12 months.
Results: During the 12-months, the CDS system issued 4,249 notifications, which led to 702 cardiology referrals. Heart failure clinic referrals showed the largest increase, rising from an average of 10 per month to more than 25. The greatest gains occurred among hospitalized high-risk heart failure patients who had no cardiology involvement during their admission.
Conclusion: The CDS system improved identification of high-risk cardiac patients and increased successful cardiology follow-up across inpatient and outpatient settings. Automated care-gap detection paired with navigator review enhanced continuity and represents a scalable approach for chronic disease management.
Speaker(s):
Leigh Anne Goodman, DO
University of Arizona College of Medicine - Phoenix, Banner Health
Author(s):
Timothy Shimon, MD - Banner Health; Leigh Anne Goodman, DO - University of Arizona College of Medicine - Phoenix, Banner Health;
Leigh Anne
Goodman,
DO - University of Arizona College of Medicine - Phoenix, Banner Health
FairED: A Multicenter Benchmark for Fairness Evaluation and Bias Mitigation in Emergency Department Predictive Modeling
Presentation Type: Poster Invite - Regular
Poster Number: 120
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
FairED provides a transparent, reproducible multicenter benchmark for population-level surveillance of algorithmic bias in emergency department prediction models. By harmonizing triage data and standardizing fairness metrics, FairED enables cross-site assessment of AI performance across demographic groups. Applied to over 550,000 ED encounters, the framework identifies persistent disparities and quantifies the stability of mitigation. FairED transforms fairness from a conceptual into a measurable quality signal that supports AI governance and advances equity in emergency care.
Speaker(s):
Xinnie Mai, Master
University of Minnesota
Author(s):
Yunqian Liu, Master - Duke University; Yilin Ning, PhD - Duke-NUS Medical School; David Wacker, MD, PhD - University of Minnesota Medical School; Mingquan Lin, PhD - University of Minnesota; Nan Liu, PhD - National University of Singapore; Michael Puskarich, MD MSCR - University of Minnesota Department of Emergency Medicine; Feng Xie, PhD - University of Minnesota;
Presentation Type: Poster Invite - Regular
Poster Number: 120
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
FairED provides a transparent, reproducible multicenter benchmark for population-level surveillance of algorithmic bias in emergency department prediction models. By harmonizing triage data and standardizing fairness metrics, FairED enables cross-site assessment of AI performance across demographic groups. Applied to over 550,000 ED encounters, the framework identifies persistent disparities and quantifies the stability of mitigation. FairED transforms fairness from a conceptual into a measurable quality signal that supports AI governance and advances equity in emergency care.
Speaker(s):
Xinnie Mai, Master
University of Minnesota
Author(s):
Yunqian Liu, Master - Duke University; Yilin Ning, PhD - Duke-NUS Medical School; David Wacker, MD, PhD - University of Minnesota Medical School; Mingquan Lin, PhD - University of Minnesota; Nan Liu, PhD - National University of Singapore; Michael Puskarich, MD MSCR - University of Minnesota Department of Emergency Medicine; Feng Xie, PhD - University of Minnesota;
Xinnie
Mai,
Master - University of Minnesota
Health IT Influence That Lasts: Leveraging Physician Builders and SmartUsers after Honor Roll Success
Presentation Type: Poster Invite - Regular
Poster Number: 121
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinician Well-Being, Education and Training, Leadership and Strategy, Human Factors and Usability
Primary Track: Advancing Wellness for Providers and Community with Consideration of Human Factors
Large health systems often struggle to translate operational successes into sustained frontline behavior change. NYU Langone Health organized a cohort of advanced EHR skills providers and created the Health IT Influencer (HITI) group. Embedded across diverse clinical areas, the group drives operational feedback and real-world testing of IT implementations like Epic upgrades and training tools. Their efforts have improved EHR education programs, operational momentum, and expanded clinical informatics initiatives for trainees
Speaker(s):
James Davydov, MSc
NYU Langone Health
Author(s):
Nusrat Jahan, MD - NYU; Gavriil Ilizarov, DO - NYU Langone Medical Center; Nicholas Genes, MD, PhD - NYU Grossman School of Medicine;
Presentation Type: Poster Invite - Regular
Poster Number: 121
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinician Well-Being, Education and Training, Leadership and Strategy, Human Factors and Usability
Primary Track: Advancing Wellness for Providers and Community with Consideration of Human Factors
Large health systems often struggle to translate operational successes into sustained frontline behavior change. NYU Langone Health organized a cohort of advanced EHR skills providers and created the Health IT Influencer (HITI) group. Embedded across diverse clinical areas, the group drives operational feedback and real-world testing of IT implementations like Epic upgrades and training tools. Their efforts have improved EHR education programs, operational momentum, and expanded clinical informatics initiatives for trainees
Speaker(s):
James Davydov, MSc
NYU Langone Health
Author(s):
Nusrat Jahan, MD - NYU; Gavriil Ilizarov, DO - NYU Langone Medical Center; Nicholas Genes, MD, PhD - NYU Grossman School of Medicine;
James
Davydov,
MSc - NYU Langone Health
Electronic Health Record Use and Workflow Factors Associated with Work Intentions among Primary Care Physicians
Presentation Type: Poster Invite - Regular
Poster Number: 122
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinician Well-Being, Workforce Automation, Communication, and Workflow Efficiency, Human Factors and Usability
Primary Track: Advancing Wellness for Providers and Community with Consideration of Human Factors
Intent to leave (ITL) a clinical job and intent to reduce (ITR) clinical effort are prevalent among U.S. physicians. However, these phenomena are typically assessed via survey responses, which are a lagging indicator of physician intentions. In this study, we sought to understand how EHR (electronic health record) use patterns and EHR workflow perceptions among PCPs are associated with reported work intentions, and thus whether they could serve as more advance indicators of physicians’ work intentions. Using the yearly physician’s UCSF STEP survey and physician EHR activity pattern data from the Epic Signal database, we assessed the association of actual EHR use patterns and physicians’ perceptions of EHR workflows with the outcomes of ITL and ITR. While EHR workflow perceptions were associated with ITL and ITR, neither absolute EHR activity measures nor changes in these measures were associated with stated work intentions. These findings emphasize the potential disconnect between perceived and true EHR burden. They suggest the importance of asking clinicians about their perceptions of EHR workflows and addressing perceived or reported challenges.
Speaker(s):
Lisa Rotenstein, MD, MBA, MSc
UCSF
Author(s):
Lisa Rotenstein, MD, MBA, MSc - UCSF; Christopher Toretsky, MPH - University of California, San Francisco; A J Holmgren, PhD - University of California, San Francisco; Clark Williamson, MD - UCSF; Rachel Willard-Grace, MPH - UCSF;
Presentation Type: Poster Invite - Regular
Poster Number: 122
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinician Well-Being, Workforce Automation, Communication, and Workflow Efficiency, Human Factors and Usability
Primary Track: Advancing Wellness for Providers and Community with Consideration of Human Factors
Intent to leave (ITL) a clinical job and intent to reduce (ITR) clinical effort are prevalent among U.S. physicians. However, these phenomena are typically assessed via survey responses, which are a lagging indicator of physician intentions. In this study, we sought to understand how EHR (electronic health record) use patterns and EHR workflow perceptions among PCPs are associated with reported work intentions, and thus whether they could serve as more advance indicators of physicians’ work intentions. Using the yearly physician’s UCSF STEP survey and physician EHR activity pattern data from the Epic Signal database, we assessed the association of actual EHR use patterns and physicians’ perceptions of EHR workflows with the outcomes of ITL and ITR. While EHR workflow perceptions were associated with ITL and ITR, neither absolute EHR activity measures nor changes in these measures were associated with stated work intentions. These findings emphasize the potential disconnect between perceived and true EHR burden. They suggest the importance of asking clinicians about their perceptions of EHR workflows and addressing perceived or reported challenges.
Speaker(s):
Lisa Rotenstein, MD, MBA, MSc
UCSF
Author(s):
Lisa Rotenstein, MD, MBA, MSc - UCSF; Christopher Toretsky, MPH - University of California, San Francisco; A J Holmgren, PhD - University of California, San Francisco; Clark Williamson, MD - UCSF; Rachel Willard-Grace, MPH - UCSF;
Lisa
Rotenstein,
MD, MBA, MSc - UCSF
Inbox Inequity: Age and Gender-Based Disparities in Patient-Initiated Messaging and Clinician Inbasket Burden
Presentation Type: Poster - Regular
Poster Number: 123
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinician Well-Being, Workforce Automation, Communication, and Workflow Efficiency, Outcomes Improvement and Equity
Primary Track: Advancing Wellness for Providers and Community with Consideration of Human Factors
Epic’s MyChart has transformed patient–clinician communication by enabling asynchronous messaging, but it has also amplified clinician in-basket workload. Emerging evidence suggests that patient demographic factors—including age and gender—may influence message volume. Our hypothesis is that this potentially produces inequitable burdens on subgroups of clinicians. In this retrospective cross-sectional study, we analyze 129,424 UC Davis Health encounters and 63.8 million encounters from Epic Cosmos to quantify age- and gender-based disparities in patient-initiated Medical Advice Requests (PMARs) and their downstream effect on clinician in-basket load. We hypothesize that female patients send significantly more messages than age-matched male patients. We further evaluate whether clinician demographics interact with patient-level patterns to generate uneven workload distribution. Findings from this study will inform the design of more equitable digital workflows and support strategies to mitigate clinician burnout resulting from inbox burden.
Speaker(s):
Jill Rushton-Miller, MD
Sutter Health (SMGR)
Author(s):
Jonathan Chiang, DO - UC Davis; Mark Moubarek, MD - UC Davis Health; John Graff, DO, FCAP - UC Davis Health;
Presentation Type: Poster - Regular
Poster Number: 123
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinician Well-Being, Workforce Automation, Communication, and Workflow Efficiency, Outcomes Improvement and Equity
Primary Track: Advancing Wellness for Providers and Community with Consideration of Human Factors
Epic’s MyChart has transformed patient–clinician communication by enabling asynchronous messaging, but it has also amplified clinician in-basket workload. Emerging evidence suggests that patient demographic factors—including age and gender—may influence message volume. Our hypothesis is that this potentially produces inequitable burdens on subgroups of clinicians. In this retrospective cross-sectional study, we analyze 129,424 UC Davis Health encounters and 63.8 million encounters from Epic Cosmos to quantify age- and gender-based disparities in patient-initiated Medical Advice Requests (PMARs) and their downstream effect on clinician in-basket load. We hypothesize that female patients send significantly more messages than age-matched male patients. We further evaluate whether clinician demographics interact with patient-level patterns to generate uneven workload distribution. Findings from this study will inform the design of more equitable digital workflows and support strategies to mitigate clinician burnout resulting from inbox burden.
Speaker(s):
Jill Rushton-Miller, MD
Sutter Health (SMGR)
Author(s):
Jonathan Chiang, DO - UC Davis; Mark Moubarek, MD - UC Davis Health; John Graff, DO, FCAP - UC Davis Health;
Jill
Rushton-Miller,
MD - Sutter Health (SMGR)
ASQ-PHI: An Adversarial Synthetic Benchmark for Clinical PHI De-Identification
Presentation Type: Poster - Student
Poster Number: 124
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Data Privacy, Cybersecurity, Reliability, and Security, Generative AI in Clinical Workflow: Ambient Listening, Chart Summarization, Automated Response with LLM, Analytical Artificial Intelligence: ML, Digital Pathology, Imaging AI, Predictive Analytics, Governance, Standards, Terminology, and Interoperability, TEFCA, FHIR, Infrastructure and Cloud Computing, Quality Informatics and Lean
Primary Track: Big Data for Health
Hospitals are starting to deploy HIPAA-compliant Business Associate Agreement (BAA) large language models (LLMs) for clinical work. When these systems call external tools such as live web search or Model Context Protocol servers, a “safe handoff” problem appears: the clinician’s Patient Health Information (PHI)-containing query must be de-identified before it leaves the BAA boundary, without destroying the intent of the question. Existing de-identification corpora are built from long narrative notes, not the short, compressed queries clinicians type into LLM chat interfaces, so they cannot directly measure safe-handoff failures such as PHI leakage or over-redaction of PHI-free input.
ASQ-PHI (Adversarial Synthetic Queries for Protected Health Information) is a fully synthetic benchmark of 1,051 single-turn clinical search queries designed for this setting. Each record is a single-line query plus a PHI_TAGS block of JSON objects labeling 13 textual HIPAA Safe Harbor identifier types. The dataset includes 832 PHI-positive queries and 219 carefully engineered hard negatives that mimic PHI-like structure while containing no identifiers, enabling simultaneous measurement of missed PHI and unnecessary over-redaction. Queries were generated using an adversarial few-shot Azure OpenAI GPT-4o pipeline with automated quality checks and a three-clinician audit (n=300) for plausibility and label accuracy.
We will present the dataset structure, validation results, and baseline performance of a commercial detector (Amazon Comprehend Medical DetectPHI), and show how informatics teams can use ASQ-PHI to test and tune their own de-identification workflows for clinical LLM deployments.
Speaker(s):
James Weatherhead, MD-PhD Student
University of Texas Medical Branch (UTMB)
Author(s):
James Weatherhead, MD-PhD Student - University of Texas Medical Branch (UTMB); Peter McCaffrey, MD - UTMB; George Golovko, PhD - University of Texas Medical Branch;
Presentation Type: Poster - Student
Poster Number: 124
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Data Privacy, Cybersecurity, Reliability, and Security, Generative AI in Clinical Workflow: Ambient Listening, Chart Summarization, Automated Response with LLM, Analytical Artificial Intelligence: ML, Digital Pathology, Imaging AI, Predictive Analytics, Governance, Standards, Terminology, and Interoperability, TEFCA, FHIR, Infrastructure and Cloud Computing, Quality Informatics and Lean
Primary Track: Big Data for Health
Hospitals are starting to deploy HIPAA-compliant Business Associate Agreement (BAA) large language models (LLMs) for clinical work. When these systems call external tools such as live web search or Model Context Protocol servers, a “safe handoff” problem appears: the clinician’s Patient Health Information (PHI)-containing query must be de-identified before it leaves the BAA boundary, without destroying the intent of the question. Existing de-identification corpora are built from long narrative notes, not the short, compressed queries clinicians type into LLM chat interfaces, so they cannot directly measure safe-handoff failures such as PHI leakage or over-redaction of PHI-free input.
ASQ-PHI (Adversarial Synthetic Queries for Protected Health Information) is a fully synthetic benchmark of 1,051 single-turn clinical search queries designed for this setting. Each record is a single-line query plus a PHI_TAGS block of JSON objects labeling 13 textual HIPAA Safe Harbor identifier types. The dataset includes 832 PHI-positive queries and 219 carefully engineered hard negatives that mimic PHI-like structure while containing no identifiers, enabling simultaneous measurement of missed PHI and unnecessary over-redaction. Queries were generated using an adversarial few-shot Azure OpenAI GPT-4o pipeline with automated quality checks and a three-clinician audit (n=300) for plausibility and label accuracy.
We will present the dataset structure, validation results, and baseline performance of a commercial detector (Amazon Comprehend Medical DetectPHI), and show how informatics teams can use ASQ-PHI to test and tune their own de-identification workflows for clinical LLM deployments.
Speaker(s):
James Weatherhead, MD-PhD Student
University of Texas Medical Branch (UTMB)
Author(s):
James Weatherhead, MD-PhD Student - University of Texas Medical Branch (UTMB); Peter McCaffrey, MD - UTMB; George Golovko, PhD - University of Texas Medical Branch;
James
Weatherhead,
MD-PhD Student - University of Texas Medical Branch (UTMB)
REDCap in TX-CTRN: Enhancing Health Data Quality through Data Management
Presentation Type: Poster - Regular
Poster Number: 125
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Data Privacy, Cybersecurity, Reliability, and Security, Clinical Decision Support and Care Pathways, Health Data Science
Primary Track: Advancing Wellness for Providers and Community with Consideration of Human Factors
TX-CTRN employs a multi-phase REDCap workflow to ensure high-quality, consistent data across sites. Pre-entry branching logic reduces common errors, while real-time checks and duplicate-entry reconciliation address issues during data entry. Post-entry audits using custom Data Quality Rules and R-based review resolve remaining discrepancies. Site-specific summaries and targeted queries enable timely corrections, establishing a scalable, iterative framework for comprehensive data quality assurance.
Speaker(s):
Fei Teng, Ph.D.
The University of Texas at Austin
Author(s):
Alexis Larraga, MS - University of Texas at Austin; Luke Klima, BBA - Dell Medical School, University of Texas at Austin; Liza Hoke, MS - University of Texas at Austin; Pat Scherer, MBA - Texas Advanced Computing Center; Vladislav Krendelev, MS - Texas Advanced Computing Center; Tomislav Urban, MGIS - Texas Advanced Computing Center; Jeffrey Newport, M.D. - University of Texas at Austin; Karen Wagner, MD, PhD - University of Texas Medical Branch; Charles Nemeroff, M.D., Ph.D. - University of Texas at Austin; Justin Rousseau, MD, MMSc - University of Texas Southwestern Medical Center;
Presentation Type: Poster - Regular
Poster Number: 125
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Data Privacy, Cybersecurity, Reliability, and Security, Clinical Decision Support and Care Pathways, Health Data Science
Primary Track: Advancing Wellness for Providers and Community with Consideration of Human Factors
TX-CTRN employs a multi-phase REDCap workflow to ensure high-quality, consistent data across sites. Pre-entry branching logic reduces common errors, while real-time checks and duplicate-entry reconciliation address issues during data entry. Post-entry audits using custom Data Quality Rules and R-based review resolve remaining discrepancies. Site-specific summaries and targeted queries enable timely corrections, establishing a scalable, iterative framework for comprehensive data quality assurance.
Speaker(s):
Fei Teng, Ph.D.
The University of Texas at Austin
Author(s):
Alexis Larraga, MS - University of Texas at Austin; Luke Klima, BBA - Dell Medical School, University of Texas at Austin; Liza Hoke, MS - University of Texas at Austin; Pat Scherer, MBA - Texas Advanced Computing Center; Vladislav Krendelev, MS - Texas Advanced Computing Center; Tomislav Urban, MGIS - Texas Advanced Computing Center; Jeffrey Newport, M.D. - University of Texas at Austin; Karen Wagner, MD, PhD - University of Texas Medical Branch; Charles Nemeroff, M.D., Ph.D. - University of Texas at Austin; Justin Rousseau, MD, MMSc - University of Texas Southwestern Medical Center;
Fei
Teng,
Ph.D. - The University of Texas at Austin
The LIFT Strategy: Transforming Informatics into a Data-Driven Healthcare World
Presentation Type: Poster - Regular
Poster Number: 126
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Education and Training, Leadership and Strategy, Health Data Science, Clinical Decision Support and Care Pathways, Population Health, Digital Therapeutics, Remote Patient Monitoring (RPM), and Digital Engagement, Outcomes Improvement and Equity, Change Management, Workforce Automation, Communication, and Workflow Efficiency
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
As healthcare technology evolves, it’s critical that clinical leaders have an elevated perspective on the intersection of technology, operations, and medicine. Leveraging Informatics for Transformation (LIFT), is an eight-month course that teaches informatics concepts to enhance patient care, strategic planning, and digital experiences. Through didactics, mentorship, networking, and application training, participants show statistically significant improvement in self‑reported informatics competencies and demonstrate the ability to solve complex technology-driven healthcare problems through presentations at program's conclusion.
Speaker(s):
Shellon Blackman-Lees, PhD
Northwell
Author(s):
Shellon Blackman-Lees, PhD, MS - Northwell Health; Kimberly Velez, MSN, RN - Northwell Health; Anncy Thomas, D.O. - Northwell Health; Michael Cassara, DO MSED - Northwell Health; Kerriann Latten, FNP - Northwell;
Presentation Type: Poster - Regular
Poster Number: 126
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Education and Training, Leadership and Strategy, Health Data Science, Clinical Decision Support and Care Pathways, Population Health, Digital Therapeutics, Remote Patient Monitoring (RPM), and Digital Engagement, Outcomes Improvement and Equity, Change Management, Workforce Automation, Communication, and Workflow Efficiency
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
As healthcare technology evolves, it’s critical that clinical leaders have an elevated perspective on the intersection of technology, operations, and medicine. Leveraging Informatics for Transformation (LIFT), is an eight-month course that teaches informatics concepts to enhance patient care, strategic planning, and digital experiences. Through didactics, mentorship, networking, and application training, participants show statistically significant improvement in self‑reported informatics competencies and demonstrate the ability to solve complex technology-driven healthcare problems through presentations at program's conclusion.
Speaker(s):
Shellon Blackman-Lees, PhD
Northwell
Author(s):
Shellon Blackman-Lees, PhD, MS - Northwell Health; Kimberly Velez, MSN, RN - Northwell Health; Anncy Thomas, D.O. - Northwell Health; Michael Cassara, DO MSED - Northwell Health; Kerriann Latten, FNP - Northwell;
Shellon
Blackman-Lees,
PhD - Northwell
Nursing Informatics Transforming Health Systems and Clinical Care
Presentation Type: Poster - Regular
Poster Number: 127
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Education and Training, Human Factors and Usability, Workforce Automation, Communication, and Workflow Efficiency, Analytics, Registries, and the Digital Command Center
Working Group: Nursing Informatics Working Group
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Since 2021, 139 Healthcare Informatics Master’s students from CU Nursing completed practicums affecting healthcare quality and efficiency across the United States and abroad. A review of available practicum projects (N=81) indicates students influenced processes and outcomes in >112 hospitals and 4,800 clinical care sites among 26,800 nurses and 5,500 providers who care for >25,000,000 patients. This highlights the importance of the Healthcare Informatics program and its contributions to nursing informatics nationally and internationally.
Speaker(s):
Sharon Giarrizzo-Wilson, PhD, RN, CNOR, NI-BC, FAAN
University of Colorado College of Nursing
Author(s):
Samantha Stonbraker, PhD, MPH, RN, FAAN, FAMIA - University of Colorado Anschutz Medical Campus; Christina Baker, PhD,RN, NCSN, NI-BC - University of Colorado, Anschutz, College of Nursing;
Presentation Type: Poster - Regular
Poster Number: 127
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Education and Training, Human Factors and Usability, Workforce Automation, Communication, and Workflow Efficiency, Analytics, Registries, and the Digital Command Center
Working Group: Nursing Informatics Working Group
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Since 2021, 139 Healthcare Informatics Master’s students from CU Nursing completed practicums affecting healthcare quality and efficiency across the United States and abroad. A review of available practicum projects (N=81) indicates students influenced processes and outcomes in >112 hospitals and 4,800 clinical care sites among 26,800 nurses and 5,500 providers who care for >25,000,000 patients. This highlights the importance of the Healthcare Informatics program and its contributions to nursing informatics nationally and internationally.
Speaker(s):
Sharon Giarrizzo-Wilson, PhD, RN, CNOR, NI-BC, FAAN
University of Colorado College of Nursing
Author(s):
Samantha Stonbraker, PhD, MPH, RN, FAAN, FAMIA - University of Colorado Anschutz Medical Campus; Christina Baker, PhD,RN, NCSN, NI-BC - University of Colorado, Anschutz, College of Nursing;
Sharon
Giarrizzo-Wilson,
PhD, RN, CNOR, NI-BC, FAAN - University of Colorado College of Nursing
Clinician Editing Patterns in AI-Generated vs Human-Generated Clinical Documentation: A Content-Level Comparison Using LLM Classification
Presentation Type: Poster - Student
Poster Number: 128
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Generative AI in Clinical Workflow: Ambient Listening, Chart Summarization, Automated Response with LLM, Workforce Automation, Communication, and Workflow Efficiency, Health Data Science, Innovation Partnerships, Implementation Science, and Learning Health Systems
Primary Track: Big Data for Health
As health systems adopt LLMs for clinical documentation, the types of content clinicians most often revise remain unknown. Analyzing paired AI-generated and human hospital courses, we used a high-agreement LLM classifier to categorize thousands of clinician edits into an eight-category clinical schema. AI drafts required fewer changes but showed disproportionate corrections to treatment details and extraneous conditions. This approach highlights predictable weaknesses in AI-generated notes and establishes feasibility of an LLM-as-a-judge to target model refinement.
Speaker(s):
Aditya Jain, BA
NYU Grossman School of Medicine
Author(s):
Aditya Jain, BA - NYU Grossman School of Medicine; Ujwal Srivastava, BS - NYU Grossman School of Medicine; Vincent Major, PhD - NYU Grossman School of Medicine; Katherine Hochman, MD, MBA - NYU Langone Health; Jonathan Austrian, MD - NYU Langone Health; Jonah Feldman, MD, FACP - NYU Langone Health; William Small, MD, MBA - NYU Langone Health;
Presentation Type: Poster - Student
Poster Number: 128
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Generative AI in Clinical Workflow: Ambient Listening, Chart Summarization, Automated Response with LLM, Workforce Automation, Communication, and Workflow Efficiency, Health Data Science, Innovation Partnerships, Implementation Science, and Learning Health Systems
Primary Track: Big Data for Health
As health systems adopt LLMs for clinical documentation, the types of content clinicians most often revise remain unknown. Analyzing paired AI-generated and human hospital courses, we used a high-agreement LLM classifier to categorize thousands of clinician edits into an eight-category clinical schema. AI drafts required fewer changes but showed disproportionate corrections to treatment details and extraneous conditions. This approach highlights predictable weaknesses in AI-generated notes and establishes feasibility of an LLM-as-a-judge to target model refinement.
Speaker(s):
Aditya Jain, BA
NYU Grossman School of Medicine
Author(s):
Aditya Jain, BA - NYU Grossman School of Medicine; Ujwal Srivastava, BS - NYU Grossman School of Medicine; Vincent Major, PhD - NYU Grossman School of Medicine; Katherine Hochman, MD, MBA - NYU Langone Health; Jonathan Austrian, MD - NYU Langone Health; Jonah Feldman, MD, FACP - NYU Langone Health; William Small, MD, MBA - NYU Langone Health;
Aditya
Jain,
BA - NYU Grossman School of Medicine
LLM-Based Automated Phenotyping of Undiagnosed Diseases Network Cases
Presentation Type: Poster - Regular
Poster Number: 129
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Generative AI in Clinical Workflow: Ambient Listening, Chart Summarization, Automated Response with LLM, Workforce Automation, Communication, and Workflow Efficiency, Clinical Decision Support and Care Pathways
Primary Track: Big Data for Health
We explored the use of Large Language Models (GPT-4 Omni) to automate information extraction from electronic health records for the Undiagnosed Diseases Network. By processing genetics notes and procedure reports, the model captured approximately 50% of the data necessary for case evaluations, with ROUGE scores indicating moderate recall accuracy. Qualitative assessment by clinicians supported the feasibility of this approach, highlighting the potential to accelerate evaluations for genetic disorders with further integration of additional data sources.
Speaker(s):
Inez Oh, PhD
Institute for Informatics at Washington University in St. Louis, School of Medicine
Author(s):
Aditi Gupta - Washington University in St. Louis; Kathleen Sisco, RN, CPNP - Washington University School of Medicine; Erin McRoy, MS, CGC - Washington University School of Medicine; Dana Kiley, BS - Washington University School of Medicine; Jennifer Wambach, MD, MS - Washington University School of Medicine; Patricia Dickson, MD - Washington University School of Medicine;
Presentation Type: Poster - Regular
Poster Number: 129
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Generative AI in Clinical Workflow: Ambient Listening, Chart Summarization, Automated Response with LLM, Workforce Automation, Communication, and Workflow Efficiency, Clinical Decision Support and Care Pathways
Primary Track: Big Data for Health
We explored the use of Large Language Models (GPT-4 Omni) to automate information extraction from electronic health records for the Undiagnosed Diseases Network. By processing genetics notes and procedure reports, the model captured approximately 50% of the data necessary for case evaluations, with ROUGE scores indicating moderate recall accuracy. Qualitative assessment by clinicians supported the feasibility of this approach, highlighting the potential to accelerate evaluations for genetic disorders with further integration of additional data sources.
Speaker(s):
Inez Oh, PhD
Institute for Informatics at Washington University in St. Louis, School of Medicine
Author(s):
Aditi Gupta - Washington University in St. Louis; Kathleen Sisco, RN, CPNP - Washington University School of Medicine; Erin McRoy, MS, CGC - Washington University School of Medicine; Dana Kiley, BS - Washington University School of Medicine; Jennifer Wambach, MD, MS - Washington University School of Medicine; Patricia Dickson, MD - Washington University School of Medicine;
Inez
Oh,
PhD - Institute for Informatics at Washington University in St. Louis, School of Medicine
Multilingual Validation of PCP-Bot: An LLM-based Chatbot for Pre-Visit Planning in Primary Care
Presentation Type: Poster - Student
Poster Number: 130
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Generative AI in Clinical Workflow: Ambient Listening, Chart Summarization, Automated Response with LLM, Workforce Automation, Communication, and Workflow Efficiency, Human Factors and Usability
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Pre-visit planning can reduce documentation burden, yet multilingual chatbot performance remains understudied. We evaluated PCP-Bot across English, Mandarin, Spanish, and Hindi in 31 bilingual participants completing 310 interactions. Mandarin and Hindi achieved usability parity with English, whereas Spanish showed significant usability and summary quality gaps. Trust and workload remained consistent across languages. Our findings that LLM translation capabilities can enable effective deployment beyond English following appropriate performance validation.
Speaker(s):
Pei-Lun Chen, Master of Science in Bioengineering
University of Pennsylvania
Author(s):
Amogh Ananda Rao, MBBS, MS - University of Pennsylvania; Sydney Pugh, Computer and Information Science - University of Pennsylvania; Kevin Johnson, MD, MS - University of Pennsylvania;
Presentation Type: Poster - Student
Poster Number: 130
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Generative AI in Clinical Workflow: Ambient Listening, Chart Summarization, Automated Response with LLM, Workforce Automation, Communication, and Workflow Efficiency, Human Factors and Usability
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Pre-visit planning can reduce documentation burden, yet multilingual chatbot performance remains understudied. We evaluated PCP-Bot across English, Mandarin, Spanish, and Hindi in 31 bilingual participants completing 310 interactions. Mandarin and Hindi achieved usability parity with English, whereas Spanish showed significant usability and summary quality gaps. Trust and workload remained consistent across languages. Our findings that LLM translation capabilities can enable effective deployment beyond English following appropriate performance validation.
Speaker(s):
Pei-Lun Chen, Master of Science in Bioengineering
University of Pennsylvania
Author(s):
Amogh Ananda Rao, MBBS, MS - University of Pennsylvania; Sydney Pugh, Computer and Information Science - University of Pennsylvania; Kevin Johnson, MD, MS - University of Pennsylvania;
Pei-Lun
Chen,
Master of Science in Bioengineering - University of Pennsylvania
A Proof-of-Concept AI Agent for Automated Sedation Triage in Endoscopy
Presentation Type: Poster - Student
Poster Number: 132
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Generative AI in Clinical Workflow: Ambient Listening, Chart Summarization, Automated Response with LLM, Clinical Decision Support and Care Pathways, Workforce Automation, Communication, and Workflow Efficiency
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Generative AI, particularly large language models, offers transformative potential in clinical workflows. This study introduces “GIVersa-Endoscopy,” a customized AI agent leveraging retrieval-augmented generation to automate sedation triage for endoscopy patients. Using institution-specific guidelines and EHR patient data, the GenAI agent matched or exceeded clinician recommendations in 91% of cases, prioritizing patient safety. This proof-of-concept highlights a practical framework for developing and validating GenAI tools in clinical workflows, paving the way for broader applications across specialties.
Speaker(s):
Lakshmipriya Subbaraj, MD
UCSF
Author(s):
Jin Ge, MD - UCSF; Albert Dang, MFA - UCSF; Steve Sun, BS - UCSF; Kendall Beck, MD - UCSF;
Presentation Type: Poster - Student
Poster Number: 132
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Generative AI in Clinical Workflow: Ambient Listening, Chart Summarization, Automated Response with LLM, Clinical Decision Support and Care Pathways, Workforce Automation, Communication, and Workflow Efficiency
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Generative AI, particularly large language models, offers transformative potential in clinical workflows. This study introduces “GIVersa-Endoscopy,” a customized AI agent leveraging retrieval-augmented generation to automate sedation triage for endoscopy patients. Using institution-specific guidelines and EHR patient data, the GenAI agent matched or exceeded clinician recommendations in 91% of cases, prioritizing patient safety. This proof-of-concept highlights a practical framework for developing and validating GenAI tools in clinical workflows, paving the way for broader applications across specialties.
Speaker(s):
Lakshmipriya Subbaraj, MD
UCSF
Author(s):
Jin Ge, MD - UCSF; Albert Dang, MFA - UCSF; Steve Sun, BS - UCSF; Kendall Beck, MD - UCSF;
Lakshmipriya
Subbaraj,
MD - UCSF
Impact of an Inpatient Chart Summarization Tool in a Real-World Clinical Setting
Presentation Type: Poster - Student
Poster Number: 133
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Generative AI in Clinical Workflow: Ambient Listening, Chart Summarization, Automated Response with LLM, Workforce Automation, Communication, and Workflow Efficiency, Innovation Partnerships, Implementation Science, and Learning Health Systems
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Clinicians must quickly gain a thorough understanding of admitted patients on their service in order to provide the best care. Here, we evaluate the use of a commercially available generative AI tool that summarizes aspects of a patient’s hospitalization. Our results from a one-month pilot of this tool in clinical practice show low level of harm and high accuracy, completeness, reliability, concision, and utility.
Speaker(s):
Catherine Blebea, MD
University of California San Francisco
Author(s):
Catherine Blebea, MD - University of California San Francisco; Alireza Ebrahimvandi, PhD, MSc - University of California San Francisco; Aris Oates, MD - UCSF Health; Sara Murray, MD, MAS - UCSF; Smitha Ganeshan, MD, MBA - UCSF Health;
Presentation Type: Poster - Student
Poster Number: 133
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Generative AI in Clinical Workflow: Ambient Listening, Chart Summarization, Automated Response with LLM, Workforce Automation, Communication, and Workflow Efficiency, Innovation Partnerships, Implementation Science, and Learning Health Systems
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Clinicians must quickly gain a thorough understanding of admitted patients on their service in order to provide the best care. Here, we evaluate the use of a commercially available generative AI tool that summarizes aspects of a patient’s hospitalization. Our results from a one-month pilot of this tool in clinical practice show low level of harm and high accuracy, completeness, reliability, concision, and utility.
Speaker(s):
Catherine Blebea, MD
University of California San Francisco
Author(s):
Catherine Blebea, MD - University of California San Francisco; Alireza Ebrahimvandi, PhD, MSc - University of California San Francisco; Aris Oates, MD - UCSF Health; Sara Murray, MD, MAS - UCSF; Smitha Ganeshan, MD, MBA - UCSF Health;
Catherine
Blebea,
MD - University of California San Francisco
Derivation and Validation of Acceptable Provider Documentation Summarization Quality Instrument (PDSQI-9) Thresholds for Evaluating Large Language Model-Based Clinical Documentation
Presentation Type: Poster - Student
Poster Number: 134
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Generative AI in Clinical Workflow: Ambient Listening, Chart Summarization, Automated Response with LLM, Human Factors and Usability, Health Data Science
Primary Track: Big Data for Health
This study establishes validated thresholds for the Physician Documentation Summarization Quality Instrument (PDSQI-9) to evaluate LLM-generated clinical documentation. Applying latent class analysis and ROC optimization to 200 summaries, we identified an acceptability threshold of 3.81 achieving sensitivity of 0.88 (95% CI: 0.68-0.97), specificity of 1.00 (95% CI: 0.76-1.00), and accuracy of 0.92 (95% CI: 0.79-0.98). This framework enables safe deployment of AI documentation systems with empirically-grounded decision rules aligned with clinician assessments.
Speaker(s):
Emma Croxford, PhD Student
University of Wisconsin Madison
Author(s):
Emma Croxford, PhD Student - University of Wisconsin Madison; Yanjun Gao, PhD - University of Colorado; Elliot First - Epic; Nicholas Pellegrino, BS - Epic; Miranda Schnier, BA - Epic; John Caskey - University of Wisconsin-Madison; Graham Wills, PhD - UW Health; Guanhua Chen, PhD - University of Wisconsin - Madison; Dmitriy Dligach, Ph.D. - Loyola University Chicago; Matthew Churpek, MD, MPH, PhD - University of Wisconsin-Madison; Anoop Mayampurath, PhD - University of Wisconsin - Madison; Frank Liao, PhD - University of Wisconsin, Madison - UW Health; Karen Wong, MD, MPH, MIDS - Epic Systems; Brian Patterson, MD MPH - University of Wisconsin-Madison; Majid Afshar, MD, MSCR - University of Wisconsin - Madison;
Presentation Type: Poster - Student
Poster Number: 134
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Generative AI in Clinical Workflow: Ambient Listening, Chart Summarization, Automated Response with LLM, Human Factors and Usability, Health Data Science
Primary Track: Big Data for Health
This study establishes validated thresholds for the Physician Documentation Summarization Quality Instrument (PDSQI-9) to evaluate LLM-generated clinical documentation. Applying latent class analysis and ROC optimization to 200 summaries, we identified an acceptability threshold of 3.81 achieving sensitivity of 0.88 (95% CI: 0.68-0.97), specificity of 1.00 (95% CI: 0.76-1.00), and accuracy of 0.92 (95% CI: 0.79-0.98). This framework enables safe deployment of AI documentation systems with empirically-grounded decision rules aligned with clinician assessments.
Speaker(s):
Emma Croxford, PhD Student
University of Wisconsin Madison
Author(s):
Emma Croxford, PhD Student - University of Wisconsin Madison; Yanjun Gao, PhD - University of Colorado; Elliot First - Epic; Nicholas Pellegrino, BS - Epic; Miranda Schnier, BA - Epic; John Caskey - University of Wisconsin-Madison; Graham Wills, PhD - UW Health; Guanhua Chen, PhD - University of Wisconsin - Madison; Dmitriy Dligach, Ph.D. - Loyola University Chicago; Matthew Churpek, MD, MPH, PhD - University of Wisconsin-Madison; Anoop Mayampurath, PhD - University of Wisconsin - Madison; Frank Liao, PhD - University of Wisconsin, Madison - UW Health; Karen Wong, MD, MPH, MIDS - Epic Systems; Brian Patterson, MD MPH - University of Wisconsin-Madison; Majid Afshar, MD, MSCR - University of Wisconsin - Madison;
Emma
Croxford,
PhD Student - University of Wisconsin Madison
Leveraging Open-Source Large Language Models to Detect Cardiorespiratory Symptoms in Medical Records
Presentation Type: Poster - Student
Poster Number: 135
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Generative AI in Clinical Workflow: Ambient Listening, Chart Summarization, Automated Response with LLM, Clinical Decision Support and Care Pathways, Health Data Science, Analytical Artificial Intelligence: ML, Digital Pathology, Imaging AI, Predictive Analytics, Governance
Primary Track: Big Data for Health
Understanding cardiorespiratory signs and symptoms (S&S) in free-text clinical notes is critical for early recognition of cardiopulmonary and oncologic complications, but manual coding is time-consuming and inconsistent. Open-source large language models (LLMs) offer a privacy-preserving way to automate S&S extraction and ICD-10-CM coding when deployed locally.
We evaluated Llama 3.3-70B, deployed via Ollama, to identify cardiorespiratory S&S and assign ICD-10-CM codes from de-identified medical notes in MTSamples. A clinician-generated gold standard of S&S and corresponding ICD-10-CM codes served as reference. We compared four prompting strategies: (1) instruction-only extraction and coding, (2) retrieval-augmented generation using ICD-10-CM definitions, (3) an “assumption-free” prompt discouraging unsupported inferences, and (4) a multi-agent pipeline separating extraction, coding, and post–hoc consistency checking. Performance was assessed using code-level precision, recall, F1, and hallucination rates based on non-existent or description-mismatched codes.
Prompt refinement markedly improved accuracy. The instruction-only setting yielded high recall but low precision. Incorporating ICD-10-CM definitions improved both, while the assumption-free prompt further balanced performance. The multi-agent pipeline with post-processing achieved the highest precision and recall and minimized hallucinations, substantially reducing non-existent and semantically incorrect codes.
Locally deployed open-source LLMs, when paired with structured prompts, external code definitions, and multi-agent orchestration, can reliably identify and code cardiorespiratory S&S from clinical text. This approach supports scalable, privacy-preserving symptom phenotyping for cancer and cardiorespiratory informatics.
Speaker(s):
Yunbing Bai, MS
University of Arizona
Author(s):
Yunbing Bai, MS - University of Arizona; Joseph Finkelstein, MD, PhD - University of Arizona;
Presentation Type: Poster - Student
Poster Number: 135
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Generative AI in Clinical Workflow: Ambient Listening, Chart Summarization, Automated Response with LLM, Clinical Decision Support and Care Pathways, Health Data Science, Analytical Artificial Intelligence: ML, Digital Pathology, Imaging AI, Predictive Analytics, Governance
Primary Track: Big Data for Health
Understanding cardiorespiratory signs and symptoms (S&S) in free-text clinical notes is critical for early recognition of cardiopulmonary and oncologic complications, but manual coding is time-consuming and inconsistent. Open-source large language models (LLMs) offer a privacy-preserving way to automate S&S extraction and ICD-10-CM coding when deployed locally.
We evaluated Llama 3.3-70B, deployed via Ollama, to identify cardiorespiratory S&S and assign ICD-10-CM codes from de-identified medical notes in MTSamples. A clinician-generated gold standard of S&S and corresponding ICD-10-CM codes served as reference. We compared four prompting strategies: (1) instruction-only extraction and coding, (2) retrieval-augmented generation using ICD-10-CM definitions, (3) an “assumption-free” prompt discouraging unsupported inferences, and (4) a multi-agent pipeline separating extraction, coding, and post–hoc consistency checking. Performance was assessed using code-level precision, recall, F1, and hallucination rates based on non-existent or description-mismatched codes.
Prompt refinement markedly improved accuracy. The instruction-only setting yielded high recall but low precision. Incorporating ICD-10-CM definitions improved both, while the assumption-free prompt further balanced performance. The multi-agent pipeline with post-processing achieved the highest precision and recall and minimized hallucinations, substantially reducing non-existent and semantically incorrect codes.
Locally deployed open-source LLMs, when paired with structured prompts, external code definitions, and multi-agent orchestration, can reliably identify and code cardiorespiratory S&S from clinical text. This approach supports scalable, privacy-preserving symptom phenotyping for cancer and cardiorespiratory informatics.
Speaker(s):
Yunbing Bai, MS
University of Arizona
Author(s):
Yunbing Bai, MS - University of Arizona; Joseph Finkelstein, MD, PhD - University of Arizona;
Yunbing
Bai,
MS - University of Arizona
Toward Global Large Language Models in Medicine
Presentation Type: Poster - Regular
Poster Number: 136
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Generative AI in Clinical Workflow: Ambient Listening, Chart Summarization, Automated Response with LLM, Health Data Science, Population Health, Digital Therapeutics, Remote Patient Monitoring (RPM), and Digital Engagement
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Despite continuous advances in medical technology, the global distribution of health care resources remains inequitable. The development of large language models (LLMs) has transformed medicine and holds promise for improving health care quality and expanding access to medical information globally. However, existing LLMs are primarily trained on high-resource languages, limiting their applicability in global medical scenarios. To mitigate this issue, we constructed GlobMed, the largest multilingual medical dataset to date, containing over 600,000 entries spanning 15 languages, including 5 low-resource languages. On top of this, we established GlobMed-Bench, which systematically assesses 56 state-of-the-art proprietary and open-weight LLMs across multilingual medical benchmarks. Additionally, we introduced GlobMed- LLMs, a suite of multilingual medical LLMs trained on GlobMed, scaling from 1.7B to 8B parameters. GlobMed-LLMs achieved over 40% average performance improvement compared with vanilla LLMs and more than threefold gains in low-resource languages. These resources provide an important foundation for advancing the equitable development of LLMs globally, enabling broader language communities to benefit from technological advances.
Speaker(s):
Rui Yang, Master
Duke-NUS Medical School
Author(s):
Rui Yang, Master - Duke-NUS Medical School; Nan Liu, PhD - National University of Singapore;
Presentation Type: Poster - Regular
Poster Number: 136
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Generative AI in Clinical Workflow: Ambient Listening, Chart Summarization, Automated Response with LLM, Health Data Science, Population Health, Digital Therapeutics, Remote Patient Monitoring (RPM), and Digital Engagement
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Despite continuous advances in medical technology, the global distribution of health care resources remains inequitable. The development of large language models (LLMs) has transformed medicine and holds promise for improving health care quality and expanding access to medical information globally. However, existing LLMs are primarily trained on high-resource languages, limiting their applicability in global medical scenarios. To mitigate this issue, we constructed GlobMed, the largest multilingual medical dataset to date, containing over 600,000 entries spanning 15 languages, including 5 low-resource languages. On top of this, we established GlobMed-Bench, which systematically assesses 56 state-of-the-art proprietary and open-weight LLMs across multilingual medical benchmarks. Additionally, we introduced GlobMed- LLMs, a suite of multilingual medical LLMs trained on GlobMed, scaling from 1.7B to 8B parameters. GlobMed-LLMs achieved over 40% average performance improvement compared with vanilla LLMs and more than threefold gains in low-resource languages. These resources provide an important foundation for advancing the equitable development of LLMs globally, enabling broader language communities to benefit from technological advances.
Speaker(s):
Rui Yang, Master
Duke-NUS Medical School
Author(s):
Rui Yang, Master - Duke-NUS Medical School; Nan Liu, PhD - National University of Singapore;
Rui
Yang,
Master - Duke-NUS Medical School
Introduction of same-day work session note closure rate as a new measure of clinical efficacy in participants using an artificial intelligence note writing scribe system.
Presentation Type: Poster - Student
Poster Number: 137
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Generative AI in Clinical Workflow: Ambient Listening, Chart Summarization, Automated Response with LLM, Workforce Automation, Communication, and Workflow Efficiency, Clinician Well-Being, Human Factors and Usability
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Physician burnout is driven by administrative burden, EHR inefficiencies, and increasing documentation demands, with national burnout rates exceeding 60%. Physicians spend nearly twice as much time on the EHR as on patient care and often complete 1–2 hours of after-hours charting, elevating burnout risk2. Ambient AI scribes may reduce documentation burden, though evidence is mixed. Existing EHR metrics may not reflect true efficiency; we propose same-day work session note closure rate as a more meaningful measure.
Speaker(s):
Minh Phan, M.D.
University of Arizona
Author(s):
Vikeen Patel, MD, MBA - Bannerhealth & University of Arizona COMP; Vinay Shah, MD - University of Arizona;
Presentation Type: Poster - Student
Poster Number: 137
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Generative AI in Clinical Workflow: Ambient Listening, Chart Summarization, Automated Response with LLM, Workforce Automation, Communication, and Workflow Efficiency, Clinician Well-Being, Human Factors and Usability
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Physician burnout is driven by administrative burden, EHR inefficiencies, and increasing documentation demands, with national burnout rates exceeding 60%. Physicians spend nearly twice as much time on the EHR as on patient care and often complete 1–2 hours of after-hours charting, elevating burnout risk2. Ambient AI scribes may reduce documentation burden, though evidence is mixed. Existing EHR metrics may not reflect true efficiency; we propose same-day work session note closure rate as a more meaningful measure.
Speaker(s):
Minh Phan, M.D.
University of Arizona
Author(s):
Vikeen Patel, MD, MBA - Bannerhealth & University of Arizona COMP; Vinay Shah, MD - University of Arizona;
Minh
Phan,
M.D. - University of Arizona
Preliminary development and evaluation of infographics to enhance HIV-prevention education with a large language model
Presentation Type: Poster - Regular
Poster Number: 138
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Generative AI in Clinical Workflow: Ambient Listening, Chart Summarization, Automated Response with LLM, Outcomes Improvement and Equity, Human Factors and Usability
Working Group: Nursing Informatics Working Group
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
While infographics are a powerful tool to improve communication and patient outcomes, current methods to generate infographics are slow and resource-heavy, limiting scalability. This study presents the preliminary training of a large language model (LLM) to create visually appealing and clinically accurate infographics and their subsequent evaluation. Results show that although the LLM is on track to produce the intended infographics, there is much remaining work needed to refine the LLM and infographics before clinical use.
Speaker(s):
Samantha Stonbraker, PhD, MPH, RN, FAAN, FAMIA
University of Colorado Anschutz Medical Campus
Author(s):
Samantha Stonbraker, PhD, MPH, RN, FAAN, FAMIA - University of Colorado Anschutz Medical Campus; Kellie Hawkins, MD, MPH - Denver Health and Hospital Authority; Katherine Frasca, MD - UCHealth; Yanjun Gao, PhD - University of Colorado;
Presentation Type: Poster - Regular
Poster Number: 138
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Generative AI in Clinical Workflow: Ambient Listening, Chart Summarization, Automated Response with LLM, Outcomes Improvement and Equity, Human Factors and Usability
Working Group: Nursing Informatics Working Group
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
While infographics are a powerful tool to improve communication and patient outcomes, current methods to generate infographics are slow and resource-heavy, limiting scalability. This study presents the preliminary training of a large language model (LLM) to create visually appealing and clinically accurate infographics and their subsequent evaluation. Results show that although the LLM is on track to produce the intended infographics, there is much remaining work needed to refine the LLM and infographics before clinical use.
Speaker(s):
Samantha Stonbraker, PhD, MPH, RN, FAAN, FAMIA
University of Colorado Anschutz Medical Campus
Author(s):
Samantha Stonbraker, PhD, MPH, RN, FAAN, FAMIA - University of Colorado Anschutz Medical Campus; Kellie Hawkins, MD, MPH - Denver Health and Hospital Authority; Katherine Frasca, MD - UCHealth; Yanjun Gao, PhD - University of Colorado;
Samantha
Stonbraker,
PhD, MPH, RN, FAAN, FAMIA - University of Colorado Anschutz Medical Campus
Generative AI in Admission Notes and Diagnostic Completeness
Presentation Type: Poster - Regular
Poster Number: 139
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Generative AI in Clinical Workflow: Ambient Listening, Chart Summarization, Automated Response with LLM, Quality Informatics and Lean, Diagnostics
Primary Track: Driving Change at Scale through Effective Leadership and Governance
Generative AI tools for admission history and physical notes may improve diagnostic completeness, but their real-world performance in inpatient care is unclear. We retrospectively compared AI-generated admission notes with paired provider-authored notes at a large academic medical center. AI drafts showed high diagnostic accuracy (96%) and surfaced additional clinically important, quality-impacting secondary diagnoses absent from provider notes, supporting a human-in-the-loop model for AI-assisted documentation.
Speaker(s):
Alfredo Camargo Rodrigues, Md
University of Iowa
Author(s):
Alfredo Camargo Rodrigues, Md - University of Iowa; Jason Misurac, MD, MS - The University of Iowa Health Care; Lindsey Knake, MD, MS - The University of Iowa Carver College of Medicine; Kevin Barker, Rn - University of Iowa Health Care; James Blum, MD, FCCM, CDH-E - University of Iowa;
Presentation Type: Poster - Regular
Poster Number: 139
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Generative AI in Clinical Workflow: Ambient Listening, Chart Summarization, Automated Response with LLM, Quality Informatics and Lean, Diagnostics
Primary Track: Driving Change at Scale through Effective Leadership and Governance
Generative AI tools for admission history and physical notes may improve diagnostic completeness, but their real-world performance in inpatient care is unclear. We retrospectively compared AI-generated admission notes with paired provider-authored notes at a large academic medical center. AI drafts showed high diagnostic accuracy (96%) and surfaced additional clinically important, quality-impacting secondary diagnoses absent from provider notes, supporting a human-in-the-loop model for AI-assisted documentation.
Speaker(s):
Alfredo Camargo Rodrigues, Md
University of Iowa
Author(s):
Alfredo Camargo Rodrigues, Md - University of Iowa; Jason Misurac, MD, MS - The University of Iowa Health Care; Lindsey Knake, MD, MS - The University of Iowa Carver College of Medicine; Kevin Barker, Rn - University of Iowa Health Care; James Blum, MD, FCCM, CDH-E - University of Iowa;
Alfredo
Camargo Rodrigues,
Md - University of Iowa
AI-Enhanced Handoff Assessment: Reducing Physician Burden and Improving Resident Education
Presentation Type: Poster - Student
Poster Number: 140
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Generative AI in Clinical Workflow: Ambient Listening, Chart Summarization, Automated Response with LLM, Workforce Automation, Communication, and Workflow Efficiency, Education and Training
Primary Track: Advancing Wellness for Providers and Community with Consideration of Human Factors
Successful implementation of the I-PASS handoff framework has consistently led to reductions in medical errors, ¹ but observation and feedback are essential to success. ² Live observations increase supervisors’ workload and vary in quality. We developed a novel AI tool that automates I-PASS element detection in verbal handoffs, reducing cognitive load on observers while providing scalable, consistent feedback for iterative improvement and real-time educational intervention.
Speaker(s):
Josh Pankin, MD
Baystate Medical Center
Author(s):
Josh Pankin, MD - Baystate Medical Center; Vishal Pallerla, B.A. - Intersystems; Samantha Pendleton, D.O - Baystate Medical Center; Hannah Miller, M.D. - Baystate Medical Center; Don Woodlock, B.Sc. - InterSystems; Jonathan Teich, MD, PhD - InterSystems Corporation; Qi Li, M.D. - Intersystems; Molly Senn-McNally, M.D. - Baystate Medical Center; Christopher Landrigan, M.D., M.P.H. - Boston Children's Hospital; Amy Starmer, MD, MPH - Baystate Health;
Presentation Type: Poster - Student
Poster Number: 140
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Generative AI in Clinical Workflow: Ambient Listening, Chart Summarization, Automated Response with LLM, Workforce Automation, Communication, and Workflow Efficiency, Education and Training
Primary Track: Advancing Wellness for Providers and Community with Consideration of Human Factors
Successful implementation of the I-PASS handoff framework has consistently led to reductions in medical errors, ¹ but observation and feedback are essential to success. ² Live observations increase supervisors’ workload and vary in quality. We developed a novel AI tool that automates I-PASS element detection in verbal handoffs, reducing cognitive load on observers while providing scalable, consistent feedback for iterative improvement and real-time educational intervention.
Speaker(s):
Josh Pankin, MD
Baystate Medical Center
Author(s):
Josh Pankin, MD - Baystate Medical Center; Vishal Pallerla, B.A. - Intersystems; Samantha Pendleton, D.O - Baystate Medical Center; Hannah Miller, M.D. - Baystate Medical Center; Don Woodlock, B.Sc. - InterSystems; Jonathan Teich, MD, PhD - InterSystems Corporation; Qi Li, M.D. - Intersystems; Molly Senn-McNally, M.D. - Baystate Medical Center; Christopher Landrigan, M.D., M.P.H. - Boston Children's Hospital; Amy Starmer, MD, MPH - Baystate Health;
Josh
Pankin,
MD - Baystate Medical Center
Implementing AI Documentation in a Safety-Net Health System
Presentation Type: Poster - Regular
Poster Number: 141
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Generative AI in Clinical Workflow: Ambient Listening, Chart Summarization, Automated Response with LLM, Outcomes Improvement and Equity, Clinician Well-Being, Quality Informatics and Lean, Change Management, Human Factors and Usability, Education and Training
Primary Track: Advancing Wellness for Providers and Community with Consideration of Human Factors
Documentation burden and after-hours EHR work drive clinician burnout and disproportionately affect safety-net systems with complex patient needs and limited resources. We conducted a 60-day AI documentation pilot with 84 clinicians across primary care, ambulatory clinics, and the emergency department. Over 20,000 notes were generated. EHR efficiency measures showed less time in notes and pajama time, improved HCC gap closure, scribe/dictation substitution, improved patient satisfaction, acceptance, and increased productivity.
Speaker(s):
Shashank Nayak, MD, MPH
Case Western MetroHealth
Author(s):
Eman Jammali, MD - MetroHealth System/Case Western Reserve University; Muhammad Alghanem, DO - MetroHealth Medical Center; Ellen Gelles, MD, FACP - MetroHealth System, Case Western Reserve School of Medicine; David Kaelber, MD, PhD, MPH - Case Western Reserve University/The MetroHealth System;
Presentation Type: Poster - Regular
Poster Number: 141
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Generative AI in Clinical Workflow: Ambient Listening, Chart Summarization, Automated Response with LLM, Outcomes Improvement and Equity, Clinician Well-Being, Quality Informatics and Lean, Change Management, Human Factors and Usability, Education and Training
Primary Track: Advancing Wellness for Providers and Community with Consideration of Human Factors
Documentation burden and after-hours EHR work drive clinician burnout and disproportionately affect safety-net systems with complex patient needs and limited resources. We conducted a 60-day AI documentation pilot with 84 clinicians across primary care, ambulatory clinics, and the emergency department. Over 20,000 notes were generated. EHR efficiency measures showed less time in notes and pajama time, improved HCC gap closure, scribe/dictation substitution, improved patient satisfaction, acceptance, and increased productivity.
Speaker(s):
Shashank Nayak, MD, MPH
Case Western MetroHealth
Author(s):
Eman Jammali, MD - MetroHealth System/Case Western Reserve University; Muhammad Alghanem, DO - MetroHealth Medical Center; Ellen Gelles, MD, FACP - MetroHealth System, Case Western Reserve School of Medicine; David Kaelber, MD, PhD, MPH - Case Western Reserve University/The MetroHealth System;
Shashank
Nayak,
MD, MPH - Case Western MetroHealth
EHR Secure Messaging Before and After Wrong-Patient Order Errors
Presentation Type: Poster - Student
Poster Number: 142
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Health Data Science, Clinical Decision Support and Care Pathways, Human Factors and Usability, Workforce Automation, Communication, and Workflow Efficiency, Clinician Well-Being
Primary Track: Big Data for Health
Using secure messaging data and EHR audit logs from clinicians across 14 hospitals, we examined the association between messages immediately preceding or following an order and the likelihood of retract-and-reorder events. Messaging within 10 minutes before an order did not increase error likelihood, and presence of an error was not associated with messaging within 10 minutes after, suggesting that neither pre-error cognitive load nor post-error communication explains the association between messaging and wrong-patient order errors.
Speaker(s):
Joanne Wang, BA
WashU Medicine
Author(s):
Joanne Wang, BA - WashU Medicine; Elise Eiden, MS - WashU Medicine; Laura Baratta, MD/PhD - Washington University School of Medicine in St. Louis; Linn Xia, BS - Washington University in St. Louis; Bailey Osweiler, B.A. - Washington University in St. Louis; Thomas Kannampallil, PhD - Washington University School of Medicine; Sunny Lou, MD, PhD - Washington University, St. Louis; Daphne Lew, PhD, MPH - Washington University School of Medicine;
Presentation Type: Poster - Student
Poster Number: 142
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Health Data Science, Clinical Decision Support and Care Pathways, Human Factors and Usability, Workforce Automation, Communication, and Workflow Efficiency, Clinician Well-Being
Primary Track: Big Data for Health
Using secure messaging data and EHR audit logs from clinicians across 14 hospitals, we examined the association between messages immediately preceding or following an order and the likelihood of retract-and-reorder events. Messaging within 10 minutes before an order did not increase error likelihood, and presence of an error was not associated with messaging within 10 minutes after, suggesting that neither pre-error cognitive load nor post-error communication explains the association between messaging and wrong-patient order errors.
Speaker(s):
Joanne Wang, BA
WashU Medicine
Author(s):
Joanne Wang, BA - WashU Medicine; Elise Eiden, MS - WashU Medicine; Laura Baratta, MD/PhD - Washington University School of Medicine in St. Louis; Linn Xia, BS - Washington University in St. Louis; Bailey Osweiler, B.A. - Washington University in St. Louis; Thomas Kannampallil, PhD - Washington University School of Medicine; Sunny Lou, MD, PhD - Washington University, St. Louis; Daphne Lew, PhD, MPH - Washington University School of Medicine;
Joanne
Wang,
BA - WashU Medicine
Lessons learned from building a natural language-to-SQL interface to comprehend and query Epic Clarity and Caboodle data models
Presentation Type: Poster - Regular
Poster Number: 143
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Health Data Science, Analytics, Registries, and the Digital Command Center, Analytical Artificial Intelligence: ML, Digital Pathology, Imaging AI, Predictive Analytics, Governance
Primary Track: Big Data for Health
We share our experiences developing a novel approach to translate natural language prompts into SQL code tailored for Epic Systems Corporation relational databases—leveraging three key elements: (1) Epic metadata summarized with generative AI, (2) primary and foreign key column relationships to join related tables, and (3) function-calling capabilities of the Azure OpenAI API. While still in development, this has proved valuable for analytics exploration, and multiple Epic users are collaborating on an open-source repository.
Speaker(s):
Peter Hong, MD
Boston Children's Hospital
Author(s):
Jeff Clark, MS, MBA - Advocate Health; Peter Hong, MD - Boston Children's Hospital;
Presentation Type: Poster - Regular
Poster Number: 143
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Health Data Science, Analytics, Registries, and the Digital Command Center, Analytical Artificial Intelligence: ML, Digital Pathology, Imaging AI, Predictive Analytics, Governance
Primary Track: Big Data for Health
We share our experiences developing a novel approach to translate natural language prompts into SQL code tailored for Epic Systems Corporation relational databases—leveraging three key elements: (1) Epic metadata summarized with generative AI, (2) primary and foreign key column relationships to join related tables, and (3) function-calling capabilities of the Azure OpenAI API. While still in development, this has proved valuable for analytics exploration, and multiple Epic users are collaborating on an open-source repository.
Speaker(s):
Peter Hong, MD
Boston Children's Hospital
Author(s):
Jeff Clark, MS, MBA - Advocate Health; Peter Hong, MD - Boston Children's Hospital;
Peter
Hong,
MD - Boston Children's Hospital
Assessment of Computerized Provider Order Entry Using Real-World Orders in a Large Health System
Presentation Type: Poster - Regular
Poster Number: 144
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Health Data Science, Clinical Decision Support and Care Pathways, Outcomes Improvement and Equity, Innovation Partnerships, Implementation Science, and Learning Health Systems
Primary Track: Big Data for Health
We developed automated interactive tools to measure Veterans Health Administration (VHA) hospitals’
computerized provider order entry system performance in 17.4 million orders over two years. The system
successfully alerted ordering providers in 89% of 253 potentially unsafe drug interaction orders. Providers
responded by canceling 86% of the orders following drug interaction alerts. A system-wide knowledge-sharing
platform includes tools for facilities to receive safety feedback, investigate non-alerting orders, optimize
performance, and share best practices.
Speaker(s):
Amber Trickey, PhD
VETERANS AFFAIRS
Author(s):
Aaron Dietz, PhD - Department of Veterans Affairs; Phillip Ng, PharmD - Veterans Affairs; Angela Laurio, DrPh, RN - VETERANS AFFAIRS- Office of Health Informatics; David Classen, MD - University of Utah School of Medicine; Dea Hughes, MPH - Clinical Informatics and Data Management Office (CIDMO), Office of Health Informatics (OHI), Department of Veterans Affairs; Jonathan Nebeker, MD, MS - Department of Veterans Affairs;
Presentation Type: Poster - Regular
Poster Number: 144
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Health Data Science, Clinical Decision Support and Care Pathways, Outcomes Improvement and Equity, Innovation Partnerships, Implementation Science, and Learning Health Systems
Primary Track: Big Data for Health
We developed automated interactive tools to measure Veterans Health Administration (VHA) hospitals’
computerized provider order entry system performance in 17.4 million orders over two years. The system
successfully alerted ordering providers in 89% of 253 potentially unsafe drug interaction orders. Providers
responded by canceling 86% of the orders following drug interaction alerts. A system-wide knowledge-sharing
platform includes tools for facilities to receive safety feedback, investigate non-alerting orders, optimize
performance, and share best practices.
Speaker(s):
Amber Trickey, PhD
VETERANS AFFAIRS
Author(s):
Aaron Dietz, PhD - Department of Veterans Affairs; Phillip Ng, PharmD - Veterans Affairs; Angela Laurio, DrPh, RN - VETERANS AFFAIRS- Office of Health Informatics; David Classen, MD - University of Utah School of Medicine; Dea Hughes, MPH - Clinical Informatics and Data Management Office (CIDMO), Office of Health Informatics (OHI), Department of Veterans Affairs; Jonathan Nebeker, MD, MS - Department of Veterans Affairs;
Amber
Trickey,
PhD - VETERANS AFFAIRS
A Multi-Omic Data Ecosystem Supporting Disease-Focused Biomedical Research: Infrastructure and Impact of the Alzheimer’s Disease Knowledge Portal
Presentation Type: Poster - Regular
Poster Number: 145
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Health Data Science, Precision Medicine, Multi-Omics, and Pharmacology Integration, Standards, Terminology, and Interoperability, TEFCA, FHIR
Primary Track: Big Data for Health
Collaboration among researchers and responsible data sharing are crucial for advancing our understanding of and ability to treat disease. The NIH-funded Alzheimer’s Disease (AD) Knowledge Portal (adknowledgeportal.org) is a prime example of a framework to support an international research community using an open-science approach that prioritizes patient privacy. This resource enables broad access to both raw and processed omics data, analytical methodology, and research tools generated by the Alzheimer’s Disease research community.
Speaker(s):
Trisha Zintel, Ph.D.
Sage Bionetworks
Author(s):
Jessica Malenfant, MPH - Sage Bionetworks; Jo Scanlan, BS - Sage Bionetworks; Tiara Adams, BS - Sage Bionetworks; Samia Ahmed, BS - Sage Bionetworks; Ram Ayyala, MS - Sage Bionetworks; Victor Baham, BA - Sage Bionetworks; Jaclyn Beck, PhD - Sage Bionetworks; Jessica S. Britton, PhD - Sage Bionetworks; Susheel Varma, PhD MBA FBCS - Sage Bionetworks; Laura Heath, PhD - Sage Bionetworks;
Presentation Type: Poster - Regular
Poster Number: 145
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Health Data Science, Precision Medicine, Multi-Omics, and Pharmacology Integration, Standards, Terminology, and Interoperability, TEFCA, FHIR
Primary Track: Big Data for Health
Collaboration among researchers and responsible data sharing are crucial for advancing our understanding of and ability to treat disease. The NIH-funded Alzheimer’s Disease (AD) Knowledge Portal (adknowledgeportal.org) is a prime example of a framework to support an international research community using an open-science approach that prioritizes patient privacy. This resource enables broad access to both raw and processed omics data, analytical methodology, and research tools generated by the Alzheimer’s Disease research community.
Speaker(s):
Trisha Zintel, Ph.D.
Sage Bionetworks
Author(s):
Jessica Malenfant, MPH - Sage Bionetworks; Jo Scanlan, BS - Sage Bionetworks; Tiara Adams, BS - Sage Bionetworks; Samia Ahmed, BS - Sage Bionetworks; Ram Ayyala, MS - Sage Bionetworks; Victor Baham, BA - Sage Bionetworks; Jaclyn Beck, PhD - Sage Bionetworks; Jessica S. Britton, PhD - Sage Bionetworks; Susheel Varma, PhD MBA FBCS - Sage Bionetworks; Laura Heath, PhD - Sage Bionetworks;
Trisha
Zintel,
Ph.D. - Sage Bionetworks
Inpatient Population Substructure effects on CMI adjusted LOS
Presentation Type: Poster - Regular
Poster Number: 146
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Health Policy, Reimbursement and Affordability, and Sustainability, Leadership and Strategy, Analytical Artificial Intelligence: ML, Digital Pathology, Imaging AI, Predictive Analytics, Governance, Social Determinants of Health (SDoH)
Primary Track: Big Data for Health
Background:
Case-Mix Index (CMI)–adjusted Length of Stay (LOS), operationalized at Vanderbilt University Medical Center (VUMC) as CARLOS, is a core performance metric for Hospital Medicine (HM), representing average reimbursement per inpatient bed day. Traditional performance analyses assume a relatively homogeneous inpatient cohort. This study examines whether population substructure—specifically a statistically distinct class of “Mega-Outliers” (MOs)—significantly distorts CARLOS and obscures realistic improvement opportunities.
Methods:
Retrospective utilization and operational data were extracted from VUMC’s enterprise PowerBI analytics environment for all HM discharges between July 1, 2024 and April 1, 2025. MOs were defined as admissions approximately five standard deviations above the mean CARLOS (CARLOS > 25). Clinical characteristics, discharge patterns, ICU utilization, and contribution to overall CARLOS performance were quantified. A sensitivity analysis recalculated CARLOS after capping MO values at 4, representing discharge at the institutional target.
Results:
Although MOs represented only 0.6% of the HM population (≈32 patients), they accounted for nearly 4% of total inpatient days. These patients required prolonged ICU care, extensive interdisciplinary coordination, and discharge planning influenced by system-level constraints—factors largely outside HM’s direct control. Censoring MOs at CARLOS = 4 improved the HM mean from 4.18 to 4.04, surpassing the modeled theoretical “reasonable best” (4.06) and yielding a performance delta unlikely to be achieved through bedside practice alone.
Discussion:
An identifiable inpatient subgroup exerts disproportionate influence on CARLOS. Informatics-driven early detection, automated signal generation, and targeted escalation—such as specialized case management—may provide greater operational benefit than broad process change. Ongoing work includes deployment of a machine-learning model for real-time MO flagging and intervention.
Speaker(s):
Jacob Franco, MD
Vanderbilt University Medical Center
Author(s):
Jacob Franco, MD - Vanderbilt University Medical Center;
Presentation Type: Poster - Regular
Poster Number: 146
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Health Policy, Reimbursement and Affordability, and Sustainability, Leadership and Strategy, Analytical Artificial Intelligence: ML, Digital Pathology, Imaging AI, Predictive Analytics, Governance, Social Determinants of Health (SDoH)
Primary Track: Big Data for Health
Background:
Case-Mix Index (CMI)–adjusted Length of Stay (LOS), operationalized at Vanderbilt University Medical Center (VUMC) as CARLOS, is a core performance metric for Hospital Medicine (HM), representing average reimbursement per inpatient bed day. Traditional performance analyses assume a relatively homogeneous inpatient cohort. This study examines whether population substructure—specifically a statistically distinct class of “Mega-Outliers” (MOs)—significantly distorts CARLOS and obscures realistic improvement opportunities.
Methods:
Retrospective utilization and operational data were extracted from VUMC’s enterprise PowerBI analytics environment for all HM discharges between July 1, 2024 and April 1, 2025. MOs were defined as admissions approximately five standard deviations above the mean CARLOS (CARLOS > 25). Clinical characteristics, discharge patterns, ICU utilization, and contribution to overall CARLOS performance were quantified. A sensitivity analysis recalculated CARLOS after capping MO values at 4, representing discharge at the institutional target.
Results:
Although MOs represented only 0.6% of the HM population (≈32 patients), they accounted for nearly 4% of total inpatient days. These patients required prolonged ICU care, extensive interdisciplinary coordination, and discharge planning influenced by system-level constraints—factors largely outside HM’s direct control. Censoring MOs at CARLOS = 4 improved the HM mean from 4.18 to 4.04, surpassing the modeled theoretical “reasonable best” (4.06) and yielding a performance delta unlikely to be achieved through bedside practice alone.
Discussion:
An identifiable inpatient subgroup exerts disproportionate influence on CARLOS. Informatics-driven early detection, automated signal generation, and targeted escalation—such as specialized case management—may provide greater operational benefit than broad process change. Ongoing work includes deployment of a machine-learning model for real-time MO flagging and intervention.
Speaker(s):
Jacob Franco, MD
Vanderbilt University Medical Center
Author(s):
Jacob Franco, MD - Vanderbilt University Medical Center;
Jacob
Franco,
MD - Vanderbilt University Medical Center
A Scoping Review of Strategies for Tailoring Patient-centered Technologies to Underserved Populations across the Cancer Continuum
Presentation Type: Poster - Regular
Poster Number: 147
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Human Factors and Usability, Outcomes Improvement and Equity, Population Health, Digital Therapeutics, Remote Patient Monitoring (RPM), and Digital Engagement
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Health information technologies (HIT) can bridge the gap between health inequities and cancer care; however, technologies are not always developed with populations of interest in mind. This study finds that tailoring strategies such as engaging community members, providing content that is culturally appropriate, and implementing HIT in the preferred mode of delivery improves engagement and acceptability. We provide a framework for understanding the most effective methods and modes of delivery to use when engaging populations.
Speaker(s):
Willi Tarver, DrPH, MLIS
The Ohio State University
Author(s):
Diamond Boyd, BS - Central State University; Pallavi Jonnalagadda, MBBS, DrPH - The Ohio State University College of Medicine; Mireille Bitangacha, BA - The Ohio State University College of Medicine; Timothy Pawlik, MD, PhD - The Ohio State University College of Medicine; Elizabeth Palmer Kelly, PhD - The Ohio State University Comprehensive Cancer Center; Electra Paskett, PhD - The Ohio State University College of Medicine;
Presentation Type: Poster - Regular
Poster Number: 147
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Human Factors and Usability, Outcomes Improvement and Equity, Population Health, Digital Therapeutics, Remote Patient Monitoring (RPM), and Digital Engagement
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Health information technologies (HIT) can bridge the gap between health inequities and cancer care; however, technologies are not always developed with populations of interest in mind. This study finds that tailoring strategies such as engaging community members, providing content that is culturally appropriate, and implementing HIT in the preferred mode of delivery improves engagement and acceptability. We provide a framework for understanding the most effective methods and modes of delivery to use when engaging populations.
Speaker(s):
Willi Tarver, DrPH, MLIS
The Ohio State University
Author(s):
Diamond Boyd, BS - Central State University; Pallavi Jonnalagadda, MBBS, DrPH - The Ohio State University College of Medicine; Mireille Bitangacha, BA - The Ohio State University College of Medicine; Timothy Pawlik, MD, PhD - The Ohio State University College of Medicine; Elizabeth Palmer Kelly, PhD - The Ohio State University Comprehensive Cancer Center; Electra Paskett, PhD - The Ohio State University College of Medicine;
Willi
Tarver,
DrPH, MLIS - The Ohio State University
Applying Human Factors Principles to Clinical Staff Pick Lists: Automating Measurement and Reporting of Potentially Inappropriate Additional Signer Entries in the Veterans Health Administration
Presentation Type: Poster - Regular
Poster Number: 148
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Human Factors and Usability, Clinician Well-Being, Quality Informatics and Lean
Primary Track: Advancing Wellness for Providers and Community with Consideration of Human Factors
Veterans Health Administration Additional Signer (AS) lists frequently conflict with human factors principles. Our analysis of 16 sites revealed that 380–585 per 1,000 list entries were potentially inappropriate (e.g., non-clinical staff), with higher potentially inappropriate entries at lesser complexity, often small, sites. We are developing automated tools to mask these entries and visually distinguish similar names. These interventions align lists with safety frameworks, reducing cognitive burden and communication errors.
Speaker(s):
Shardool Patel
VA Salt Lake City
Author(s):
Trevor Jones, MD - Salt Lake City VA; Tania Knight, BSN - Veterans Affairs Medical Center; Lilibeth Zafico, MBA, BSN, RN - VISN 19; J Edward Maddela, MD, MBA, FAMIA - University of Arizona and U.S. Department of Veterans Affairs; Andrea Bleak, BS - VA Salt Lake City;
Presentation Type: Poster - Regular
Poster Number: 148
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Human Factors and Usability, Clinician Well-Being, Quality Informatics and Lean
Primary Track: Advancing Wellness for Providers and Community with Consideration of Human Factors
Veterans Health Administration Additional Signer (AS) lists frequently conflict with human factors principles. Our analysis of 16 sites revealed that 380–585 per 1,000 list entries were potentially inappropriate (e.g., non-clinical staff), with higher potentially inappropriate entries at lesser complexity, often small, sites. We are developing automated tools to mask these entries and visually distinguish similar names. These interventions align lists with safety frameworks, reducing cognitive burden and communication errors.
Speaker(s):
Shardool Patel
VA Salt Lake City
Author(s):
Trevor Jones, MD - Salt Lake City VA; Tania Knight, BSN - Veterans Affairs Medical Center; Lilibeth Zafico, MBA, BSN, RN - VISN 19; J Edward Maddela, MD, MBA, FAMIA - University of Arizona and U.S. Department of Veterans Affairs; Andrea Bleak, BS - VA Salt Lake City;
Shardool
Patel - VA Salt Lake City
Human-Centered Approaches for a Mobile Health App for People with Systemic Sclerosis-Associated Raynaud's Phenomenon: Usability Testing and Clinical Trial Insights
Presentation Type: Poster Invite - Regular
Poster Number: 149
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Human Factors and Usability, Population Health, Digital Therapeutics, Remote Patient Monitoring (RPM), and Digital Engagement, Data Privacy, Cybersecurity, Reliability, and Security
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
This study aims to evaluate the usability of the Raynaud app for digital self-reporting of systemic sclerosis-associated Raynaud’s phenomenon. Participants found the app useful and preferred this digital reporting method, but requested improvements related to dexterity limitations, accessibility, network connectivity, caching, and unclear prompts. Combined with findings from the published trial, these results guided key feature enhancements to support accurate patient-report outcomes and encourage confident daily reporting in real-world settings.
Speaker(s):
Haomin Hu, PhD in Rehabilitation Science
University of Pittsburgh
Author(s):
I Made Agus Setiawan, PhD - University of Pittsburgh; Maureen Laffoon, BS - University of Pittsburgh; Andi Saptono, PhD - University of Pittsburgh School of Health and Rehabilitation Services; Robyn Domsic-Degazio, MD, MPH - University of Pittsburgh; Bambang Parmanto, PhD - University of Pittsburgh;
Presentation Type: Poster Invite - Regular
Poster Number: 149
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Human Factors and Usability, Population Health, Digital Therapeutics, Remote Patient Monitoring (RPM), and Digital Engagement, Data Privacy, Cybersecurity, Reliability, and Security
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
This study aims to evaluate the usability of the Raynaud app for digital self-reporting of systemic sclerosis-associated Raynaud’s phenomenon. Participants found the app useful and preferred this digital reporting method, but requested improvements related to dexterity limitations, accessibility, network connectivity, caching, and unclear prompts. Combined with findings from the published trial, these results guided key feature enhancements to support accurate patient-report outcomes and encourage confident daily reporting in real-world settings.
Speaker(s):
Haomin Hu, PhD in Rehabilitation Science
University of Pittsburgh
Author(s):
I Made Agus Setiawan, PhD - University of Pittsburgh; Maureen Laffoon, BS - University of Pittsburgh; Andi Saptono, PhD - University of Pittsburgh School of Health and Rehabilitation Services; Robyn Domsic-Degazio, MD, MPH - University of Pittsburgh; Bambang Parmanto, PhD - University of Pittsburgh;
Haomin
Hu,
PhD in Rehabilitation Science - University of Pittsburgh
Scaling Causal Inference for a Pediatric Behavioral and Mental Health Learning Health System
Presentation Type: Poster - Regular
Poster Number: 150
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Innovation Partnerships, Implementation Science, and Learning Health Systems, Health Data Science, Outcomes Improvement and Equity, Analytics, Registries, and the Digital Command Center
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
We developed a web-based tool that automates causal inference–oriented comparative effectiveness analyses at a pediatric behavioral and mental health learning health system. Using EHR data, users define cohorts, outcomes, and subgroups, then run adjusted models with diagnostics and interactive outputs. The tool informs decisions about program expansion and resource allocation by providing timely, standardized evidence of impact on preventing ED recidivism.
Speaker(s):
Nikolay Braykov, MS
Children's Helathcare of Atlanta
Author(s):
Nikolay Braykov, MS - Children's Helathcare of Atlanta; Andrea McCarter, MS, PhD - Children's Healthcare of Atlanta; Hillary Henderson, MSc, LSSBB - Children's Healthcare of Atlanta; Afrin Jahan, MPH - Children's Healthcare of Atlanta; Xiaotao Jing, MS - Children's Healthcare of Atlanta; Laura Gillard; Kayla Mays, DNP, APRN, PMHNP-BC - Children's Healthcare of Atlanta; Gargi Mukherjee, MD - Emory University School of Medicine; William Copeland, PhD - University of Vermont; Janet Cummings, PhD - Emory University; Naveen Muthu, MD - Children's Healthcare of Atlanta; Evan Orenstein, MD - Children's Healthcare of Atlanta; John Constantino, PhD - Emory University;
Presentation Type: Poster - Regular
Poster Number: 150
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Innovation Partnerships, Implementation Science, and Learning Health Systems, Health Data Science, Outcomes Improvement and Equity, Analytics, Registries, and the Digital Command Center
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
We developed a web-based tool that automates causal inference–oriented comparative effectiveness analyses at a pediatric behavioral and mental health learning health system. Using EHR data, users define cohorts, outcomes, and subgroups, then run adjusted models with diagnostics and interactive outputs. The tool informs decisions about program expansion and resource allocation by providing timely, standardized evidence of impact on preventing ED recidivism.
Speaker(s):
Nikolay Braykov, MS
Children's Helathcare of Atlanta
Author(s):
Nikolay Braykov, MS - Children's Helathcare of Atlanta; Andrea McCarter, MS, PhD - Children's Healthcare of Atlanta; Hillary Henderson, MSc, LSSBB - Children's Healthcare of Atlanta; Afrin Jahan, MPH - Children's Healthcare of Atlanta; Xiaotao Jing, MS - Children's Healthcare of Atlanta; Laura Gillard; Kayla Mays, DNP, APRN, PMHNP-BC - Children's Healthcare of Atlanta; Gargi Mukherjee, MD - Emory University School of Medicine; William Copeland, PhD - University of Vermont; Janet Cummings, PhD - Emory University; Naveen Muthu, MD - Children's Healthcare of Atlanta; Evan Orenstein, MD - Children's Healthcare of Atlanta; John Constantino, PhD - Emory University;
Nikolay
Braykov,
MS - Children's Helathcare of Atlanta
The Internal Startup: A Scalable Framework for Clinical Tool Development and Deployment
Presentation Type: Poster - Regular
Poster Number: 151
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Innovation Partnerships, Implementation Science, and Learning Health Systems, Clinician Well-Being, Workforce Automation, Communication, and Workflow Efficiency
Primary Track: Advancing Wellness for Providers and Community with Consideration of Human Factors
To address rigid EHR workflows and slow vendor solutions, the Stanford Emerging Applications Lab (SEAL) created an agile, “embedded startup” model that builds SMART-on-FHIR tools directly within Epic. In just a short time, the team deployed thirteen applications, including a Nutrition Provision tool that reduced task time from 10–15 minutes to 5–10 minutes. Used by 1,984 clinicians across 12,927 patients, SEAL demonstrates how an embedded, rapid-development framework can meaningfully reduce friction and clinician burden.
Speaker(s):
Imran Mohiuddin, MD
Stanford Medicine
Author(s):
Imran Mohiuddin, MD - Stanford Medicine; Oluseyi Fayanju, MD - Stanford University; Stephen Ma, MD, PhD - Stanford University School of Medicine; Shivam Vedak, MD, MBA - Stanford University School of Medicine; Jennifer Tran, PharmD; Srinivasan Boosi, BE - Stanford Health Care; Lisa Gohil, MBA/MPH - Stanford Health Care; Ron Li, MD - Stanford School of Medicine; Lawrence Hoffman, MD - Stanford Health Care; Christopher Sharp, MD - Stanford University School of Medicine;
Presentation Type: Poster - Regular
Poster Number: 151
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Innovation Partnerships, Implementation Science, and Learning Health Systems, Clinician Well-Being, Workforce Automation, Communication, and Workflow Efficiency
Primary Track: Advancing Wellness for Providers and Community with Consideration of Human Factors
To address rigid EHR workflows and slow vendor solutions, the Stanford Emerging Applications Lab (SEAL) created an agile, “embedded startup” model that builds SMART-on-FHIR tools directly within Epic. In just a short time, the team deployed thirteen applications, including a Nutrition Provision tool that reduced task time from 10–15 minutes to 5–10 minutes. Used by 1,984 clinicians across 12,927 patients, SEAL demonstrates how an embedded, rapid-development framework can meaningfully reduce friction and clinician burden.
Speaker(s):
Imran Mohiuddin, MD
Stanford Medicine
Author(s):
Imran Mohiuddin, MD - Stanford Medicine; Oluseyi Fayanju, MD - Stanford University; Stephen Ma, MD, PhD - Stanford University School of Medicine; Shivam Vedak, MD, MBA - Stanford University School of Medicine; Jennifer Tran, PharmD; Srinivasan Boosi, BE - Stanford Health Care; Lisa Gohil, MBA/MPH - Stanford Health Care; Ron Li, MD - Stanford School of Medicine; Lawrence Hoffman, MD - Stanford Health Care; Christopher Sharp, MD - Stanford University School of Medicine;
Imran
Mohiuddin,
MD - Stanford Medicine
A Standardized “Sales Pitch” For New Technology Requests
Presentation Type: Poster - Regular
Poster Number: 152
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Leadership and Strategy, Education and Training, Change Management, Standards, Terminology, and Interoperability, TEFCA, FHIR
Primary Track: Driving Change at Scale through Effective Leadership and Governance
With the rapid development of technology and artificial intelligence, there has been an influx of requests to adopt new technology within the health system, so a standardized approach may be beneficial. As this is a relatively new phenomenon, standardized policies and procedures for evaluating these requests are lacking, and guidance on best practices for assessing these requests, which are rapidly increasing, is limited.
At our institution, there has not been a standardized method for clinicians to submit requests for new technology. As a result, requests have come in various formats, including Word documents and PowerPoint presentations, which may contain incomplete or insufficient information to be properly evaluated. Furthermore, the methods used to obtain supporting data have often been questionable or suboptimal, and financial projections have shown limited reliability.
A team of physician informaticists sought to standardize the technology request process for our hospital system using an iterative process to develop a structured presentation template in a “sales pitch” format to help streamline the review process and prepare applicants to deliver a concise and effective proposal. The final product was a 4-slide presentation template with 3 appendix slides containing information our executive team deemed necessary to evaluate the proposal.
Using a standardized PowerPoint presentation template may enhance completeness and also streamline workflow as the “application” is already in a format amenable to presenting to an executive team and facilitate efficient review. The “sales pitch” format minimizes extraneous information and encourages presenters to prepare a concise request with appropriately researched support.
Speaker(s):
Timothy Shimon, MD
Banner Health
Author(s):
Jennifer Fernandez, MD, RD - University of Arizona - Phoenix; Alan Weiss, MD, MBA, FACP - Self Employed;
Presentation Type: Poster - Regular
Poster Number: 152
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Leadership and Strategy, Education and Training, Change Management, Standards, Terminology, and Interoperability, TEFCA, FHIR
Primary Track: Driving Change at Scale through Effective Leadership and Governance
With the rapid development of technology and artificial intelligence, there has been an influx of requests to adopt new technology within the health system, so a standardized approach may be beneficial. As this is a relatively new phenomenon, standardized policies and procedures for evaluating these requests are lacking, and guidance on best practices for assessing these requests, which are rapidly increasing, is limited.
At our institution, there has not been a standardized method for clinicians to submit requests for new technology. As a result, requests have come in various formats, including Word documents and PowerPoint presentations, which may contain incomplete or insufficient information to be properly evaluated. Furthermore, the methods used to obtain supporting data have often been questionable or suboptimal, and financial projections have shown limited reliability.
A team of physician informaticists sought to standardize the technology request process for our hospital system using an iterative process to develop a structured presentation template in a “sales pitch” format to help streamline the review process and prepare applicants to deliver a concise and effective proposal. The final product was a 4-slide presentation template with 3 appendix slides containing information our executive team deemed necessary to evaluate the proposal.
Using a standardized PowerPoint presentation template may enhance completeness and also streamline workflow as the “application” is already in a format amenable to presenting to an executive team and facilitate efficient review. The “sales pitch” format minimizes extraneous information and encourages presenters to prepare a concise request with appropriately researched support.
Speaker(s):
Timothy Shimon, MD
Banner Health
Author(s):
Jennifer Fernandez, MD, RD - University of Arizona - Phoenix; Alan Weiss, MD, MBA, FACP - Self Employed;
Timothy
Shimon,
MD - Banner Health
Governmental Public Health Leadership: Insights from a National Survey on Informatics and Technical Training Needs
Presentation Type: Poster - Student
Poster Number: 153
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Leadership and Strategy, Education and Training, Public Surveillance and Reporting, Environmental Exposure, & Global Health
Primary Track: Driving Change at Scale through Effective Leadership and Governance
Public health leaders oversee modernization projects in public health, and some are multi-million investments. Informatics training is critical and studies that examine the skills gaps of public health leaders are limited. 2024 Public Health Workforce Interests and Needs Survey data on Tier 2 (supervisors, managers) and Tier 3 ( executives) (N=15,681) was analyzed. Findings show significant skill gap (p <.0001) in atleast one job category across six informatics-related questions evaluated and across public health settings.
Speaker(s):
Divya Rupini Gunashekar, PhD
University of Minnesota- Twin Cities
Author(s):
Divya Rupini Gunashekar, PhD - University of Minnesota- Twin Cities; Sripriya Rajamani, MBBS, MPH, PhD, FAMIA - University of Minnesota;
Presentation Type: Poster - Student
Poster Number: 153
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Leadership and Strategy, Education and Training, Public Surveillance and Reporting, Environmental Exposure, & Global Health
Primary Track: Driving Change at Scale through Effective Leadership and Governance
Public health leaders oversee modernization projects in public health, and some are multi-million investments. Informatics training is critical and studies that examine the skills gaps of public health leaders are limited. 2024 Public Health Workforce Interests and Needs Survey data on Tier 2 (supervisors, managers) and Tier 3 ( executives) (N=15,681) was analyzed. Findings show significant skill gap (p <.0001) in atleast one job category across six informatics-related questions evaluated and across public health settings.
Speaker(s):
Divya Rupini Gunashekar, PhD
University of Minnesota- Twin Cities
Author(s):
Divya Rupini Gunashekar, PhD - University of Minnesota- Twin Cities; Sripriya Rajamani, MBBS, MPH, PhD, FAMIA - University of Minnesota;
Divya Rupini
Gunashekar,
PhD - University of Minnesota- Twin Cities
Usability Assessment of a Nursing and Clinical Care Services Platform to Drive Informed Nursing Leadership Decisions
Presentation Type: Poster Invite - Regular
Poster Number: 154
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Leadership and Strategy, Analytics, Registries, and the Digital Command Center, Change Management
Primary Track: Driving Change at Scale through Effective Leadership and Governance
Healthcare leaders often face fragmented dashboards, inconsistent updates, and missing nursing-specific metrics, which limit data-driven decision-making. To address these challenges, The Children’s Hospital of Philadelphia’s Nursing Department partnered with various teams to create an enterprise nursing analytics platform. This multi-phase initiative integrates clinical, workforce, patient, family, and clinician satisfaction data into a standardized, real-time dashboard. Usability testing showed high satisfaction. Lessons learned include the need for analytics education and the use of multidisciplinary metrics to optimize nursing operations and outcomes.
Speaker(s):
Sherri Duarte, DNP, RN, NI-BC, CPHIMS, BRMP
Childrens Hospital of Philadelphia
Sherri Duarte, DNP
Children's Hospital of Philadelphia
Author(s):
Sherri Duarte, DNP, RN, NI-BC, CPHIMS, BRMP - Childrens Hospital of Philadelphia; Maria Tickner, BS - Childrens Hospital of Philadelphia; Timothy Higgins, BA - Childrens Hospital of Philadelphia; Margaret McCabe, PhD, RN - Childrens Hospital of Philadelphia; Martha A.Q. Curley, PhD, RN - Childrens Hospital of Philadelphia; Kenrick Cato, PhD, RN, CPHIMS, FAAN, FACMI - University of Pennsylvania/ Children's Hospital of Philadelphia;
Presentation Type: Poster Invite - Regular
Poster Number: 154
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Leadership and Strategy, Analytics, Registries, and the Digital Command Center, Change Management
Primary Track: Driving Change at Scale through Effective Leadership and Governance
Healthcare leaders often face fragmented dashboards, inconsistent updates, and missing nursing-specific metrics, which limit data-driven decision-making. To address these challenges, The Children’s Hospital of Philadelphia’s Nursing Department partnered with various teams to create an enterprise nursing analytics platform. This multi-phase initiative integrates clinical, workforce, patient, family, and clinician satisfaction data into a standardized, real-time dashboard. Usability testing showed high satisfaction. Lessons learned include the need for analytics education and the use of multidisciplinary metrics to optimize nursing operations and outcomes.
Speaker(s):
Sherri Duarte, DNP, RN, NI-BC, CPHIMS, BRMP
Childrens Hospital of Philadelphia
Sherri Duarte, DNP
Children's Hospital of Philadelphia
Author(s):
Sherri Duarte, DNP, RN, NI-BC, CPHIMS, BRMP - Childrens Hospital of Philadelphia; Maria Tickner, BS - Childrens Hospital of Philadelphia; Timothy Higgins, BA - Childrens Hospital of Philadelphia; Margaret McCabe, PhD, RN - Childrens Hospital of Philadelphia; Martha A.Q. Curley, PhD, RN - Childrens Hospital of Philadelphia; Kenrick Cato, PhD, RN, CPHIMS, FAAN, FACMI - University of Pennsylvania/ Children's Hospital of Philadelphia;
Sherri
Duarte,
DNP, RN, NI-BC, CPHIMS, BRMP - Childrens Hospital of Philadelphia
Sherri Duarte, DNP - Children's Hospital of Philadelphia
Sherri Duarte, DNP - Children's Hospital of Philadelphia
Digital Inclusion Screening Measures for Equitable Identification of Digital Support Need
Presentation Type: Poster - Regular
Poster Number: 155
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Outcomes Improvement and Equity, Telemedicine, Health at Home, and Virtual Care, Population Health, Digital Therapeutics, Remote Patient Monitoring (RPM), and Digital Engagement
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
We analyzed surveys and digital task assessments to identify which predictors most strongly indicate patients’ digital support needs. Although there are measures available to assess digital readiness, evidence of their predictive value is limited. Results will inform streamlined digital readiness screening workflows that flag patients at risk of digital exclusion and promote equitable engagement with portals, telehealth, and other digital health tools.
Speaker(s):
Elaine Khoong, MD, MS
University of California San Francisco
Author(s):
Maria Plascencia Mooradian, BA - UCSF; Elaine Khoong, MD, MS - University of California San Francisco; Courtney Lyles, PhD - UC Davis Center for Healthcare Policy and Research; Hyunjin Cindy Kim, MPH - UCSF; Isabel Luna, BA - UCSF; Jeannette Wong, BS - UCSF; Eric Li, BA - UCSF; Andersen Yang, MPH - UCSF; Lina Tieu, PhD - UC Davis Center for Healthcare Policy and Research;
Presentation Type: Poster - Regular
Poster Number: 155
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Outcomes Improvement and Equity, Telemedicine, Health at Home, and Virtual Care, Population Health, Digital Therapeutics, Remote Patient Monitoring (RPM), and Digital Engagement
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
We analyzed surveys and digital task assessments to identify which predictors most strongly indicate patients’ digital support needs. Although there are measures available to assess digital readiness, evidence of their predictive value is limited. Results will inform streamlined digital readiness screening workflows that flag patients at risk of digital exclusion and promote equitable engagement with portals, telehealth, and other digital health tools.
Speaker(s):
Elaine Khoong, MD, MS
University of California San Francisco
Author(s):
Maria Plascencia Mooradian, BA - UCSF; Elaine Khoong, MD, MS - University of California San Francisco; Courtney Lyles, PhD - UC Davis Center for Healthcare Policy and Research; Hyunjin Cindy Kim, MPH - UCSF; Isabel Luna, BA - UCSF; Jeannette Wong, BS - UCSF; Eric Li, BA - UCSF; Andersen Yang, MPH - UCSF; Lina Tieu, PhD - UC Davis Center for Healthcare Policy and Research;
Elaine
Khoong,
MD, MS - University of California San Francisco
Influence of Social Media on Cancer Fatalism among US adults: Rural-Urban Comparison
Presentation Type: Poster Invite - Regular
Poster Number: 156
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Population Health, Digital Therapeutics, Remote Patient Monitoring (RPM), and Digital Engagement, Social Determinants of Health (SDoH), Public Surveillance and Reporting, Environmental Exposure, & Global Health
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
This study assessed the influence of social media on cancer fatalism among US adults using population-based data. We found a statistically significant association between social media health information awareness and cancer fatalism. Those who self-reported using social media information for discussion with a doctor have higher odds of believing that everything causes cancer. Rural residence was not a factor. The findings indicate that preventive cancer behaviors are not uniform, and tailored communication strategies are needed.
Speaker(s):
Sayantani Sarkar, PhD
University of California Berkeley
Author(s):
Sayantani Sarkar, PhD - University of California Berkeley; Katherine Kim, PhD, MPH, MBA, FAMIA - University of California Davis;
Presentation Type: Poster Invite - Regular
Poster Number: 156
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Population Health, Digital Therapeutics, Remote Patient Monitoring (RPM), and Digital Engagement, Social Determinants of Health (SDoH), Public Surveillance and Reporting, Environmental Exposure, & Global Health
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
This study assessed the influence of social media on cancer fatalism among US adults using population-based data. We found a statistically significant association between social media health information awareness and cancer fatalism. Those who self-reported using social media information for discussion with a doctor have higher odds of believing that everything causes cancer. Rural residence was not a factor. The findings indicate that preventive cancer behaviors are not uniform, and tailored communication strategies are needed.
Speaker(s):
Sayantani Sarkar, PhD
University of California Berkeley
Author(s):
Sayantani Sarkar, PhD - University of California Berkeley; Katherine Kim, PhD, MPH, MBA, FAMIA - University of California Davis;
Sayantani
Sarkar,
PhD - University of California Berkeley
A Dual-Encoder Deep Learning Framework for Epilepsy Drug Repurposing Using Molecular and Gene Expression Data
Presentation Type: Poster - Student
Poster Number: 157
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Precision Medicine, Multi-Omics, and Pharmacology Integration, Analytical Artificial Intelligence: ML, Digital Pathology, Imaging AI, Predictive Analytics, Governance, Health Data Science
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
We present a dual-encoder deep learning framework integrating molecular structure and gene expression for epilepsy drug repurposing. Evaluated on 104,102 drug-condition profiles from 1,713 compounds, our approach achieved 91.6% AUROC and 76% precision at top-50 predictions, correctly identifying 38 of 50 anti-epileptic drug-condition profiles. This substantially outperformed single-modality and traditional machine learning methods. The multi-modal approach reduces experimental screening effort from 400 random to 50 targeted candidates.
Speaker(s):
RISHIK KONDADADI, MR.
EASTVIEW SENIOR HIGH SCHOOL
Author(s):
Dan McCreary, MS - Kelly-McCreary Associates; RISHIK KONDADADI, MR. - EASTVIEW SENIOR HIGH SCHOOL;
Presentation Type: Poster - Student
Poster Number: 157
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Precision Medicine, Multi-Omics, and Pharmacology Integration, Analytical Artificial Intelligence: ML, Digital Pathology, Imaging AI, Predictive Analytics, Governance, Health Data Science
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
We present a dual-encoder deep learning framework integrating molecular structure and gene expression for epilepsy drug repurposing. Evaluated on 104,102 drug-condition profiles from 1,713 compounds, our approach achieved 91.6% AUROC and 76% precision at top-50 predictions, correctly identifying 38 of 50 anti-epileptic drug-condition profiles. This substantially outperformed single-modality and traditional machine learning methods. The multi-modal approach reduces experimental screening effort from 400 random to 50 targeted candidates.
Speaker(s):
RISHIK KONDADADI, MR.
EASTVIEW SENIOR HIGH SCHOOL
Author(s):
Dan McCreary, MS - Kelly-McCreary Associates; RISHIK KONDADADI, MR. - EASTVIEW SENIOR HIGH SCHOOL;
RISHIK
KONDADADI,
MR. - EASTVIEW SENIOR HIGH SCHOOL
Modernizing Occupational Health Surveillance: A Two-Year, Single-Institute Quality Improvement Study Following Survey Tools Digitization for Evaluation of Animal Allergen Exposure
Presentation Type: Poster - Student
Poster Number: 158
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Public Surveillance and Reporting, Environmental Exposure, & Global Health, Workforce Automation, Communication, and Workflow Efficiency, Change Management
Primary Track: Driving Change at Scale through Effective Leadership and Governance
Occupational health surveillance participation at the San Francisco VA was historically low due to paper-based Animal Allergen Exposure Survey (AAES) workflows. In 2024, we modernized the AAES using REDCap, enabling features like mobile-responsive design, real-time validation, and QR-code access. Survey completion rates increased significantly from 26.4% in 2023 to 59.5% in 2025, with strong user preference for the digital format. Digitization and human-centered design substantially improved survey completion and usability, strengthening occupational health surveillance.
Speaker(s):
Sristi Sharma, M.D., M.P.H.
UCSF
Author(s):
Sara Faghihi Kashani, MD, MPH - UCSF; Doreen Bernstein, MSN, RN, PHN, AGPCNP -BC - San Francisco VA Health Care System; Sandeep Guntur, MD, MPH - University of California, San Francisco | San Francisco VA Health Care System;
Presentation Type: Poster - Student
Poster Number: 158
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Public Surveillance and Reporting, Environmental Exposure, & Global Health, Workforce Automation, Communication, and Workflow Efficiency, Change Management
Primary Track: Driving Change at Scale through Effective Leadership and Governance
Occupational health surveillance participation at the San Francisco VA was historically low due to paper-based Animal Allergen Exposure Survey (AAES) workflows. In 2024, we modernized the AAES using REDCap, enabling features like mobile-responsive design, real-time validation, and QR-code access. Survey completion rates increased significantly from 26.4% in 2023 to 59.5% in 2025, with strong user preference for the digital format. Digitization and human-centered design substantially improved survey completion and usability, strengthening occupational health surveillance.
Speaker(s):
Sristi Sharma, M.D., M.P.H.
UCSF
Author(s):
Sara Faghihi Kashani, MD, MPH - UCSF; Doreen Bernstein, MSN, RN, PHN, AGPCNP -BC - San Francisco VA Health Care System; Sandeep Guntur, MD, MPH - University of California, San Francisco | San Francisco VA Health Care System;
Sristi
Sharma,
M.D., M.P.H. - UCSF
Standardizing Documentation of Medical Decision Making in Hospital Medicine: A Quality Improvement Initiative
Presentation Type: Poster Invite - Regular
Poster Number: 159
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Quality Informatics and Lean, Change Management, Leadership and Strategy, Human Factors and Usability, Clinical Decision Support and Care Pathways
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Accurate documentation of medical decision making is essential for compliance and billing under 2023 CMS guidelines, yet variability in note structure leads to systematic undercoding. We implemented two standardized EHR tools with education and incentives, increasing adoption from 18% to >90% and high-complexity CPT coding from 17% to 52%. Template use strongly correlated with coding accuracy (R²=0.89). Metadata showed more meaningful manual text with equal or shorter editing times, demonstrating improved documentation quality and efficiency.
Speaker(s):
Loukya Kanakamedala, DO
CHOP
Author(s):
Peter Zhang, MD, MS - Children's Hospital of Philadelphia / University of Pennsylvania;
Presentation Type: Poster Invite - Regular
Poster Number: 159
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Quality Informatics and Lean, Change Management, Leadership and Strategy, Human Factors and Usability, Clinical Decision Support and Care Pathways
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Accurate documentation of medical decision making is essential for compliance and billing under 2023 CMS guidelines, yet variability in note structure leads to systematic undercoding. We implemented two standardized EHR tools with education and incentives, increasing adoption from 18% to >90% and high-complexity CPT coding from 17% to 52%. Template use strongly correlated with coding accuracy (R²=0.89). Metadata showed more meaningful manual text with equal or shorter editing times, demonstrating improved documentation quality and efficiency.
Speaker(s):
Loukya Kanakamedala, DO
CHOP
Author(s):
Peter Zhang, MD, MS - Children's Hospital of Philadelphia / University of Pennsylvania;
Loukya
Kanakamedala,
DO - CHOP
Rapid Patient Feedback on Patient Experience: Leveraging HCAHPS to Gather Data on Patient-Facing Interventions
Presentation Type: Poster - Student
Poster Number: 160
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Quality Informatics and Lean, Innovation Partnerships, Implementation Science, and Learning Health Systems, Public Surveillance and Reporting, Environmental Exposure, & Global Health
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
We demonstrate how Hospital Consumer Assessment of Healthcare Providers and Systems (HCAHPS) surveys can rapidly evaluate patient-facing interventions. By analyzing 15,734 patients' responses to the 21st Century Cures Act's immediate test result release policy, we found 88% preferred continued immediate access. Viewing results significantly reduced anxiety (58.1% decreased vs 4.6% increased). This scalable methodology enables timely assessment of policy impacts on patients.
Speaker(s):
Uday Suresh, MS
Vanderbilt University Department of Biomedical Informatics
Author(s):
Jessica Ancker, MPH, PhD, FACMI - Vanderbilt University Medical Center; Bryan Steitz, PhD - Vanderbilt University Medical Center;
Presentation Type: Poster - Student
Poster Number: 160
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Quality Informatics and Lean, Innovation Partnerships, Implementation Science, and Learning Health Systems, Public Surveillance and Reporting, Environmental Exposure, & Global Health
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
We demonstrate how Hospital Consumer Assessment of Healthcare Providers and Systems (HCAHPS) surveys can rapidly evaluate patient-facing interventions. By analyzing 15,734 patients' responses to the 21st Century Cures Act's immediate test result release policy, we found 88% preferred continued immediate access. Viewing results significantly reduced anxiety (58.1% decreased vs 4.6% increased). This scalable methodology enables timely assessment of policy impacts on patients.
Speaker(s):
Uday Suresh, MS
Vanderbilt University Department of Biomedical Informatics
Author(s):
Jessica Ancker, MPH, PhD, FACMI - Vanderbilt University Medical Center; Bryan Steitz, PhD - Vanderbilt University Medical Center;
Uday
Suresh,
MS - Vanderbilt University Department of Biomedical Informatics
Wearable Use and Health Outcomes Among Older Adults with Chronic Conditions
Presentation Type: Poster - Regular
Poster Number: 161
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Social Determinants of Health (SDoH), Wearable Health Technology, Analytical Artificial Intelligence: ML, Digital Pathology, Imaging AI, Predictive Analytics, Governance, Human Factors and Usability
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Wearable devices may support health behaviors and self-efficacy among older adults with multiple chronic conditions (MCC), but evidence is limited. The aim of this study was to examine associations between wearable device use, health-related self-efficacy, and physical activity composition among older adults with MCC. This was a cross-sectional analysis of 910 adults aged 65+ with ≥2 chronic conditions from HINTS 6 (2022-2023). Survey-weighted linear regression adjusted for sociodemographic factors was used to analyze the data. Wearable device use was associated with higher physical activity composition (β=0.129, p=0.042) but not health-related self-efficacy (β=-0.055, p=0.637). Findings demonstrate differential effects: wearables modestly support physical activity but show neutral effects on health-related self-efficacy, suggesting technology alone may be insufficient to improve health perceptions in this population.
Speaker(s):
Dante Anthony Tolentino, PhD, RN-BC
UCLA School of Nursing
Author(s):
Dante Anthony Tolentino, PhD, RN-BC - UCLA School of Nursing; Paul Boy, MSN, MPH - University of California, Los Angeles; Yuriko Matsuo, MSN - University of California, Los Angeles;
Presentation Type: Poster - Regular
Poster Number: 161
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Social Determinants of Health (SDoH), Wearable Health Technology, Analytical Artificial Intelligence: ML, Digital Pathology, Imaging AI, Predictive Analytics, Governance, Human Factors and Usability
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Wearable devices may support health behaviors and self-efficacy among older adults with multiple chronic conditions (MCC), but evidence is limited. The aim of this study was to examine associations between wearable device use, health-related self-efficacy, and physical activity composition among older adults with MCC. This was a cross-sectional analysis of 910 adults aged 65+ with ≥2 chronic conditions from HINTS 6 (2022-2023). Survey-weighted linear regression adjusted for sociodemographic factors was used to analyze the data. Wearable device use was associated with higher physical activity composition (β=0.129, p=0.042) but not health-related self-efficacy (β=-0.055, p=0.637). Findings demonstrate differential effects: wearables modestly support physical activity but show neutral effects on health-related self-efficacy, suggesting technology alone may be insufficient to improve health perceptions in this population.
Speaker(s):
Dante Anthony Tolentino, PhD, RN-BC
UCLA School of Nursing
Author(s):
Dante Anthony Tolentino, PhD, RN-BC - UCLA School of Nursing; Paul Boy, MSN, MPH - University of California, Los Angeles; Yuriko Matsuo, MSN - University of California, Los Angeles;
Dante Anthony
Tolentino,
PhD, RN-BC - UCLA School of Nursing
Turning Point-of-Care Electronic Health Record-Captured mCODE-aligned Structured Data into Trial-Grade Real World Data (RWD) – Two-Site Proof of Concept
Presentation Type: Poster Invite - Regular
Poster Number: 162
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Standards, Terminology, and Interoperability, TEFCA, FHIR, Workforce Automation, Communication, and Workflow Efficiency, Outcomes Improvement and Equity, Quality Informatics and Lean
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
We implemented mCODE-aligned structured data elements (SDEs) within routine oncology note templates at two sites and evaluated whether point-of-care structured data can serve as fit-for-purpose clinical trial real-world data (RWD). In a two-site proof-of-concept (n=100), baseline cancer stage, baseline performance status and cycles 1-4 performance/disease status showed high electronic health record (EHR) → electronic case report form (eCRF) concordance. EHR data were available shortly after note signature, preceding eCRF entry in most cases, supporting future EHR→ electronic data capture (EDC) integration.
Speaker(s):
Hamid Emamekhoo, MD
University of Wisconsin Madison
Author(s):
Hamid Emamekhoo, MD - University of Wisconsin Madison;
Presentation Type: Poster Invite - Regular
Poster Number: 162
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Standards, Terminology, and Interoperability, TEFCA, FHIR, Workforce Automation, Communication, and Workflow Efficiency, Outcomes Improvement and Equity, Quality Informatics and Lean
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
We implemented mCODE-aligned structured data elements (SDEs) within routine oncology note templates at two sites and evaluated whether point-of-care structured data can serve as fit-for-purpose clinical trial real-world data (RWD). In a two-site proof-of-concept (n=100), baseline cancer stage, baseline performance status and cycles 1-4 performance/disease status showed high electronic health record (EHR) → electronic case report form (eCRF) concordance. EHR data were available shortly after note signature, preceding eCRF entry in most cases, supporting future EHR→ electronic data capture (EDC) integration.
Speaker(s):
Hamid Emamekhoo, MD
University of Wisconsin Madison
Author(s):
Hamid Emamekhoo, MD - University of Wisconsin Madison;
Hamid
Emamekhoo,
MD - University of Wisconsin Madison
Evaluation of a mapping tool using the WHO-FIC Mapping Principles
Presentation Type: Poster - Regular
Poster Number: 163
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Standards, Terminology, and Interoperability, TEFCA, FHIR, Health Data Science, Workforce Automation, Communication, and Workflow Efficiency
Primary Track: Big Data for Health
Mappings between terminologies and between local concepts and terminologies ensure consistent semantic meaning and the process of creating mappings is streamlined by using an automated mapping tool. We used the WHO-FIC Mapping Principles and Best Practice recommendations to validate our mapping tool and to identify gaps needing to be addressed. Overall, our mapping tool aligned with the WHO-FIC recommendations. We have identified valuable enhancements which will be addressed in future iterations of the mapping tool.
Speaker(s):
Karen Bavuso, RN, MSN
Semedy, Inc.
Author(s):
Karen Bavuso, RN, MSN - Semedy, Inc.; Ali Daowd, MD, PhD - Semedy, Inc.; Roberto Rocha, MD, PhD, FACMI - Semedy, Inc.;
Presentation Type: Poster - Regular
Poster Number: 163
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Standards, Terminology, and Interoperability, TEFCA, FHIR, Health Data Science, Workforce Automation, Communication, and Workflow Efficiency
Primary Track: Big Data for Health
Mappings between terminologies and between local concepts and terminologies ensure consistent semantic meaning and the process of creating mappings is streamlined by using an automated mapping tool. We used the WHO-FIC Mapping Principles and Best Practice recommendations to validate our mapping tool and to identify gaps needing to be addressed. Overall, our mapping tool aligned with the WHO-FIC recommendations. We have identified valuable enhancements which will be addressed in future iterations of the mapping tool.
Speaker(s):
Karen Bavuso, RN, MSN
Semedy, Inc.
Author(s):
Karen Bavuso, RN, MSN - Semedy, Inc.; Ali Daowd, MD, PhD - Semedy, Inc.; Roberto Rocha, MD, PhD, FACMI - Semedy, Inc.;
Karen
Bavuso,
RN, MSN - Semedy, Inc.
Connecting Health and Community: Developing a Model for Health and Social Service Integration via Health Information Exchange
Presentation Type: Poster Invite - Regular
Poster Number: 164
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Standards, Terminology, and Interoperability, TEFCA, FHIR, Outcomes Improvement and Equity, Social Determinants of Health (SDoH)
Primary Track: Innovation through Industry, Public Health, Non-profit and Commercial Partnerships
To address barriers in WIC enrollment and care coordination, we implemented an innovative electronic referral linking a safety-net health system with the Minnesota WIC program via Health Information Exchange (HIE). This solution leveraged the 360X protocol and existing infrastructure to automate referrals and streamline clinical data sharing. Early results demonstrate strong adoption, improved referral volumes—particularly among pregnant women—and successful integration into provider workflows. This model offers a scalable, cross-sector interoperability solution for social service referrals.
Speaker(s):
Justine Mrosak, MD
Hennepin Healthcare
Author(s):
Chad Peterson, BS - Koble; Joni Geppert, MPH, RDN, LN - Minnesota Department of Health, WIC; Rebecca Gruenes, MS, RD - Minnesota Department of Health, WIC; Kate Franken, MPH, RD - Minnesota Department of Health, WIC;
Presentation Type: Poster Invite - Regular
Poster Number: 164
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Standards, Terminology, and Interoperability, TEFCA, FHIR, Outcomes Improvement and Equity, Social Determinants of Health (SDoH)
Primary Track: Innovation through Industry, Public Health, Non-profit and Commercial Partnerships
To address barriers in WIC enrollment and care coordination, we implemented an innovative electronic referral linking a safety-net health system with the Minnesota WIC program via Health Information Exchange (HIE). This solution leveraged the 360X protocol and existing infrastructure to automate referrals and streamline clinical data sharing. Early results demonstrate strong adoption, improved referral volumes—particularly among pregnant women—and successful integration into provider workflows. This model offers a scalable, cross-sector interoperability solution for social service referrals.
Speaker(s):
Justine Mrosak, MD
Hennepin Healthcare
Author(s):
Chad Peterson, BS - Koble; Joni Geppert, MPH, RDN, LN - Minnesota Department of Health, WIC; Rebecca Gruenes, MS, RD - Minnesota Department of Health, WIC; Kate Franken, MPH, RD - Minnesota Department of Health, WIC;
Justine
Mrosak,
MD - Hennepin Healthcare
Association of digital health technology use with engagement in physical activity in older adults
Presentation Type: Poster - Regular
Poster Number: 165
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Telemedicine, Health at Home, and Virtual Care, Population Health, Digital Therapeutics, Remote Patient Monitoring (RPM), and Digital Engagement, Outcomes Improvement and Equity
Working Group: Public Health Informatics Working Group
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Background: Engaging in walking and vigorous physical activities has many positive health benefits. Although older adults are utilizing digital health technology, whether this technology influences their engagement in walking and vigorous physical activity remains unclear.
Methods: This cross-sectional study used data from the National Health and Aging Trends Study (NHATS, Year 2022) involving older adults (65 years and older, N= 4651). Digital health technology included a) making telehealth visits, b) managing Medicare or insurance data, c) scheduling medical appointments/ filling prescriptions, and d) accessing online health information. Participants were asked about their engagement in walking as exercise and vigorous physical activities in the previous month. The association of digital health technology use with walking and vigorous physical activity was examined using multivariable logistic regression, adjusting for sociodemographic and clinical characteristics. We used R 4.4.2 to conduct these statistical analyses.
Results: Digital health technology included making telehealth visits (51.9%), managing insurance data (42.5%), scheduling medical appointments or filling prescriptions (30.2%), and accessing online health information (25.8%). Making telehealth visits (aOR = 1.36; 95% CI = 1.16-1.60) and managing health insurance (aOR = 1.29; 95% CI = 1.08-1.53) were positively associated with engagement in walking among older adults, after adjusting for covariates. Assessing online health information was positively associated with engagement in vigorous activities (aOR=1.29, 95% CI=1.09-1.53) among older adults, after adjusting for covariates.
Conclusion: Digital health technology can be utilized to deliver personalized health messages to promote engagement in walking and vigorous physical activities among older adults.
Speaker(s):
Shamatree Shakya, Assistant Professor/ PhD
University of North Carolina Wilmington
Author(s):
Jeeyae Choi, PhD - University of North Carolina Wilmington; Seema Das, MPH - Emory University;
Presentation Type: Poster - Regular
Poster Number: 165
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Telemedicine, Health at Home, and Virtual Care, Population Health, Digital Therapeutics, Remote Patient Monitoring (RPM), and Digital Engagement, Outcomes Improvement and Equity
Working Group: Public Health Informatics Working Group
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Background: Engaging in walking and vigorous physical activities has many positive health benefits. Although older adults are utilizing digital health technology, whether this technology influences their engagement in walking and vigorous physical activity remains unclear.
Methods: This cross-sectional study used data from the National Health and Aging Trends Study (NHATS, Year 2022) involving older adults (65 years and older, N= 4651). Digital health technology included a) making telehealth visits, b) managing Medicare or insurance data, c) scheduling medical appointments/ filling prescriptions, and d) accessing online health information. Participants were asked about their engagement in walking as exercise and vigorous physical activities in the previous month. The association of digital health technology use with walking and vigorous physical activity was examined using multivariable logistic regression, adjusting for sociodemographic and clinical characteristics. We used R 4.4.2 to conduct these statistical analyses.
Results: Digital health technology included making telehealth visits (51.9%), managing insurance data (42.5%), scheduling medical appointments or filling prescriptions (30.2%), and accessing online health information (25.8%). Making telehealth visits (aOR = 1.36; 95% CI = 1.16-1.60) and managing health insurance (aOR = 1.29; 95% CI = 1.08-1.53) were positively associated with engagement in walking among older adults, after adjusting for covariates. Assessing online health information was positively associated with engagement in vigorous activities (aOR=1.29, 95% CI=1.09-1.53) among older adults, after adjusting for covariates.
Conclusion: Digital health technology can be utilized to deliver personalized health messages to promote engagement in walking and vigorous physical activities among older adults.
Speaker(s):
Shamatree Shakya, Assistant Professor/ PhD
University of North Carolina Wilmington
Author(s):
Jeeyae Choi, PhD - University of North Carolina Wilmington; Seema Das, MPH - Emory University;
Shamatree
Shakya,
Assistant Professor/ PhD - University of North Carolina Wilmington
Enhancing Nurse Efficiency and Reducing Burnout: A Pilot Study on Virtual Nursing Care Integration
Presentation Type: Poster - Regular
Poster Number: 166
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Telemedicine, Health at Home, and Virtual Care, Clinician Well-Being, Workforce Automation, Communication, and Workflow Efficiency
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Bedside nurses face increasing documentation burden and burnout despite multiple workforce support strategies. To address this gap, Children’s Healthcare of Atlanta implemented a virtual nursing care model using in-room video technology to offload admission and discharge documentation. A pilot on two units incorporated usability testing, safety simulation, and task-based workflows. Ongoing evaluation includes workload measures, documentation timeliness, throughput metrics, and satisfaction scores. Early feedback suggests improved workflow support while maintaining high patient/family satisfaction.
Speaker(s):
Jasmine Jallo, RN, BSN
Children's Heathcare of Altlanta
Author(s):
Jasmine Jallo, RN, BSN - Children's Heathcare of Altlanta; Michelle Tillis, MBA, BSN, RN - Children's Healthcare of Atlanta; Sarah Thompson, MSHIMI, BSN, RN - Children's Healthcare of Atlanta; James Eagens, MSN, RN - Children's Healthcare of Atlanta; Juls Phanthavong, BSN, RN - Children's Healthcare of Atlanta; Katy Cown, RN, BSN - Children's Healthcare of Atlanta; Erica Towery, MSN, RN - Children's Healthcare of Atlanta; Michelle Tillis, MBA, BSN, RN - Children's Healthcare of Atlanta;
Presentation Type: Poster - Regular
Poster Number: 166
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Telemedicine, Health at Home, and Virtual Care, Clinician Well-Being, Workforce Automation, Communication, and Workflow Efficiency
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Bedside nurses face increasing documentation burden and burnout despite multiple workforce support strategies. To address this gap, Children’s Healthcare of Atlanta implemented a virtual nursing care model using in-room video technology to offload admission and discharge documentation. A pilot on two units incorporated usability testing, safety simulation, and task-based workflows. Ongoing evaluation includes workload measures, documentation timeliness, throughput metrics, and satisfaction scores. Early feedback suggests improved workflow support while maintaining high patient/family satisfaction.
Speaker(s):
Jasmine Jallo, RN, BSN
Children's Heathcare of Altlanta
Author(s):
Jasmine Jallo, RN, BSN - Children's Heathcare of Altlanta; Michelle Tillis, MBA, BSN, RN - Children's Healthcare of Atlanta; Sarah Thompson, MSHIMI, BSN, RN - Children's Healthcare of Atlanta; James Eagens, MSN, RN - Children's Healthcare of Atlanta; Juls Phanthavong, BSN, RN - Children's Healthcare of Atlanta; Katy Cown, RN, BSN - Children's Healthcare of Atlanta; Erica Towery, MSN, RN - Children's Healthcare of Atlanta; Michelle Tillis, MBA, BSN, RN - Children's Healthcare of Atlanta;
Jasmine
Jallo,
RN, BSN - Children's Heathcare of Altlanta
Enhancing Care Continuity Through Machine Learning Driven Workflow Prioritization for Attachment Appointments
Presentation Type: Poster - Regular
Poster Number: 167
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Workforce Automation, Communication, and Workflow Efficiency, Analytical Artificial Intelligence: ML, Digital Pathology, Imaging AI, Predictive Analytics, Governance, Clinical Decision Support and Care Pathways
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Addressing a critical 73% non-attendance rate in post-discharge appointments, this project leveraged machine learning models and feature engineering to predict non-attendance risk (Random Forest F1 0.86, Recall 0.81). We operationalized these predictions into a prioritized workflow for Patient Care Advocates, utilizing SHAP values to explain risk factors including lead time, distance, behavior history, and tailor interventions. This work demonstrates the application of explainable machine learning to improve continuity of care in a safety-net health system.
Speaker(s):
Kamal Babaei Sonbolabadi, MS
Jackson Health System
Author(s):
David Stephen, MS - Jackson Health System; Vyshnavi Mattapalli, MS - Jackson Health System; Valmarie Montes De Oca, MSHA - Jackson Health System; Paola Nayar, BE - Jackson Health System; George Rosello, MS - Jackson Health System; Kristofer Rato, MBA - Jackson Health System; Cristina Flores, AS - Jackson Health System;
Presentation Type: Poster - Regular
Poster Number: 167
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Workforce Automation, Communication, and Workflow Efficiency, Analytical Artificial Intelligence: ML, Digital Pathology, Imaging AI, Predictive Analytics, Governance, Clinical Decision Support and Care Pathways
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Addressing a critical 73% non-attendance rate in post-discharge appointments, this project leveraged machine learning models and feature engineering to predict non-attendance risk (Random Forest F1 0.86, Recall 0.81). We operationalized these predictions into a prioritized workflow for Patient Care Advocates, utilizing SHAP values to explain risk factors including lead time, distance, behavior history, and tailor interventions. This work demonstrates the application of explainable machine learning to improve continuity of care in a safety-net health system.
Speaker(s):
Kamal Babaei Sonbolabadi, MS
Jackson Health System
Author(s):
David Stephen, MS - Jackson Health System; Vyshnavi Mattapalli, MS - Jackson Health System; Valmarie Montes De Oca, MSHA - Jackson Health System; Paola Nayar, BE - Jackson Health System; George Rosello, MS - Jackson Health System; Kristofer Rato, MBA - Jackson Health System; Cristina Flores, AS - Jackson Health System;
Kamal
Babaei Sonbolabadi,
MS - Jackson Health System
Evaluation of Barcode Enabled Medication Delivery Tracking Software: Impact on Workflow Efficiency and Cost Reduction
Presentation Type: Poster - Student
Poster Number: 168
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Workforce Automation, Communication, and Workflow Efficiency, Human Factors and Usability, Health Data Science
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
This multicenter retrospective pre–post evaluation examined the impact of MedEx TubeSafe® implementation across eight hospitals at Baptist Health. Rate of medication redispenses, redispense turnaround times, delivery timeliness, and nursing-to-pharmacy communication burden are compared between pre-implementation and at 1-, 3-, and 6- months post-implementation. Preliminary findings show a 4% reduction in redispensed doses with medication request volumes and reasons remaining consistent.
Speaker(s):
Samantha Cossin, PharmD
Baptist Health South Florida
Author(s):
Samantha Cossin, PharmD - Baptist Health South Florida; Kristina Lee, PharmD, BCPS - Baptist Health South; Jesus Fernandez, PharmD, CPHIMS - Baptist Health South Florida; Javed Umar, PharmD - Baptist Health South Florida;
Presentation Type: Poster - Student
Poster Number: 168
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Workforce Automation, Communication, and Workflow Efficiency, Human Factors and Usability, Health Data Science
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
This multicenter retrospective pre–post evaluation examined the impact of MedEx TubeSafe® implementation across eight hospitals at Baptist Health. Rate of medication redispenses, redispense turnaround times, delivery timeliness, and nursing-to-pharmacy communication burden are compared between pre-implementation and at 1-, 3-, and 6- months post-implementation. Preliminary findings show a 4% reduction in redispensed doses with medication request volumes and reasons remaining consistent.
Speaker(s):
Samantha Cossin, PharmD
Baptist Health South Florida
Author(s):
Samantha Cossin, PharmD - Baptist Health South Florida; Kristina Lee, PharmD, BCPS - Baptist Health South; Jesus Fernandez, PharmD, CPHIMS - Baptist Health South Florida; Javed Umar, PharmD - Baptist Health South Florida;
Samantha
Cossin,
PharmD - Baptist Health South Florida
Modernizing Patient Safety Reporting with Large Language Models: A Comparative Study of Fine-Tuned and Zero-Shot Approaches
Presentation Type: Poster - Regular
Poster Number: 169
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Workforce Automation, Communication, and Workflow Efficiency, Outcomes Improvement and Equity, Generative AI in Clinical Workflow: Ambient Listening, Chart Summarization, Automated Response with LLM
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Background: Incident reporting systems now receive tens to hundreds of thousands of reports annually - far beyond what healthcare systems can effectively evaluate. Patient safety event rates remain unchanged. LLMs' potential to extract insights from free-text incident reports remains untapped.
Methods: With 144,701 patient safety reports from UC San Diego Health (2014 - 2023), we cleaned and normalized the text, and split into training (19.3%), validation (0.7%), and test (80%) sets. The Amazon Titan Text Express model was fine-tuned using training and validation data to answer eight questions about each incident. Separately, Meta’s Llama 3.1 70B model was used in a zero-shot setting to perform the same classification tasks.
Results: The fine-tuned model predicted severity with 74% accuracy, sensitivities up to 84%, and PPVs up to 80%, outperforming the zero-shot Llama model. For contributing factors, the fine-tuned model showed higher sensitivity and precision than the zero-shot model, and overall lower accuracies. Performance of the fine-tuned model was consistent across race, sex, and ethnicity.
Discussion: We developed two LLM-based pipelines to analyze over unstructured incident reports, accurately labeling severity levels. While fine-tuned models showed stronger precision, zero-shot models demonstrated impressive adaptability, highlighting a balance between customization and rapid deployment. Despite promising results, limitations remain, especially in detecting high-severity events and fully capturing contributing factors, and real-world impact on patient safety has yet to be evaluated. These automated tools hold potential to speed up incident analysis, improve learning, and reduce harm and costs, but further integration and assessment in clinical settings are needed.
Speaker(s):
Eileen Kim, MD
University of California San Diego Health
Author(s):
Eileen Kim, MD - University of California San Diego Health; Jie Cao, PhD - University of California San Diego; Lily Poursoltan, PhD - University of California San Diego; Arvin Bhattacharya, (n/a) - Xpertech; Aaron Boussina; Andrew Chua, PhD - University of California San Diego; Jeff Pan, (n/a) - University of California San Diego; Robert El-Kareh, MD MPH MS - UC San Diego School of Medicine; Chad van den Berg, (n/a) - University of California San Diego; Chris Longhurst, MD - UC San Diego Health; Karandeep Singh, MD, MMSc - University of California, San Diego;
Presentation Type: Poster - Regular
Poster Number: 169
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Workforce Automation, Communication, and Workflow Efficiency, Outcomes Improvement and Equity, Generative AI in Clinical Workflow: Ambient Listening, Chart Summarization, Automated Response with LLM
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Background: Incident reporting systems now receive tens to hundreds of thousands of reports annually - far beyond what healthcare systems can effectively evaluate. Patient safety event rates remain unchanged. LLMs' potential to extract insights from free-text incident reports remains untapped.
Methods: With 144,701 patient safety reports from UC San Diego Health (2014 - 2023), we cleaned and normalized the text, and split into training (19.3%), validation (0.7%), and test (80%) sets. The Amazon Titan Text Express model was fine-tuned using training and validation data to answer eight questions about each incident. Separately, Meta’s Llama 3.1 70B model was used in a zero-shot setting to perform the same classification tasks.
Results: The fine-tuned model predicted severity with 74% accuracy, sensitivities up to 84%, and PPVs up to 80%, outperforming the zero-shot Llama model. For contributing factors, the fine-tuned model showed higher sensitivity and precision than the zero-shot model, and overall lower accuracies. Performance of the fine-tuned model was consistent across race, sex, and ethnicity.
Discussion: We developed two LLM-based pipelines to analyze over unstructured incident reports, accurately labeling severity levels. While fine-tuned models showed stronger precision, zero-shot models demonstrated impressive adaptability, highlighting a balance between customization and rapid deployment. Despite promising results, limitations remain, especially in detecting high-severity events and fully capturing contributing factors, and real-world impact on patient safety has yet to be evaluated. These automated tools hold potential to speed up incident analysis, improve learning, and reduce harm and costs, but further integration and assessment in clinical settings are needed.
Speaker(s):
Eileen Kim, MD
University of California San Diego Health
Author(s):
Eileen Kim, MD - University of California San Diego Health; Jie Cao, PhD - University of California San Diego; Lily Poursoltan, PhD - University of California San Diego; Arvin Bhattacharya, (n/a) - Xpertech; Aaron Boussina; Andrew Chua, PhD - University of California San Diego; Jeff Pan, (n/a) - University of California San Diego; Robert El-Kareh, MD MPH MS - UC San Diego School of Medicine; Chad van den Berg, (n/a) - University of California San Diego; Chris Longhurst, MD - UC San Diego Health; Karandeep Singh, MD, MMSc - University of California, San Diego;
Eileen
Kim,
MD - University of California San Diego Health
Characterizing Secure Chat Feature Use as a Foundation for Evaluating Clinical Communication Burden
Presentation Type: Poster - Student
Poster Number: 170
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Workforce Automation, Communication, and Workflow Efficiency, Innovation Partnerships, Implementation Science, and Learning Health Systems, Clinician Well-Being
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
EHR messaging tools like Epic Secure Chat have rapidly become the predominant communication channel in hospitals, with vendors introducing advanced features intended to streamline use, but little is known about how often these features are used or by whom. We analyzed Secure Chat metadata from Internal Medicine and Hospitalist physicians to examine feature use relative to inbound message volume. Overall utilization was low, with the bottom 75% showing high zero-use rates across most features and the top quartile demonstrating only modestly higher engagement. These patterns highlight opportunities for targeted training and future evaluation of secure chat’s impact on clinician communication burden.
Speaker(s):
Nusrat Jahan, MD
NYU Langone Health
Author(s):
Nusrat Jahan, MD - NYU Langone Health; William Small, MD, MBA - NYU Langone Health; John Will, MPA - NYU Langone Health; Jesse Burk-Rafel, MD - NYU Langone Health; Jonathan Austrian, MD - NYU Langone Health; Nicholas Genes, MD, PhD - NYU Grossman School of Medicine;
Presentation Type: Poster - Student
Poster Number: 170
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Workforce Automation, Communication, and Workflow Efficiency, Innovation Partnerships, Implementation Science, and Learning Health Systems, Clinician Well-Being
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
EHR messaging tools like Epic Secure Chat have rapidly become the predominant communication channel in hospitals, with vendors introducing advanced features intended to streamline use, but little is known about how often these features are used or by whom. We analyzed Secure Chat metadata from Internal Medicine and Hospitalist physicians to examine feature use relative to inbound message volume. Overall utilization was low, with the bottom 75% showing high zero-use rates across most features and the top quartile demonstrating only modestly higher engagement. These patterns highlight opportunities for targeted training and future evaluation of secure chat’s impact on clinician communication burden.
Speaker(s):
Nusrat Jahan, MD
NYU Langone Health
Author(s):
Nusrat Jahan, MD - NYU Langone Health; William Small, MD, MBA - NYU Langone Health; John Will, MPA - NYU Langone Health; Jesse Burk-Rafel, MD - NYU Langone Health; Jonathan Austrian, MD - NYU Langone Health; Nicholas Genes, MD, PhD - NYU Grossman School of Medicine;
Nusrat
Jahan,
MD - NYU Langone Health
Accelerating Cohort Selection and EHR Data Extraction via an Agent-based Interactive Visual Analytics System
Presentation Type: Poster - Regular
Poster Number: 171
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Electronic health records (EHRs) offer tremendous potential as a data source for clinical research, observational studies, and real-world evidence generation. To conduct such research, investigators must extract relevant data, such as diagnoses, lab values, treatments, outcomes, etc., from the EHR. However, a substantial portion of clinically relevant information remains locked in unstructured clinical notes, free-text reports, or across complex combinations of structured data. Moreover, for many research questions, the relevant cohort spans a large population: one must identify eligible patients from the entire EHR, before extracting individual-level data. Thus, there is a strong need for a system that can 1) reliably define complex inclusion/exclusion criteria over large EHR populations to identify the cohort; and 2) extract both structured and unstructured data in a scalable, reproducible, and user-friendly manner. To address these challenges, we propose an agent-based system that supports interactive visual cohort creation and automated data extraction from both structured and unstructured EHR sources.
Speaker(s):
Huan He, Ph.D.
Yale University
Author(s):
Huan He, Ph.D. - Yale University; Lingfei Qian, PHD - Yale University; Vipina K. Keloth, PhD - Yale University; Vincent Zhang, MS - Yale University; Yujia Zhou, M.S. - Yale University; Na Hong, PhD - Yale University; Hua Xu, Ph.D - Yale University;
Presentation Type: Poster - Regular
Poster Number: 171
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Electronic health records (EHRs) offer tremendous potential as a data source for clinical research, observational studies, and real-world evidence generation. To conduct such research, investigators must extract relevant data, such as diagnoses, lab values, treatments, outcomes, etc., from the EHR. However, a substantial portion of clinically relevant information remains locked in unstructured clinical notes, free-text reports, or across complex combinations of structured data. Moreover, for many research questions, the relevant cohort spans a large population: one must identify eligible patients from the entire EHR, before extracting individual-level data. Thus, there is a strong need for a system that can 1) reliably define complex inclusion/exclusion criteria over large EHR populations to identify the cohort; and 2) extract both structured and unstructured data in a scalable, reproducible, and user-friendly manner. To address these challenges, we propose an agent-based system that supports interactive visual cohort creation and automated data extraction from both structured and unstructured EHR sources.
Speaker(s):
Huan He, Ph.D.
Yale University
Author(s):
Huan He, Ph.D. - Yale University; Lingfei Qian, PHD - Yale University; Vipina K. Keloth, PhD - Yale University; Vincent Zhang, MS - Yale University; Yujia Zhou, M.S. - Yale University; Na Hong, PhD - Yale University; Hua Xu, Ph.D - Yale University;
Huan
He,
Ph.D. - Yale University
CDEAtlas: Interactive Exploration of the NIH Common Data Element Landscape
Presentation Type: Poster - Regular
Poster Number: 172
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
CDEAtlas is an interactive visualization tool designed to help researchers, data curators, and policy-makers explore the rapidly expanding landscape of Common Data Elements (CDEs). Although CDEs play a critical role in enabling data standardization and interoperability, navigating existing repositories remains challenging due to their scale, heterogeneity, and limited exploratory interfaces.
To address these barriers, we developed CDEAtlas using a human-centered and data-driven design process. The system integrates three major components: (1) a data pipeline that preprocesses CDE metadata and generates semantic embeddings from CDE text, definitions, and permissible values; (2) an interactive 2D semantic map that visualizes more than 22,000 CDEs using Three.js, enabling dynamic zooming, hovering, cluster inspection, and link-outs to source records; and (3) an in-browser search tool that provides fast concept retrieval and supports seamless transitions between global patterns and detailed CDE information.
CDEAtlas also incorporates a temporal density panel that allows users to filter CDEs by publication year, enabling longitudinal exploration. This supports identification of historical evolution, emerging domains, and shifts in development priorities across NIH organizations.
Our results demonstrate how integrating embeddings-based visualization with interactive UI design enables multi-level insight into a complex metadata ecosystem. CDEAtlas supports global landscape understanding, rapid concept lookup, and discovery of similar or related CDEs. Future work includes deeper visualization of fine-grained semantic relationships and a formal usability evaluation to assess how the tool supports researcher workflows.
Speaker(s):
Huan He, Ph.D.
Yale University
Author(s):
Ruey-Ling Weng, MS. - Yale University; Huan He, Ph.D. - Yale University; Vincent Zhang, MS - Yale University; Yujia Zhou, M.S. - Yale University; Lingfei Qian, PHD - Yale University; Hua Xu, Ph.D - Yale University; Na Hong, PhD - Yale University;
Presentation Type: Poster - Regular
Poster Number: 172
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
CDEAtlas is an interactive visualization tool designed to help researchers, data curators, and policy-makers explore the rapidly expanding landscape of Common Data Elements (CDEs). Although CDEs play a critical role in enabling data standardization and interoperability, navigating existing repositories remains challenging due to their scale, heterogeneity, and limited exploratory interfaces.
To address these barriers, we developed CDEAtlas using a human-centered and data-driven design process. The system integrates three major components: (1) a data pipeline that preprocesses CDE metadata and generates semantic embeddings from CDE text, definitions, and permissible values; (2) an interactive 2D semantic map that visualizes more than 22,000 CDEs using Three.js, enabling dynamic zooming, hovering, cluster inspection, and link-outs to source records; and (3) an in-browser search tool that provides fast concept retrieval and supports seamless transitions between global patterns and detailed CDE information.
CDEAtlas also incorporates a temporal density panel that allows users to filter CDEs by publication year, enabling longitudinal exploration. This supports identification of historical evolution, emerging domains, and shifts in development priorities across NIH organizations.
Our results demonstrate how integrating embeddings-based visualization with interactive UI design enables multi-level insight into a complex metadata ecosystem. CDEAtlas supports global landscape understanding, rapid concept lookup, and discovery of similar or related CDEs. Future work includes deeper visualization of fine-grained semantic relationships and a formal usability evaluation to assess how the tool supports researcher workflows.
Speaker(s):
Huan He, Ph.D.
Yale University
Author(s):
Ruey-Ling Weng, MS. - Yale University; Huan He, Ph.D. - Yale University; Vincent Zhang, MS - Yale University; Yujia Zhou, M.S. - Yale University; Lingfei Qian, PHD - Yale University; Hua Xu, Ph.D - Yale University; Na Hong, PhD - Yale University;
Huan
He,
Ph.D. - Yale University
Causal Discovery Analysis Using Composite AI Visualizations: From NMDoH Pathways to Precision Policies
Presentation Type: Poster Invite - Regular
Poster Number: 173
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Numerous studies on non-medical drivers of health (NMDoH) have established their prevalence, co-occurrence, and risks, with the goal of enabling interventions that improve overall health. However, while several theoretical models have emphasized the importance of identifying NMDoH causal pathways and their outcome-based risks, few studies have quantitatively analyzed them using a wide range of NMDoH variables at the individual level. Such an understanding could enable the identification of upstream factors and downstream effects enabling the design of more precise policies for making efficient and effective use of federal and state resources tailored to individual needs. Here we demonstrate the use of a novel method called Composite AI Visualization (CAIV) which visually integrates results from causal discovery, subtyping, and comparative risk analyses, and discuss how it enabled a stakeholder team to infer three NMDoH pathways and their translation to health policy. We conclude with the strengths and limitations of our approach and potential future improvements.
Speaker(s):
Suresh Bhavnani, PhD, Presidential Leadership Scholar
University of Texas Medical Branch
Author(s):
Susanne Schmidt, PhD - University of Texas San Antonio; Weibin Zhang, PhD - University of Texas Medical Branch; Monique Pappadis, PhD - University of Texas Medical Branch; Vibhuti Gupta, PhD - Meharry Medical College; Timothy Reistetter, PhD - University of Texas San Antonio; Christopher Kulesza, PhD - Rice University; Erich Kummerfeld, PhD - University of Minnesota;
Presentation Type: Poster Invite - Regular
Poster Number: 173
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Numerous studies on non-medical drivers of health (NMDoH) have established their prevalence, co-occurrence, and risks, with the goal of enabling interventions that improve overall health. However, while several theoretical models have emphasized the importance of identifying NMDoH causal pathways and their outcome-based risks, few studies have quantitatively analyzed them using a wide range of NMDoH variables at the individual level. Such an understanding could enable the identification of upstream factors and downstream effects enabling the design of more precise policies for making efficient and effective use of federal and state resources tailored to individual needs. Here we demonstrate the use of a novel method called Composite AI Visualization (CAIV) which visually integrates results from causal discovery, subtyping, and comparative risk analyses, and discuss how it enabled a stakeholder team to infer three NMDoH pathways and their translation to health policy. We conclude with the strengths and limitations of our approach and potential future improvements.
Speaker(s):
Suresh Bhavnani, PhD, Presidential Leadership Scholar
University of Texas Medical Branch
Author(s):
Susanne Schmidt, PhD - University of Texas San Antonio; Weibin Zhang, PhD - University of Texas Medical Branch; Monique Pappadis, PhD - University of Texas Medical Branch; Vibhuti Gupta, PhD - Meharry Medical College; Timothy Reistetter, PhD - University of Texas San Antonio; Christopher Kulesza, PhD - Rice University; Erich Kummerfeld, PhD - University of Minnesota;
Suresh
Bhavnani,
PhD, Presidential Leadership Scholar - University of Texas Medical Branch
Using Electronic Health Records in Texas Safety Net Clinic Settings to Increase HPV Vaccinations for Disinvested Populations
Presentation Type: Poster - Regular
Poster Number: 175
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Translation Bioinformatics/Precision Medicine
Using the electronic health record (EHR) is crucial for increasing Human Papillomavirus (HPV) vaccination rates through three strategies: Provider Assessment and Feedback (A&F), Provider Prompts, and Patient Reminders. Optimizing safety net clinics’ EHRs automates these strategies, enhancing provider knowledge of HPV vaccination rates, enabling real-time ordering and patient education, and using a reminder system to notify parents of eligible patients due or overdue for vaccination.
Speaker(s):
Susan Fenton, PhD
UTHealth Houston McWilliams School of Biomedical Informatics
Author(s):
Diana Canales, Associate of Science - The University of Texas Health Science Center at Houston D. Bradley McWilliams School of Biomedical Informatics - The Center for Quality Health IT Improvement; Alexa Poole, BA - The University of Texas Health Science Center at Houston McWilliams School of Biomedical Informatics; Lara Savas, PhD - The University of Texas Health Science Center at Houston School of Public Health; Catherine Healy, MD - Baylor College of Medicine; Laura Thormaehlen, MPH - The University of Texas Health Science Center at Houston School of Public Health; Ross Shegog, PhD - The University of Texas Health Science Center at Houston School of Public Health;
Presentation Type: Poster - Regular
Poster Number: 175
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Translation Bioinformatics/Precision Medicine
Using the electronic health record (EHR) is crucial for increasing Human Papillomavirus (HPV) vaccination rates through three strategies: Provider Assessment and Feedback (A&F), Provider Prompts, and Patient Reminders. Optimizing safety net clinics’ EHRs automates these strategies, enhancing provider knowledge of HPV vaccination rates, enabling real-time ordering and patient education, and using a reminder system to notify parents of eligible patients due or overdue for vaccination.
Speaker(s):
Susan Fenton, PhD
UTHealth Houston McWilliams School of Biomedical Informatics
Author(s):
Diana Canales, Associate of Science - The University of Texas Health Science Center at Houston D. Bradley McWilliams School of Biomedical Informatics - The Center for Quality Health IT Improvement; Alexa Poole, BA - The University of Texas Health Science Center at Houston McWilliams School of Biomedical Informatics; Lara Savas, PhD - The University of Texas Health Science Center at Houston School of Public Health; Catherine Healy, MD - Baylor College of Medicine; Laura Thormaehlen, MPH - The University of Texas Health Science Center at Houston School of Public Health; Ross Shegog, PhD - The University of Texas Health Science Center at Houston School of Public Health;
Susan
Fenton,
PhD - UTHealth Houston McWilliams School of Biomedical Informatics
Implementation of a Pilot Grant Program to Encourage Novel Informatics and Artificial Intelligence Solutions for Clinical and Translational Science
Presentation Type: Poster - Regular
Poster Number: 176
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
The Michigan Institute for Clinical and Health Research (MICHR), in collaboration with the University of Michigan Office of Research and the Michigan Institute for Data & AI in Society (MIDAS), launched a targeted pilot grant program to catalyze innovative informatics and artificial intelligence (AI) solutions in Clinical and Translational Science (CTS). Designed to attract novel approaches that address CTS challenges and advance research processes, the program invited letters of intent from University of Michigan teams. Ten proposals were reviewed using a standardized rubric; three were invited to submit full applications, each assessed for innovation, significance, interdisciplinary collaboration, potential impact, and alignment with MICHR’s mission. The top-rated proposal, focusing on AI-driven mixed methods analyses, was selected for funding. The initiative demonstrated that small pilot grants are an effective mechanism to stimulate interest and progress in applying informatics and AI to CTS, while highlighting the need for clearer education around CTS concepts. Expansion of such programs may enhance research capacity and translational impact across diverse domains.
Speaker(s):
David Hanauer, MD
University of Michigan
Author(s):
David Hanauer, MD - University of Michigan; James Maszatics - MICHR/University of Michigan; Jing Liu, PhD - University of Michigan; H.V. Jagadish, PhD - University of Michigan; Elizabeth LaPensee, PhD - University of Michigan; Julie Lumeng, MD - University of Michigan;
Presentation Type: Poster - Regular
Poster Number: 176
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
The Michigan Institute for Clinical and Health Research (MICHR), in collaboration with the University of Michigan Office of Research and the Michigan Institute for Data & AI in Society (MIDAS), launched a targeted pilot grant program to catalyze innovative informatics and artificial intelligence (AI) solutions in Clinical and Translational Science (CTS). Designed to attract novel approaches that address CTS challenges and advance research processes, the program invited letters of intent from University of Michigan teams. Ten proposals were reviewed using a standardized rubric; three were invited to submit full applications, each assessed for innovation, significance, interdisciplinary collaboration, potential impact, and alignment with MICHR’s mission. The top-rated proposal, focusing on AI-driven mixed methods analyses, was selected for funding. The initiative demonstrated that small pilot grants are an effective mechanism to stimulate interest and progress in applying informatics and AI to CTS, while highlighting the need for clearer education around CTS concepts. Expansion of such programs may enhance research capacity and translational impact across diverse domains.
Speaker(s):
David Hanauer, MD
University of Michigan
Author(s):
David Hanauer, MD - University of Michigan; James Maszatics - MICHR/University of Michigan; Jing Liu, PhD - University of Michigan; H.V. Jagadish, PhD - University of Michigan; Elizabeth LaPensee, PhD - University of Michigan; Julie Lumeng, MD - University of Michigan;
David
Hanauer,
MD - University of Michigan
Large-Scale Data Validation Software for Multimodal EEG and Wearable Sensor Data
Presentation Type: Poster - Student
Poster Number: 177
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
We developed automated software to validate, standardize, and align long-term clinical EEG and wearable sensor data from pediatric epilepsy patients undergoing video-EEG monitoring at Boston Children’s Hospital while wearing Empatica E4 devices. The system detects errors, evaluates data completeness, verifies seizure coverage, and enables efficient visual review of available data. Applied to 16,700 EEG recording hours, it produced a reliable multimodal dataset supporting reproducible research and scalable informatics applications in pediatric epilepsy.
Speaker(s):
Doroteja Dragovic, M.S.
Chicago College of Osteopathic Medicine
Author(s):
Doroteja Dragovic, M.S. - Chicago College of Osteopathic Medicine; Navaneethakrishna Makaram, PhD - Boston Children’s Hospital; Edeline Jean Baptiste, BS - Boston Children’s Hospital; Michele Jackson, BA - Boston Children’s Hospital; Paulina Moehrle, Cand. Med. - Boston Children’s Hospital; David Gromes, Cand. Med. - Boston Children’s Hospital; Tanuj Hasija, PhD - Paderborn University; William Bosl, PhD, FAMIA, FACNS - University of San Francisco; Tobias Loddenkemper;
Presentation Type: Poster - Student
Poster Number: 177
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
We developed automated software to validate, standardize, and align long-term clinical EEG and wearable sensor data from pediatric epilepsy patients undergoing video-EEG monitoring at Boston Children’s Hospital while wearing Empatica E4 devices. The system detects errors, evaluates data completeness, verifies seizure coverage, and enables efficient visual review of available data. Applied to 16,700 EEG recording hours, it produced a reliable multimodal dataset supporting reproducible research and scalable informatics applications in pediatric epilepsy.
Speaker(s):
Doroteja Dragovic, M.S.
Chicago College of Osteopathic Medicine
Author(s):
Doroteja Dragovic, M.S. - Chicago College of Osteopathic Medicine; Navaneethakrishna Makaram, PhD - Boston Children’s Hospital; Edeline Jean Baptiste, BS - Boston Children’s Hospital; Michele Jackson, BA - Boston Children’s Hospital; Paulina Moehrle, Cand. Med. - Boston Children’s Hospital; David Gromes, Cand. Med. - Boston Children’s Hospital; Tanuj Hasija, PhD - Paderborn University; William Bosl, PhD, FAMIA, FACNS - University of San Francisco; Tobias Loddenkemper;
Doroteja
Dragovic,
M.S. - Chicago College of Osteopathic Medicine
A Best Practice for Reproducible Data Reuse at Scale
Presentation Type: Poster - Regular
Poster Number: 178
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
This work describes the data quality approach we developed to archive results from a survey aiming to address gaps in understanding the pathophysiology of Gulf War Illness as part of the Million Veteran Program (MVP). Specifically, we implemented a two-stage curation process, first cataloging errors in the data set and second resolving them in a customized way for each downstream analysis. Our error catalog annotated 108 variables with values outside their plausible range, 271 variables that did not agree with related sub/super-questions, and 386 variables with missing values. Based on the success of this method we recommend including an error catalog when submitting data to a public archive.
Speaker(s):
Kelson Zawack, PhD
Yale University
Author(s):
Kelson Zawack, PhD - Veterans Affairs Cooperative Studies Program Clinical Epidemiology Research Center (CSP-CERC); Renee Chang, MS - Veterans Affairs Cooperative Studies Program Clinical Epidemiology Research Center (CSP-CERC), Veterans Affairs Connecticut Healthcare System; Rachel Quaden, MA - Boston VA Medical Center; Sarah Ahmed, PhD - Baylor College of Medicine; Mihaela Aslan, PhD - 1Veterans Affairs Cooperative Studies Program Clinical Epidemiology Research Center (CSP-CERC), Veterans Affairs Connecticut Healthcare System; Drew Helmer, md - MEDVAMC; Elizabeth Hauser, M.H.S., M.S., Ph. D - Durham VA Medical Center/ Duke University; Kei-Hoi Cheung, PhD - Yale University;
Presentation Type: Poster - Regular
Poster Number: 178
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
This work describes the data quality approach we developed to archive results from a survey aiming to address gaps in understanding the pathophysiology of Gulf War Illness as part of the Million Veteran Program (MVP). Specifically, we implemented a two-stage curation process, first cataloging errors in the data set and second resolving them in a customized way for each downstream analysis. Our error catalog annotated 108 variables with values outside their plausible range, 271 variables that did not agree with related sub/super-questions, and 386 variables with missing values. Based on the success of this method we recommend including an error catalog when submitting data to a public archive.
Speaker(s):
Kelson Zawack, PhD
Yale University
Author(s):
Kelson Zawack, PhD - Veterans Affairs Cooperative Studies Program Clinical Epidemiology Research Center (CSP-CERC); Renee Chang, MS - Veterans Affairs Cooperative Studies Program Clinical Epidemiology Research Center (CSP-CERC), Veterans Affairs Connecticut Healthcare System; Rachel Quaden, MA - Boston VA Medical Center; Sarah Ahmed, PhD - Baylor College of Medicine; Mihaela Aslan, PhD - 1Veterans Affairs Cooperative Studies Program Clinical Epidemiology Research Center (CSP-CERC), Veterans Affairs Connecticut Healthcare System; Drew Helmer, md - MEDVAMC; Elizabeth Hauser, M.H.S., M.S., Ph. D - Durham VA Medical Center/ Duke University; Kei-Hoi Cheung, PhD - Yale University;
Kelson
Zawack,
PhD - Yale University
Rare Disease Clinical Trial Data Extraction Approach Comparison
Presentation Type: Poster Invite - Regular
Poster Number: 179
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Objective: To conduct the first head-to-head comparison of BioMCP and direct ClinicalTrials.gov API v2 approaches for extracting rare disease clinical trial data
Methods: We extracted clinical trial data for 7,156 rare diseases using : (1) BioMCP's integrated trial searcher tool, and (2) direct ClinicalTrials.gov API v2. We compared trial retrieval time, drug identification accuracy, disease coverage, data quality metrics.Using statistical analysis to assess the overlap and effect size calculations.
Results: BioMCP retrieved 22,404 trials (11,159 unique NCT IDs) across 6,954 rare diseases with drug entity extraction (5,774 unique drugs), the direct API approach identified 183,065 trials (83,795 unique NCT IDs) across 7,143 diseases with comprehensive metadata coverage (36,162 unique drugs). Trial overlap was only 12.3% (10,402 trials). Direct API demonstrated superior comprehensive coverage (7.5 times more unique trials) and processing speed (21.9 times faster: 11.87 vs 260.41 minutes), and BioMCP identified 757 unique trials (0.9% of total) that Direct API missed. Drug overlap was 15.6%, with both methods contributing unique drug discoveries. Statistical tests confirmed significant differences in trials per disease (Mann-Whitney U: p < 0.001, Cohen's d = -0.280) and strong correlation for diseases covered by both methods (Spearman ρ = 0.917, p < 0.001). The hybrid approach combining both methods achieved complete coverage of all 84,552 discoverable trials.
Conclusions: Neither approach alone can provide complete coverage for rare disease clinical trial data extraction. Only 12.3% overlap demonstrates that single-method approaches systematically miss 87.7% of discoverable.
Speaker(s):
jinlian wang, PhD
UTHealth
Author(s):
jinlian wang, PhD - UTHealth;
Presentation Type: Poster Invite - Regular
Poster Number: 179
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Objective: To conduct the first head-to-head comparison of BioMCP and direct ClinicalTrials.gov API v2 approaches for extracting rare disease clinical trial data
Methods: We extracted clinical trial data for 7,156 rare diseases using : (1) BioMCP's integrated trial searcher tool, and (2) direct ClinicalTrials.gov API v2. We compared trial retrieval time, drug identification accuracy, disease coverage, data quality metrics.Using statistical analysis to assess the overlap and effect size calculations.
Results: BioMCP retrieved 22,404 trials (11,159 unique NCT IDs) across 6,954 rare diseases with drug entity extraction (5,774 unique drugs), the direct API approach identified 183,065 trials (83,795 unique NCT IDs) across 7,143 diseases with comprehensive metadata coverage (36,162 unique drugs). Trial overlap was only 12.3% (10,402 trials). Direct API demonstrated superior comprehensive coverage (7.5 times more unique trials) and processing speed (21.9 times faster: 11.87 vs 260.41 minutes), and BioMCP identified 757 unique trials (0.9% of total) that Direct API missed. Drug overlap was 15.6%, with both methods contributing unique drug discoveries. Statistical tests confirmed significant differences in trials per disease (Mann-Whitney U: p < 0.001, Cohen's d = -0.280) and strong correlation for diseases covered by both methods (Spearman ρ = 0.917, p < 0.001). The hybrid approach combining both methods achieved complete coverage of all 84,552 discoverable trials.
Conclusions: Neither approach alone can provide complete coverage for rare disease clinical trial data extraction. Only 12.3% overlap demonstrates that single-method approaches systematically miss 87.7% of discoverable.
Speaker(s):
jinlian wang, PhD
UTHealth
Author(s):
jinlian wang, PhD - UTHealth;
jinlian
wang,
PhD - UTHealth
Open-DialogueDDx: An Open-Source Multi-Turn Sequential Diagnostic Evaluation of Large Language Model Performance from Presentation to Diagnosis
Presentation Type: Poster - Student
Poster Number: 180
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
We introduce Open-DialogueDDx, an agentic simulation framework for assessing LLMs in an iterative clinical diagnostic workflow representative of real-world clinical encounters. Testing multiple LLMs on PMC and MIMIC-IV showed strong overall diagnostic accuracy, but significant variation in performance and cost. Larger, more expensive models were generally more efficient with higher accuracy. These findings highlight the need for further sequential clinical decision-making analysis of AI's abilities in clinical contexts.
Speaker(s):
Sam Moghaddam, B.S. Candidate - Biomedical Computation
Stanford University
Author(s):
Sam Moghaddam, B.S. Candidate - Biomedical Computation - Stanford University; Amit Kumthekar, B.S. Candidate - Biomedical Computatoin - Stanford University; Kevin Wu, PhD - Stanford University;
Presentation Type: Poster - Student
Poster Number: 180
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
We introduce Open-DialogueDDx, an agentic simulation framework for assessing LLMs in an iterative clinical diagnostic workflow representative of real-world clinical encounters. Testing multiple LLMs on PMC and MIMIC-IV showed strong overall diagnostic accuracy, but significant variation in performance and cost. Larger, more expensive models were generally more efficient with higher accuracy. These findings highlight the need for further sequential clinical decision-making analysis of AI's abilities in clinical contexts.
Speaker(s):
Sam Moghaddam, B.S. Candidate - Biomedical Computation
Stanford University
Author(s):
Sam Moghaddam, B.S. Candidate - Biomedical Computation - Stanford University; Amit Kumthekar, B.S. Candidate - Biomedical Computatoin - Stanford University; Kevin Wu, PhD - Stanford University;
Sam
Moghaddam,
B.S. Candidate - Biomedical Computation - Stanford University
Enhancing Fast-Food Research Data Service Processes with Uncompromised Reduction
Presentation Type: Poster - Regular
Poster Number: 181
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
Research data service teams who broker real-world data are increasingly challenged to accommodate researchers with varying deadlines, funding, and expertise. Previously, we created a framework to support research data service teams in balancing institutional goals with tradeoffs between efficiency and service. Following our framework, the Regenstrief Institute is implementing and evaluating changes to the current data request workflow, with the goal of producing higher quality data at minimal increases in costs.
Speaker(s):
Leigh Anne Tang, PhD
Indiana University/Regenstrief Institute
Author(s):
Leigh Anne Tang, PhD - Indiana University/Regenstrief Institute; Jennifer Gatz, PhD - Regenstrief Institute; Diane Kuhn, MD, PhD - Indiana University School of Medicine; Jiang Bian, PhD - Indiana University/Regenstrief Institute; Christopher Harle, PhD - Indiana University;
Presentation Type: Poster - Regular
Poster Number: 181
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
Research data service teams who broker real-world data are increasingly challenged to accommodate researchers with varying deadlines, funding, and expertise. Previously, we created a framework to support research data service teams in balancing institutional goals with tradeoffs between efficiency and service. Following our framework, the Regenstrief Institute is implementing and evaluating changes to the current data request workflow, with the goal of producing higher quality data at minimal increases in costs.
Speaker(s):
Leigh Anne Tang, PhD
Indiana University/Regenstrief Institute
Author(s):
Leigh Anne Tang, PhD - Indiana University/Regenstrief Institute; Jennifer Gatz, PhD - Regenstrief Institute; Diane Kuhn, MD, PhD - Indiana University School of Medicine; Jiang Bian, PhD - Indiana University/Regenstrief Institute; Christopher Harle, PhD - Indiana University;
Leigh Anne
Tang,
PhD - Indiana University/Regenstrief Institute
Data Discovery for a Clinical Data Commons at Roswell Park
Presentation Type: Poster - Regular
Poster Number: 182
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
Roswell Park is developing a Clinical Data Commons to unify and standardize clinical data into a federated system inspired by FAIR principles. Surveys and key stakeholder interviews highlighted barriers to creation, management, and quality of databases, gaps in data governance, and valuable data repository assets. Our findings may serve to identify challenges associated with transforming existing databases and developing high-quality databases across diverse disease areas and disciplines within academic institutions.
Speaker(s):
Sarah Mullin, PhD
Roswell Park Comprehensive Cancer Center
Author(s):
Jessica Michnik-Rubinstein, MS - Roswell Park Comprehensive Cancer Center; Kara Kelly, MD - Roswell Park Comprehensive Cancer Center;
Presentation Type: Poster - Regular
Poster Number: 182
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
Roswell Park is developing a Clinical Data Commons to unify and standardize clinical data into a federated system inspired by FAIR principles. Surveys and key stakeholder interviews highlighted barriers to creation, management, and quality of databases, gaps in data governance, and valuable data repository assets. Our findings may serve to identify challenges associated with transforming existing databases and developing high-quality databases across diverse disease areas and disciplines within academic institutions.
Speaker(s):
Sarah Mullin, PhD
Roswell Park Comprehensive Cancer Center
Author(s):
Jessica Michnik-Rubinstein, MS - Roswell Park Comprehensive Cancer Center; Kara Kelly, MD - Roswell Park Comprehensive Cancer Center;
Sarah
Mullin,
PhD - Roswell Park Comprehensive Cancer Center
Common Data Elements in Pediatric Critical Care: Two Use Cases
Presentation Type: Poster - Regular
Poster Number: 185
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
A gap analysis was conducted against two use cases to examine the state of common data elements for pediatric critical care research. We found less than 3% of elements were represented by extant CDE, highlighting the crucial need for CDE development in this domain.
Speaker(s):
Katherine Sward, PhD
University of Utah
Author(s):
Jia-Wen Guo, PhD, RN, FAMIA - University of Utah;
Presentation Type: Poster - Regular
Poster Number: 185
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
A gap analysis was conducted against two use cases to examine the state of common data elements for pediatric critical care research. We found less than 3% of elements were represented by extant CDE, highlighting the crucial need for CDE development in this domain.
Speaker(s):
Katherine Sward, PhD
University of Utah
Author(s):
Jia-Wen Guo, PhD, RN, FAMIA - University of Utah;
Katherine
Sward,
PhD - University of Utah
Evaluating the use of CDEMapper and Local RAG model in Mapping Real-World Research data
Presentation Type: Poster - Regular
Poster Number: 186
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Common data elements (CDEs) can standardize research data collection, thereby improving research reproducibility across studies and centers. CDE adoptions have been hindered due to variability in representation and granularity of CDE, as well as limitedy and implementat generalizabilition barriers in current tools. In this study, we evaluate the use of CDEMapper, the most recent CDE mapping tool, and a local RAG model in mapping real-world research items to CDE. We evaluated those two tools using 50 annotated research elements extracted from the ACTIV-4 study. CDEMapper achieved sensitivity and precision of 0.36 and 0.58, respectively. Our RAG model had similar performance metrics with 0.32 and 0.59 for sensitivity and precision, respectively. Our model can be implemented as a feature in REDCap which can lead to possible wide-spreaded adotion in organization that uses REDCap without concerns of sharing OpenAI API keys or sensitive data to third party API such as CDEMapper.
Speaker(s):
Lina Sulieman, PhD
Vanderbilt University Medical Center
Author(s):
Lina Sulieman, PhD - Vanderbilt University Medical Center; Alex Cheng, PhD - Vanderbilt University Medical Center; Stephany Duda, PhD - Vanderbilt University; Hua Xu, Ph.D - Yale University; Paul Harris, PhD - Vanderbilt University; Na Hong, PhD - Yale University; Vincent Zhang, MS - Yale University; Michelle Jones, MEd - Vanderbilt University Medical Center;
Presentation Type: Poster - Regular
Poster Number: 186
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Common data elements (CDEs) can standardize research data collection, thereby improving research reproducibility across studies and centers. CDE adoptions have been hindered due to variability in representation and granularity of CDE, as well as limitedy and implementat generalizabilition barriers in current tools. In this study, we evaluate the use of CDEMapper, the most recent CDE mapping tool, and a local RAG model in mapping real-world research items to CDE. We evaluated those two tools using 50 annotated research elements extracted from the ACTIV-4 study. CDEMapper achieved sensitivity and precision of 0.36 and 0.58, respectively. Our RAG model had similar performance metrics with 0.32 and 0.59 for sensitivity and precision, respectively. Our model can be implemented as a feature in REDCap which can lead to possible wide-spreaded adotion in organization that uses REDCap without concerns of sharing OpenAI API keys or sensitive data to third party API such as CDEMapper.
Speaker(s):
Lina Sulieman, PhD
Vanderbilt University Medical Center
Author(s):
Lina Sulieman, PhD - Vanderbilt University Medical Center; Alex Cheng, PhD - Vanderbilt University Medical Center; Stephany Duda, PhD - Vanderbilt University; Hua Xu, Ph.D - Yale University; Paul Harris, PhD - Vanderbilt University; Na Hong, PhD - Yale University; Vincent Zhang, MS - Yale University; Michelle Jones, MEd - Vanderbilt University Medical Center;
Lina
Sulieman,
PhD - Vanderbilt University Medical Center
Assessing the Implementation of CDIS to Extract FHIR Data and its Transformation to OMOP: CDIS–FHIR–OMOP Integration Repository (CHOIR) Pipeline
Presentation Type: Poster - Regular
Poster Number: 187
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
OMOP Common Data Model (CDM) has become one of the most widely adopted frameworks for building biomedical research data repositories. CDMs facilitate replication and reproducibility of research across organizations. However, some organizations, especially those with limited resources, may have difficulty converting their EHRs into OMOP. We propose the CDIS–FHIR–OMOP Integration Repository (CHOIR) that utilizes REDCap CDIS to extract EHR data as FHIR payload and then transform it into OMOP. We focused on transforming five CDIS files: vitals, labs, conditions, procedures, and encounters into four OMOP tables: measurements, condition_occurrence, procedure_occurrence, and visit_occurrence. In measurements, we were able to convert all measurements into OMOP entries. In visits, we were able to transform 98% of entries into OMOP concepts. In procedures, we were able to transform 95% of CDIS entries to OMOP. This pilot study demonstrates the feasibility of using CDIS to generate a FHIR payload that can be transformed into the OMOP.
Speaker(s):
Lina Sulieman, PhD
Vanderbilt University Medical Center
Author(s):
Lina Sulieman, PhD - Vanderbilt University Medical Center; Alex Cheng, PhD - Vanderbilt University Medical Center; Paul Harris, PhD - Vanderbilt University; Michelle Jones, MEd - VUMC; Adam Lewis, MS - VUMC; Francesco Delacqua, MS - VUMC;
Presentation Type: Poster - Regular
Poster Number: 187
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
OMOP Common Data Model (CDM) has become one of the most widely adopted frameworks for building biomedical research data repositories. CDMs facilitate replication and reproducibility of research across organizations. However, some organizations, especially those with limited resources, may have difficulty converting their EHRs into OMOP. We propose the CDIS–FHIR–OMOP Integration Repository (CHOIR) that utilizes REDCap CDIS to extract EHR data as FHIR payload and then transform it into OMOP. We focused on transforming five CDIS files: vitals, labs, conditions, procedures, and encounters into four OMOP tables: measurements, condition_occurrence, procedure_occurrence, and visit_occurrence. In measurements, we were able to convert all measurements into OMOP entries. In visits, we were able to transform 98% of entries into OMOP concepts. In procedures, we were able to transform 95% of CDIS entries to OMOP. This pilot study demonstrates the feasibility of using CDIS to generate a FHIR payload that can be transformed into the OMOP.
Speaker(s):
Lina Sulieman, PhD
Vanderbilt University Medical Center
Author(s):
Lina Sulieman, PhD - Vanderbilt University Medical Center; Alex Cheng, PhD - Vanderbilt University Medical Center; Paul Harris, PhD - Vanderbilt University; Michelle Jones, MEd - VUMC; Adam Lewis, MS - VUMC; Francesco Delacqua, MS - VUMC;
Lina
Sulieman,
PhD - Vanderbilt University Medical Center
Prediction of 6-Month Mortality Among Solid Cancer Patients Using Explainable Machine Learning Models
Presentation Type: Poster Invite - Regular
Poster Number: 188
Presentation Time: 05:00 PM - 06:30 PM
Working Group: Clinical Research Informatics Working Group
Primary Track: Data Science/Artificial Intelligence
challenge. This study evaluated machine-learning (ML) models using routinely collected electronic health record (EHR) data to predict 6-month all-cause mortality after oncology visits. We performed a retrospective cohort study using Avera Health EHR data from 2021–2024, including adults with ICD-10 C00–C80 solid tumors. Each post-diagnosis encounter was analyzed as an index visit. The dataset comprised 774,888 visits from 44,960 patients, of whom 3,012 died within 6 months. Predictors included demographics, comorbidities, laboratory results, medications, and derived biomarkers. Five ML models (GLMnet, XGBoost, Random Forest, Decision Tree, and Neural Network) were trained using a 75/25 patient-level split. Model performance was assessed using AUROC, AUPRC, accuracy, precision, recall, F1 score, and Brier score. SHAP values quantified feature importance.
The overall 6-month mortality rate was 5.59%. XGBoost achieved the strongest performance (AUROC 0.85; AUPRC 0.56; recall 0.76; Brier 0.15). Across models, important predictors included the Elixhauser weighted index, albumin level, cancer grouping, hemoglobin, and neutrophil-to-lymphocyte ratio. Lower albumin, elevated inflammatory markers, and higher comorbidity burden were consistently associated with increased mortality risk. Cancer distribution in the cohort was dominated by melanoma/skin (28%), secondary/ill-defined sites (15%), male genital organs (14%), breast (12%), and gastrointestinal cancers (7%).
Explainability analysis demonstrated that ML models not only provided strong discrimination but also identified clinically interpretable risk factors aligned with known prognostic signals. These findings support the feasibility of deploying explainable ML tools in oncology workflows to assist with early identification of high-risk patients and to inform timely palliative care and treatment planning.
Speaker(s):
Emmanuel Barton Odro, Phd in Mathematical Biology
Avera health
Author(s):
Tobias Meissner, PhD - Avera Cancer Institute; Dmitry Khomyakov, PhD - Avera Health; Rachel Elsey, PharmD - Avera Cancer Institute; McKenna Perrin, MPH - Avera Cancer Institute; Padmapriya Swaminathan, PhD - Avera Cancer Institute; Bing Xu, phD - Avera;
Presentation Type: Poster Invite - Regular
Poster Number: 188
Presentation Time: 05:00 PM - 06:30 PM
Working Group: Clinical Research Informatics Working Group
Primary Track: Data Science/Artificial Intelligence
challenge. This study evaluated machine-learning (ML) models using routinely collected electronic health record (EHR) data to predict 6-month all-cause mortality after oncology visits. We performed a retrospective cohort study using Avera Health EHR data from 2021–2024, including adults with ICD-10 C00–C80 solid tumors. Each post-diagnosis encounter was analyzed as an index visit. The dataset comprised 774,888 visits from 44,960 patients, of whom 3,012 died within 6 months. Predictors included demographics, comorbidities, laboratory results, medications, and derived biomarkers. Five ML models (GLMnet, XGBoost, Random Forest, Decision Tree, and Neural Network) were trained using a 75/25 patient-level split. Model performance was assessed using AUROC, AUPRC, accuracy, precision, recall, F1 score, and Brier score. SHAP values quantified feature importance.
The overall 6-month mortality rate was 5.59%. XGBoost achieved the strongest performance (AUROC 0.85; AUPRC 0.56; recall 0.76; Brier 0.15). Across models, important predictors included the Elixhauser weighted index, albumin level, cancer grouping, hemoglobin, and neutrophil-to-lymphocyte ratio. Lower albumin, elevated inflammatory markers, and higher comorbidity burden were consistently associated with increased mortality risk. Cancer distribution in the cohort was dominated by melanoma/skin (28%), secondary/ill-defined sites (15%), male genital organs (14%), breast (12%), and gastrointestinal cancers (7%).
Explainability analysis demonstrated that ML models not only provided strong discrimination but also identified clinically interpretable risk factors aligned with known prognostic signals. These findings support the feasibility of deploying explainable ML tools in oncology workflows to assist with early identification of high-risk patients and to inform timely palliative care and treatment planning.
Speaker(s):
Emmanuel Barton Odro, Phd in Mathematical Biology
Avera health
Author(s):
Tobias Meissner, PhD - Avera Cancer Institute; Dmitry Khomyakov, PhD - Avera Health; Rachel Elsey, PharmD - Avera Cancer Institute; McKenna Perrin, MPH - Avera Cancer Institute; Padmapriya Swaminathan, PhD - Avera Cancer Institute; Bing Xu, phD - Avera;
Emmanuel
Barton Odro,
Phd in Mathematical Biology - Avera health
Harmonizing Governance for Linked Health Data: A Proof-of-Concept Common Governance Framework to Accelerate Approvals and Access to Linked Data for Research
Presentation Type: Poster - Regular
Poster Number: 189
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Translation Bioinformatics/Precision Medicine
Linking individual-level health data enables novel insights but is delayed by fragmented governance processes, duplicate reviews, and conflicting governance rules. Multiple HHS agencies developed a proof-of-concept Common Governance Framework (CGF) that establishes a Common Governance Process with a coordinating body, a single request package, synchronized review, and governance rule harmonization. The CGF preserves dataset holder authority while delivering a streamlined process, reduced administrative burden, appropriate privacy/security, and responsible data use to accelerate scientific discovery.
Speaker(s):
HEATHER K BASEHORE, PhD
NCI
Author(s):
Emily Boja, PhD - National Institutes of Health / National Cancer Institute; Granger Sutton, PhD - National Institutes of Health / National Cancer Institute; HEATHER K BASEHORE, PhD - NCI; Emily Kraus, MPH PHD - MITRE Organization; Gwynne Jenkins, MPH PhD - MITRE;
Presentation Type: Poster - Regular
Poster Number: 189
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Translation Bioinformatics/Precision Medicine
Linking individual-level health data enables novel insights but is delayed by fragmented governance processes, duplicate reviews, and conflicting governance rules. Multiple HHS agencies developed a proof-of-concept Common Governance Framework (CGF) that establishes a Common Governance Process with a coordinating body, a single request package, synchronized review, and governance rule harmonization. The CGF preserves dataset holder authority while delivering a streamlined process, reduced administrative burden, appropriate privacy/security, and responsible data use to accelerate scientific discovery.
Speaker(s):
HEATHER K BASEHORE, PhD
NCI
Author(s):
Emily Boja, PhD - National Institutes of Health / National Cancer Institute; Granger Sutton, PhD - National Institutes of Health / National Cancer Institute; HEATHER K BASEHORE, PhD - NCI; Emily Kraus, MPH PHD - MITRE Organization; Gwynne Jenkins, MPH PhD - MITRE;
HEATHER K
BASEHORE,
PhD - NCI
Ontology-based FAIRification of Europe Cancer Image (EUCAIM) data sets applied to prostate and liver cancers.
Presentation Type: Poster Invite - Regular
Poster Number: 190
Presentation Time: 05:00 PM - 06:30 PM
Working Group: Clinical Research Informatics Working Group
Primary Track: Clinical Research Informatics
High-quality, interoperable datasets are key to advancing biomedical research. This study describes the use of the ontology developed within the EUCAIM (Cancer Image Europe) project at AP-HP, one of the pilot sites, for supporting the FAIRification process conducted by data holders. The ontology played a central role in 16 out of the 41 indicators of the FAIR evaluation tool used in the project as illustrated in the colorectal cancer domain. Furthermore, the alignment between local models and the EUCAIM common data model contributed to enrich the ontology and to assess its extensibility. FAIR-compliant, well-founded domain ontologies that explicitly model domain concepts and support logical reasoning are key component of federated platforms not only facilitating data sharing but also enhancing the reliability of downstream applications. Robust methodologies are required to ensure their capability to integrate new medical knowledge from real-world datasets.
Speaker(s):
Christel DANIEL, MD, PhD
AP-HP
Author(s):
Mirna EL GHOSH, PhD - Sorbonne Université; Varvara KALOKYRI, PhD - Institute of Computer Science, Foundation of Research and Technology Hellas, Heraklion; Laure FOURNIER, MD, PhD - Université Paris Cité, AP-HP, Hôpital Européen Georges Pompidou, PARCC, INSERM, Paris, France; Gianna TSAKOU, PhD - MAGGIOLI S.P.A., Research and Development Lab, Marousi, Greece; Manolis TSIKNAKIS, PhD - Institute of Computer Science, Foundation of Research and Technology Hellas, Heraklion;
Presentation Type: Poster Invite - Regular
Poster Number: 190
Presentation Time: 05:00 PM - 06:30 PM
Working Group: Clinical Research Informatics Working Group
Primary Track: Clinical Research Informatics
High-quality, interoperable datasets are key to advancing biomedical research. This study describes the use of the ontology developed within the EUCAIM (Cancer Image Europe) project at AP-HP, one of the pilot sites, for supporting the FAIRification process conducted by data holders. The ontology played a central role in 16 out of the 41 indicators of the FAIR evaluation tool used in the project as illustrated in the colorectal cancer domain. Furthermore, the alignment between local models and the EUCAIM common data model contributed to enrich the ontology and to assess its extensibility. FAIR-compliant, well-founded domain ontologies that explicitly model domain concepts and support logical reasoning are key component of federated platforms not only facilitating data sharing but also enhancing the reliability of downstream applications. Robust methodologies are required to ensure their capability to integrate new medical knowledge from real-world datasets.
Speaker(s):
Christel DANIEL, MD, PhD
AP-HP
Author(s):
Mirna EL GHOSH, PhD - Sorbonne Université; Varvara KALOKYRI, PhD - Institute of Computer Science, Foundation of Research and Technology Hellas, Heraklion; Laure FOURNIER, MD, PhD - Université Paris Cité, AP-HP, Hôpital Européen Georges Pompidou, PARCC, INSERM, Paris, France; Gianna TSAKOU, PhD - MAGGIOLI S.P.A., Research and Development Lab, Marousi, Greece; Manolis TSIKNAKIS, PhD - Institute of Computer Science, Foundation of Research and Technology Hellas, Heraklion;
Christel
DANIEL,
MD, PhD - AP-HP
Drug-drug interaction prediction using proteomic scale signatures within the CANDO platform
Presentation Type: Poster - Student
Poster Number: 191
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Translation Bioinformatics/Precision Medicine
Co-prescription of multiple pharmaceuticals frequently produce drug side effects and adverse drug reactions (ADRs). Current clinical decision support tools fail to provide physicians with actionable insights into how these unintended effects occur. In this study we developed a prediction module in the CANDO drug discovery platform for ADR-induced drug-drug interactions (DDIs). CANDO-DDI ranks ADRs based on compound pair proteomic signature similarity and shows excellent performance compared to random control.
Speaker(s):
Lucille Tomin, B.S.
SUNY University at Buffalo
Author(s):
Lucille Tomin, B.S. - SUNY University at Buffalo; Sarah Mullin, PhD - Roswell Park Comprehensive Cancer Center; Ram Samudrala, PhD - University at Buffalo; Zackary Falls, PhD - University at Buffalo, Jacobs School of Medicine and Biomedical Sciences;
Presentation Type: Poster - Student
Poster Number: 191
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Translation Bioinformatics/Precision Medicine
Co-prescription of multiple pharmaceuticals frequently produce drug side effects and adverse drug reactions (ADRs). Current clinical decision support tools fail to provide physicians with actionable insights into how these unintended effects occur. In this study we developed a prediction module in the CANDO drug discovery platform for ADR-induced drug-drug interactions (DDIs). CANDO-DDI ranks ADRs based on compound pair proteomic signature similarity and shows excellent performance compared to random control.
Speaker(s):
Lucille Tomin, B.S.
SUNY University at Buffalo
Author(s):
Lucille Tomin, B.S. - SUNY University at Buffalo; Sarah Mullin, PhD - Roswell Park Comprehensive Cancer Center; Ram Samudrala, PhD - University at Buffalo; Zackary Falls, PhD - University at Buffalo, Jacobs School of Medicine and Biomedical Sciences;
Lucille
Tomin,
B.S. - SUNY University at Buffalo
Repurposing Drugs to Treat Colorectal Cancer Using the CANDO Platform
Presentation Type: Poster - Student
Poster Number: 192
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Translation Bioinformatics/Precision Medicine
This study used the CANDO drug repurposing platform to identify new therapeutic candidates for colorectal cancer (CRC) by comparing drug–proteome interaction signatures. Benchmarking showed strong predictive performance for CRC. Three top candidates, cabozantinib, nilotinib, and pazopanib, were identified, all tyrosine kinase inhibitors with similar mechanism of action to three approved CRC drugs. These findings highlight the potential of computational multitarget approaches to reveal repurposed treatments for CRC.
Speaker(s):
Asma Jerbi, MD
University at Buffalo
Author(s):
Asma Jerbi, MD - University at Buffalo; Katherine Elefteriou, MSc - State University of New York at buffalo; William Mangione, PhD - University at Buffalo; Ram Samudrala, PhD - University at Buffalo; Zackary Falls, PhD - University at Buffalo, Jacobs School of Medicine and Biomedical Sciences;
Presentation Type: Poster - Student
Poster Number: 192
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Translation Bioinformatics/Precision Medicine
This study used the CANDO drug repurposing platform to identify new therapeutic candidates for colorectal cancer (CRC) by comparing drug–proteome interaction signatures. Benchmarking showed strong predictive performance for CRC. Three top candidates, cabozantinib, nilotinib, and pazopanib, were identified, all tyrosine kinase inhibitors with similar mechanism of action to three approved CRC drugs. These findings highlight the potential of computational multitarget approaches to reveal repurposed treatments for CRC.
Speaker(s):
Asma Jerbi, MD
University at Buffalo
Author(s):
Asma Jerbi, MD - University at Buffalo; Katherine Elefteriou, MSc - State University of New York at buffalo; William Mangione, PhD - University at Buffalo; Ram Samudrala, PhD - University at Buffalo; Zackary Falls, PhD - University at Buffalo, Jacobs School of Medicine and Biomedical Sciences;
Asma
Jerbi,
MD - University at Buffalo
Implementing an Emergency Department Dementia Detection Algorithm Across Health Systems
Presentation Type: Poster - Student
Poster Number: 193
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
Dementia is frequently under-recognized in emergency department (ED) care. Using retrospective electronic health record data from a second health system, the Emergency Department Dementia Algorithm maintained strong discrimination (AUC 0.87-0.89) and high negative predictive value after local retraining, supporting consistent performance across ED sites and informing future clinical use.
Speaker(s):
Inessa Cohen, MPH
Yale University
Author(s):
Inessa Cohen, MPH - Yale University; Natalia Sifnugel, MPH - New York University School of Medicine; R. Andrew Taylor, MD, MHS - University of Virginia School of Medicine; Mark Iscoe, MD, MHS; Daniella Meeker, PhD - Yale School of Medicine; Ula Hwang, MD, MPH - New York University School of Medicine;
Presentation Type: Poster - Student
Poster Number: 193
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
Dementia is frequently under-recognized in emergency department (ED) care. Using retrospective electronic health record data from a second health system, the Emergency Department Dementia Algorithm maintained strong discrimination (AUC 0.87-0.89) and high negative predictive value after local retraining, supporting consistent performance across ED sites and informing future clinical use.
Speaker(s):
Inessa Cohen, MPH
Yale University
Author(s):
Inessa Cohen, MPH - Yale University; Natalia Sifnugel, MPH - New York University School of Medicine; R. Andrew Taylor, MD, MHS - University of Virginia School of Medicine; Mark Iscoe, MD, MHS; Daniella Meeker, PhD - Yale School of Medicine; Ula Hwang, MD, MPH - New York University School of Medicine;
Inessa
Cohen,
MPH - Yale University
LLM-Driven Identification of Heart Failure Patients Without Structured Codes
Presentation Type: Poster - Regular
Poster Number: 194
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
In this study, we explored identifying heart failure (HF) patients using a large language model (LLM) among patients without HF ICD codes. A total of 4,519 Mass General Brigham’s Biobank patients without HF codes were selected. The history of HF diagnosis was identified from the patients’ notes by GPT-4o using a Retrieval-Augmented Generation (RAG) framework. A physician validated the results and performed error analysis. GPT-4o identified 29 patients as having a history of HF and demonstrated a precision of 25.93%. Error analysis showed that RAG was successful in retrieving keywords relevant to heart failure, but GPT-4o often failed to infer whether the patient had an HF diagnosis or if the symptoms were caused by HF. Although improvement is needed, this study is meaningful in that the LLM demonstrated the ability to identify a patient cohort without codified information.
Speaker(s):
Heekyong Park, PhD
Mass General Brigham
Author(s):
Martin Rees, BS - Mass General Brigham; Kenshiro Fuse, MD, MS - Harvard T.H. Chan School of Public Health; Nils Kruger, MD - Brigham and Women’s Hospital; Victor Castro, MS - Mass General Brigham; Vivian Gainer, MS - Mass General Brigham; Nich Wattanasin, MS - Mass General Brigham; Kavishwar Wagholikar, MD, PhD - Harvard Medical School /MGH; Shawn Murphy, MD, Ph.D. - Massachusetts General Hospital;
Presentation Type: Poster - Regular
Poster Number: 194
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
In this study, we explored identifying heart failure (HF) patients using a large language model (LLM) among patients without HF ICD codes. A total of 4,519 Mass General Brigham’s Biobank patients without HF codes were selected. The history of HF diagnosis was identified from the patients’ notes by GPT-4o using a Retrieval-Augmented Generation (RAG) framework. A physician validated the results and performed error analysis. GPT-4o identified 29 patients as having a history of HF and demonstrated a precision of 25.93%. Error analysis showed that RAG was successful in retrieving keywords relevant to heart failure, but GPT-4o often failed to infer whether the patient had an HF diagnosis or if the symptoms were caused by HF. Although improvement is needed, this study is meaningful in that the LLM demonstrated the ability to identify a patient cohort without codified information.
Speaker(s):
Heekyong Park, PhD
Mass General Brigham
Author(s):
Martin Rees, BS - Mass General Brigham; Kenshiro Fuse, MD, MS - Harvard T.H. Chan School of Public Health; Nils Kruger, MD - Brigham and Women’s Hospital; Victor Castro, MS - Mass General Brigham; Vivian Gainer, MS - Mass General Brigham; Nich Wattanasin, MS - Mass General Brigham; Kavishwar Wagholikar, MD, PhD - Harvard Medical School /MGH; Shawn Murphy, MD, Ph.D. - Massachusetts General Hospital;
Heekyong
Park,
PhD - Mass General Brigham
Towards a Shapelet-Based Primitives Library for Exposure Health Machine Learning
Presentation Type: Poster - Student
Poster Number: 195
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Traditional epidemiological research frequently rely on simplistic air pollution metrics such as daily averages, which mask important short-term temporal characteristics. To provide interpretable, motif-based exposure primitives for advanced health studies, we created a national, multi-pollutant exposure shapelet library. Shapelets are highly discriminative subsequences in time series data. We built a library consisting of US EPA monitor data covering PM2.5, CO2, NO2, PM10. Time-series shapelet algorithm was applied across rolling 7- and 30-day window to derive over 40 million short-term motifs. The resulting library features a user-centered design, allows querying and downloading based on pollutant, geography, duration, and data quality.
To validate the usability of this library, we linked it to a deidentified cohort of 168 HP patients (95 fibrotic) from the University of Utah Health. For each patient we identified a multi-pollutant exposure motif vector by identifying nearby monitors (20,50,100 miles) and extracted the shapelets two years prior to the HP event. We then used k-means clustering on these vectors to classify the patients based on fibrotic or non-fibrotic cases. Clustering identified four distinct cluster patterns based on PM2.5 spikes and elevated NO2 levels. This was not observable using conventional methods. This work demonstrates that shapelet library offers a reusable and scalable resource for machine learning application in clinical applications.
Speaker(s):
Sukrut Shishupal, MS
University of Utah
Author(s):
Mollie Cummins, PhD, RN, FAAN, FACMI - University of Utah; Julio Facelli, PhD - University of Utah; Ram Gouripeddi, MD - University of Utah; Katherine Sward, PhD - University of Utah; Matthew Glick, MD - University of Utah; Cheryl Pirozzi, MD, MSCI - University of Utah; Abdelrahman Badawy, BS - University of Utah;
Presentation Type: Poster - Student
Poster Number: 195
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Traditional epidemiological research frequently rely on simplistic air pollution metrics such as daily averages, which mask important short-term temporal characteristics. To provide interpretable, motif-based exposure primitives for advanced health studies, we created a national, multi-pollutant exposure shapelet library. Shapelets are highly discriminative subsequences in time series data. We built a library consisting of US EPA monitor data covering PM2.5, CO2, NO2, PM10. Time-series shapelet algorithm was applied across rolling 7- and 30-day window to derive over 40 million short-term motifs. The resulting library features a user-centered design, allows querying and downloading based on pollutant, geography, duration, and data quality.
To validate the usability of this library, we linked it to a deidentified cohort of 168 HP patients (95 fibrotic) from the University of Utah Health. For each patient we identified a multi-pollutant exposure motif vector by identifying nearby monitors (20,50,100 miles) and extracted the shapelets two years prior to the HP event. We then used k-means clustering on these vectors to classify the patients based on fibrotic or non-fibrotic cases. Clustering identified four distinct cluster patterns based on PM2.5 spikes and elevated NO2 levels. This was not observable using conventional methods. This work demonstrates that shapelet library offers a reusable and scalable resource for machine learning application in clinical applications.
Speaker(s):
Sukrut Shishupal, MS
University of Utah
Author(s):
Mollie Cummins, PhD, RN, FAAN, FACMI - University of Utah; Julio Facelli, PhD - University of Utah; Ram Gouripeddi, MD - University of Utah; Katherine Sward, PhD - University of Utah; Matthew Glick, MD - University of Utah; Cheryl Pirozzi, MD, MSCI - University of Utah; Abdelrahman Badawy, BS - University of Utah;
Sukrut
Shishupal,
MS - University of Utah
NLP-Driven, Citation-Aware Automation of Sensor Metadata Extraction for Exposure Health Research
Presentation Type: Poster - Regular
Poster Number: 196
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
This study presents a citation-aware NLP-based pipeline that extracts sensor metadata and captures citation markers surrounding any mention of a sensing device in a base document, and follows those citations to the referenced papers to extract additional metadata which may not be in the primary document and that is then used to enrich and complete the metadata obtained from the base document. This pipeline enables scalable and reproducible metadata extraction for exposure health research.
Speaker(s):
Fatemeh Shah-Mohammadi, PhD
University of Utah
Author(s):
Fatemeh Shah-Mohammadi, PhD - University of Utah; Sunho Im, Doctoral Degree - University of Utah College of Nursing; Julio Facelli, PhD - University of Utah; Mollie Cummins, PhD, RN, FAAN, FACMI - University of Utah; Ram Gouripeddi, MD - University of Utah;
Presentation Type: Poster - Regular
Poster Number: 196
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
This study presents a citation-aware NLP-based pipeline that extracts sensor metadata and captures citation markers surrounding any mention of a sensing device in a base document, and follows those citations to the referenced papers to extract additional metadata which may not be in the primary document and that is then used to enrich and complete the metadata obtained from the base document. This pipeline enables scalable and reproducible metadata extraction for exposure health research.
Speaker(s):
Fatemeh Shah-Mohammadi, PhD
University of Utah
Author(s):
Fatemeh Shah-Mohammadi, PhD - University of Utah; Sunho Im, Doctoral Degree - University of Utah College of Nursing; Julio Facelli, PhD - University of Utah; Mollie Cummins, PhD, RN, FAAN, FACMI - University of Utah; Ram Gouripeddi, MD - University of Utah;
Fatemeh
Shah-Mohammadi,
PhD - University of Utah
Evaluating Age and Sex Biases in Acoustic Diagnostic Models for Dysphonia
Presentation Type: Poster - Student
Poster Number: 197
Presentation Time: 05:00 PM - 06:30 PM
Working Group: Ethical, Legal, and Social Issues Working Group
Primary Track: Data Science/Artificial Intelligence
With the widespread application of AI in acoustic diagnosis, model fairness has drawn increasing attention. This study systematically evaluated the effectiveness and fairness of Gaussian mixture models (GMMs) and convolutional neural networks (CNNs) for dysphonia detection across age and sex groups. Using real-world datasets from 661 individuals under varying vowel conditions, we found that CNNs outperformed GMMs in both effectiveness and fairness. This study establishes a framework to support fair and robust acoustic diagnostic systems.
Speaker(s):
Sihan Xie, Bachelor
Shanghaitech University
Author(s):
Sihan Xie, Bachelor - Shanghaitech University; Yuhang Guo, Graduate - Shanghaitech University; Jiayi Wang, Bachelor - ShanghaiTech University; Yike Li, Ph.D., M.D. - Vanderbilt University Medical Center; Zhiyu Wan, PhD - ShanghaiTech University;
Presentation Type: Poster - Student
Poster Number: 197
Presentation Time: 05:00 PM - 06:30 PM
Working Group: Ethical, Legal, and Social Issues Working Group
Primary Track: Data Science/Artificial Intelligence
With the widespread application of AI in acoustic diagnosis, model fairness has drawn increasing attention. This study systematically evaluated the effectiveness and fairness of Gaussian mixture models (GMMs) and convolutional neural networks (CNNs) for dysphonia detection across age and sex groups. Using real-world datasets from 661 individuals under varying vowel conditions, we found that CNNs outperformed GMMs in both effectiveness and fairness. This study establishes a framework to support fair and robust acoustic diagnostic systems.
Speaker(s):
Sihan Xie, Bachelor
Shanghaitech University
Author(s):
Sihan Xie, Bachelor - Shanghaitech University; Yuhang Guo, Graduate - Shanghaitech University; Jiayi Wang, Bachelor - ShanghaiTech University; Yike Li, Ph.D., M.D. - Vanderbilt University Medical Center; Zhiyu Wan, PhD - ShanghaiTech University;
Sihan
Xie,
Bachelor - Shanghaitech University
TRIO-AI Uncovers APOE-LRP1-Driven Macrophage States in Public Liver Injury Atlases
Presentation Type: Poster - Regular
Poster Number: 198
Presentation Time: 05:00 PM - 06:30 PM
Working Group: Genomics and Translational Bioinformatics Working Group
Primary Track: Translation Bioinformatics/Precision Medicine
Abstract. Resolving dynamic cellular transitions at single-cell resolution is essential for understanding complex biological processes in development, disease, and regeneration, yet public atlases often lack integrated temporal modeling for rare transitional states. Here, we present TRIO-AI, a hybrid computational framework synergistically integrating Temporal Graph Neural Networks for branching detection, Neural Ordinary Differential Equations for continuous state flow modeling, and Time-Variational Autoencoders for density-based transitional state identification. We applied TRIO-AI to the public multimodal atlas GSE223561, encompassing single-nucleus RNA-seq and Visium spatial transcriptomics from mouse acetaminophen (APAP)-induced liver injury (timepoints: 0h baseline, 3h, 24h, 48h post-injury) and human acute liver failure (ALF) explants. TRIO-AI comprehensively characterized hepatic macrophage dynamics, identifying a distinct transitional macrophage population (MPs_3) peaking at 48h post-injury with coordinated signatures for lipid handling (LRP1, LRP6, ABCA1, LDLR, SCARB1), efferocytosis (MARCO, MERTK, MSR1), and extracellular matrix (ECM) engagement (ITGA9, ITGB1, SDC2). MPs_3 integrates multicellular pro-reparative signals via APOE-LRP1 and FN1-ITGA9 axes. Spatial transcriptomics from GSE223561 validated MPs_3 localization to peri-necrotic borders, with APOE-LRP1 co-localization enriched at 48h (19 pairs) vs. 24h (3 pairs). Comparative analysis against five state-of-the-art methods showed TRIO-AI's superiority in detecting transitional states and reconstructing branching trajectories in public data. TRIO-AI provides a reusable platform for public single-cell atlases, revealing macrophage-mediated repair mechanisms in liver regeneration.
Speaker(s):
Hui Li, Phd
University of Texas Health Science Center at Houston
Author(s):
Hui Li, Phd - University of Texas Health Science Center at Houston;
Presentation Type: Poster - Regular
Poster Number: 198
Presentation Time: 05:00 PM - 06:30 PM
Working Group: Genomics and Translational Bioinformatics Working Group
Primary Track: Translation Bioinformatics/Precision Medicine
Abstract. Resolving dynamic cellular transitions at single-cell resolution is essential for understanding complex biological processes in development, disease, and regeneration, yet public atlases often lack integrated temporal modeling for rare transitional states. Here, we present TRIO-AI, a hybrid computational framework synergistically integrating Temporal Graph Neural Networks for branching detection, Neural Ordinary Differential Equations for continuous state flow modeling, and Time-Variational Autoencoders for density-based transitional state identification. We applied TRIO-AI to the public multimodal atlas GSE223561, encompassing single-nucleus RNA-seq and Visium spatial transcriptomics from mouse acetaminophen (APAP)-induced liver injury (timepoints: 0h baseline, 3h, 24h, 48h post-injury) and human acute liver failure (ALF) explants. TRIO-AI comprehensively characterized hepatic macrophage dynamics, identifying a distinct transitional macrophage population (MPs_3) peaking at 48h post-injury with coordinated signatures for lipid handling (LRP1, LRP6, ABCA1, LDLR, SCARB1), efferocytosis (MARCO, MERTK, MSR1), and extracellular matrix (ECM) engagement (ITGA9, ITGB1, SDC2). MPs_3 integrates multicellular pro-reparative signals via APOE-LRP1 and FN1-ITGA9 axes. Spatial transcriptomics from GSE223561 validated MPs_3 localization to peri-necrotic borders, with APOE-LRP1 co-localization enriched at 48h (19 pairs) vs. 24h (3 pairs). Comparative analysis against five state-of-the-art methods showed TRIO-AI's superiority in detecting transitional states and reconstructing branching trajectories in public data. TRIO-AI provides a reusable platform for public single-cell atlases, revealing macrophage-mediated repair mechanisms in liver regeneration.
Speaker(s):
Hui Li, Phd
University of Texas Health Science Center at Houston
Author(s):
Hui Li, Phd - University of Texas Health Science Center at Houston;
Hui
Li,
Phd - University of Texas Health Science Center at Houston
Distinct but Complementary Signals: Comparing Evo2 and Annovar in Alzheimer’s Disease Variants
Presentation Type: Poster - Regular
Poster Number: 199
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Evo2, a DNA language model quantifying evolutionary novelty, poses the possibility that evolutionary signatures could be correlated to allele frequency. This project investigates whether Evo2 scores reflect variant prevalence by analyzing 211 APOE-associated variants. However, our results extracted with the ANNOVAR tool show no correlation between minor allele frequency (MAF) and Evo2 scores, suggesting the scores capture biological impact rather than variant prevalence.
Speaker(s):
Ivy Liang, PhD
Yale University
Author(s):
Ivy Liang, PhD - Yale University; Rui Zhu, Ph.D - Yale University; Xiaopu Zhou, PhD - Hospital for Sick Children; Lucila Ohno-Machado, M.D., PhD, MBA - Yale University;
Presentation Type: Poster - Regular
Poster Number: 199
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Evo2, a DNA language model quantifying evolutionary novelty, poses the possibility that evolutionary signatures could be correlated to allele frequency. This project investigates whether Evo2 scores reflect variant prevalence by analyzing 211 APOE-associated variants. However, our results extracted with the ANNOVAR tool show no correlation between minor allele frequency (MAF) and Evo2 scores, suggesting the scores capture biological impact rather than variant prevalence.
Speaker(s):
Ivy Liang, PhD
Yale University
Author(s):
Ivy Liang, PhD - Yale University; Rui Zhu, Ph.D - Yale University; Xiaopu Zhou, PhD - Hospital for Sick Children; Lucila Ohno-Machado, M.D., PhD, MBA - Yale University;
Ivy
Liang,
PhD - Yale University
Engineering FAIR Privacy-preserving Applications that Learn Histories of Disease
Presentation Type: Poster Invite - Regular
Poster Number: 200
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
A recent report on "Learning the natural history of human disease with generative transformers" created an
opportunity to assess the engineering challenge of delivering user-facing Generative AI applications in privacy-
sensitive domains. The application of these models, particularly for personalized healthcare tasks like predicting
individual morbidity risk, is typically constrained by data privacy concerns. This project was accordingly designed
as an in-browser model deployment exercise (an “App”) testing the architectural boundaries of client-side inference
generation (no downloads or installations). We relied exclusively on the documentation provided in the reference
report to develop the model, specifically testing the "R" component of the FAIR data principles: Findability,
Accessibility, Interoperability, and Reusability. The successful model deployment, leveraging ONNX and a custom
JavaScript SDK, establishes a secure, high-performance architectural blueprint for the future of private generative
AI in medicine.
Speaker(s):
Ines Duarte, MS
NCI - National Cancer Institute, National Institutes of Health
Author(s):
Praphulla Bhawsar, MS; Lee Mason; Jeya Balaji Balasubramanian, PhD - National Cancer Institute; Daniel Russ, Ph.D.; Arlindo Oliveira, PhD - Instituto Superior Tecnico; Jonas Almeida, PhD - NIH / National Cancer Institute / Division of Cancer Epidemiology and Genetics;
Presentation Type: Poster Invite - Regular
Poster Number: 200
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
A recent report on "Learning the natural history of human disease with generative transformers" created an
opportunity to assess the engineering challenge of delivering user-facing Generative AI applications in privacy-
sensitive domains. The application of these models, particularly for personalized healthcare tasks like predicting
individual morbidity risk, is typically constrained by data privacy concerns. This project was accordingly designed
as an in-browser model deployment exercise (an “App”) testing the architectural boundaries of client-side inference
generation (no downloads or installations). We relied exclusively on the documentation provided in the reference
report to develop the model, specifically testing the "R" component of the FAIR data principles: Findability,
Accessibility, Interoperability, and Reusability. The successful model deployment, leveraging ONNX and a custom
JavaScript SDK, establishes a secure, high-performance architectural blueprint for the future of private generative
AI in medicine.
Speaker(s):
Ines Duarte, MS
NCI - National Cancer Institute, National Institutes of Health
Author(s):
Praphulla Bhawsar, MS; Lee Mason; Jeya Balaji Balasubramanian, PhD - National Cancer Institute; Daniel Russ, Ph.D.; Arlindo Oliveira, PhD - Instituto Superior Tecnico; Jonas Almeida, PhD - NIH / National Cancer Institute / Division of Cancer Epidemiology and Genetics;
Ines
Duarte,
MS - NCI - National Cancer Institute, National Institutes of Health
How Much Reasoning Do LLMs Need? A Case Study in Immunotherapy Adverse Event Extraction
Presentation Type: Poster - Regular
Poster Number: 201
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
In this study, we assess how much “reasoning” a local GPT-OSS:20b needs to extract immune-related adverse events (irAEs) from real-world oncology notes. We compared Low, Medium, and High reasoning settings on a curated reference set. Reported F1 / median time per document are: Low 56.5%/23.3s; Medium 58.7%/28.3s; High 56.3%/40.5s; e.g. Medium demonstrated the most balanced performance. More computation is not always better; rigorous, local evaluation is still essential.
Speaker(s):
Dmitry Khomyakov, PhD
Avera Health
Author(s):
Dmitry Khomyakov, PhD - Avera Health; Yuliang Sun, MD/PhD - Avera Cancer Institute; Rachel Elsey, PharmD - Avera Cancer Institute; Tobias Meissner, PhD - Avera Cancer Institute;
Presentation Type: Poster - Regular
Poster Number: 201
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
In this study, we assess how much “reasoning” a local GPT-OSS:20b needs to extract immune-related adverse events (irAEs) from real-world oncology notes. We compared Low, Medium, and High reasoning settings on a curated reference set. Reported F1 / median time per document are: Low 56.5%/23.3s; Medium 58.7%/28.3s; High 56.3%/40.5s; e.g. Medium demonstrated the most balanced performance. More computation is not always better; rigorous, local evaluation is still essential.
Speaker(s):
Dmitry Khomyakov, PhD
Avera Health
Author(s):
Dmitry Khomyakov, PhD - Avera Health; Yuliang Sun, MD/PhD - Avera Cancer Institute; Rachel Elsey, PharmD - Avera Cancer Institute; Tobias Meissner, PhD - Avera Cancer Institute;
Dmitry
Khomyakov,
PhD - Avera Health
User Perspectives on Integrating Large Language Models into EMERSE
Presentation Type: Poster - Regular
Poster Number: 202
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
Large language models (LLMs) have the potential to significantly enhance clinical and research workflows, including tools like the Electronic Medical Record Search Engine (EMERSE). To guide responsible LLM integration, we surveyed 668 EMERSE users at the University of Michigan; 61 responded (9% response rate). Respondents comprised diverse roles, including faculty, staff, clinicians, researchers, and students. Most users were satisfied with EMERSE’s current functionality and primarily used it for research and clinical applications. Users expressed high interest in LLM-enhanced features, particularly for search and data extraction, but also voiced concerns about cost, accuracy, privacy, and reliability. Addressing these issues through ongoing stakeholder engagement and iterative development will be essential for successful LLM adoption in EMERSE.
Speaker(s):
David Hanauer, MD
University of Michigan
Author(s):
David Hanauer, MD - University of Michigan; Kai Zheng, PhD - University of California, Irvine;
Presentation Type: Poster - Regular
Poster Number: 202
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
Large language models (LLMs) have the potential to significantly enhance clinical and research workflows, including tools like the Electronic Medical Record Search Engine (EMERSE). To guide responsible LLM integration, we surveyed 668 EMERSE users at the University of Michigan; 61 responded (9% response rate). Respondents comprised diverse roles, including faculty, staff, clinicians, researchers, and students. Most users were satisfied with EMERSE’s current functionality and primarily used it for research and clinical applications. Users expressed high interest in LLM-enhanced features, particularly for search and data extraction, but also voiced concerns about cost, accuracy, privacy, and reliability. Addressing these issues through ongoing stakeholder engagement and iterative development will be essential for successful LLM adoption in EMERSE.
Speaker(s):
David Hanauer, MD
University of Michigan
Author(s):
David Hanauer, MD - University of Michigan; Kai Zheng, PhD - University of California, Irvine;
David
Hanauer,
MD - University of Michigan
Time-Adjusted Citation Analysis: A Pilot Study to Assess the Change of Biomedical Research Impact Over Time
Presentation Type: Poster Invite - Regular
Poster Number: 203
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
We introduce a novel measure, time-adjusted citation count (TACC), to address the temporal bias in traditional citation metrics. TACC incorporates citation timing to weigh recent citations more heavily, providing a more balanced assessment of scholarly impact. We applied TACC to two biomedical domains with markedly different lifetimes: COVID-19 variants (2020–2025), representing a recently developed, rapidly evolving field, and smallpox (1784–2025), representing a long-established, slow-evolving domain. For both datasets, TACC differed significantly from traditional citation counts (p < 0.0001). The level of disagreement between two ranked lists of publications based on these metrics was 4.3% for COVID-19 variants and 42% for smallpox. These results demonstrate the initial effectiveness of TACC in providing a more nuanced understanding of research impact that better reflects the current scholarly influence in rapidly evolving fields. Future research is required to further validate the findings from this study.
Speaker(s):
Tanuj Singh Shekhawat, MS
Arizona State University
Author(s):
Meredith Abrams, MS, MPH - N/A; Dongwen Wang, PhD - Arizona State University;
Presentation Type: Poster Invite - Regular
Poster Number: 203
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
We introduce a novel measure, time-adjusted citation count (TACC), to address the temporal bias in traditional citation metrics. TACC incorporates citation timing to weigh recent citations more heavily, providing a more balanced assessment of scholarly impact. We applied TACC to two biomedical domains with markedly different lifetimes: COVID-19 variants (2020–2025), representing a recently developed, rapidly evolving field, and smallpox (1784–2025), representing a long-established, slow-evolving domain. For both datasets, TACC differed significantly from traditional citation counts (p < 0.0001). The level of disagreement between two ranked lists of publications based on these metrics was 4.3% for COVID-19 variants and 42% for smallpox. These results demonstrate the initial effectiveness of TACC in providing a more nuanced understanding of research impact that better reflects the current scholarly influence in rapidly evolving fields. Future research is required to further validate the findings from this study.
Speaker(s):
Tanuj Singh Shekhawat, MS
Arizona State University
Author(s):
Meredith Abrams, MS, MPH - N/A; Dongwen Wang, PhD - Arizona State University;
Tanuj Singh
Shekhawat,
MS - Arizona State University
Computer Vision and AI in Endoscopy: A Systematic Review
Presentation Type: Poster - Regular
Poster Number: 204
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
We systematically reviewed 82 studies applying computer vision to endoscopic procedures across nine medical specialties. We characterized the distribution of studies and analyzed methodological rigor, validation design, and model performance using standardized criteria. Results highlight strong progress in gastroenterology, emerging adoption in otorhinolaryngology, urology, and general surgery, and persistent gaps in reproducibility practices and real-time validation across clinical AI research.
Speaker(s):
Elena Karras, MD
Ochsner Medical Center
Author(s):
Kayla E. Baker, MD - Ochsner Medical Center; Elena Karras, MD - Ochsner Medical Center; Elena D. Bartolone, MD - Ochsner Medical Center; Dhara Patel, MD - Ochsner Medical Center; Ashley Conley, MD - Ochsner Medical Center; Valentina Vargas Marin, MD - Ochsner Medical Center; Quinn O'Malley, MD - Ochsner Medical Center; Akio Fujiwara, MD - Ochsner Medical Center; Thomas J. Mundy, BS - Ochsner Medical Center; Dipesh Gyawali, MS - Ochsner Medical Center; Duncan Green, PhD - Ochsner Medical Center; Edward D. McCoul, MD, MPH - Ochsner Medical Center; Jonathan Bidwell, PhD - Ochsner Medical Center;
Presentation Type: Poster - Regular
Poster Number: 204
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
We systematically reviewed 82 studies applying computer vision to endoscopic procedures across nine medical specialties. We characterized the distribution of studies and analyzed methodological rigor, validation design, and model performance using standardized criteria. Results highlight strong progress in gastroenterology, emerging adoption in otorhinolaryngology, urology, and general surgery, and persistent gaps in reproducibility practices and real-time validation across clinical AI research.
Speaker(s):
Elena Karras, MD
Ochsner Medical Center
Author(s):
Kayla E. Baker, MD - Ochsner Medical Center; Elena Karras, MD - Ochsner Medical Center; Elena D. Bartolone, MD - Ochsner Medical Center; Dhara Patel, MD - Ochsner Medical Center; Ashley Conley, MD - Ochsner Medical Center; Valentina Vargas Marin, MD - Ochsner Medical Center; Quinn O'Malley, MD - Ochsner Medical Center; Akio Fujiwara, MD - Ochsner Medical Center; Thomas J. Mundy, BS - Ochsner Medical Center; Dipesh Gyawali, MS - Ochsner Medical Center; Duncan Green, PhD - Ochsner Medical Center; Edward D. McCoul, MD, MPH - Ochsner Medical Center; Jonathan Bidwell, PhD - Ochsner Medical Center;
Elena
Karras,
MD - Ochsner Medical Center
Deep Blood Phenotyping via Transformer-Based Autoencoders
Presentation Type: Poster - Student
Poster Number: 205
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
The complete blood count (CBC) is the most ordered test globally, yet its clinical use relies on coarse cell types such as differentials. In this study, we applied transformer-based autoencoders on CBC single-cell flow cytometry data and extracted embeddings that capture information beyond simple blood counts. These embeddings showed statistically significant associations with inpatient admission and 30-day mortality. Our findings underscore the need for continued research on single-cell data to advance deep blood phenotyping.
Speaker(s):
Ya-Lin Chen, PharmD
University of Washington
Author(s):
Ya-Lin Chen, PharmD - University of Washington; Jennifer Hadlock, MD - Institute For Systems Biology; Brody Foy, DPhil - University of Washington;
Presentation Type: Poster - Student
Poster Number: 205
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
The complete blood count (CBC) is the most ordered test globally, yet its clinical use relies on coarse cell types such as differentials. In this study, we applied transformer-based autoencoders on CBC single-cell flow cytometry data and extracted embeddings that capture information beyond simple blood counts. These embeddings showed statistically significant associations with inpatient admission and 30-day mortality. Our findings underscore the need for continued research on single-cell data to advance deep blood phenotyping.
Speaker(s):
Ya-Lin Chen, PharmD
University of Washington
Author(s):
Ya-Lin Chen, PharmD - University of Washington; Jennifer Hadlock, MD - Institute For Systems Biology; Brody Foy, DPhil - University of Washington;
Ya-Lin
Chen,
PharmD - University of Washington
Advancing AI Research and Development at NLM
Presentation Type: Poster - Regular
Poster Number: 206
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
The National Library of Medicine (NLM), a pioneer in biomedical informatics, has been at the forefront of artificial intelligence (AI) research since the 1960s. Today, NLM continues to lead both efforts to advance AI methods and approaches in biomedical and data science research. This poster summarizes NLM’s activities surrounding development, use, and application of AI methods and approaches in NLM-funded research and biomedical information products and services.
Speaker(s):
Brittany Chao, DPhil
NIH
Author(s):
Liz Amos, MLIS - National Library of Medicine; Brittany Chao, DPhil - NIH; Aida Tessema, MS - National Library of Medicine; Teresa Zayas Cabán, PhD - National Library of Medicine, National Institutes of Health;
Presentation Type: Poster - Regular
Poster Number: 206
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
The National Library of Medicine (NLM), a pioneer in biomedical informatics, has been at the forefront of artificial intelligence (AI) research since the 1960s. Today, NLM continues to lead both efforts to advance AI methods and approaches in biomedical and data science research. This poster summarizes NLM’s activities surrounding development, use, and application of AI methods and approaches in NLM-funded research and biomedical information products and services.
Speaker(s):
Brittany Chao, DPhil
NIH
Author(s):
Liz Amos, MLIS - National Library of Medicine; Brittany Chao, DPhil - NIH; Aida Tessema, MS - National Library of Medicine; Teresa Zayas Cabán, PhD - National Library of Medicine, National Institutes of Health;
Brittany
Chao,
DPhil - NIH
Sustained Burnout Improvements One Year After System-Wide Rollout of Ambient Documentation Technology
Presentation Type: Poster Invite - Regular
Poster Number: 207
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
We conducted a one-year follow-up survey after an initial pilot evaluating ambient documentation technology at Mass General Brigham with 683 respondents (20.2% response rate). Adoption remained high, with 55% using ADT in ≥50% of encounters. Recommendations and performance ratings were also strong. Burnout levels were lowest among high-use clinicians, with a significant dose–response relationship (p=0.028). These findings indicate sustained benefits of ADT for clinician experience.
Speaker(s):
Erik Holbrook, MD
Mass General Brigham
Author(s):
Erik Holbrook, MD - Mass General Brigham; Jacqueline You, MD, MBI - Mass General Brigham; Laura Angelo, BA - Mass General Brigham; Rebecca Mishuris, MD, MS, MPH - Mass General Brigham;
Presentation Type: Poster Invite - Regular
Poster Number: 207
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
We conducted a one-year follow-up survey after an initial pilot evaluating ambient documentation technology at Mass General Brigham with 683 respondents (20.2% response rate). Adoption remained high, with 55% using ADT in ≥50% of encounters. Recommendations and performance ratings were also strong. Burnout levels were lowest among high-use clinicians, with a significant dose–response relationship (p=0.028). These findings indicate sustained benefits of ADT for clinician experience.
Speaker(s):
Erik Holbrook, MD
Mass General Brigham
Author(s):
Erik Holbrook, MD - Mass General Brigham; Jacqueline You, MD, MBI - Mass General Brigham; Laura Angelo, BA - Mass General Brigham; Rebecca Mishuris, MD, MS, MPH - Mass General Brigham;
Erik
Holbrook,
MD - Mass General Brigham
Implementation of Cost-Effective Health System Scale Semantic Search
Presentation Type: Poster Invite - Regular
Poster Number: 209
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Large language models excel across a variety of tasks, but their usefulness in patient -specific settings is limited by their lack of access to patients’ medical records. Retrieval augmented generation can address this gap, yet solutions for deploying semantic search at health-system scale are lacking. We built a HIPAA compliant semantic search pipeline spanning 1.68 million patients and 243 million notes enabling sub-second retrieval of patient context. To balance retrieval quality with computational cost, we developed a patient-specific multiple-choice benchmark that allowed us to systematically evaluate embedding models and chunking strategies. By leveraging a storage-optimized vector database, we substantially reduced infrastructure costs while maintaining high quality retrieval and sub-second latency. This work demonstrates that fast, cost-efficient semantic search system is achievable at health-system scale, providing a robust platform to support downstream clinical applications.
Speaker(s):
Faith Wavinya Mutinda, PhD
Children's Hospital of Philadelphia
Author(s):
Faith Wavinya Mutinda, PhD - Children's Hospital of Philadelphia; Spandana Makeneni, PhD - Children's Hospital of Philadelphia; Shivaji Dutta, BS - Google; Prashant Jayaraman, MS - Google; Patrick Dibussolo, BFA - Children’s Hospital of Philadelphia; Shivani Kamath Belman, MS - Children’s Hospital of Philadelphia; Robert Grundmeier, MD - Children's Hospital of Philadelphia; Scott Haag, PhD - Children’s Hospital of Philadelphia; Jeffery Miller, MS - Children’s Hospital of Philadelphia; Ian Campbell, MD, PhD - Children's Hospital of Philadelphia;
Presentation Type: Poster Invite - Regular
Poster Number: 209
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Large language models excel across a variety of tasks, but their usefulness in patient -specific settings is limited by their lack of access to patients’ medical records. Retrieval augmented generation can address this gap, yet solutions for deploying semantic search at health-system scale are lacking. We built a HIPAA compliant semantic search pipeline spanning 1.68 million patients and 243 million notes enabling sub-second retrieval of patient context. To balance retrieval quality with computational cost, we developed a patient-specific multiple-choice benchmark that allowed us to systematically evaluate embedding models and chunking strategies. By leveraging a storage-optimized vector database, we substantially reduced infrastructure costs while maintaining high quality retrieval and sub-second latency. This work demonstrates that fast, cost-efficient semantic search system is achievable at health-system scale, providing a robust platform to support downstream clinical applications.
Speaker(s):
Faith Wavinya Mutinda, PhD
Children's Hospital of Philadelphia
Author(s):
Faith Wavinya Mutinda, PhD - Children's Hospital of Philadelphia; Spandana Makeneni, PhD - Children's Hospital of Philadelphia; Shivaji Dutta, BS - Google; Prashant Jayaraman, MS - Google; Patrick Dibussolo, BFA - Children’s Hospital of Philadelphia; Shivani Kamath Belman, MS - Children’s Hospital of Philadelphia; Robert Grundmeier, MD - Children's Hospital of Philadelphia; Scott Haag, PhD - Children’s Hospital of Philadelphia; Jeffery Miller, MS - Children’s Hospital of Philadelphia; Ian Campbell, MD, PhD - Children's Hospital of Philadelphia;
Faith Wavinya
Mutinda,
PhD - Children's Hospital of Philadelphia
Interpretable Disease Risk Modeling via Unsupervised Learning of Feature-Space Structure
Presentation Type: Poster - Regular
Poster Number: 211
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
This study developed an unsupervised learning approach to characterize feature-space structure and improve the interpretability of risk prediction models. Using data from the Health and Retirement Study, we showed that this method can substantially reduce the number of predictors used in a Random Forest model for Alzheimer’s disease risk without sacrificing performance. The resulting compact model revealed clear monotonic and non-monotonic patterns in predictor–outcome relationships, enhancing interpretability and guiding directions for future research.
Speaker(s):
Jinying Chen, PhD
Boston University
Author(s):
Dhruvi Joshi, MS - Boston University; Rhoda Au, PhD - Boston University Chobanian & Avedisian School of Medicine; Jinying Chen, PhD - Boston University;
Presentation Type: Poster - Regular
Poster Number: 211
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
This study developed an unsupervised learning approach to characterize feature-space structure and improve the interpretability of risk prediction models. Using data from the Health and Retirement Study, we showed that this method can substantially reduce the number of predictors used in a Random Forest model for Alzheimer’s disease risk without sacrificing performance. The resulting compact model revealed clear monotonic and non-monotonic patterns in predictor–outcome relationships, enhancing interpretability and guiding directions for future research.
Speaker(s):
Jinying Chen, PhD
Boston University
Author(s):
Dhruvi Joshi, MS - Boston University; Rhoda Au, PhD - Boston University Chobanian & Avedisian School of Medicine; Jinying Chen, PhD - Boston University;
Jinying
Chen,
PhD - Boston University
Leveraging Machine Learning to Detect Fraud, Waste, and Abuse in Genetic Testing Claims: Development, Iteration, and Performance of a Random Forest Model
Presentation Type: Poster - Regular
Poster Number: 212
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Fraud, waste, and abuse (FWA) in genetic testing is a growing challenge. A supervised machine learning model using a Random Forest classifier was trained on 750,000 post-adjudicated claims from 600+ laboratories, incorporating multi-domain features and composite “fingerprints” of atypical billing behavior. Since 2021, the model has achieved precision up to 71.0%, supporting targeted investigations without impacting claims payment. Iterative updates and bias testing ensure adaptability, fairness, and sustained performance.
Speaker(s):
Amanda Zaleski, PhD, MS, FACSM
Clinical Evidence Development, CVS Health
Author(s):
Elisea Avalos, PhD - CVS Health; Min Zheng, PhD - CVS Health; Ed Wang, MS - CVS Health; Linda Delaney, AHFI - CVS Health; Tyanne Ryan, BA - CVS Health; Amisha Shah Punj, MSc, MS, CGC - CVS Health; Patricia Serio, MS, CPC - CVS Health; Kelly Jean Craig, PhD - CVS Health; Joanne Armstrong, MD, MPH - CVS Health;
Presentation Type: Poster - Regular
Poster Number: 212
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Fraud, waste, and abuse (FWA) in genetic testing is a growing challenge. A supervised machine learning model using a Random Forest classifier was trained on 750,000 post-adjudicated claims from 600+ laboratories, incorporating multi-domain features and composite “fingerprints” of atypical billing behavior. Since 2021, the model has achieved precision up to 71.0%, supporting targeted investigations without impacting claims payment. Iterative updates and bias testing ensure adaptability, fairness, and sustained performance.
Speaker(s):
Amanda Zaleski, PhD, MS, FACSM
Clinical Evidence Development, CVS Health
Author(s):
Elisea Avalos, PhD - CVS Health; Min Zheng, PhD - CVS Health; Ed Wang, MS - CVS Health; Linda Delaney, AHFI - CVS Health; Tyanne Ryan, BA - CVS Health; Amisha Shah Punj, MSc, MS, CGC - CVS Health; Patricia Serio, MS, CPC - CVS Health; Kelly Jean Craig, PhD - CVS Health; Joanne Armstrong, MD, MPH - CVS Health;
Amanda
Zaleski,
PhD, MS, FACSM - Clinical Evidence Development, CVS Health
Small-Sized Reasoning Language Models for Linguistic Screening of Alzheimer’s Disease
Presentation Type: Poster Invite - Regular
Poster Number: 215
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Alzheimer’s disease (AD) is increasing in prevalence, and early detection is essential for timely care. Clinical services face growing demand, leading to delays in diagnostic appointments and increasing the risk of disease progression before evaluation. This work examines artificial intelligence (AI) methods for assessing cognitive status from linguistic features. The proposed architecture uses small language models (SLMs) to analyze speech patterns, and its compact design allows deployment on mobile devices. Recent reasoning-focused models, including Deepseek-R1 and Llama, were evaluated for dementia classification. Multiple fine-tuning strategies were compared, and the best model achieved 91% accuracy and an F1 score. The findings show that AI systems built on SLMs can achieve performance comparable to large language models, indicating their potential as efficient tools that may support health care providers through accessible pre-clinical screening for AD.
Speaker(s):
Venkatanand ram Addepalli, PhD
University of Missouri
Author(s):
Knoo Lee, PhD RN FAMIA - University of Missouri - Columbia; Praveen Rao, PhD - University of Missouri; Guy Hembroff, PhD - Michigan Tech University; Erich Kummerfeld, PhD - University of Minnesota; Nader Abdalnabi, MBA/MSB - University of Missouri; Andrew M. Kiselica, PhD, ABPP-CN - Institute of Gerontology, Health Policy & Management, University of Georgia.;
Presentation Type: Poster Invite - Regular
Poster Number: 215
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Alzheimer’s disease (AD) is increasing in prevalence, and early detection is essential for timely care. Clinical services face growing demand, leading to delays in diagnostic appointments and increasing the risk of disease progression before evaluation. This work examines artificial intelligence (AI) methods for assessing cognitive status from linguistic features. The proposed architecture uses small language models (SLMs) to analyze speech patterns, and its compact design allows deployment on mobile devices. Recent reasoning-focused models, including Deepseek-R1 and Llama, were evaluated for dementia classification. Multiple fine-tuning strategies were compared, and the best model achieved 91% accuracy and an F1 score. The findings show that AI systems built on SLMs can achieve performance comparable to large language models, indicating their potential as efficient tools that may support health care providers through accessible pre-clinical screening for AD.
Speaker(s):
Venkatanand ram Addepalli, PhD
University of Missouri
Author(s):
Knoo Lee, PhD RN FAMIA - University of Missouri - Columbia; Praveen Rao, PhD - University of Missouri; Guy Hembroff, PhD - Michigan Tech University; Erich Kummerfeld, PhD - University of Minnesota; Nader Abdalnabi, MBA/MSB - University of Missouri; Andrew M. Kiselica, PhD, ABPP-CN - Institute of Gerontology, Health Policy & Management, University of Georgia.;
Venkatanand ram
Addepalli,
PhD - University of Missouri
RareGraph-AgenticAI: A Multimodal Knowledge Graph and Multi-Agent LLM Framework for Rare Disease Evaluation and Gene Prioritization
Presentation Type: Poster Invite - Student
Poster Number: 216
Presentation Time: 05:00 PM - 06:30 PM
Working Group: Natural Language Processing Working Group
Primary Track: Data Science/Artificial Intelligence
Rare diseases are difficult to diagnose due to heterogeneous presentations, inconsistent documentation, and critical biomedical knowledge scattered across unstructured sources. Key information—such as phenotype frequency, phenotype–mechanism relevance, inheritance, demographics, variant mechanisms, genotype–phenotype correlations, and gene–disease evidence—is incomplete or buried in long texts, limiting the effectiveness of downstream AI systems, especially when phenotypes are imprecise or expressed through synonyms. To address this, we developed RareGraph, a two-stage system that first constructs a rare-disease Knowledge Graph (KG) using LLMs to extract structured evidence from MONDO, OMIM, Orphanet, HPO, and GeneReviews, and then applies a multi-agent LLM pipeline that analyzes patient notes and images, maps extracted entities onto the KG, retrieves relevant relationships, and ranks candidate diseases and genes through a composite scoring framework. In preliminary evaluations, on PhenoPackets-derived synthetic clinical notes (n=8181), RareGraph achieved 23% top-1, 50% top-10, and 76% top-50 disease prediction accuracy. On GestaltMatcher-derived synthetic notes (GMDB) (n=4138), performance reached 16% top-1, 45% top-10, and 66% top-50. For gene identification, the model reached 30% top-1, 52% top-10, and 66% top-50 accuracy on PhenoPackets, and 27% top-1, 46% top-10, and 56% top-50 on GMDB. RareGraph markedly outperformed foundation LLaMA models, which yielded under 10% accuracy for both disease and gene prediction across datasets. Beyond numerical performance, the pipeline generated interpretable, traceable reasoning paths by surfacing phenotype–gene–disease relationships directly from the KG. The KG also surfaced emergent patterns, including underreported phenotype frequencies and cross-database inconsistencies. Together, these results show that RareGraph provides accurate, interpretable, and scalable graph-guided diagnostic reasoning for rare diseases.
Speaker(s):
Quan Nguyen, Doctorate
University of Pennsylvania
Author(s):
Kai Wang, PhD - Children's Hospital of Philadelphia;
Presentation Type: Poster Invite - Student
Poster Number: 216
Presentation Time: 05:00 PM - 06:30 PM
Working Group: Natural Language Processing Working Group
Primary Track: Data Science/Artificial Intelligence
Rare diseases are difficult to diagnose due to heterogeneous presentations, inconsistent documentation, and critical biomedical knowledge scattered across unstructured sources. Key information—such as phenotype frequency, phenotype–mechanism relevance, inheritance, demographics, variant mechanisms, genotype–phenotype correlations, and gene–disease evidence—is incomplete or buried in long texts, limiting the effectiveness of downstream AI systems, especially when phenotypes are imprecise or expressed through synonyms. To address this, we developed RareGraph, a two-stage system that first constructs a rare-disease Knowledge Graph (KG) using LLMs to extract structured evidence from MONDO, OMIM, Orphanet, HPO, and GeneReviews, and then applies a multi-agent LLM pipeline that analyzes patient notes and images, maps extracted entities onto the KG, retrieves relevant relationships, and ranks candidate diseases and genes through a composite scoring framework. In preliminary evaluations, on PhenoPackets-derived synthetic clinical notes (n=8181), RareGraph achieved 23% top-1, 50% top-10, and 76% top-50 disease prediction accuracy. On GestaltMatcher-derived synthetic notes (GMDB) (n=4138), performance reached 16% top-1, 45% top-10, and 66% top-50. For gene identification, the model reached 30% top-1, 52% top-10, and 66% top-50 accuracy on PhenoPackets, and 27% top-1, 46% top-10, and 56% top-50 on GMDB. RareGraph markedly outperformed foundation LLaMA models, which yielded under 10% accuracy for both disease and gene prediction across datasets. Beyond numerical performance, the pipeline generated interpretable, traceable reasoning paths by surfacing phenotype–gene–disease relationships directly from the KG. The KG also surfaced emergent patterns, including underreported phenotype frequencies and cross-database inconsistencies. Together, these results show that RareGraph provides accurate, interpretable, and scalable graph-guided diagnostic reasoning for rare diseases.
Speaker(s):
Quan Nguyen, Doctorate
University of Pennsylvania
Author(s):
Kai Wang, PhD - Children's Hospital of Philadelphia;
Quan
Nguyen,
Doctorate - University of Pennsylvania
Predicting Diabetes Diagnosis and Onset Dates from Clinical Free-text Notes using RAG
Presentation Type: Poster Invite - Regular
Poster Number: 217
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Disease diagnosis date is typically not a structured EHR field and extraction requires manual chart review or NLP approaches. In this work, we use an off-the-shelf LLM (GPT-4.1-nano) with RAG to classify patients as diabetic vs. non-diabetic from clinical free-text notes and, if diabetic, identify onset date. Our work shows very promising results on diabetes status prediction with 96% precision, 90% recall, and 93% accuracy.
Speaker(s):
Rabins Wosti, Doctorate
University of South Carolina
Author(s):
Rabins Wosti, Doctorate - University of South Carolina; Paul Heider, PhD - Medical University of South Carolina; April Heyward, ABD - Medical University of South Carolina; Chad Arledge, PhD - Medical University of South Carolina; Caroline Rudisill, PhD, MSc - University of South Carolina; Bo Cai, PhD - University of South Carolina; Angela Liese, PhD - University of South Carolina; Jihad Obeid, MD - Medical University of South Carolina;
Presentation Type: Poster Invite - Regular
Poster Number: 217
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Disease diagnosis date is typically not a structured EHR field and extraction requires manual chart review or NLP approaches. In this work, we use an off-the-shelf LLM (GPT-4.1-nano) with RAG to classify patients as diabetic vs. non-diabetic from clinical free-text notes and, if diabetic, identify onset date. Our work shows very promising results on diabetes status prediction with 96% precision, 90% recall, and 93% accuracy.
Speaker(s):
Rabins Wosti, Doctorate
University of South Carolina
Author(s):
Rabins Wosti, Doctorate - University of South Carolina; Paul Heider, PhD - Medical University of South Carolina; April Heyward, ABD - Medical University of South Carolina; Chad Arledge, PhD - Medical University of South Carolina; Caroline Rudisill, PhD, MSc - University of South Carolina; Bo Cai, PhD - University of South Carolina; Angela Liese, PhD - University of South Carolina; Jihad Obeid, MD - Medical University of South Carolina;
Rabins
Wosti,
Doctorate - University of South Carolina
Evaluating Retrieval-Augmented Generation vs. Long-Context Input for Antibiotic Timeline Extraction from EHRs
Presentation Type: Poster Invite - Regular
Poster Number: 218
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Retrieval-augmented generation (RAG) retrieves task-relevant text passages to provide to large language models. In this study, we evaluate a simple RAG method for extracting therapeutic antibiotic timelines from across clinical notes, emulating part of the workflow of Infectious Disease clinicians. RAG achieved accuracy comparable to providing recent clinical notes up to LLMs’ full context length, while requiring far fewer tokens, demonstrating its promise for efficient longitudinal EHR reasoning.
Speaker(s):
Skatje Myers, PhD
University of Wisconsin-Madison
Author(s):
Skatje Myers, PhD - University of Wisconsin-Madison; Dmitriy Dligach, Ph.D. - Loyola University Chicago; Timothy A. Miller, PhD - Harvard Medical School; Samantha Barr, BS - University of Wisconsin-Madison; Yanjun Gao, PhD - University of Colorado-Anchutz; Matthew Churpek, MD, MPH, PhD - University of Wisconsin-Madison; Anoop Mayampurath, PhD - University of Wisconsin - Madison; Majid Afshar, MD, MSCR - University of Wisconsin - Madison;
Presentation Type: Poster Invite - Regular
Poster Number: 218
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Retrieval-augmented generation (RAG) retrieves task-relevant text passages to provide to large language models. In this study, we evaluate a simple RAG method for extracting therapeutic antibiotic timelines from across clinical notes, emulating part of the workflow of Infectious Disease clinicians. RAG achieved accuracy comparable to providing recent clinical notes up to LLMs’ full context length, while requiring far fewer tokens, demonstrating its promise for efficient longitudinal EHR reasoning.
Speaker(s):
Skatje Myers, PhD
University of Wisconsin-Madison
Author(s):
Skatje Myers, PhD - University of Wisconsin-Madison; Dmitriy Dligach, Ph.D. - Loyola University Chicago; Timothy A. Miller, PhD - Harvard Medical School; Samantha Barr, BS - University of Wisconsin-Madison; Yanjun Gao, PhD - University of Colorado-Anchutz; Matthew Churpek, MD, MPH, PhD - University of Wisconsin-Madison; Anoop Mayampurath, PhD - University of Wisconsin - Madison; Majid Afshar, MD, MSCR - University of Wisconsin - Madison;
Skatje
Myers,
PhD - University of Wisconsin-Madison
Improving Automated ICD-10 Coding Through Finetuning of Smaller LLMs
Presentation Type: Poster - Regular
Poster Number: 219
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Clinical notes must be mapped to diagnosis codes to facilitate tracking healthcare statistics and billing. Large language models (LLMs) can potentially support or automate parts of the coding workflow. In this study, we evaluated three LLMs on their ability to generate ICD-10 billing codes from clinical notes as well as the effect of fine-tuning two of these models on this task. We found fine-tuning smaller open models to outperform the larger proprietary GPT 4.1 model.
Speaker(s):
Skatje Myers, PhD
University of Wisconsin-Madison
Author(s):
Skatje Myers, PhD - University of Wisconsin-Madison; Dmitriy Dligach, Ph.D. - Loyola University Chicago; Timothy A. Miller, PhD - Harvard Medical School; Graham Wills, PhD - UW Health; Matthew Churpek, MD, MPH, PhD - University of Wisconsin-Madison; Anoop Mayampurath, PhD - University of Wisconsin - Madison; Majid Afshar, MD, MSCR - University of Wisconsin - Madison;
Presentation Type: Poster - Regular
Poster Number: 219
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Clinical notes must be mapped to diagnosis codes to facilitate tracking healthcare statistics and billing. Large language models (LLMs) can potentially support or automate parts of the coding workflow. In this study, we evaluated three LLMs on their ability to generate ICD-10 billing codes from clinical notes as well as the effect of fine-tuning two of these models on this task. We found fine-tuning smaller open models to outperform the larger proprietary GPT 4.1 model.
Speaker(s):
Skatje Myers, PhD
University of Wisconsin-Madison
Author(s):
Skatje Myers, PhD - University of Wisconsin-Madison; Dmitriy Dligach, Ph.D. - Loyola University Chicago; Timothy A. Miller, PhD - Harvard Medical School; Graham Wills, PhD - UW Health; Matthew Churpek, MD, MPH, PhD - University of Wisconsin-Madison; Anoop Mayampurath, PhD - University of Wisconsin - Madison; Majid Afshar, MD, MSCR - University of Wisconsin - Madison;
Skatje
Myers,
PhD - University of Wisconsin-Madison
KG-RAG Knowledge Guided Retrieval-Augmented Generation Rare Disease Diagnosis
Presentation Type: Poster Invite - Regular
Poster Number: 220
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
The diagnosis of rare disease is very challenging due to the rarity and lack of scientific knowledge. Many patients with rare diseases take years to get diagnosed and many stay misdiagnosed or even are not diagnosed. Comparing with traditional diagnosis prediction task, rare-disease detection poses unique difficulties, including extreme rarity, sparse training data, and incomplete understanding of disease phenotypes.
With the rapid development of large language models (LLMs), an increasing number of studies have explored their potential to assist in rare-disease diagnosis. In this work, we investigate how LLMs can leverage retrieval-augmented generation (RAG) with patient record databases and external medical knowledge to support more accurate and informed rare-disease diagnostic reasoning.
Speaker(s):
Rui Li, Phd
UT health
Author(s):
Liwei Wang, MD, PhD - UTHealth; jinlian wang, PhD - UTHealth; Andrew Wen, MS - University of Texas Health Sciences Center at Houston; Xin Li, Master of science - UTHealth Houston; Hongfang Liu, PhD - University of Texas Health Science Center at Houston;
Presentation Type: Poster Invite - Regular
Poster Number: 220
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
The diagnosis of rare disease is very challenging due to the rarity and lack of scientific knowledge. Many patients with rare diseases take years to get diagnosed and many stay misdiagnosed or even are not diagnosed. Comparing with traditional diagnosis prediction task, rare-disease detection poses unique difficulties, including extreme rarity, sparse training data, and incomplete understanding of disease phenotypes.
With the rapid development of large language models (LLMs), an increasing number of studies have explored their potential to assist in rare-disease diagnosis. In this work, we investigate how LLMs can leverage retrieval-augmented generation (RAG) with patient record databases and external medical knowledge to support more accurate and informed rare-disease diagnostic reasoning.
Speaker(s):
Rui Li, Phd
UT health
Author(s):
Liwei Wang, MD, PhD - UTHealth; jinlian wang, PhD - UTHealth; Andrew Wen, MS - University of Texas Health Sciences Center at Houston; Xin Li, Master of science - UTHealth Houston; Hongfang Liu, PhD - University of Texas Health Science Center at Houston;
Rui
Li,
Phd - UT health
LLM-Powered Personalized Glycemic Assessment in Type 2 Diabetes with Wearable Sensor Data
Presentation Type: Poster - Regular
Poster Number: 222
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Type 2 Diabetes (T2D) poses an increasing global health threat, demanding effective wearable analytics for improved diabetes care. Emerging machine learning approaches can offer clinical support in this area. However, existing approaches fail to adequately integrate wearable sensor data along with patients' information to achieve effective analysis. The rise of large language models (LLMs) has advanced natural language understanding and broader tasks across diverse healthcare applications. Building on this success, in this work, we propose a LLM framework that can integrate wearable data and static metadata to deliver comprehensive analysis in T2D. Experiments on two related tasks on the AI-READI dataset demonstrate that our model outperforms baselines by an average of 2.91% RMSE for glucose forecasting and 1.32% OvR-AUROC for diabetes severity classification. Our work presents a promising step towards harnessing the power of LLMs for wearable data analytics in T2D healthcare.
Speaker(s):
Yifan Gao, B.S.
University of Texas at San Antonio
Author(s):
Yifan Gao, B.S. - University of Texas at San Antonio; Yuanxiong Guo, Ph.D. - The University of Texas at San Antonio;
Presentation Type: Poster - Regular
Poster Number: 222
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Type 2 Diabetes (T2D) poses an increasing global health threat, demanding effective wearable analytics for improved diabetes care. Emerging machine learning approaches can offer clinical support in this area. However, existing approaches fail to adequately integrate wearable sensor data along with patients' information to achieve effective analysis. The rise of large language models (LLMs) has advanced natural language understanding and broader tasks across diverse healthcare applications. Building on this success, in this work, we propose a LLM framework that can integrate wearable data and static metadata to deliver comprehensive analysis in T2D. Experiments on two related tasks on the AI-READI dataset demonstrate that our model outperforms baselines by an average of 2.91% RMSE for glucose forecasting and 1.32% OvR-AUROC for diabetes severity classification. Our work presents a promising step towards harnessing the power of LLMs for wearable data analytics in T2D healthcare.
Speaker(s):
Yifan Gao, B.S.
University of Texas at San Antonio
Author(s):
Yifan Gao, B.S. - University of Texas at San Antonio; Yuanxiong Guo, Ph.D. - The University of Texas at San Antonio;
Yifan
Gao,
B.S. - University of Texas at San Antonio
Guiding Social Determinants of Health Annotation and Challenges
Presentation Type: Poster - Student
Poster Number: 223
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
Social determinants of health (SDoH) drive chronic disease health outcomes, yet most SDoH information exists in unstructured clinical notes. We developed a large-scale annotation guide mapping 118 child-level ICD-10-CM Z-codes (Z55–Z65) to adverse SDoH mentions with token-level instructions. Four medical students completed five annotation rounds which improved Gamma (γ) inter-annotator agreement (IAA) from 0.65 to 0.94. This resource enables precise, billable SDoH extraction to support machine learning and disease risk stratification applications.
Speaker(s):
Jeffrey Fung, MA
University of Illinois College of Medicine at Peoria
Author(s):
Jeffrey Fung, MA - University of Illinois College of Medicine at Peoria; Paul Landes; Jennifer King, BS - University of Illinois College of Medicine at Peoria; Lauren O'Neil, BS - University of Illinois College of Medicine at Peoria; Mya Watts, BS - University of Illinois College of Medicine at Peoria; Jimeng Sun - University of Illinois at Urbana Champaign; Adam Cross, MD - University of Illinois College of Medicine at Peoria;
Presentation Type: Poster - Student
Poster Number: 223
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
Social determinants of health (SDoH) drive chronic disease health outcomes, yet most SDoH information exists in unstructured clinical notes. We developed a large-scale annotation guide mapping 118 child-level ICD-10-CM Z-codes (Z55–Z65) to adverse SDoH mentions with token-level instructions. Four medical students completed five annotation rounds which improved Gamma (γ) inter-annotator agreement (IAA) from 0.65 to 0.94. This resource enables precise, billable SDoH extraction to support machine learning and disease risk stratification applications.
Speaker(s):
Jeffrey Fung, MA
University of Illinois College of Medicine at Peoria
Author(s):
Jeffrey Fung, MA - University of Illinois College of Medicine at Peoria; Paul Landes; Jennifer King, BS - University of Illinois College of Medicine at Peoria; Lauren O'Neil, BS - University of Illinois College of Medicine at Peoria; Mya Watts, BS - University of Illinois College of Medicine at Peoria; Jimeng Sun - University of Illinois at Urbana Champaign; Adam Cross, MD - University of Illinois College of Medicine at Peoria;
Jeffrey
Fung,
MA - University of Illinois College of Medicine at Peoria
LLM-enabled natural history study analysis to support rare disease research
Presentation Type: Poster Invite - Regular
Poster Number: 224
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
Natural history studies (NHS) can help us comprehend rare disease time courses, but it can be challenging to gain insights from free-text data at scale. We compared the output of three open-source large language models (LLMs) across accuracy and efficiency metrics for NHS information extraction. Llama-3.1-70B-Instruct demonstrated the strongest overall performance after a manual review of results, and we observed that LLMs can reliably identify key NHS variables.
Speaker(s):
Kevin Li, MS
National Institutes of Health
Author(s):
Kevin Li, MS - National Institutes of Health; Eric Sid, M.D. - National Institutes of Health; Qian Zhu - National Institutes of Health;
Presentation Type: Poster Invite - Regular
Poster Number: 224
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
Natural history studies (NHS) can help us comprehend rare disease time courses, but it can be challenging to gain insights from free-text data at scale. We compared the output of three open-source large language models (LLMs) across accuracy and efficiency metrics for NHS information extraction. Llama-3.1-70B-Instruct demonstrated the strongest overall performance after a manual review of results, and we observed that LLMs can reliably identify key NHS variables.
Speaker(s):
Kevin Li, MS
National Institutes of Health
Author(s):
Kevin Li, MS - National Institutes of Health; Eric Sid, M.D. - National Institutes of Health; Qian Zhu - National Institutes of Health;
Kevin
Li,
MS - National Institutes of Health
An evaluation of DeepSeek Models in Biomedical Natural Language Processing
Presentation Type: Poster Invite - Student
Poster Number: 225
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
The advancement of Large Language Models (LLMs) has significantly impacted biomedical Natural Language Processing (NLP), enhancing tasks such as named entity recognition, relation extraction, event extraction, and text classification. In this context, the DeepSeek series of models have shown promising potential in general NLP tasks, yet their capabilities in the biomedical domain remain underexplored. This study evaluates multiple DeepSeek models (Distilled-DeepSeek-R1 series and Deepseek-LLMs) across four key biomedical NLP tasks using 12 datasets, benchmarking them against state-of-the-art alternatives (Llama3-8B, Qwen2.5-7B, Mistral-7B, Phi-4-14B, Gemma-2-9B). Our results reveal that while DeepSeek models perform competitively in named entity recognition and text classification, challenges persist in event and relation extraction due to precision-recall trade-offs. We provide task-specific model recommendations and highlight future research directions. This evaluation underscores the strengths and limitations of DeepSeek models in biomedical NLP, guiding their future deployment and optimization.
Speaker(s):
Zaifu Zhan, MS
University of Minnesota twin cities
Author(s):
Zaifu Zhan, MS - University of Minnesota twin cities; Shuang Zhou, PhD - University of Minnesota Twin Cities; Huixue Zhou, PhD - University of Minnesota; Jiawen Deng, High school - University of Minnesota; Yu Hou, PhD - University of Minnesota; Jeremy Yeung, Master of Information and Data Science - University of Minnesota; Rui Zhang, PhD, FACMI, FAMIA, FIAHSI - University of Minnesota, Twin Cities;
Presentation Type: Poster Invite - Student
Poster Number: 225
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
The advancement of Large Language Models (LLMs) has significantly impacted biomedical Natural Language Processing (NLP), enhancing tasks such as named entity recognition, relation extraction, event extraction, and text classification. In this context, the DeepSeek series of models have shown promising potential in general NLP tasks, yet their capabilities in the biomedical domain remain underexplored. This study evaluates multiple DeepSeek models (Distilled-DeepSeek-R1 series and Deepseek-LLMs) across four key biomedical NLP tasks using 12 datasets, benchmarking them against state-of-the-art alternatives (Llama3-8B, Qwen2.5-7B, Mistral-7B, Phi-4-14B, Gemma-2-9B). Our results reveal that while DeepSeek models perform competitively in named entity recognition and text classification, challenges persist in event and relation extraction due to precision-recall trade-offs. We provide task-specific model recommendations and highlight future research directions. This evaluation underscores the strengths and limitations of DeepSeek models in biomedical NLP, guiding their future deployment and optimization.
Speaker(s):
Zaifu Zhan, MS
University of Minnesota twin cities
Author(s):
Zaifu Zhan, MS - University of Minnesota twin cities; Shuang Zhou, PhD - University of Minnesota Twin Cities; Huixue Zhou, PhD - University of Minnesota; Jiawen Deng, High school - University of Minnesota; Yu Hou, PhD - University of Minnesota; Jeremy Yeung, Master of Information and Data Science - University of Minnesota; Rui Zhang, PhD, FACMI, FAMIA, FIAHSI - University of Minnesota, Twin Cities;
Zaifu
Zhan,
MS - University of Minnesota twin cities
Drug or Pokemon? An analysis of large language models' ability to discern fabricated medications
Presentation Type: Poster Invite - Regular
Poster Number: 226
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
This study aimed to determine the effect of adversarial attacks on large language models by embedding one fabricated drug into a list of existing drug. Hallucination rates for the drug dosing prompt ranged from 77.47–96.4% across models, while hallucination rates for the drug indication prompt ranged from 37.07–92%. The best-performing model, Llama-3-70b, demonstrated hallucination rates spanning 1.07–85.2%. Model improvement is imperative before routinely using LLMs in the medical field can be considered safe.
Speaker(s):
Andrea Sikora, PharmD, MSCR
University of Colorado
Author(s):
Xingmeng Zhao, PhD - CU; Brooke Smith, PharmD - Wellstar; Kaitlin Blotske, PharmD - CU; Brian Murray, PharmD - CU; Yanjun Gao, PhD - University of Colorado; Susan Smith, PharmD - UGA; Kelli Henry, PharmD, MHA - CU;
Presentation Type: Poster Invite - Regular
Poster Number: 226
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
This study aimed to determine the effect of adversarial attacks on large language models by embedding one fabricated drug into a list of existing drug. Hallucination rates for the drug dosing prompt ranged from 77.47–96.4% across models, while hallucination rates for the drug indication prompt ranged from 37.07–92%. The best-performing model, Llama-3-70b, demonstrated hallucination rates spanning 1.07–85.2%. Model improvement is imperative before routinely using LLMs in the medical field can be considered safe.
Speaker(s):
Andrea Sikora, PharmD, MSCR
University of Colorado
Author(s):
Xingmeng Zhao, PhD - CU; Brooke Smith, PharmD - Wellstar; Kaitlin Blotske, PharmD - CU; Brian Murray, PharmD - CU; Yanjun Gao, PhD - University of Colorado; Susan Smith, PharmD - UGA; Kelli Henry, PharmD, MHA - CU;
Andrea
Sikora,
PharmD, MSCR - University of Colorado
Assessing NLP approaches for Melanoma pathology reports: A real-world evaluation of rules, large language models, and hybrid approaches
Presentation Type: Poster - Regular
Poster Number: 227
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
This study compares a rule-based NLP system, a Large Language Model (LLM), and a hybrid system for extracting melanoma features from pathology reports. Performance was measured against a manually annotated dataset. The LLM performed comparably to the rule-based system, and the rules + LLM hybrid system provided the overall best performance.
Speaker(s):
Johnathan Stanley, Biomedical Informatics
Department of Veterans Affairs
Author(s):
Mengke Hu, PHD - University of Utah; Jianlin Shi, MD, PhD - The Division of Epidemiology, School of Medicine, University of Utah; VA Salt Lake City Healthcare System; Qiwei Gan; Tiffany Quilter, PT, DPT - University of Utah; Lacy Castleton, RN, BSN - University of Utah; Elizabeth Hanchrow, RN, MSN - Veterans Affairs and WIVR; Sudarshan Karki, PhD - Portland VA Medical Center; Julie Lynch, PhD, RN, MBA - Veterans Health Administration; Wesley Yu, MD - Portland VA Medical Center; Oregon Health & Science University; Patrick Alba, MS - United States Department of Veterans Affairs;
Presentation Type: Poster - Regular
Poster Number: 227
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
This study compares a rule-based NLP system, a Large Language Model (LLM), and a hybrid system for extracting melanoma features from pathology reports. Performance was measured against a manually annotated dataset. The LLM performed comparably to the rule-based system, and the rules + LLM hybrid system provided the overall best performance.
Speaker(s):
Johnathan Stanley, Biomedical Informatics
Department of Veterans Affairs
Author(s):
Mengke Hu, PHD - University of Utah; Jianlin Shi, MD, PhD - The Division of Epidemiology, School of Medicine, University of Utah; VA Salt Lake City Healthcare System; Qiwei Gan; Tiffany Quilter, PT, DPT - University of Utah; Lacy Castleton, RN, BSN - University of Utah; Elizabeth Hanchrow, RN, MSN - Veterans Affairs and WIVR; Sudarshan Karki, PhD - Portland VA Medical Center; Julie Lynch, PhD, RN, MBA - Veterans Health Administration; Wesley Yu, MD - Portland VA Medical Center; Oregon Health & Science University; Patrick Alba, MS - United States Department of Veterans Affairs;
Johnathan
Stanley,
Biomedical Informatics - Department of Veterans Affairs
A Prompt Library for Efficient Clinical Entity Recognition Using Large Language Models
Presentation Type: Poster Invite - Regular
Poster Number: 228
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Large Language Models (LLMs) hold promise for clinical information extraction (IE), yet evaluation remains limited by prompt engineering and annotated data. We developed an automated pipeline that extracts entity-level schema information (definitions, examples, and guidelines) from published clinical IE studies to construct structured prompts. From 70 PubMed articles spanning 44 diseases and over 100 entities, we generated task-specific prompts for evaluating GPT-4o, Llama-3.1-8B, and Llama-3.3-70B under few-shot and QLoRA fine-tuning settings on ten public datasets. Bio_ClinicalBERT and generic-prompted Llama-3.1-8B served as baselines. GPT-4o achieved strict F1 above 0.7 on four datasets. Fine-tuned Llama-3.1-8B with schema-rich prompts reached strict F1 above 0.9 on five datasets, surpassing prior state-of-the-art. Baseline models underperformed without structured prompts. Our results highlight the value of schema-derived prompting in enabling LLMs to generalize across diverse clinical IE tasks. This framework supports reproducible, scalable benchmarking for LLM evaluation in clinical IE.
Speaker(s):
Yang Ren, Ph.D.
Yale University
Author(s):
Yang Ren, Ph.D. - Yale University; Vipina K. Keloth, PhD - Yale University; Yan Hu, MS - UTHealth Science Center Houston; Hua Xu, Ph.D - Yale University;
Presentation Type: Poster Invite - Regular
Poster Number: 228
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Large Language Models (LLMs) hold promise for clinical information extraction (IE), yet evaluation remains limited by prompt engineering and annotated data. We developed an automated pipeline that extracts entity-level schema information (definitions, examples, and guidelines) from published clinical IE studies to construct structured prompts. From 70 PubMed articles spanning 44 diseases and over 100 entities, we generated task-specific prompts for evaluating GPT-4o, Llama-3.1-8B, and Llama-3.3-70B under few-shot and QLoRA fine-tuning settings on ten public datasets. Bio_ClinicalBERT and generic-prompted Llama-3.1-8B served as baselines. GPT-4o achieved strict F1 above 0.7 on four datasets. Fine-tuned Llama-3.1-8B with schema-rich prompts reached strict F1 above 0.9 on five datasets, surpassing prior state-of-the-art. Baseline models underperformed without structured prompts. Our results highlight the value of schema-derived prompting in enabling LLMs to generalize across diverse clinical IE tasks. This framework supports reproducible, scalable benchmarking for LLM evaluation in clinical IE.
Speaker(s):
Yang Ren, Ph.D.
Yale University
Author(s):
Yang Ren, Ph.D. - Yale University; Vipina K. Keloth, PhD - Yale University; Yan Hu, MS - UTHealth Science Center Houston; Hua Xu, Ph.D - Yale University;
Yang
Ren,
Ph.D. - Yale University
A Project-Agnostic FAIR Data Lifecycle Framework for Biomedical Research
Presentation Type: Poster Invite - Regular
Poster Number: 229
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Translation Bioinformatics/Precision Medicine
We present a project-agnostic FAIR data lifecycle framework, developed at Sage Bionetworks and deployed across NIH-funded consortia on the Synapse platform. The framework integrates standards-based metadata modeling, GitHub-driven schema management, automated R/Python validation pipelines, and contributor workflows via Jira Service Desk and Synapse-powered Knowledge Portals. By embedding these practices in an open-science ecosystem, we improve metadata completeness, reduce manual effort, and enable interoperable reuse of heterogeneous multi-omics, imaging, clinical, and behavioral datasets.
Speaker(s):
Jessica Malenfant, MPH
Sage Bionetworks
Author(s):
Jessica Malenfant, MPH - Sage Bionetworks; Susheel Varma, PhD MBA FBCS - Sage Bionetworks; Trisha Zintel, Ph.D. - Sage Bionetworks; Thomas Yu, BS - Sage Bionetworks; Chelsea Nayan, MS - Sage Bionetworks; Amelia Kallaher, MS - Sage Bionetworks; Melissa Klein, MPH - Sage Bionetworks; Jordan Driscoll, MS - Sage Bionetworks; Ram Ayyala, MS - Sage Bionetworks;
Presentation Type: Poster Invite - Regular
Poster Number: 229
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Translation Bioinformatics/Precision Medicine
We present a project-agnostic FAIR data lifecycle framework, developed at Sage Bionetworks and deployed across NIH-funded consortia on the Synapse platform. The framework integrates standards-based metadata modeling, GitHub-driven schema management, automated R/Python validation pipelines, and contributor workflows via Jira Service Desk and Synapse-powered Knowledge Portals. By embedding these practices in an open-science ecosystem, we improve metadata completeness, reduce manual effort, and enable interoperable reuse of heterogeneous multi-omics, imaging, clinical, and behavioral datasets.
Speaker(s):
Jessica Malenfant, MPH
Sage Bionetworks
Author(s):
Jessica Malenfant, MPH - Sage Bionetworks; Susheel Varma, PhD MBA FBCS - Sage Bionetworks; Trisha Zintel, Ph.D. - Sage Bionetworks; Thomas Yu, BS - Sage Bionetworks; Chelsea Nayan, MS - Sage Bionetworks; Amelia Kallaher, MS - Sage Bionetworks; Melissa Klein, MPH - Sage Bionetworks; Jordan Driscoll, MS - Sage Bionetworks; Ram Ayyala, MS - Sage Bionetworks;
Jessica
Malenfant,
MPH - Sage Bionetworks
Development and validation of machine learning algorithm in predicting intracranial hemorrhage among antithrombotic drug users
Presentation Type: Poster Invite - Student
Poster Number: 230
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Translation Bioinformatics/Precision Medicine
Machine learning models have shown promise for predicting range of adverse events in the field of cardiovascular therapies. We aimed to develop and validate the performance of various ML models to predict intracranial hemorrhage (ICH) among users of antithrombotic medications. Marketscan® healthcare claims database enriched with Micromedex® drug database from 2013-01-01 to 2021-12-31 was used. Logistic regression, random forest, and XGBoost were trained, validated and tested. Model development included feature selection via ElasticNetCV and hyperparameter tuning using GridSearch CV. The best classification threshold was selected based on Youden’s statistics. Performance was primarily assessed using AUC, F1. Overall, tuned models achieved an average accuracy of 0.72 and AUC of 0.70.
Speaker(s):
Prajwal Pradhan, MPH
University of Minnesota Institute for Health Informatics
Author(s):
Gyorgy Simon, PhD; Steven G. Johnson, PhD - University of Minnesota; Erich Kummerfeld, PhD - University of Minnesota; Terrence Adam, MD, PhD - Sioux Falls Veterans Administration;
Presentation Type: Poster Invite - Student
Poster Number: 230
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Translation Bioinformatics/Precision Medicine
Machine learning models have shown promise for predicting range of adverse events in the field of cardiovascular therapies. We aimed to develop and validate the performance of various ML models to predict intracranial hemorrhage (ICH) among users of antithrombotic medications. Marketscan® healthcare claims database enriched with Micromedex® drug database from 2013-01-01 to 2021-12-31 was used. Logistic regression, random forest, and XGBoost were trained, validated and tested. Model development included feature selection via ElasticNetCV and hyperparameter tuning using GridSearch CV. The best classification threshold was selected based on Youden’s statistics. Performance was primarily assessed using AUC, F1. Overall, tuned models achieved an average accuracy of 0.72 and AUC of 0.70.
Speaker(s):
Prajwal Pradhan, MPH
University of Minnesota Institute for Health Informatics
Author(s):
Gyorgy Simon, PhD; Steven G. Johnson, PhD - University of Minnesota; Erich Kummerfeld, PhD - University of Minnesota; Terrence Adam, MD, PhD - Sioux Falls Veterans Administration;
Prajwal
Pradhan,
MPH - University of Minnesota Institute for Health Informatics
Potential of Patient-Facing Technologies: A REDCap-Based Case Study
Presentation Type: Poster Invite - Student
Poster Number: 231
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
Patient-facing technologies (PFTs) present new opportunities for enhancing self-management and patient engagement, yet sustainable and generalizable design architectures remain underdeveloped. This study introduces MedCap, a REDCap-based PFT architecture that is designed to support cancer patients’ concerns about signs, symptoms, and medication management during transitions of care. Guided by the Chronic Care Model (CCM) and User-Centered Design (UCD) principles and centered on three key user roles: patients, administrative (clinical) staff, and technical personnel, the system development outlines a reproducible approach to building PFT platforms, detailing core functionalities and implementation challenges. Patient representatives participated in two rounds of interviews, providing feedback that guided iterative refinements and highlighted usability barriers. Leveraging the flexible capabilities of REDCap, this work proposes a generalizable design strategy to inform future PFT development and advance integration into patient-centered, interoperable health information systems.
Speaker(s):
Yuheng Shi, MS
UTHealth Houston
Author(s):
Yuheng Shi, MS - UTHealth Houston; Zhengcan Xie, MS - UT Health Houston; Eric Yang, MS - UTHealth Houston; Katie Gahn, BS - University of Michigan; Heidi Mason, DNP, ACNP-BC - University of Michigan; Yun Jiang, PhD, MS, RN, FAMIA - University of Michigan; Yang Gong, MD, PhD - UTHealth Houston;
Presentation Type: Poster Invite - Student
Poster Number: 231
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
Patient-facing technologies (PFTs) present new opportunities for enhancing self-management and patient engagement, yet sustainable and generalizable design architectures remain underdeveloped. This study introduces MedCap, a REDCap-based PFT architecture that is designed to support cancer patients’ concerns about signs, symptoms, and medication management during transitions of care. Guided by the Chronic Care Model (CCM) and User-Centered Design (UCD) principles and centered on three key user roles: patients, administrative (clinical) staff, and technical personnel, the system development outlines a reproducible approach to building PFT platforms, detailing core functionalities and implementation challenges. Patient representatives participated in two rounds of interviews, providing feedback that guided iterative refinements and highlighted usability barriers. Leveraging the flexible capabilities of REDCap, this work proposes a generalizable design strategy to inform future PFT development and advance integration into patient-centered, interoperable health information systems.
Speaker(s):
Yuheng Shi, MS
UTHealth Houston
Author(s):
Yuheng Shi, MS - UTHealth Houston; Zhengcan Xie, MS - UT Health Houston; Eric Yang, MS - UTHealth Houston; Katie Gahn, BS - University of Michigan; Heidi Mason, DNP, ACNP-BC - University of Michigan; Yun Jiang, PhD, MS, RN, FAMIA - University of Michigan; Yang Gong, MD, PhD - UTHealth Houston;
Yuheng
Shi,
MS - UTHealth Houston
A Pharmacogenomic-Informed Representation Improves Multimodal EHR Survival Prediction
Presentation Type: Poster Invite - Regular
Poster Number: 232
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Survival prediction in lung cancer requires integrating clinical and molecular signals, yet multimodal EHR models remain limited by the information recorded in clinical data. Drug–gene interactions, which strongly influence treatment response, are not encoded in structured EHR features and therefore cannot be incorporated into existing survival models. External pharmacogenomic datasets quantify these interactions, but they have rarely been translated into real-world prediction. Recent large language models enable learning pharmacogenomic relationships directly from drug structure and mutation context.
We develop a representation-transfer strategy that finetunes a large language model on experimental drug–mutation sensitivity pairs to generate a pharmacogenomics-informed embedding (PGx embedding). This embedding provides mechanistic signal absent from clinical modalities and is used as complementary context within multimodal EHR prediction models.
Using a curated lung cancer cohort with mutation, medication, and laboratory features, we evaluated four previously published survival architectures. Adding PGx embeddings significantly improved AUROC across all models (p < 0.001), demonstrating that PGx supplies information not captured by existing EHR features. Modality-specific ablation showed the largest gains in lab-only models and the smallest in gene-plus-drug models, indicating a systematic pattern of complementary value.
Pharmacogenomic embeddings introduce mechanistic signal that enhances multimodal EHR prediction and provide a generalizable strategy for integrating external biological knowledge into clinical AI systems.
Speaker(s):
Mun Hwan Lee
Mayo Clinic
Author(s):
Mun Hwan Lee - Mayo Clinic; Mun Hwan Lee, Ph.D - Mayo Clinic; Yang Xiao, MS - Mayo Clinic; Xiaodi Li, Ph.D. - Mayo Clinic; Eric Klee, Ph.D - Mayo Clinic; Ping Yang, M.D., Ph.D. - Mayo Clinic; Terence Sio, MD, PhD - Mayo Clinic; Liewei Wang, MD, PhD - Mayo Clinic; James Cerhan, MD, PhD - Mayo Clinic; Nansu Zong, Ph.D. - Mayo Clinic;
Presentation Type: Poster Invite - Regular
Poster Number: 232
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Survival prediction in lung cancer requires integrating clinical and molecular signals, yet multimodal EHR models remain limited by the information recorded in clinical data. Drug–gene interactions, which strongly influence treatment response, are not encoded in structured EHR features and therefore cannot be incorporated into existing survival models. External pharmacogenomic datasets quantify these interactions, but they have rarely been translated into real-world prediction. Recent large language models enable learning pharmacogenomic relationships directly from drug structure and mutation context.
We develop a representation-transfer strategy that finetunes a large language model on experimental drug–mutation sensitivity pairs to generate a pharmacogenomics-informed embedding (PGx embedding). This embedding provides mechanistic signal absent from clinical modalities and is used as complementary context within multimodal EHR prediction models.
Using a curated lung cancer cohort with mutation, medication, and laboratory features, we evaluated four previously published survival architectures. Adding PGx embeddings significantly improved AUROC across all models (p < 0.001), demonstrating that PGx supplies information not captured by existing EHR features. Modality-specific ablation showed the largest gains in lab-only models and the smallest in gene-plus-drug models, indicating a systematic pattern of complementary value.
Pharmacogenomic embeddings introduce mechanistic signal that enhances multimodal EHR prediction and provide a generalizable strategy for integrating external biological knowledge into clinical AI systems.
Speaker(s):
Mun Hwan Lee
Mayo Clinic
Author(s):
Mun Hwan Lee - Mayo Clinic; Mun Hwan Lee, Ph.D - Mayo Clinic; Yang Xiao, MS - Mayo Clinic; Xiaodi Li, Ph.D. - Mayo Clinic; Eric Klee, Ph.D - Mayo Clinic; Ping Yang, M.D., Ph.D. - Mayo Clinic; Terence Sio, MD, PhD - Mayo Clinic; Liewei Wang, MD, PhD - Mayo Clinic; James Cerhan, MD, PhD - Mayo Clinic; Nansu Zong, Ph.D. - Mayo Clinic;
Mun Hwan
Lee - Mayo Clinic
Moving into the Modern Era of the Emerging Infections Program (EIP)
Presentation Type: Poster - Regular
Poster Number: 233
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
The Emerging Infections Program (EIP) has entered a new era with a revitalized focus on data modernization (DM). Four key deliverables: (DM strategy, roadmap, monitoring plan, and dashboard) were developed using a collaborative, network-level approach. The strategy outlines aims and activities, while the roadmap details specific steps for implementation. The monitoring plan tracks progress across activities, and the dashboard visualizes network metrics, facilitating ongoing evaluation of modernization efforts across the network.
Speaker(s):
Alyssa Harvey, MPH
CDC
Author(s):
Megan Mueller, MPH - Centers for Disease Control and Prevention; Tricia Aden, MT(ASCP) - Centers for Disease Control and Prevention;
Presentation Type: Poster - Regular
Poster Number: 233
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
The Emerging Infections Program (EIP) has entered a new era with a revitalized focus on data modernization (DM). Four key deliverables: (DM strategy, roadmap, monitoring plan, and dashboard) were developed using a collaborative, network-level approach. The strategy outlines aims and activities, while the roadmap details specific steps for implementation. The monitoring plan tracks progress across activities, and the dashboard visualizes network metrics, facilitating ongoing evaluation of modernization efforts across the network.
Speaker(s):
Alyssa Harvey, MPH
CDC
Author(s):
Megan Mueller, MPH - Centers for Disease Control and Prevention; Tricia Aden, MT(ASCP) - Centers for Disease Control and Prevention;
Alyssa
Harvey,
MPH - CDC
Reducing Human Burden in Legal Epidemiology Through Large Language Models: Identifying Substance Use Disorder Treatment Policies
Presentation Type: Poster - Regular
Poster Number: 234
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Identifying relevant legal documents in legal epidemiology is both labor-intensive and time-sensitive. This study examines the capability of Large Language Models (LLMs) to identify substance use disorder policies for emergency department practices in Massachusetts. From a pool of 99 documents, we compared LLM-based screening with human expert review. The LLM identified three documents found by humans plus two additional ones, demonstrating its potential to reduce human burden. Future work will focus on optimizing human–LLM collaboration.
Speaker(s):
Jaeyoung Park, PhD
University of Central Florida
Author(s):
Jaeyoung Park, PhD - University of Central Florida; Barbara Andraka-Christou, JD, PhD - University of Central Florida; Fatema Ahmed, PhD, MHA - University of Central Florida; Suhas Shewale, PhD - University of Central Florida; Tahira Yeasmeen, MPH, PhD - University of Central Florida; Thuy Nguyen, PhD - University of Michigan;
Presentation Type: Poster - Regular
Poster Number: 234
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Identifying relevant legal documents in legal epidemiology is both labor-intensive and time-sensitive. This study examines the capability of Large Language Models (LLMs) to identify substance use disorder policies for emergency department practices in Massachusetts. From a pool of 99 documents, we compared LLM-based screening with human expert review. The LLM identified three documents found by humans plus two additional ones, demonstrating its potential to reduce human burden. Future work will focus on optimizing human–LLM collaboration.
Speaker(s):
Jaeyoung Park, PhD
University of Central Florida
Author(s):
Jaeyoung Park, PhD - University of Central Florida; Barbara Andraka-Christou, JD, PhD - University of Central Florida; Fatema Ahmed, PhD, MHA - University of Central Florida; Suhas Shewale, PhD - University of Central Florida; Tahira Yeasmeen, MPH, PhD - University of Central Florida; Thuy Nguyen, PhD - University of Michigan;
Jaeyoung
Park,
PhD - University of Central Florida
Geographic Disparities in Rheumatoid Arthritis Prevalence, Healthcare Costs, and Quality of Care Among US Adult Medicare Beneficiaries
Presentation Type: Poster Invite - Regular
Poster Number: 235
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
This study investigates rural–urban disparities in rheumatoid arthritis (RA) care among U.S. Medicare beneficiaries from 2022–2023, focusing on service costs, provider distribution, and quality of care. By integrating Medicare Utilization data with CDC PLACES, RUCA codes, NUCC taxonomy, and QPP metrics, the analysis evaluates how provider location and performance shape cost differences across regions. Rural beneficiaries have access to far fewer providers and often receive care from urban clinicians, resulting in substantially higher average service costs—up to twice those of similar urban beneficiaries in low-risk groups. Urban providers show higher performance, with a mean MIPS score of 83.11 versus 73.33 for rural providers and deliver more services per beneficiary (100 vs. 22), reflecting stronger infrastructure. Despite this, rural beneficiaries continue to bear higher costs when accessing urban-based care. These findings highlight the need for targeted rural incentives to improve equitable RA management.
Speaker(s):
Tonushree Dutta, BSc in Electrical and Electronic Engineering
University of Minnesota
Author(s):
Tonushree Dutta, BSc in Electrical and Electronic Engineering - University of Minnesota; Sourav Kumar Ghosh, MS in Electrical and Computer Engineering - University of Minnesota; Ingrid R. Aragon, MS in Bioinformatics and Computational Biology - University of Minnesota;
Presentation Type: Poster Invite - Regular
Poster Number: 235
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
This study investigates rural–urban disparities in rheumatoid arthritis (RA) care among U.S. Medicare beneficiaries from 2022–2023, focusing on service costs, provider distribution, and quality of care. By integrating Medicare Utilization data with CDC PLACES, RUCA codes, NUCC taxonomy, and QPP metrics, the analysis evaluates how provider location and performance shape cost differences across regions. Rural beneficiaries have access to far fewer providers and often receive care from urban clinicians, resulting in substantially higher average service costs—up to twice those of similar urban beneficiaries in low-risk groups. Urban providers show higher performance, with a mean MIPS score of 83.11 versus 73.33 for rural providers and deliver more services per beneficiary (100 vs. 22), reflecting stronger infrastructure. Despite this, rural beneficiaries continue to bear higher costs when accessing urban-based care. These findings highlight the need for targeted rural incentives to improve equitable RA management.
Speaker(s):
Tonushree Dutta, BSc in Electrical and Electronic Engineering
University of Minnesota
Author(s):
Tonushree Dutta, BSc in Electrical and Electronic Engineering - University of Minnesota; Sourav Kumar Ghosh, MS in Electrical and Computer Engineering - University of Minnesota; Ingrid R. Aragon, MS in Bioinformatics and Computational Biology - University of Minnesota;
Tonushree
Dutta,
BSc in Electrical and Electronic Engineering - University of Minnesota
Multi-layer Co-expression Networks applied to Three Brain Regions of Alzheimer’s Donors
Presentation Type: Poster Invite - Regular
Poster Number: 238
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Translation Bioinformatics/Precision Medicine
With growing multi-omics data, multi-layer network tools are needed to handle weaker inter-layer edges and donors
missing layers. We present a multi-layer co-expression framework applied to RNA-seq data from three brain regions
of 464 ROSMAP donors. Our framework uses dynamic soft-thresholding and statistical distance matching to
integrate inter and intra-layer subnetworks, yielding 390 multi-region modules and a generalizable adjacency matrix.
Speaker(s):
Judith Somekh, PhD
University of Haifa
Author(s):
Eden Eldar, MSc - University of Haifa; David Bennett, MD - Rush University Medical Center; Judith Somekh, PhD - University of Haifa;
Presentation Type: Poster Invite - Regular
Poster Number: 238
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Translation Bioinformatics/Precision Medicine
With growing multi-omics data, multi-layer network tools are needed to handle weaker inter-layer edges and donors
missing layers. We present a multi-layer co-expression framework applied to RNA-seq data from three brain regions
of 464 ROSMAP donors. Our framework uses dynamic soft-thresholding and statistical distance matching to
integrate inter and intra-layer subnetworks, yielding 390 multi-region modules and a generalizable adjacency matrix.
Speaker(s):
Judith Somekh, PhD
University of Haifa
Author(s):
Eden Eldar, MSc - University of Haifa; David Bennett, MD - Rush University Medical Center; Judith Somekh, PhD - University of Haifa;
Judith
Somekh,
PhD - University of Haifa
Poster Session 1 and Reception – Sponsored by University of Colorado Anschutz Center for Health AI
Description
Custom CSS
double-click to edit, do not edit in source