- Home
- 2026 Amplify Informatics Conference Gallery
- Leveraging EHR Data to Identify Gaps in Osteoporosis Screening for Patients with Rheumatoid Arthritis and/or Chronic Glucocorticoids
Custom CSS
double-click to edit, do not edit in source
5/20/2026 |
5:00 PM – 6:30 PM |
Aspen Ballroom
Poster Session 2 and Reception
Presentation Type: Poster
Breast Cancer Restoration Surgery Complication Prediction and Identification of Risk Factors
Presentation Type: Poster - Regular
Poster Number: 100
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, Health Data Science, Population Health, Digital Therapeutics, Remote Patient Monitoring (RPM), and Digital Engagement
Primary Track: Big Data for Health
Women with varying comorbidities and characteristics experience different postoperative complications in the year following breast reconstruction. Understanding and predicting the risk associated with each complication is essential for better patient care. In this work, we utilize data of 5,620 patients from Clalit Health Services, to train and validate a prediction model for post-reconstruction complications.. We further employ machine learning to predict the nuanced risk for five specific complication types. Building on this model, we performed feature analysis, and found that high BMI, Charlson comorbidity score, and Total Cholesterol levels were strongly correlated with elevated risk. This study aims to contribute to our understanding of post breast reconstruction risks and provide better, personalized patient care.
Speaker(s):
Nadav Rappoport, Ph.D.
Ben-Gurion University of the Negev
Author(s):
Idan Cohen Zada, Msc. - Ben-Gurion University of the Negev; Eitan Bachmat, PhD - Ben-Gurion University of the Negev; Nadav Rappoport, Ph.D. - Ben-Gurion University of the Negev; Elizaveta Kouniavski, M.D - Kaplan Hospital, Department of Plastic Surgery;
Presentation Type: Poster - Regular
Poster Number: 100
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, Health Data Science, Population Health, Digital Therapeutics, Remote Patient Monitoring (RPM), and Digital Engagement
Primary Track: Big Data for Health
Women with varying comorbidities and characteristics experience different postoperative complications in the year following breast reconstruction. Understanding and predicting the risk associated with each complication is essential for better patient care. In this work, we utilize data of 5,620 patients from Clalit Health Services, to train and validate a prediction model for post-reconstruction complications.. We further employ machine learning to predict the nuanced risk for five specific complication types. Building on this model, we performed feature analysis, and found that high BMI, Charlson comorbidity score, and Total Cholesterol levels were strongly correlated with elevated risk. This study aims to contribute to our understanding of post breast reconstruction risks and provide better, personalized patient care.
Speaker(s):
Nadav Rappoport, Ph.D.
Ben-Gurion University of the Negev
Author(s):
Idan Cohen Zada, Msc. - Ben-Gurion University of the Negev; Eitan Bachmat, PhD - Ben-Gurion University of the Negev; Nadav Rappoport, Ph.D. - Ben-Gurion University of the Negev; Elizaveta Kouniavski, M.D - Kaplan Hospital, Department of Plastic Surgery;
Nadav
Rappoport,
Ph.D. - Ben-Gurion University of the Negev
MedDent AI: Analytical Feasibility of an AI-enabled SaMD for Image–based Oral Disease Screening
Presentation Type: Poster - Student
Poster Number: 102
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, Innovation Partnerships, Implementation Science, and Learning Health Systems, Human Factors and Usability, Clinical Decision Support and Care Pathways, Outcomes Improvement and Equity
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Oral diseases can delay critical medical care. Two software-as-a-medical-device (SaMD) image classifiers screened de-identified intraoral photos using separate dental caries (DC; N=737) and periodontal disease (PD; N=222) models. On held-out tests (DC n=223; PD n=68), area under the receiver operating characteristic curve was 0.99 (95% confidence interval 0.98-1.00) for DC and 0.82 (0.72-0.90) for PD. DC referral sensitivity/specificity was 0.96/0.87; few healthy PD test images (n=13) limited precision.
Speaker(s):
Georgia Dounis, JHU BIDS MS PROGRAM
JHU STUDENT
Author(s):
Kiki Dounis, JHU BIDS Program - JHU; Frank Jones, DDS, MBA - University of Los Angeles, CA; Adler Archer, J.D., M.S., M.Sc., M.P.S. - Johns Hopkins University; Alex Zhu, BS, MSE - Whiting School of Engineering, GR, Johns Hopkins University;
Presentation Type: Poster - Student
Poster Number: 102
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, Innovation Partnerships, Implementation Science, and Learning Health Systems, Human Factors and Usability, Clinical Decision Support and Care Pathways, Outcomes Improvement and Equity
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Oral diseases can delay critical medical care. Two software-as-a-medical-device (SaMD) image classifiers screened de-identified intraoral photos using separate dental caries (DC; N=737) and periodontal disease (PD; N=222) models. On held-out tests (DC n=223; PD n=68), area under the receiver operating characteristic curve was 0.99 (95% confidence interval 0.98-1.00) for DC and 0.82 (0.72-0.90) for PD. DC referral sensitivity/specificity was 0.96/0.87; few healthy PD test images (n=13) limited precision.
Speaker(s):
Georgia Dounis, JHU BIDS MS PROGRAM
JHU STUDENT
Author(s):
Kiki Dounis, JHU BIDS Program - JHU; Frank Jones, DDS, MBA - University of Los Angeles, CA; Adler Archer, J.D., M.S., M.Sc., M.P.S. - Johns Hopkins University; Alex Zhu, BS, MSE - Whiting School of Engineering, GR, Johns Hopkins University;
Georgia
Dounis,
JHU BIDS MS PROGRAM - JHU STUDENT
Flawed Questions, Flawed Logic: A Taxonomy and Correction of Reasoning Errors in Medical LLMs Using Sparse Autoencoders
Presentation Type: Poster - Regular
Click to View Presentation
Poster Number: 103
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Analytical Artificial Intelligence: ML, Digital Pathology, Imaging AI, Predictive Analytics, Governance, Data Privacy, Cybersecurity, Reliability, and Security, Analytics, Registries, and the Digital Command Center
Primary Track: Big Data for Health
We investigated reasoning failures in medical LLMs by auditing benchmark quality and developing a clinically-informed error taxonomy. Evaluating OpenAI o1 on MedQA revealed that 41% of apparent errors reflected benchmark flaws, including missing figures and ambiguities. Our taxonomy classified genuine failures into four categories validated across multiple models. Using sparse autoencoders to steer reasoning-specific features significantly improved accuracy across three benchmarks, suggesting mechanistic interventions could advance interpretable, process-aware clinical AI.
Speaker(s):
Siru Liu, PhD
Vanderbilt University Medical Center
Author(s):
Siru Liu, PhD - Vanderbilt University Medical Center; Jialin Liu, MD - West China Hospital Sichuan University; Adam Wright, PhD - Vanderbilt University Medical Center;
Presentation Type: Poster - Regular
Click to View Presentation
Poster Number: 103
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Analytical Artificial Intelligence: ML, Digital Pathology, Imaging AI, Predictive Analytics, Governance, Data Privacy, Cybersecurity, Reliability, and Security, Analytics, Registries, and the Digital Command Center
Primary Track: Big Data for Health
We investigated reasoning failures in medical LLMs by auditing benchmark quality and developing a clinically-informed error taxonomy. Evaluating OpenAI o1 on MedQA revealed that 41% of apparent errors reflected benchmark flaws, including missing figures and ambiguities. Our taxonomy classified genuine failures into four categories validated across multiple models. Using sparse autoencoders to steer reasoning-specific features significantly improved accuracy across three benchmarks, suggesting mechanistic interventions could advance interpretable, process-aware clinical AI.
Speaker(s):
Siru Liu, PhD
Vanderbilt University Medical Center
Author(s):
Siru Liu, PhD - Vanderbilt University Medical Center; Jialin Liu, MD - West China Hospital Sichuan University; Adam Wright, PhD - Vanderbilt University Medical Center;
Siru
Liu,
PhD - Vanderbilt University Medical Center
“An Extra Set of Eyes,” Not a Replacement: Patient Views of Diverse Artificial Intelligence Tools Across the Cancer Care Continuum
Presentation Type: Poster - Regular
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, Human Factors and Usability, Clinical Decision Support and Care Pathways, Diagnostics, Outcomes Improvement and Equity
Primary Track: Big Data for Health
Artificial intelligence (AI) is increasingly integrated into oncology, yet patient perspectives on its varied applications remain underexplored. To inform patient-centered implementation, we examined views of multiple AI tools across the cancer care continuum. Three virtual, 60-minute focus groups at a high-volume cancer center were recorded, transcribed, and analyzed thematically. Participants responded to five AI use cases: diagnostic/screening support, treatment-planning assistance, triage of portal messages about side effects, clinical-trial matching, and survivorship risk assessment. Twenty participants (average age 60; 14 female; 16 white) expressed cautiously positive views, emphasizing that AI should augment—not replace—clinical expertise. Diagnostic AI was seen as an “extra set of eyes” that could reduce error and increase efficiency, but participants insisted that clinicians verify results. For treatment planning, they appreciated AI’s ability to analyze large datasets but wanted transparency about data sources, assurance that information was up to date and non-biased, and clinician interpretation tailored to the individual. AI triage of side-effect messages elicited the strongest reservations. While some saw potential to speed care access, most worried about errors and wanted clinician involvement, especially for serious symptoms. AI-supported clinical-trial matching was viewed as promising for expanding access and identifying overlooked opportunities, though some raised equity concerns. Survivorship risk-assessment tools were acceptable only with clear clinician oversight and validation. Overall, patients supported AI when it enhances accuracy, efficiency, and access while preserving human judgment, transparency, and communication. Ensuring clinician oversight and addressing concerns about equity and data integrity are essential for trustworthy, patient-centered AI in cancer care.
Speaker(s):
Fernanda Polubriaginof, MD PhD
Memorial Sloan Kettering Cancer Center
Author(s):
SUSAN CHIMONAS, PhD - Memorial Sloan Kettering Cancer Center; Allison Lipitz-Snyderman, PhD - Memorial Sloan Kettering Cancer Center; Jennifer Prey Dawson, PhD - Geisinger; Fernanda Polubriaginof, MD PhD - Memorial Sloan Kettering Cancer Center;
Presentation Type: Poster - Regular
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, Human Factors and Usability, Clinical Decision Support and Care Pathways, Diagnostics, Outcomes Improvement and Equity
Primary Track: Big Data for Health
Artificial intelligence (AI) is increasingly integrated into oncology, yet patient perspectives on its varied applications remain underexplored. To inform patient-centered implementation, we examined views of multiple AI tools across the cancer care continuum. Three virtual, 60-minute focus groups at a high-volume cancer center were recorded, transcribed, and analyzed thematically. Participants responded to five AI use cases: diagnostic/screening support, treatment-planning assistance, triage of portal messages about side effects, clinical-trial matching, and survivorship risk assessment. Twenty participants (average age 60; 14 female; 16 white) expressed cautiously positive views, emphasizing that AI should augment—not replace—clinical expertise. Diagnostic AI was seen as an “extra set of eyes” that could reduce error and increase efficiency, but participants insisted that clinicians verify results. For treatment planning, they appreciated AI’s ability to analyze large datasets but wanted transparency about data sources, assurance that information was up to date and non-biased, and clinician interpretation tailored to the individual. AI triage of side-effect messages elicited the strongest reservations. While some saw potential to speed care access, most worried about errors and wanted clinician involvement, especially for serious symptoms. AI-supported clinical-trial matching was viewed as promising for expanding access and identifying overlooked opportunities, though some raised equity concerns. Survivorship risk-assessment tools were acceptable only with clear clinician oversight and validation. Overall, patients supported AI when it enhances accuracy, efficiency, and access while preserving human judgment, transparency, and communication. Ensuring clinician oversight and addressing concerns about equity and data integrity are essential for trustworthy, patient-centered AI in cancer care.
Speaker(s):
Fernanda Polubriaginof, MD PhD
Memorial Sloan Kettering Cancer Center
Author(s):
SUSAN CHIMONAS, PhD - Memorial Sloan Kettering Cancer Center; Allison Lipitz-Snyderman, PhD - Memorial Sloan Kettering Cancer Center; Jennifer Prey Dawson, PhD - Geisinger; Fernanda Polubriaginof, MD PhD - Memorial Sloan Kettering Cancer Center;
Fernanda
Polubriaginof,
MD PhD - Memorial Sloan Kettering Cancer Center
Mitigating Health Disparities with AI
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, Social Determinants of Health (SDoH), Outcomes Improvement and Equity, Health Data Science
Primary Track: Big Data for Health
Social Determinants of Health (SDOH) significantly impact health outcomes and risks, and identifying individuals with related needs is crucial for intervention. This research project aims to leverage Artificial Intelligence (AI) and Machine Learning (ML) using a large, diverse dataset to identify patients who need community resources but may have been missed by traditional screening methods. The goal is to identify those without a positive SDOH screen flag who have profiles similar to those who did screen positive.
Speaker(s):
Kendria Hall, MD
University at Buffalo School of Medicine and Biomedical Sciences
Author(s):
Kendria Hall, MD - University at Buffalo School of Medicine and Biomedical Sciences;
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, Social Determinants of Health (SDoH), Outcomes Improvement and Equity, Health Data Science
Primary Track: Big Data for Health
Social Determinants of Health (SDOH) significantly impact health outcomes and risks, and identifying individuals with related needs is crucial for intervention. This research project aims to leverage Artificial Intelligence (AI) and Machine Learning (ML) using a large, diverse dataset to identify patients who need community resources but may have been missed by traditional screening methods. The goal is to identify those without a positive SDOH screen flag who have profiles similar to those who did screen positive.
Speaker(s):
Kendria Hall, MD
University at Buffalo School of Medicine and Biomedical Sciences
Author(s):
Kendria Hall, MD - University at Buffalo School of Medicine and Biomedical Sciences;
Kendria
Hall,
MD - University at Buffalo School of Medicine and Biomedical Sciences
Health-System Stewardship of Deployed AI: A Scoping Review of Real-World Practices and Gaps
Presentation Type: Poster - Regular
Poster Number: 106
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Analytical Artificial Intelligence: ML, Digital Pathology, Imaging AI, Predictive Analytics, Governance, Innovation Partnerships, Implementation Science, and Learning Health Systems, Leadership and Strategy
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
We conducted a PRISMA-ScR–aligned scoping review (PubMed, 2015-present) to characterize post-deployment monitoring of AI-enabled digital health in real-world health systems. Six included studies yielded five recommendation areas (performance monitoring/case reporting, post-market surveillance, adverse-event dissemination, end-user training, and data standardization/documentation). Six cross-cutting themes emerged: fragmented oversight, inconsistent monitoring, challenges with bias and drift, economic sustainability, limited de-implementation strategies, and workflow integration gaps, highlighting the need for a harmonized, scalable stewardship framework.
Speaker(s):
Chenyu Gai, Master
Mayo Clinic
Author(s):
Chenyu Gai, Master - Mayo Clinic; Chung Wi, MD - Mayo Clinic; Momin Malik, PhD in Societal Computing - Mayo Clinic; Young Juhn, M.D., M.P.H. - Mayo Clinic;
Presentation Type: Poster - Regular
Poster Number: 106
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Analytical Artificial Intelligence: ML, Digital Pathology, Imaging AI, Predictive Analytics, Governance, Innovation Partnerships, Implementation Science, and Learning Health Systems, Leadership and Strategy
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
We conducted a PRISMA-ScR–aligned scoping review (PubMed, 2015-present) to characterize post-deployment monitoring of AI-enabled digital health in real-world health systems. Six included studies yielded five recommendation areas (performance monitoring/case reporting, post-market surveillance, adverse-event dissemination, end-user training, and data standardization/documentation). Six cross-cutting themes emerged: fragmented oversight, inconsistent monitoring, challenges with bias and drift, economic sustainability, limited de-implementation strategies, and workflow integration gaps, highlighting the need for a harmonized, scalable stewardship framework.
Speaker(s):
Chenyu Gai, Master
Mayo Clinic
Author(s):
Chenyu Gai, Master - Mayo Clinic; Chung Wi, MD - Mayo Clinic; Momin Malik, PhD in Societal Computing - Mayo Clinic; Young Juhn, M.D., M.P.H. - Mayo Clinic;
Chenyu
Gai,
Master - Mayo Clinic
Order from Chaos: Pediatric Oncology Epic Implementation
Presentation Type: Poster - Regular
Poster Number: 107
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Change Management, Leadership and Strategy, Education and Training
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
The implementation of EPIC for Pediatric Hematology/Oncology (PHO) in large health systems requires reeducation of clinical and informatic leadership. Managing the complexities of PHO including body surface area-based dosing and coordination of integrated send-out pathology requires complicated, non-standardized communication and data exchange among interdisciplinary teams. To tackle these challenges, we engaged stakeholders to facilitate building a new change management order set implementation. The performance of this implementation maintained effective workflows and patient safety.
Speaker(s):
Mark Atlas, MD
Northwell Health
Author(s):
Alfred Caligiuri, PA, MBA;
Presentation Type: Poster - Regular
Poster Number: 107
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Change Management, Leadership and Strategy, Education and Training
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
The implementation of EPIC for Pediatric Hematology/Oncology (PHO) in large health systems requires reeducation of clinical and informatic leadership. Managing the complexities of PHO including body surface area-based dosing and coordination of integrated send-out pathology requires complicated, non-standardized communication and data exchange among interdisciplinary teams. To tackle these challenges, we engaged stakeholders to facilitate building a new change management order set implementation. The performance of this implementation maintained effective workflows and patient safety.
Speaker(s):
Mark Atlas, MD
Northwell Health
Author(s):
Alfred Caligiuri, PA, MBA;
Mark
Atlas,
MD - Northwell Health
Implementing Real-World Change in Electronic Preventative Health Tools and Care Pathways
Presentation Type: Poster - Regular
Poster Number: 108
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Change Management, Clinical Decision Support and Care Pathways, Workforce Automation, Communication, and Workflow Efficiency
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
This project streamlined over 900 clinical reminders (CR’s) in the VA’s electronic health record through systematic review and governance development. By removing redundant, outdated, and misconfigured CR’s, and establishing a dedicated committee and procedures, we enhanced clinical decision support and care pathways. The integration and collaboration of informatics and clinical expertise improved workflow efficiency, data accuracy, and patient outcomes, reducing CR’s to 64 mandatory National, 4 Regional, and 3 Local.
Speaker(s):
Tania Knight, BSN
Veterans Affairs Medical Center
Author(s):
Tania Knight, BSN - Veterans Affairs Medical Center; Trevor Jones, MD - Salt Lake City VA; Nathan Erickson, BS - Veterans Affairs Medical Center;
Presentation Type: Poster - Regular
Poster Number: 108
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Change Management, Clinical Decision Support and Care Pathways, Workforce Automation, Communication, and Workflow Efficiency
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
This project streamlined over 900 clinical reminders (CR’s) in the VA’s electronic health record through systematic review and governance development. By removing redundant, outdated, and misconfigured CR’s, and establishing a dedicated committee and procedures, we enhanced clinical decision support and care pathways. The integration and collaboration of informatics and clinical expertise improved workflow efficiency, data accuracy, and patient outcomes, reducing CR’s to 64 mandatory National, 4 Regional, and 3 Local.
Speaker(s):
Tania Knight, BSN
Veterans Affairs Medical Center
Author(s):
Tania Knight, BSN - Veterans Affairs Medical Center; Trevor Jones, MD - Salt Lake City VA; Nathan Erickson, BS - Veterans Affairs Medical Center;
Tania
Knight,
BSN - Veterans Affairs Medical Center
Optimizing Community Pharmacy Workflows With Enhanced CDS Tools
Presentation Type: Poster - Regular
Poster Number: 109
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support and Care Pathways, Standards, Terminology, and Interoperability, TEFCA, FHIR, Clinician Well-Being, Innovation Partnerships, Implementation Science, and Learning Health Systems
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Outpatient pharmacists are inundated with electronic drug-utilization alerts and do not always have the time or resources to address all the alerts they receive. This busy workflow leaves less time for patient interactions and possibility of missed interventions.
We examined the impact of employing a single clinical concept warning across multiple content areas such as dosing, drug-drug interaction, duplicate therapy, and drug-disease contraindications on reducing the number of clinical alerts screened by the outpatient pharmacist.
Speaker(s):
Usha Desiraju, PharmD
First Databank
Author(s):
Mike Silver, PharmD - First Databank;
Presentation Type: Poster - Regular
Poster Number: 109
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support and Care Pathways, Standards, Terminology, and Interoperability, TEFCA, FHIR, Clinician Well-Being, Innovation Partnerships, Implementation Science, and Learning Health Systems
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Outpatient pharmacists are inundated with electronic drug-utilization alerts and do not always have the time or resources to address all the alerts they receive. This busy workflow leaves less time for patient interactions and possibility of missed interventions.
We examined the impact of employing a single clinical concept warning across multiple content areas such as dosing, drug-drug interaction, duplicate therapy, and drug-disease contraindications on reducing the number of clinical alerts screened by the outpatient pharmacist.
Speaker(s):
Usha Desiraju, PharmD
First Databank
Author(s):
Mike Silver, PharmD - First Databank;
Usha
Desiraju,
PharmD - First Databank
SEE-Diabetes: Leveraging Digital Engagement for Personalized Self-Care
Presentation Type: Poster - Regular
Poster Number: 110
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support and Care Pathways, Education and Training, Human Factors and Usability
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Diabetes affects nearly 25% of Americans over 65, yet fewer than 10% seek self-management education in the first year, highlighting a gap in patient-centered support. SEE-Diabetes (Support-Engage-Empower-Diabetes), developed over five years by an interdisciplinary team, enhances continuous self-care education, fosters digital engagement, and supports shared decision-making through personalized materials for older adults. This pilot study examined patient perceptions of the educational materials’ value and evaluated how providers leveraged digital engagement through the tool to facilitate patient-centered dialogue and collaborative goal setting during clinical encounters.
Speaker(s):
Min Soon Kim, PhD
University of Missouri
Author(s):
Min Soon Kim, PhD - University of Missouri; Uzma Khan, MD - University of Missouri-Columbia; Margaret Day, MD - University of Missouri - Columbia; Suzanne Boren, PhD, MHA, FACMI, FAMIA - University of Missouri; Eduardo Simoes - University of Missouri;
Presentation Type: Poster - Regular
Poster Number: 110
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support and Care Pathways, Education and Training, Human Factors and Usability
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Diabetes affects nearly 25% of Americans over 65, yet fewer than 10% seek self-management education in the first year, highlighting a gap in patient-centered support. SEE-Diabetes (Support-Engage-Empower-Diabetes), developed over five years by an interdisciplinary team, enhances continuous self-care education, fosters digital engagement, and supports shared decision-making through personalized materials for older adults. This pilot study examined patient perceptions of the educational materials’ value and evaluated how providers leveraged digital engagement through the tool to facilitate patient-centered dialogue and collaborative goal setting during clinical encounters.
Speaker(s):
Min Soon Kim, PhD
University of Missouri
Author(s):
Min Soon Kim, PhD - University of Missouri; Uzma Khan, MD - University of Missouri-Columbia; Margaret Day, MD - University of Missouri - Columbia; Suzanne Boren, PhD, MHA, FACMI, FAMIA - University of Missouri; Eduardo Simoes - University of Missouri;
Min Soon
Kim,
PhD - University of Missouri
Supporting Antibiotic Stewardship in Neonates Through an EHR-Integrated Late-Onset Sepsis Pathway
Presentation Type: Poster - Regular
Poster Number: 111
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support and Care Pathways, Innovation Partnerships, Implementation Science, and Learning Health Systems, Education and Training, Change Management, Human Factors and Usability, Outcomes Improvement and Equity
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
An EHR-integrated late-onset sepsis pathway clinical decision support tool was implemented across four level 3 NICUs to support antibiotic stewardship efforts aiming to optimize vancomycin use. The pathway provides step-by-step guidance for initial antibiotic selection, decision support for when to consider broad-spectrum antimicrobials, and management of positive blood cultures, while enabling direct ordering to streamline workflow. Early evidence indicates improved adherence to evidence-based guidelines and reductions in empiric vancomycin use.
Speaker(s):
Alex Ruan, MD
Children's Hospital of Philadelphia
Author(s):
Miren Dhudasia, MBBS, MPH - Children's Hospital of Philadelphia; Emilia Flores, PhD - Penn Medicine; Nikhil Mull, MD; Carly Gartner, PA-C - Penn Medicine; Lori Christ, MD - Children's Hospital of Philadelphia/Penn Medicine; Samuel Garber, MD - Perelman School of Medicine at the University of Pennsylvania; El Noh, DO, MPH - Children's Hospital of Philadelphia/Penn Medicine; Catherine Muhumuza, MD - Children's Hospital of Philadelphia/Penn Medicine; Sagori Mukhopadhyay, MD, MMSc - Children's Hospital of Philadelphia/Penn Medicine;
Presentation Type: Poster - Regular
Poster Number: 111
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support and Care Pathways, Innovation Partnerships, Implementation Science, and Learning Health Systems, Education and Training, Change Management, Human Factors and Usability, Outcomes Improvement and Equity
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
An EHR-integrated late-onset sepsis pathway clinical decision support tool was implemented across four level 3 NICUs to support antibiotic stewardship efforts aiming to optimize vancomycin use. The pathway provides step-by-step guidance for initial antibiotic selection, decision support for when to consider broad-spectrum antimicrobials, and management of positive blood cultures, while enabling direct ordering to streamline workflow. Early evidence indicates improved adherence to evidence-based guidelines and reductions in empiric vancomycin use.
Speaker(s):
Alex Ruan, MD
Children's Hospital of Philadelphia
Author(s):
Miren Dhudasia, MBBS, MPH - Children's Hospital of Philadelphia; Emilia Flores, PhD - Penn Medicine; Nikhil Mull, MD; Carly Gartner, PA-C - Penn Medicine; Lori Christ, MD - Children's Hospital of Philadelphia/Penn Medicine; Samuel Garber, MD - Perelman School of Medicine at the University of Pennsylvania; El Noh, DO, MPH - Children's Hospital of Philadelphia/Penn Medicine; Catherine Muhumuza, MD - Children's Hospital of Philadelphia/Penn Medicine; Sagori Mukhopadhyay, MD, MMSc - Children's Hospital of Philadelphia/Penn Medicine;
Alex
Ruan,
MD - Children's Hospital of Philadelphia
Nudging Transitions from Pediatrics to Adult Providers : A Mixed-Methods QI Study
Presentation Type: Poster - Regular
Poster Number: 112
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support and Care Pathways, Quality Informatics and Lean, Outcomes Improvement and Equity, Innovation Partnerships, Implementation Science, and Learning Health Systems, Leadership and Strategy
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Pediatricians often struggle to initiate transition discussions with teenagers. In a mixed-methods QI study at an urban pediatric clinic, interviews identified barriers, and a baseline chart review showed only 7.5% of well visits included documented transition counseling. An EHR intervention embedding choice-architecture prompts into note templates as a nudge increased documentation to 89% within one month. This low-burden approach improved rates of transition counseling; future phases will explore clinician experiences and opportunities to strengthen transition discussions.
Speaker(s):
William Vervilles, MD,MPH
Hospital of the University of Pennsylvania
Author(s):
Margery Schonfeld, MD, FAAP - Children's Hospital of Philadelphia; Adam Greenberg, MSN, CRNP - Children's Hospital of Philadelphia; Jessica Hart, MD, MHQS - Children's Hospital of Philadelphia;
Presentation Type: Poster - Regular
Poster Number: 112
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support and Care Pathways, Quality Informatics and Lean, Outcomes Improvement and Equity, Innovation Partnerships, Implementation Science, and Learning Health Systems, Leadership and Strategy
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Pediatricians often struggle to initiate transition discussions with teenagers. In a mixed-methods QI study at an urban pediatric clinic, interviews identified barriers, and a baseline chart review showed only 7.5% of well visits included documented transition counseling. An EHR intervention embedding choice-architecture prompts into note templates as a nudge increased documentation to 89% within one month. This low-burden approach improved rates of transition counseling; future phases will explore clinician experiences and opportunities to strengthen transition discussions.
Speaker(s):
William Vervilles, MD,MPH
Hospital of the University of Pennsylvania
Author(s):
Margery Schonfeld, MD, FAAP - Children's Hospital of Philadelphia; Adam Greenberg, MSN, CRNP - Children's Hospital of Philadelphia; Jessica Hart, MD, MHQS - Children's Hospital of Philadelphia;
William
Vervilles,
MD,MPH - Hospital of the University of Pennsylvania
Improving MASLD Screening by Implementation of a Computerized Clinical Decision Support Tool: A Midpoint Analysis
Presentation Type: Poster - Student
Poster Number: 113
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support and Care Pathways, Change Management, Diagnostics
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
This computerized clinical decision support tool was implemented to improve screening for liver fibrosis in the primary care setting through targeted interruptive alert. It utilized common lab values to calculate the Fibrosis-4 (FIB-4) score to risk stratify patients. Midpoint analysis of the 6 months implementation revealed successful firing of the tool but poor response rate from providers, complicated by inappropriate dismissal selections.
Speaker(s):
James Miller, M.D.
Wake Forest University School of Medicine
Author(s):
James Miller, M.D. - Wake Forest University School of Medicine; Ted Xiao, MD - Emory University School of Medicine; Bradley Rowland; Richa Bundy, MPH - Wake Forest Baptist Health; Corey Obermiller, MAS Applied Statistics - Wake Forest School of Medicine; Lauren Witek, MStat - Advocate Health; Adam Moses, MHA, PMP, FAMIA - Wake Forest Baptist Medical Center; Sean Rudnick, MD - Wake Forest University School of Medicine; Ajay Dharod, MD, FACP, FAMIA - Advocate Health - Wake Forest University 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, Change Management, Diagnostics
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
This computerized clinical decision support tool was implemented to improve screening for liver fibrosis in the primary care setting through targeted interruptive alert. It utilized common lab values to calculate the Fibrosis-4 (FIB-4) score to risk stratify patients. Midpoint analysis of the 6 months implementation revealed successful firing of the tool but poor response rate from providers, complicated by inappropriate dismissal selections.
Speaker(s):
James Miller, M.D.
Wake Forest University School of Medicine
Author(s):
James Miller, M.D. - Wake Forest University School of Medicine; Ted Xiao, MD - Emory University School of Medicine; Bradley Rowland; Richa Bundy, MPH - Wake Forest Baptist Health; Corey Obermiller, MAS Applied Statistics - Wake Forest School of Medicine; Lauren Witek, MStat - Advocate Health; Adam Moses, MHA, PMP, FAMIA - Wake Forest Baptist Medical Center; Sean Rudnick, MD - Wake Forest University School of Medicine; Ajay Dharod, MD, FACP, FAMIA - Advocate Health - Wake Forest University School of Medicine;
James
Miller,
M.D. - Wake Forest University School of Medicine
Beyond Automation: The Constraints of Clinical Decision Support in Addressing Pressure Injuries
Presentation Type: Poster - Regular
Poster Number: 114
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support and Care Pathways, Human Factors and Usability, Quality Informatics and Lean
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Pressure injuries are an international problem that persists despite decades of research and recommendations on prevention. This study aimed to increase adherence to pressure injury prevention by using clinical decision support (CDS) to improve repositioning documentation compliance. Utilizing a quasi-experimental design, the study observed an increase in turn orders per ICU admission and a reduction in late turns across both intervention and control units, although pressure injury rates remained unchanged. Nurse survey results showed mixed reactions, with most users finding the tasks easy and useful but some noting workflow integration challenges. These findings highlight that while CDS can streamline electronic health record processes and support standard care, they are not sufficient alone to drive behavior change or improve clinical outcomes.
Speaker(s):
Sarah Thompson, MSHIMI, BSN, RN
Children's Healthcare of Atlanta
Author(s):
Tenia White; Brittany Brennan, MSN, PNP-AC - CHOA; Trish Burdett, BSN - Children's Healthcare of Atlanta; Swaminathan Kandaswamy, PhD - Emory University School of Medicine; Julia Yarahuan, MD, MBI - Children's Healthcare of Atlanta/Emory University;
Presentation Type: Poster - Regular
Poster Number: 114
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support and Care Pathways, Human Factors and Usability, Quality Informatics and Lean
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Pressure injuries are an international problem that persists despite decades of research and recommendations on prevention. This study aimed to increase adherence to pressure injury prevention by using clinical decision support (CDS) to improve repositioning documentation compliance. Utilizing a quasi-experimental design, the study observed an increase in turn orders per ICU admission and a reduction in late turns across both intervention and control units, although pressure injury rates remained unchanged. Nurse survey results showed mixed reactions, with most users finding the tasks easy and useful but some noting workflow integration challenges. These findings highlight that while CDS can streamline electronic health record processes and support standard care, they are not sufficient alone to drive behavior change or improve clinical outcomes.
Speaker(s):
Sarah Thompson, MSHIMI, BSN, RN
Children's Healthcare of Atlanta
Author(s):
Tenia White; Brittany Brennan, MSN, PNP-AC - CHOA; Trish Burdett, BSN - Children's Healthcare of Atlanta; Swaminathan Kandaswamy, PhD - Emory University School of Medicine; Julia Yarahuan, MD, MBI - Children's Healthcare of Atlanta/Emory University;
Sarah
Thompson,
MSHIMI, BSN, RN - Children's Healthcare of Atlanta
Barriers and Facilitators to Adoption of AgileMD Clinical Pathways in the Walk-In Clinic Setting
Presentation Type: Poster - Student
Poster Number: 115
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support and Care Pathways, Innovation Partnerships, Implementation Science, and Learning Health Systems, Quality Informatics and Lean
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
AgileMD Clinical Pathways embed evidence-based algorithms in the EHR to standardize care. Despite implementing 13 pathways in Vanderbilt Walk-In Clinics, usage is low. We analyzed utilization data and interviewed providers to identify barriers and facilitators. Barriers include limited awareness, lack of trust, and preference for other resources; facilitators include interest in improving confidence in decisions. Findings will inform pathway improvements and adoption strategies.
Speaker(s):
Sarah Stern, MD
Vanderbilt University Medical Center
Author(s):
Uday Suresh, MS - Vanderbilt University Department of Biomedical Informatics; Kim Unertl, PhD - Vanderbilt University Medical Center;
Presentation Type: Poster - Student
Poster Number: 115
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support and Care Pathways, Innovation Partnerships, Implementation Science, and Learning Health Systems, Quality Informatics and Lean
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
AgileMD Clinical Pathways embed evidence-based algorithms in the EHR to standardize care. Despite implementing 13 pathways in Vanderbilt Walk-In Clinics, usage is low. We analyzed utilization data and interviewed providers to identify barriers and facilitators. Barriers include limited awareness, lack of trust, and preference for other resources; facilitators include interest in improving confidence in decisions. Findings will inform pathway improvements and adoption strategies.
Speaker(s):
Sarah Stern, MD
Vanderbilt University Medical Center
Author(s):
Uday Suresh, MS - Vanderbilt University Department of Biomedical Informatics; Kim Unertl, PhD - Vanderbilt University Medical Center;
Sarah
Stern,
MD - Vanderbilt University Medical Center
Lessons Learned from Implementation of Automated Endocrinology Consultation for Hospitalized Patients with Insulin Pumps
Presentation Type: Poster - Student
Poster Number: 116
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support and Care Pathways, Workforce Automation, Communication, and Workflow Efficiency, Quality Informatics and Lean, Outcomes Improvement and Equity, Change Management, Human Factors and Usability
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Patients that are admitted to the hospital with an insulin pump are at risk of unique complications, which necessitates the early consultation of endocrinology. We developed a system that utilizes natural language processing and existing workflow processes to identify these patients. Epic’s instant orders functionality is used as a failsafe to consult endocrinology if not already completed. Early results and feedback from monitoring the live implementation of this system have been resoundingly positive.
Speaker(s):
Sarah Stern, MD
Vanderbilt University Medical Center
Author(s):
Sarah Stern, MD - Vanderbilt University Medical Center; Marc Maldaver, MD - Vanderbilt University Medical Center; Dara Mize, MD, MS - 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, Workforce Automation, Communication, and Workflow Efficiency, Quality Informatics and Lean, Outcomes Improvement and Equity, Change Management, Human Factors and Usability
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Patients that are admitted to the hospital with an insulin pump are at risk of unique complications, which necessitates the early consultation of endocrinology. We developed a system that utilizes natural language processing and existing workflow processes to identify these patients. Epic’s instant orders functionality is used as a failsafe to consult endocrinology if not already completed. Early results and feedback from monitoring the live implementation of this system have been resoundingly positive.
Speaker(s):
Sarah Stern, MD
Vanderbilt University Medical Center
Author(s):
Sarah Stern, MD - Vanderbilt University Medical Center; Marc Maldaver, MD - Vanderbilt University Medical Center; Dara Mize, MD, MS - Vanderbilt University Medical Center;
Sarah
Stern,
MD - Vanderbilt University Medical Center
Real-Time Care Coordination for ED 30-Day Revisits Through Robotic Process Automation Triggered Secure Chat and Service-Led Governance
Presentation Type: Poster - Student
Poster Number: 117
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support and Care Pathways, Change Management, Workforce Automation, Communication, and Workflow Efficiency
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Thirty-day Emergency Department (ED) revisits are critical opportunities to prevent unnecessary admissions, yet most alerting systems are too late, routed to wrong recipients, and lack context needed to influence real-time decisions. We implemented a real-time ED revisit coordination model across three campuses using a robotic process automation trigger that sent a secure group message for patients re-presenting to the ED within 30 days of discharge. Messages included index discharge date, service, diagnosis, chief complaint, and instructions to coordinate care, with a 15-minute acknowledgment expectation. One campus added a Proactive Care Coordination Overlay, in which care management/social work monitored chats and proposed safe disposition plans. Pre-post analyses used Z-tests, Wilcoxon tests, logistic regression, and statistical control process charts.
Between October 19, 2023 and April 9, 2025, there were 27,592 thirty-day revisits. Engagement was high: 61.8% of chats received responses (median 6.1 minutes; 40.5% ≤15 minutes). Systemwide readmission rates were unchanged (49.4% vs 48.6%, p=0.149). Observation use increased from 7.0% to 8.0% (p=1.7e-3). At the Proactive Coordination campus, ED treat-and-release rose by 2.9% (p=0.024) and admissions fell by 2.8% (p=0.029). ED disposition time decreased from 345.8 to 323.8 minutes (p=0.004). Each additional message was associated with lower odds of admission (OR 0.98 per message, 95% CI 0.97–0.99, p<0.001). Readmissions showed improved weekly stability on SPC charts.
Secure chat alone was insufficient to shift outcomes, but paired with proactive care coordination produced meaningful operational improvements. This real-time coordination model demonstrates how informatics interventions can influence ED disposition when supported by accountable workflows.
Speaker(s):
Priyanka Solanki, MD
NYU
Author(s):
Priyanka Solanki, MD - NYU; William Small, MD, MBA - NYU Langone Health; Jaya Sondhi, BS, RN - NYU; Jonathan Austrian, MD - NYU Langone Health;
Presentation Type: Poster - Student
Poster Number: 117
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support and Care Pathways, Change Management, Workforce Automation, Communication, and Workflow Efficiency
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Thirty-day Emergency Department (ED) revisits are critical opportunities to prevent unnecessary admissions, yet most alerting systems are too late, routed to wrong recipients, and lack context needed to influence real-time decisions. We implemented a real-time ED revisit coordination model across three campuses using a robotic process automation trigger that sent a secure group message for patients re-presenting to the ED within 30 days of discharge. Messages included index discharge date, service, diagnosis, chief complaint, and instructions to coordinate care, with a 15-minute acknowledgment expectation. One campus added a Proactive Care Coordination Overlay, in which care management/social work monitored chats and proposed safe disposition plans. Pre-post analyses used Z-tests, Wilcoxon tests, logistic regression, and statistical control process charts.
Between October 19, 2023 and April 9, 2025, there were 27,592 thirty-day revisits. Engagement was high: 61.8% of chats received responses (median 6.1 minutes; 40.5% ≤15 minutes). Systemwide readmission rates were unchanged (49.4% vs 48.6%, p=0.149). Observation use increased from 7.0% to 8.0% (p=1.7e-3). At the Proactive Coordination campus, ED treat-and-release rose by 2.9% (p=0.024) and admissions fell by 2.8% (p=0.029). ED disposition time decreased from 345.8 to 323.8 minutes (p=0.004). Each additional message was associated with lower odds of admission (OR 0.98 per message, 95% CI 0.97–0.99, p<0.001). Readmissions showed improved weekly stability on SPC charts.
Secure chat alone was insufficient to shift outcomes, but paired with proactive care coordination produced meaningful operational improvements. This real-time coordination model demonstrates how informatics interventions can influence ED disposition when supported by accountable workflows.
Speaker(s):
Priyanka Solanki, MD
NYU
Author(s):
Priyanka Solanki, MD - NYU; William Small, MD, MBA - NYU Langone Health; Jaya Sondhi, BS, RN - NYU; Jonathan Austrian, MD - NYU Langone Health;
Priyanka
Solanki,
MD - NYU
Improving Emergency Department Triage Accuracy for Febrile Sickle Cell Disease Patients Using Targeted Informatics Tools
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, Human Factors and Usability
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Fever in sickle cell disease (SCD) requires rapid ED evaluation, making accurate triage identification essential. We implemented two EHR interventions to reduce misidentification: a disappearing navigator section and an interruptive alert. Before implementation, misidentification occurred twice monthly. In the 5 months since release, alerts fired 19 times for 13 encounters, all appropriately corrected or addressed, and no antibiotic delays have occurred since. These targeted informatics tools improved triage accuracy for a high-risk population.
Speaker(s):
Andrew Fowler, MD
Emory School of Medicine
Author(s):
Andrew Fowler, MD - Emory School of Medicine; Brett Slagh, BS - Children's Healthcare of Atlanta; Rebekah Carter, BSN, RN, CPEN - Children's Healthcare of Atlanta; Julia Yarahuan, MD, MBI - Children's Healthcare of Atlanta/Emory University;
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, Human Factors and Usability
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Fever in sickle cell disease (SCD) requires rapid ED evaluation, making accurate triage identification essential. We implemented two EHR interventions to reduce misidentification: a disappearing navigator section and an interruptive alert. Before implementation, misidentification occurred twice monthly. In the 5 months since release, alerts fired 19 times for 13 encounters, all appropriately corrected or addressed, and no antibiotic delays have occurred since. These targeted informatics tools improved triage accuracy for a high-risk population.
Speaker(s):
Andrew Fowler, MD
Emory School of Medicine
Author(s):
Andrew Fowler, MD - Emory School of Medicine; Brett Slagh, BS - Children's Healthcare of Atlanta; Rebekah Carter, BSN, RN, CPEN - Children's Healthcare of Atlanta; Julia Yarahuan, MD, MBI - Children's Healthcare of Atlanta/Emory University;
Andrew
Fowler,
MD - Emory School of Medicine
Indication-Based CDS Panels to Promote Evidence-Based Diagnostic Stewardship for Blood Cultures, Urine Cultures, and Respiratory Pathogen Testing
Presentation Type: Poster - Regular
Poster Number: 120
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support and Care Pathways, Outcomes Improvement and Equity, Clinical Decision Support and Care Pathways
Primary Track: Advancing Wellness for Providers and Community with Consideration of Human Factors
This initiative describes the implementation of indication-based Clinical Decision Support (CDS) panels to improve appropriateness of microbiologic testing for blood cultures, urine cultures, and respiratory pathogen panels (RPP) across clinical settings. Developed through multidisciplinary collaboration and aligned with evidence-based guidance, the interventions reduced unnecessary testing, including a 28% decrease in blood culture orders and a 37% reduction in RPP utilization. These changes supported more appropriate antibiotic use and generated over $500,000 reagent cost savings.
Speaker(s):
Joy Iocca, DNP
Penn Medicine
Author(s):
Christopher Ware, MD, FAAEM, CPHIMS - Penn Medicine Chester County Hospital; Mika Epps, MSN - Penn Medicine; Joy Iocca, DNP - Penn Medicine; Nikhil K Mull, MD - University of Pennsylvania Health System; Judith O'Donnell, MD - Penn Presbyterian Medical Center; Kathleen Degnan, MD - Hospital of the University of Pennsylvania; Laurel Glaser, MD - Hospital of the University of Pennsylvania; Kyle Rodino, PhD, D (ABMM) - Hospital of the University of Pennsylvania; Erika Flores, PhD - University of Pennsylvania Perelman School of Medicine;
Presentation Type: Poster - Regular
Poster Number: 120
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support and Care Pathways, Outcomes Improvement and Equity, Clinical Decision Support and Care Pathways
Primary Track: Advancing Wellness for Providers and Community with Consideration of Human Factors
This initiative describes the implementation of indication-based Clinical Decision Support (CDS) panels to improve appropriateness of microbiologic testing for blood cultures, urine cultures, and respiratory pathogen panels (RPP) across clinical settings. Developed through multidisciplinary collaboration and aligned with evidence-based guidance, the interventions reduced unnecessary testing, including a 28% decrease in blood culture orders and a 37% reduction in RPP utilization. These changes supported more appropriate antibiotic use and generated over $500,000 reagent cost savings.
Speaker(s):
Joy Iocca, DNP
Penn Medicine
Author(s):
Christopher Ware, MD, FAAEM, CPHIMS - Penn Medicine Chester County Hospital; Mika Epps, MSN - Penn Medicine; Joy Iocca, DNP - Penn Medicine; Nikhil K Mull, MD - University of Pennsylvania Health System; Judith O'Donnell, MD - Penn Presbyterian Medical Center; Kathleen Degnan, MD - Hospital of the University of Pennsylvania; Laurel Glaser, MD - Hospital of the University of Pennsylvania; Kyle Rodino, PhD, D (ABMM) - Hospital of the University of Pennsylvania; Erika Flores, PhD - University of Pennsylvania Perelman School of Medicine;
Joy
Iocca,
DNP - Penn Medicine
Burden Reduction Emergency Documentation: Development of the American College of Emergency Physicians’ BRED Measures
Presentation Type: Poster - Regular
Poster Number: 121
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinician Well-Being, Human Factors and Usability, Standards, Terminology, and Interoperability, TEFCA, FHIR
Primary Track: Advancing Wellness for Providers and Community with Consideration of Human Factors
Despite widespread work in evaluating documentation burden, alignment of measurement techniques is inconsistent and not emergency medicine specific. The American College of Emergency Medicine Health Information Technology Committee convened a consensus committee with the objective of studying and improving documentation burden. A team of seven expert emergency physician informaticians created twelve top-level measures and sixty-nine sub-measures. Measurement and reporting should allow for health systems to optimize workload to easy and measure ED EHR burden.
Speaker(s):
Jeffrey Nielson, MD, MS
Kettering Health
Author(s):
Jeffrey Nielson, MD, MS - Kettering Health; Colton Hood, Physician - GW Medical Faculty Associates; Christopher Alban, MD - Epic; Jonathan Siff, MD, MBA, FACEP, FAMIA - The MetroHealth System;
Presentation Type: Poster - Regular
Poster Number: 121
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinician Well-Being, Human Factors and Usability, Standards, Terminology, and Interoperability, TEFCA, FHIR
Primary Track: Advancing Wellness for Providers and Community with Consideration of Human Factors
Despite widespread work in evaluating documentation burden, alignment of measurement techniques is inconsistent and not emergency medicine specific. The American College of Emergency Medicine Health Information Technology Committee convened a consensus committee with the objective of studying and improving documentation burden. A team of seven expert emergency physician informaticians created twelve top-level measures and sixty-nine sub-measures. Measurement and reporting should allow for health systems to optimize workload to easy and measure ED EHR burden.
Speaker(s):
Jeffrey Nielson, MD, MS
Kettering Health
Author(s):
Jeffrey Nielson, MD, MS - Kettering Health; Colton Hood, Physician - GW Medical Faculty Associates; Christopher Alban, MD - Epic; Jonathan Siff, MD, MBA, FACEP, FAMIA - The MetroHealth System;
Jeffrey
Nielson,
MD, MS - Kettering Health
Characterization of Interprofessional Electronic Health Record-Based Secure Messaging Related to Oncology Inpatients
Presentation Type: Poster - Student
Poster Number: 122
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinician Well-Being, Health Data Science, Workforce Automation, Communication, and Workflow Efficiency
Primary Track: Big Data for Health
This retrospective cohort study analyzed secure messaging metadata for 2,480 oncology inpatient admissions across solid tumor, malignant hematology, and bone marrow transplant (BMT) services at three hospitals. There were 905,806 sent messages, with the three services each exceeding internal medicine messaging load by over 40%. To improve care quality, focus should shift to the disproportionately high messaging load on BMT Nurses and APPs and the limited involvement of primary oncologists on hematology services.
Speaker(s):
William Small, MD, MBA
NYU Langone Health
Author(s):
Daniel Wronski, BS - NYU Grossman Long Island School of Medicine; William Nguyen, BS - NYU Grossman Long Island School of Medicine; Marc Braunstein, MD PhD - NYU Grossman Long Island School of Medicine; Mario Lacoutre, MD - NYU Grossman Long Island School of Medicine; William Small, MD, MBA - NYU Langone Health;
Presentation Type: Poster - Student
Poster Number: 122
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinician Well-Being, Health Data Science, Workforce Automation, Communication, and Workflow Efficiency
Primary Track: Big Data for Health
This retrospective cohort study analyzed secure messaging metadata for 2,480 oncology inpatient admissions across solid tumor, malignant hematology, and bone marrow transplant (BMT) services at three hospitals. There were 905,806 sent messages, with the three services each exceeding internal medicine messaging load by over 40%. To improve care quality, focus should shift to the disproportionately high messaging load on BMT Nurses and APPs and the limited involvement of primary oncologists on hematology services.
Speaker(s):
William Small, MD, MBA
NYU Langone Health
Author(s):
Daniel Wronski, BS - NYU Grossman Long Island School of Medicine; William Nguyen, BS - NYU Grossman Long Island School of Medicine; Marc Braunstein, MD PhD - NYU Grossman Long Island School of Medicine; Mario Lacoutre, MD - NYU Grossman Long Island School of Medicine; William Small, MD, MBA - NYU Langone Health;
William
Small,
MD, MBA - NYU Langone Health
FAMES: Federated additive model using piecewise exponential survival data
Presentation Type: Poster - Regular
Poster Number: 123
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Data Privacy, Cybersecurity, Reliability, and Security, Analytical Artificial Intelligence: ML, Digital Pathology, Imaging AI, Predictive Analytics, Governance, Clinical Decision Support and Care Pathways, Precision Medicine, Multi-Omics, and Pharmacology Integration, Health Data Science, Analytics, Registries, and the Digital Command Center, Innovation Partnerships, Implementation Science, and Learning Health Systems, Standards, Terminology, and Interoperability, TEFCA, FHIR
Primary Track: Innovation through Industry, Public Health, Non-profit and Commercial Partnerships
We have developed a privacy-preserving penalized additive model to quantify the relationship between clinical features and time-to-event outcomes. This model estimates both smooth and fixed effect parameters using site-specific summary statistics, thereby obviating the need to pool individual patient data. Its numerical performance is comparable to that of pooled analysis.
Speaker(s):
Nazmul Islam, PhD
RefinedScience
Author(s):
Steven Bair, MD - Department of Medicine, University of Colorado Anschutz Campus, Aurora, Colorado, USA; Andrew Kent, MD, PhD - Department of Medicine, University of Colorado Anschutz Campus, Aurora, Colorado, USA; Grant Weller, PhD - RefinedScience;
Presentation Type: Poster - Regular
Poster Number: 123
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Data Privacy, Cybersecurity, Reliability, and Security, Analytical Artificial Intelligence: ML, Digital Pathology, Imaging AI, Predictive Analytics, Governance, Clinical Decision Support and Care Pathways, Precision Medicine, Multi-Omics, and Pharmacology Integration, Health Data Science, Analytics, Registries, and the Digital Command Center, Innovation Partnerships, Implementation Science, and Learning Health Systems, Standards, Terminology, and Interoperability, TEFCA, FHIR
Primary Track: Innovation through Industry, Public Health, Non-profit and Commercial Partnerships
We have developed a privacy-preserving penalized additive model to quantify the relationship between clinical features and time-to-event outcomes. This model estimates both smooth and fixed effect parameters using site-specific summary statistics, thereby obviating the need to pool individual patient data. Its numerical performance is comparable to that of pooled analysis.
Speaker(s):
Nazmul Islam, PhD
RefinedScience
Author(s):
Steven Bair, MD - Department of Medicine, University of Colorado Anschutz Campus, Aurora, Colorado, USA; Andrew Kent, MD, PhD - Department of Medicine, University of Colorado Anschutz Campus, Aurora, Colorado, USA; Grant Weller, PhD - RefinedScience;
Nazmul
Islam,
PhD - RefinedScience
Evolving Practices and Regulations in Healthcare Data De-Identification
Presentation Type: Poster - Regular
Poster Number: 124
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Data Privacy, Cybersecurity, Reliability, and Security, Change Management, Health Data Science
Primary Track: Big Data for Health
This review analyzed global developments in healthcare data de-identification between 2020–2025. A total of 435 records were screened, and 24 full-text articles included. Key topics included advances in privacy-preserving techniques (homomorphic encryption, secure multiparty computation) and tools, significant impacts from AI-driven de-identification and synthetic data development, increasing use of quantitative risk metrics, regulatory updates across multiple countries.
Speaker(s):
Olga Vovk, MSc
Tallinn University of Technology
Author(s):
Gunnar Piho, PhD - Tallinn University of Technology; Peeter Ross, PhD - Tallinn University of Technology;
Presentation Type: Poster - Regular
Poster Number: 124
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Data Privacy, Cybersecurity, Reliability, and Security, Change Management, Health Data Science
Primary Track: Big Data for Health
This review analyzed global developments in healthcare data de-identification between 2020–2025. A total of 435 records were screened, and 24 full-text articles included. Key topics included advances in privacy-preserving techniques (homomorphic encryption, secure multiparty computation) and tools, significant impacts from AI-driven de-identification and synthetic data development, increasing use of quantitative risk metrics, regulatory updates across multiple countries.
Speaker(s):
Olga Vovk, MSc
Tallinn University of Technology
Author(s):
Gunnar Piho, PhD - Tallinn University of Technology; Peeter Ross, PhD - Tallinn University of Technology;
Olga
Vovk,
MSc - Tallinn University of Technology
Program Director Decisions on Ambient AI Documentation for Medical Trainees
Presentation Type: Poster - Student
Poster Number: 126
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Education and Training, Generative AI in Clinical Workflow: Ambient Listening, Chart Summarization, Automated Response with LLM, Leadership and Strategy, Clinician Well-Being, Innovation Partnerships, Implementation Science, and Learning Health Systems
Primary Track: Advancing Wellness for Providers and Community with Consideration of Human Factors
AI scribes are reshaping documentation and training. We surveyed Stanford Medicine program directors to characterize decisions on trainee use. Thirty-three percent of programs have policies on AI scribe use. Among respondents, 40% of residency and 85% fellowship programs permitted use across all trainee levels; 81% of all programs permitted unrestricted use, with assessment/plan being the most restricted SmartSection. Free-text themes balances enthusiasm with concerns about implications on medical training, informing AI integration in GME.
Speaker(s):
Bethel Mieso, MD
Stanford
Author(s):
Bethel Mieso, MD - Stanford; April Liang, MD - Stanford University; Stephen Ma, MD, PhD - Stanford University School of Medicine; Shreya Shah, MD - Stanford University; Kirsten Murtagah, MA - Stanford University School of Medicine; Jie Li, PhD - Stanford Health Care; Caroline Hsia, MS - Stanford Health Care; Clarissa Delahaie, LCA - Stanford Health Care; Nat Taylor, MA - Stanford Medicine Children's Health; Rosalia Sandoval, MA, PMP - Stanford Medicine Children's Health; Christopher Sharp, MD - Stanford University School of Medicine; Natalie Pageler, MD, MEd - Stanford University; Rebecca Blankenburg, MD, MPH - Stanford University School of Medicine; Lindsay Stevens, MD - Stanford University; Patricia Garcia, MD - Stanford University, School of Medicine;
Presentation Type: Poster - Student
Poster Number: 126
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Education and Training, Generative AI in Clinical Workflow: Ambient Listening, Chart Summarization, Automated Response with LLM, Leadership and Strategy, Clinician Well-Being, Innovation Partnerships, Implementation Science, and Learning Health Systems
Primary Track: Advancing Wellness for Providers and Community with Consideration of Human Factors
AI scribes are reshaping documentation and training. We surveyed Stanford Medicine program directors to characterize decisions on trainee use. Thirty-three percent of programs have policies on AI scribe use. Among respondents, 40% of residency and 85% fellowship programs permitted use across all trainee levels; 81% of all programs permitted unrestricted use, with assessment/plan being the most restricted SmartSection. Free-text themes balances enthusiasm with concerns about implications on medical training, informing AI integration in GME.
Speaker(s):
Bethel Mieso, MD
Stanford
Author(s):
Bethel Mieso, MD - Stanford; April Liang, MD - Stanford University; Stephen Ma, MD, PhD - Stanford University School of Medicine; Shreya Shah, MD - Stanford University; Kirsten Murtagah, MA - Stanford University School of Medicine; Jie Li, PhD - Stanford Health Care; Caroline Hsia, MS - Stanford Health Care; Clarissa Delahaie, LCA - Stanford Health Care; Nat Taylor, MA - Stanford Medicine Children's Health; Rosalia Sandoval, MA, PMP - Stanford Medicine Children's Health; Christopher Sharp, MD - Stanford University School of Medicine; Natalie Pageler, MD, MEd - Stanford University; Rebecca Blankenburg, MD, MPH - Stanford University School of Medicine; Lindsay Stevens, MD - Stanford University; Patricia Garcia, MD - Stanford University, School of Medicine;
Bethel
Mieso,
MD - Stanford
A Secure Large Language Model Pipeline for Identifying Real-Time High- Risk Conditions in the ED
Presentation Type: Poster Invite - 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, Innovation Partnerships, Implementation Science, and Learning Health Systems, Clinical Decision Support and Care Pathways, Workforce Automation, Communication, and Workflow Efficiency
Primary Track: Big Data for Health
We describe a secure, event-driven large language model pipeline that identifies high-risk emergency department conditions in real time using structured and unstructured EHR data. The system extracts FHIR outputs hourly, generates clinical summaries, performs automated risk stratification, and routes asynchronous secure chat notifications to treating teams. We outline the design, testing, and early performance metrics of a model looking at necrotizing fasciitis, demonstrating feasibility and safety of LLM-augmented clinical decision support
Speaker(s):
Vincent Xiao, MD
NYU Langone Health
Author(s):
Vincent Xiao, MD - NYU Langone Health; Gregory Simon, MD - NYU Langone Health; Joseph Rosenthal, BSE - NYU Langone Health; Nusrat Jahan, MD - NYU; Robert Femia, MD, MBA - NYU Langone Health; Catherine Jamin, MD - NYU Langone Health; Paul Testa, MD, JD, MPH - NYU Langone Health; Leora Horwitz, MD - NYU Langone; Jung Kim, PhD - NYU Langone Health; Nicholas Genes, MD, PhD - NYU Grossman School of Medicine; Yindalon Aphinyanaphongs, MD - NYU Langone Health;
Presentation Type: Poster Invite - 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, Innovation Partnerships, Implementation Science, and Learning Health Systems, Clinical Decision Support and Care Pathways, Workforce Automation, Communication, and Workflow Efficiency
Primary Track: Big Data for Health
We describe a secure, event-driven large language model pipeline that identifies high-risk emergency department conditions in real time using structured and unstructured EHR data. The system extracts FHIR outputs hourly, generates clinical summaries, performs automated risk stratification, and routes asynchronous secure chat notifications to treating teams. We outline the design, testing, and early performance metrics of a model looking at necrotizing fasciitis, demonstrating feasibility and safety of LLM-augmented clinical decision support
Speaker(s):
Vincent Xiao, MD
NYU Langone Health
Author(s):
Vincent Xiao, MD - NYU Langone Health; Gregory Simon, MD - NYU Langone Health; Joseph Rosenthal, BSE - NYU Langone Health; Nusrat Jahan, MD - NYU; Robert Femia, MD, MBA - NYU Langone Health; Catherine Jamin, MD - NYU Langone Health; Paul Testa, MD, JD, MPH - NYU Langone Health; Leora Horwitz, MD - NYU Langone; Jung Kim, PhD - NYU Langone Health; Nicholas Genes, MD, PhD - NYU Grossman School of Medicine; Yindalon Aphinyanaphongs, MD - NYU Langone Health;
Vincent
Xiao,
MD - NYU Langone Health
Impact of ambient AI scribe on resident physician workflow in primary care clinic
Presentation Type: Poster - Student
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, Clinician Well-Being
Primary Track: Advancing Wellness for Providers and Community with Consideration of Human Factors
Recent studies have shown that AI scribes can improve documentation burden in attending physicians; however, its impact has not been studied widely on trainees. After implementation of an ambient AI scribe tool in clinic, resident physicians spent less time in EHR overall, with a significant decrease in documentation time. This highlights the need for dissemination of AI tools for residents especially in light of significant burnout experienced during medical training.
Speaker(s):
James Lim, MD, MAS
The Ohio State University Wexner Medical Center
Author(s):
James Lim, MD, MAS - The Ohio State University Wexner Medical Center; Timothy Frommeyer, MD - The Ohio State University Wexner Medical Center; Garrett Brittain, MD - The Ohio State University Wexner Medical Center; Rich Thompson, PhD - The Ohio State University; Harrison Jackson, MD - The Ohio State University Wexner Medical Center; Philip Chang, MD - The Ohio State University Wexner Medical Center; Christopher Chiu, MD - The Ohio State University Wexner Medical Center;
Presentation Type: Poster - Student
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, Clinician Well-Being
Primary Track: Advancing Wellness for Providers and Community with Consideration of Human Factors
Recent studies have shown that AI scribes can improve documentation burden in attending physicians; however, its impact has not been studied widely on trainees. After implementation of an ambient AI scribe tool in clinic, resident physicians spent less time in EHR overall, with a significant decrease in documentation time. This highlights the need for dissemination of AI tools for residents especially in light of significant burnout experienced during medical training.
Speaker(s):
James Lim, MD, MAS
The Ohio State University Wexner Medical Center
Author(s):
James Lim, MD, MAS - The Ohio State University Wexner Medical Center; Timothy Frommeyer, MD - The Ohio State University Wexner Medical Center; Garrett Brittain, MD - The Ohio State University Wexner Medical Center; Rich Thompson, PhD - The Ohio State University; Harrison Jackson, MD - The Ohio State University Wexner Medical Center; Philip Chang, MD - The Ohio State University Wexner Medical Center; Christopher Chiu, MD - The Ohio State University Wexner Medical Center;
James
Lim,
MD, MAS - The Ohio State University Wexner Medical Center
Assessing Provider Press Ganey Scores Pre‑ and Post‑Ambient Artificial Intelligence
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, Outcomes Improvement and Equity, Workforce Automation, Communication, and Workflow Efficiency
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Ambient artificial intelligence (AI) transcription tools aim to decrease note-taking burden on clinicians, but can they also improve the patient experience? We compared Press Ganey Care Provider ratings before and after Ambient AI implementation. Paired Wilcoxon rank tests showed nonsignificant improvements in overall ratings across five patient‑experience items. In primary care settings specifically, these improvements in Care Provider ratings were statistically significant, suggesting that Ambient AI may enhance the patient experience in these clinical settings.
Speaker(s):
Carlie Hoffman, DO
University of Pittsburgh Medical Center
Author(s):
Carlie Hoffman, DO - University of Pittsburgh Medical Center; Janel Hanmer, MD, PhD - University of Pittsburgh Medical Center; Gary Fischer, MD FACP - University of Pittsburgh; Michael Curren - University of Pittsburgh Medical Center; Jonathan Arnold, MD, MS, MSE - UPMC; Jennifer Berliner, MD, MBA - Northwell Health; Lauren Giugale, MD - UPMC;
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, Outcomes Improvement and Equity, Workforce Automation, Communication, and Workflow Efficiency
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Ambient artificial intelligence (AI) transcription tools aim to decrease note-taking burden on clinicians, but can they also improve the patient experience? We compared Press Ganey Care Provider ratings before and after Ambient AI implementation. Paired Wilcoxon rank tests showed nonsignificant improvements in overall ratings across five patient‑experience items. In primary care settings specifically, these improvements in Care Provider ratings were statistically significant, suggesting that Ambient AI may enhance the patient experience in these clinical settings.
Speaker(s):
Carlie Hoffman, DO
University of Pittsburgh Medical Center
Author(s):
Carlie Hoffman, DO - University of Pittsburgh Medical Center; Janel Hanmer, MD, PhD - University of Pittsburgh Medical Center; Gary Fischer, MD FACP - University of Pittsburgh; Michael Curren - University of Pittsburgh Medical Center; Jonathan Arnold, MD, MS, MSE - UPMC; Jennifer Berliner, MD, MBA - Northwell Health; Lauren Giugale, MD - UPMC;
Carlie
Hoffman,
DO - University of Pittsburgh Medical Center
SAFE-Vet: Signalment-Conditioned Alignment and Clinical Safety Benchmarking for Veterinary Edge AI
Presentation Type: Poster - Regular
Poster Number: 131
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, Ethics
Primary Track: Big Data for Health
We present a pilot ablation study identifying a critical safety vulnerability in edge-deployed veterinary AI. Comparing "Blind" vs. "Signalment-Conditioned" small language models, we found that partial context (Species-Only tags) paradoxically triggered defensive mode collapse, rendering the model safe but clinically useless. Only multi-axis conditioning (Species+Breed+Age) resolved these contradictions. This poster highlights the non-linear relationship between data granularity and utility in resource-constrained clinical environments.
Speaker(s):
David Brundage, PhD, RHIA, CHDA
University of Wisconsin School of Veterinary Medicine
Author(s):
David Brundage, PhD, RHIA, CHDA - University of Wisconsin School of Veterinary Medicine;
Presentation Type: Poster - Regular
Poster Number: 131
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, Ethics
Primary Track: Big Data for Health
We present a pilot ablation study identifying a critical safety vulnerability in edge-deployed veterinary AI. Comparing "Blind" vs. "Signalment-Conditioned" small language models, we found that partial context (Species-Only tags) paradoxically triggered defensive mode collapse, rendering the model safe but clinically useless. Only multi-axis conditioning (Species+Breed+Age) resolved these contradictions. This poster highlights the non-linear relationship between data granularity and utility in resource-constrained clinical environments.
Speaker(s):
David Brundage, PhD, RHIA, CHDA
University of Wisconsin School of Veterinary Medicine
Author(s):
David Brundage, PhD, RHIA, CHDA - University of Wisconsin School of Veterinary Medicine;
David
Brundage,
PhD, RHIA, CHDA - University of Wisconsin School of Veterinary Medicine
Generative AI Discharge Summaries in Pediatric Hospital Medicine: Lessons from a Mixed-Methods Early Implementation Study
Presentation Type: Poster Invite - Regular
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, Clinician Well-Being, Workforce Automation, Communication, and Workflow Efficiency
Primary Track: Advancing Wellness for Providers and Community with Consideration of Human Factors
We evaluated early implementation of an EHR-integrated large language model for generating pediatric discharge summaries in a mixed-methods pilot. Adoption was modest, with clinicians accessing the tool more often than incorporating generated text, particularly for longer hospitalizations. Quantitative and qualitative findings highlighted usefulness for complex or prolonged stays, alongside workflow and content-quality barriers. Results demonstrate both the promise and challenges of integrating generative AI documentation tools into pediatric inpatient care.
Speaker(s):
Averi Wilson, MD
University of Texas Southwestern
Author(s):
Averi Wilson, MD - University of Texas Southwestern; Pooja Dave, MD - University of Texas Southwestern; Shravan Vallala, MD - University of Texas Southwestern; Andrew Bain, MD - University of Texas Southwestern Medical Center; Javier Lasa, M.D. - Children's Health; Philip Bernard, M.D. - children's health; Vineeta Mittal, MD - University of Texas Southwestern;
Presentation Type: Poster Invite - Regular
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, Clinician Well-Being, Workforce Automation, Communication, and Workflow Efficiency
Primary Track: Advancing Wellness for Providers and Community with Consideration of Human Factors
We evaluated early implementation of an EHR-integrated large language model for generating pediatric discharge summaries in a mixed-methods pilot. Adoption was modest, with clinicians accessing the tool more often than incorporating generated text, particularly for longer hospitalizations. Quantitative and qualitative findings highlighted usefulness for complex or prolonged stays, alongside workflow and content-quality barriers. Results demonstrate both the promise and challenges of integrating generative AI documentation tools into pediatric inpatient care.
Speaker(s):
Averi Wilson, MD
University of Texas Southwestern
Author(s):
Averi Wilson, MD - University of Texas Southwestern; Pooja Dave, MD - University of Texas Southwestern; Shravan Vallala, MD - University of Texas Southwestern; Andrew Bain, MD - University of Texas Southwestern Medical Center; Javier Lasa, M.D. - Children's Health; Philip Bernard, M.D. - children's health; Vineeta Mittal, MD - University of Texas Southwestern;
Averi
Wilson,
MD - University of Texas Southwestern
MedAgentBrief: Development and Prospective Real-World Evaluation of an LLM-Based Hospital Course Summarizer
Presentation Type: Poster Invite - Regular
Click to View Presentation
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, Clinician Well-Being, Workforce Automation, Communication, and Workflow Efficiency
Primary Track: Advancing Wellness for Providers and Community with Consideration of Human Factors
Accurate, timely discharge summaries improve transitions of care but composing the hospital course (HC) is time-consuming. We implemented MedAgentBrief, an LLM-based HC generator, on a single academic inpatient unit and prospectively evaluated safety and utility. Across 10 weeks and 384 discharges, physician users used the MedAgentBrief HC with edits more than half the time and rated them safe overall. Impact on time in discharge note and time to chart closure was mixed, but user-reported burnout scores decreased significantly. These results demonstrate MedAgentBrief's ability to provide safe and useful summaries and highlight its potential to mitigate clinician documentation burden and burnout.
Speaker(s):
April Liang, MD
Stanford University
Author(s):
April Liang, MD - Stanford University; Francois Grolleau, MD, PhD - Stanford University; Timothy Keyes, PhD - Stanford Health Care; Stephen Ma, MD, PhD - Stanford University School of Medicine; Thomas Lew, MD - Stanford University School of Medicine; Tridu Huynh, MD - Stanford University School of Medicine; Natasha Steele, MD, MPH - Stanford University School of Medicine; Nerissa Ambers, MPH - Stanford Health Care; Nikesh Kotecha; Emily Alsentzer, MS, PhD - Brigham and Women's Hospital; Nigam Shah, MBBS - Stanford University; Jason Hom, MD - Stanford University School of Medicine; Kevin Schulman; Jonathan Chen, MD, PhD - Stanford University Hospital;
Presentation Type: Poster Invite - Regular
Click to View Presentation
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, Clinician Well-Being, Workforce Automation, Communication, and Workflow Efficiency
Primary Track: Advancing Wellness for Providers and Community with Consideration of Human Factors
Accurate, timely discharge summaries improve transitions of care but composing the hospital course (HC) is time-consuming. We implemented MedAgentBrief, an LLM-based HC generator, on a single academic inpatient unit and prospectively evaluated safety and utility. Across 10 weeks and 384 discharges, physician users used the MedAgentBrief HC with edits more than half the time and rated them safe overall. Impact on time in discharge note and time to chart closure was mixed, but user-reported burnout scores decreased significantly. These results demonstrate MedAgentBrief's ability to provide safe and useful summaries and highlight its potential to mitigate clinician documentation burden and burnout.
Speaker(s):
April Liang, MD
Stanford University
Author(s):
April Liang, MD - Stanford University; Francois Grolleau, MD, PhD - Stanford University; Timothy Keyes, PhD - Stanford Health Care; Stephen Ma, MD, PhD - Stanford University School of Medicine; Thomas Lew, MD - Stanford University School of Medicine; Tridu Huynh, MD - Stanford University School of Medicine; Natasha Steele, MD, MPH - Stanford University School of Medicine; Nerissa Ambers, MPH - Stanford Health Care; Nikesh Kotecha; Emily Alsentzer, MS, PhD - Brigham and Women's Hospital; Nigam Shah, MBBS - Stanford University; Jason Hom, MD - Stanford University School of Medicine; Kevin Schulman; Jonathan Chen, MD, PhD - Stanford University Hospital;
April
Liang,
MD - Stanford University
Assessing the Internal Consistency of Ambient Scribe Platforms using Simulated Ambulatory Encounters
Presentation Type: Poster - Student
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, Workforce Automation, Communication, and Workflow Efficiency, Human Factors and Usability
Primary Track: Innovation through Industry, Public Health, Non-profit and Commercial Partnerships
This study evaluated the consistency of three ambient digital scribe (ADS) platforms using 14 simulated ambulatory encounters. Two subscription-based platforms demonstrated more consistent performance across replicates, while the free platform exhibited variability in potential clinical harm and lower consistency in correctly reported elements across replicates. These findings highlight concerns around the internal reliability and consistency of ADS platforms, reinforcing the need for performance evaluation and clinician oversight.
Speaker(s):
Taylor Anderson, MD
Stanford, Oregon Health & Science University
Author(s):
Vishnu Mohan, MD, MBI, FACP, FAMIA - Oregon Health & Science University; David Dorr, MD, MS, FACMI, FAMIA, FIAHSI - Oregon Health & Science University; Jeffrey Gold, MD - Oregon Health & Science University;
Presentation Type: Poster - Student
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, Workforce Automation, Communication, and Workflow Efficiency, Human Factors and Usability
Primary Track: Innovation through Industry, Public Health, Non-profit and Commercial Partnerships
This study evaluated the consistency of three ambient digital scribe (ADS) platforms using 14 simulated ambulatory encounters. Two subscription-based platforms demonstrated more consistent performance across replicates, while the free platform exhibited variability in potential clinical harm and lower consistency in correctly reported elements across replicates. These findings highlight concerns around the internal reliability and consistency of ADS platforms, reinforcing the need for performance evaluation and clinician oversight.
Speaker(s):
Taylor Anderson, MD
Stanford, Oregon Health & Science University
Author(s):
Vishnu Mohan, MD, MBI, FACP, FAMIA - Oregon Health & Science University; David Dorr, MD, MS, FACMI, FAMIA, FIAHSI - Oregon Health & Science University; Jeffrey Gold, MD - Oregon Health & Science University;
Taylor
Anderson,
MD - Stanford, Oregon Health & Science University
Enhancing Antibiotic Prescription Safety with Natural Language Processing and GPT
Presentation Type: Poster Invite - Regular
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, Clinical Decision Support and Care Pathways, Health Data Science
Primary Track: Big Data for Health
Discrepancies between documented clinical intent and electronic medication orders can cause significant patient harm, particularly in cases involving high-risk antibiotics. To address this challenge, we developed a GPT-enhanced sequential model that uses natural language processing (NLP) to detect mismatches between clinical notes and electronic orders, focusing on three high-risk antibiotics: ampicillin, meropenem, and isoniazid. The model identifies text snippets suggesting an intended prescription and employs GPT-based in-context learning to differentiate actual prescribing intent from conditional or non-current references. When discrepancies are detected upon note signing, an interruptive alert prompts providers to take corrective action.
Silent validation conducted over three months demonstrated strong model reliability, and live implementation began across five hospitals in July 2024. Over two months, the system parsed 10,000 clinical notes daily, identifying 3,658 notes with prescribing intent and generating 10 alerts with a positive predictive value (PPV) of 90%. Alerts resulted in timely corrective action in 7 out of 9 true discrepancy cases, ensuring necessary antibiotics were ordered. Two alerts prompted revisions to clinical documentation, improving record accuracy. A single false positive highlighted opportunities for refinement.
This implementation exemplifies GPT's potential to enhance clinical decision support by addressing gaps between documentation and practice. By improving antibiotic prescribing safety, the model directly impacts patient care, minimizing treatment delays and risks of antimicrobial resistance. Future efforts will focus on reducing false positives, expanding coverage to additional high-risk medications, and addressing systemic causes of missed orders through broader analysis.
Speaker(s):
Paawan Punjabi, MD, MSc
New York University School of Medicine/NYU Langone Health System
Author(s):
Paawan Punjabi, MD, MSc - New York University School of Medicine/NYU Langone Health System; Walter Wang, MSc - NYU Langone Health; Dana Mazo, MD MSc - NYU Langone Health; Jonathan Austrian, MD - NYU Langone Health; Fritz Francois, MD, MSc - NYU Langone Health; Yin Aphinyanaphongs;
Presentation Type: Poster Invite - Regular
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, Clinical Decision Support and Care Pathways, Health Data Science
Primary Track: Big Data for Health
Discrepancies between documented clinical intent and electronic medication orders can cause significant patient harm, particularly in cases involving high-risk antibiotics. To address this challenge, we developed a GPT-enhanced sequential model that uses natural language processing (NLP) to detect mismatches between clinical notes and electronic orders, focusing on three high-risk antibiotics: ampicillin, meropenem, and isoniazid. The model identifies text snippets suggesting an intended prescription and employs GPT-based in-context learning to differentiate actual prescribing intent from conditional or non-current references. When discrepancies are detected upon note signing, an interruptive alert prompts providers to take corrective action.
Silent validation conducted over three months demonstrated strong model reliability, and live implementation began across five hospitals in July 2024. Over two months, the system parsed 10,000 clinical notes daily, identifying 3,658 notes with prescribing intent and generating 10 alerts with a positive predictive value (PPV) of 90%. Alerts resulted in timely corrective action in 7 out of 9 true discrepancy cases, ensuring necessary antibiotics were ordered. Two alerts prompted revisions to clinical documentation, improving record accuracy. A single false positive highlighted opportunities for refinement.
This implementation exemplifies GPT's potential to enhance clinical decision support by addressing gaps between documentation and practice. By improving antibiotic prescribing safety, the model directly impacts patient care, minimizing treatment delays and risks of antimicrobial resistance. Future efforts will focus on reducing false positives, expanding coverage to additional high-risk medications, and addressing systemic causes of missed orders through broader analysis.
Speaker(s):
Paawan Punjabi, MD, MSc
New York University School of Medicine/NYU Langone Health System
Author(s):
Paawan Punjabi, MD, MSc - New York University School of Medicine/NYU Langone Health System; Walter Wang, MSc - NYU Langone Health; Dana Mazo, MD MSc - NYU Langone Health; Jonathan Austrian, MD - NYU Langone Health; Fritz Francois, MD, MSc - NYU Langone Health; Yin Aphinyanaphongs;
Paawan
Punjabi,
MD, MSc - New York University School of Medicine/NYU Langone Health System
From Notes to Conversations: Evaluating Automated Detection of Biased Clinical Communication
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, Health Data Science
Primary Track: Big Data for Health
To develop automated methods of assessment for patient-centered communication, we compare TRACE, a sentiment-based NLP tool for bias detection, on MIMIC clinical notes and simulated primary care conversations. We found that TRACE detects consistently higher positive and negative bias scores in notes across five sentiment dimensions, with status and ability language showing the largest differences. Findings will inform ambient clinical computing analytics, highlighting methodological challenges and opportunities in harmonizing feedback for patient-centered care.
Speaker(s):
Rohith Palli, MD, PhD
University of Washington
Author(s):
Andrea Hartzler, PhD FACMI - University of Washington; Rohith Palli, MD, PhD - University of Washington; Barbara Lam, MD - University of Washington; Aishwarya R, Phd - UW; Patrick Wedgeworth, MD, MISM - University of Washington;
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, Health Data Science
Primary Track: Big Data for Health
To develop automated methods of assessment for patient-centered communication, we compare TRACE, a sentiment-based NLP tool for bias detection, on MIMIC clinical notes and simulated primary care conversations. We found that TRACE detects consistently higher positive and negative bias scores in notes across five sentiment dimensions, with status and ability language showing the largest differences. Findings will inform ambient clinical computing analytics, highlighting methodological challenges and opportunities in harmonizing feedback for patient-centered care.
Speaker(s):
Rohith Palli, MD, PhD
University of Washington
Author(s):
Andrea Hartzler, PhD FACMI - University of Washington; Rohith Palli, MD, PhD - University of Washington; Barbara Lam, MD - University of Washington; Aishwarya R, Phd - UW; Patrick Wedgeworth, MD, MISM - University of Washington;
Rohith
Palli,
MD, PhD - University of Washington
LLM-Assisted Analysis of Clinician Feedback: A Mixed-Methods Framework for Evaluating Clinical AI Tools
Presentation Type: Poster Invite - 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, Analytical Artificial Intelligence: ML, Digital Pathology, Imaging AI, Predictive Analytics, Governance, Health Data Science, Human Factors and Usability, Quality Informatics and Lean
Primary Track: Advancing Wellness for Providers and Community with Consideration of Human Factors
We developed a mixed-methods framework for evaluating EHR-integrated AI tools, combining structured ratings with LLM-assisted thematic analysis. In an inpatient AI-generated chart summary pilot, 26 clinicians generated 783 summaries and submitted 172 reviews, of which 62% were rated helpful without errors. Our custom LLM pipeline segmented free-text feedback into 331 discrete units to quantify strengths and safety gaps. This scalable approach supports responsible AI governance by transforming narrative feedback into actionable metrics.
Speaker(s):
Jonathan Colston, MD
Stanford University
Author(s):
Jonathan Colston, MD - Stanford University; Stephen Ma, MD, PhD - Stanford University School of Medicine; April Liang, MD - Stanford University; Olivia Aparicio- Kratz, MPH, MSN, RN - Stanford Medicine; Matthew Eisenberg MD FAMIA, MD - Stanford Health Care/Stanford Medicine;
Presentation Type: Poster Invite - 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, Analytical Artificial Intelligence: ML, Digital Pathology, Imaging AI, Predictive Analytics, Governance, Health Data Science, Human Factors and Usability, Quality Informatics and Lean
Primary Track: Advancing Wellness for Providers and Community with Consideration of Human Factors
We developed a mixed-methods framework for evaluating EHR-integrated AI tools, combining structured ratings with LLM-assisted thematic analysis. In an inpatient AI-generated chart summary pilot, 26 clinicians generated 783 summaries and submitted 172 reviews, of which 62% were rated helpful without errors. Our custom LLM pipeline segmented free-text feedback into 331 discrete units to quantify strengths and safety gaps. This scalable approach supports responsible AI governance by transforming narrative feedback into actionable metrics.
Speaker(s):
Jonathan Colston, MD
Stanford University
Author(s):
Jonathan Colston, MD - Stanford University; Stephen Ma, MD, PhD - Stanford University School of Medicine; April Liang, MD - Stanford University; Olivia Aparicio- Kratz, MPH, MSN, RN - Stanford Medicine; Matthew Eisenberg MD FAMIA, MD - Stanford Health Care/Stanford Medicine;
Jonathan
Colston,
MD - Stanford University
Documentation Burden and Burnout in the Emergency Department: Baseline Results Before AI Implementation
Presentation Type: Poster - Regular
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, Clinician Well-Being, Human Factors and Usability
Primary Track: Big Data for Health
Provider burnout poses a significant threat to emergency medicine workforce sustainability, with rates as high as 71.4% among emergency physicians. Documentation burden and cognitive workload are believed to contribute substantially to burnout. This study characterizes baseline associations between burnout and documentation-related workflow factors in emergency department providers prior to implementing an AI ambient clinical documentation tool aimed at reducing these burdens.
Speaker(s):
Jessica Schumann, DO
University of Kansas Medical Center
Author(s):
Jessica Schumann, DO - University of Kansas Medical Center; Niaman Nazir, MD, MPH; Timothy Smith, MD - University of Kansas School of Medicine;
Presentation Type: Poster - Regular
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, Clinician Well-Being, Human Factors and Usability
Primary Track: Big Data for Health
Provider burnout poses a significant threat to emergency medicine workforce sustainability, with rates as high as 71.4% among emergency physicians. Documentation burden and cognitive workload are believed to contribute substantially to burnout. This study characterizes baseline associations between burnout and documentation-related workflow factors in emergency department providers prior to implementing an AI ambient clinical documentation tool aimed at reducing these burdens.
Speaker(s):
Jessica Schumann, DO
University of Kansas Medical Center
Author(s):
Jessica Schumann, DO - University of Kansas Medical Center; Niaman Nazir, MD, MPH; Timothy Smith, MD - University of Kansas School of Medicine;
Jessica
Schumann,
DO - University of Kansas Medical Center
Operationalizing Simulation-Based Human Factors Testing to Mitigate Automation Bias in Clinical Generative AI
Presentation Type: Poster Invite - 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, Human Factors and Usability, Leadership and Strategy
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Generative AI can reduce cognitive burden and improve workflows, but automation bias creates measurable safety risk when clinicians overweight AI output despite conflicting clinical evidence. We present a simulation-based evaluation strategy to detect and mitigate automation bias pre-deployment, with simulation-based testing underway through a partnership with the Mayo Clinic Multidisciplinary Simulation Center. We present (1) a literature-informed factor set organized across context, interface cognition, and organizational safeguards, and (2) a multidisciplinary simulation testing guide with defined roles, methods, and deliverables to support implementation and governance at scale across connected health workflows.
Speaker(s):
Lu Zheng, Ph.D., M.S.
Mayo Clinic
Author(s):
Lu Zheng, Ph.D., M.S. - Mayo Clinic; Joshua Ohde, PhD - Mayo Clinic; Young Juhn, M.D., M.P.H. - Mayo Clinic; Momin Malik, PhD in Societal Computing - Mayo Clinic; Deepak Sharma, MS - Mayo Clinic; Lauren Rost, PhD - Mayo Clinic; Chung Wi, MD - Mayo Clinic;
Presentation Type: Poster Invite - 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, Human Factors and Usability, Leadership and Strategy
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Generative AI can reduce cognitive burden and improve workflows, but automation bias creates measurable safety risk when clinicians overweight AI output despite conflicting clinical evidence. We present a simulation-based evaluation strategy to detect and mitigate automation bias pre-deployment, with simulation-based testing underway through a partnership with the Mayo Clinic Multidisciplinary Simulation Center. We present (1) a literature-informed factor set organized across context, interface cognition, and organizational safeguards, and (2) a multidisciplinary simulation testing guide with defined roles, methods, and deliverables to support implementation and governance at scale across connected health workflows.
Speaker(s):
Lu Zheng, Ph.D., M.S.
Mayo Clinic
Author(s):
Lu Zheng, Ph.D., M.S. - Mayo Clinic; Joshua Ohde, PhD - Mayo Clinic; Young Juhn, M.D., M.P.H. - Mayo Clinic; Momin Malik, PhD in Societal Computing - Mayo Clinic; Deepak Sharma, MS - Mayo Clinic; Lauren Rost, PhD - Mayo Clinic; Chung Wi, MD - Mayo Clinic;
Lu
Zheng,
Ph.D., M.S. - Mayo Clinic
Understanding Psychiatric Comorbidity Trajectories Surrounding Treatment-Resistant Depression Onset Using Sequence Analysis
Presentation Type: Poster - Regular
Poster Number: 142
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Health Data Science, Analytics, Registries, and the Digital Command Center, Precision Medicine, Multi-Omics, and Pharmacology Integration, Clinical Decision Support and Care Pathways, Population Health, Digital Therapeutics, Remote Patient Monitoring (RPM), and Digital Engagement, Diagnostics
Primary Track: Big Data for Health
Using electronic health records from 4,731 treatment-resistant depression (TRD) patients in the Saint Francis Health System, we constructed year-long diagnostic state sequences from seven psychiatric disorders and compared them using optimal matching. Preliminary cohort patterns suggest increasing anxiety diagnoses leading up to TRD onset and a shift toward more complex multimorbidity profiles afterward. Multidimensional scaling and clustering revealed distinct trajectory subgroups with differing comorbidity timing and burden, while association analyses suggested demographic differences across trajectories.
Speaker(s):
William Agyapong, PhD
Laureate Institute for Brain Research
Author(s):
Wesley Thompson, PhD - Laureate Institute for Brain Research; Katherine Forthman, MS - Laureate Institute for Brain Research; Martin Paulus, MD - Laureate Institute for Brain Research; Chun Chieh Fan, MD, PhD - Laureate Institute for Brain Research;
Presentation Type: Poster - Regular
Poster Number: 142
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Health Data Science, Analytics, Registries, and the Digital Command Center, Precision Medicine, Multi-Omics, and Pharmacology Integration, Clinical Decision Support and Care Pathways, Population Health, Digital Therapeutics, Remote Patient Monitoring (RPM), and Digital Engagement, Diagnostics
Primary Track: Big Data for Health
Using electronic health records from 4,731 treatment-resistant depression (TRD) patients in the Saint Francis Health System, we constructed year-long diagnostic state sequences from seven psychiatric disorders and compared them using optimal matching. Preliminary cohort patterns suggest increasing anxiety diagnoses leading up to TRD onset and a shift toward more complex multimorbidity profiles afterward. Multidimensional scaling and clustering revealed distinct trajectory subgroups with differing comorbidity timing and burden, while association analyses suggested demographic differences across trajectories.
Speaker(s):
William Agyapong, PhD
Laureate Institute for Brain Research
Author(s):
Wesley Thompson, PhD - Laureate Institute for Brain Research; Katherine Forthman, MS - Laureate Institute for Brain Research; Martin Paulus, MD - Laureate Institute for Brain Research; Chun Chieh Fan, MD, PhD - Laureate Institute for Brain Research;
William
Agyapong,
PhD - Laureate Institute for Brain Research
Validation of ICD-10 Codes for Acute Kidney Injury in Veterans Affairs Inpatient Settings
Presentation Type: Poster - Student
Click to View Presentation
Poster Number: 143
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Health Data Science, Innovation Partnerships, Implementation Science, and Learning Health Systems, Diagnostics
Primary Track: Big Data for Health
In this retrospective cohort study, the performance of ICD-10 codes for identifying acute kidney injury (AKI) was compared to lab-based KDIGO criteria. The study cohort was comprised of 1,508,862 hospital admissions from 38 Veterans Affairs (VA) facilities from 2015 to 2025. ICD-10 sensitivity was 44.3%, specificity 79.2%, positive predictive value 17.7%, and negative predictive value 93.3%. Results highlight the underestimation of AKI incidence at the VA when looking at ICD-10 codes alone.
Speaker(s):
Victoria Brooks, Pharm.D/MSHI
Veterans Health Administration
Author(s):
Anders Westanmo, PharmD, MBA - Minneapolis VAHCS; Jesse Sutton, PharmD, MS - Minneapolis VA Health Care System; Areef Ishani, MD, MS - Minneapolis VA Health Care System; Robert Foley, MD - Minneapolis VA Health Care System; Scott Reule, MD - Minneapolis VA Health Care System;
Presentation Type: Poster - Student
Click to View Presentation
Poster Number: 143
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Health Data Science, Innovation Partnerships, Implementation Science, and Learning Health Systems, Diagnostics
Primary Track: Big Data for Health
In this retrospective cohort study, the performance of ICD-10 codes for identifying acute kidney injury (AKI) was compared to lab-based KDIGO criteria. The study cohort was comprised of 1,508,862 hospital admissions from 38 Veterans Affairs (VA) facilities from 2015 to 2025. ICD-10 sensitivity was 44.3%, specificity 79.2%, positive predictive value 17.7%, and negative predictive value 93.3%. Results highlight the underestimation of AKI incidence at the VA when looking at ICD-10 codes alone.
Speaker(s):
Victoria Brooks, Pharm.D/MSHI
Veterans Health Administration
Author(s):
Anders Westanmo, PharmD, MBA - Minneapolis VAHCS; Jesse Sutton, PharmD, MS - Minneapolis VA Health Care System; Areef Ishani, MD, MS - Minneapolis VA Health Care System; Robert Foley, MD - Minneapolis VA Health Care System; Scott Reule, MD - Minneapolis VA Health Care System;
Victoria
Brooks,
Pharm.D/MSHI - Veterans Health Administration
Transforming Electronic Health Record Data to Develop a GIS-based Hospital Infection Visualization Tool: Validation and Initial Implementation
Presentation Type: Poster - Regular
Poster Number: 144
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Health Data Science, Innovation Partnerships, Implementation Science, and Learning Health Systems, Public Surveillance and Reporting, Environmental Exposure, & Global Health, Human Factors and Usability
Primary Track: Big Data for Health
This study tests the methodology and initial implementation of a geographic information system (GIS)-based application called GeoHAI that leverages electronic health record (EHR) data to visualize hospital onset C. diff infection (HO-CDI).
Speaker(s):
Thrisha Kalpatthi, undergraduate student
Pennsylvania State University
Author(s):
Thrisha Kalpatthi, undergraduate student - Pennsylvania State University; Megan Gregory, Ph.D. - University of Florida; David Kline, PhD - Wake Forest University; Courtney Hebert, MD, MS - The Ohio State University;
Presentation Type: Poster - Regular
Poster Number: 144
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Health Data Science, Innovation Partnerships, Implementation Science, and Learning Health Systems, Public Surveillance and Reporting, Environmental Exposure, & Global Health, Human Factors and Usability
Primary Track: Big Data for Health
This study tests the methodology and initial implementation of a geographic information system (GIS)-based application called GeoHAI that leverages electronic health record (EHR) data to visualize hospital onset C. diff infection (HO-CDI).
Speaker(s):
Thrisha Kalpatthi, undergraduate student
Pennsylvania State University
Author(s):
Thrisha Kalpatthi, undergraduate student - Pennsylvania State University; Megan Gregory, Ph.D. - University of Florida; David Kline, PhD - Wake Forest University; Courtney Hebert, MD, MS - The Ohio State University;
Thrisha
Kalpatthi,
undergraduate student - Pennsylvania State University
RDMatcher: Hybrid Competitive Matching for High-Dimensional Causal Inference in Large-Scale Electronic Health Records
Presentation Type: Poster - Student
Poster Number: 145
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Health Data Science, Analytics, Registries, and the Digital Command Center, Outcomes Improvement and Equity
Primary Track: Big Data for Health
RDMatcher is an efficient Python algorithm for high-dimensional rare-disease matching in EHRs. It uses a hybrid competitive strategy that prioritizes cases with scarce feasible controls, then optimizes remaining matches. In UCSF EHR data, RDMatcher achieved covariate balance (SMD <0.01) and within-set similarity (median Gower distance 0.004) for 1:10 matching. In simulations (1:1-1:10), balanced cohorts showed similar efficacy; in low-support cohorts, RDMatcher restriction before PSM may improve matches beyond PSM alone.
Speaker(s):
Noah Baker, MPH
UCSF
Author(s):
Noah Baker, MPH - UCSF; Madhumita Sushil, PhD - UCSF;
Presentation Type: Poster - Student
Poster Number: 145
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Health Data Science, Analytics, Registries, and the Digital Command Center, Outcomes Improvement and Equity
Primary Track: Big Data for Health
RDMatcher is an efficient Python algorithm for high-dimensional rare-disease matching in EHRs. It uses a hybrid competitive strategy that prioritizes cases with scarce feasible controls, then optimizes remaining matches. In UCSF EHR data, RDMatcher achieved covariate balance (SMD <0.01) and within-set similarity (median Gower distance 0.004) for 1:10 matching. In simulations (1:1-1:10), balanced cohorts showed similar efficacy; in low-support cohorts, RDMatcher restriction before PSM may improve matches beyond PSM alone.
Speaker(s):
Noah Baker, MPH
UCSF
Author(s):
Noah Baker, MPH - UCSF; Madhumita Sushil, PhD - UCSF;
Noah
Baker,
MPH - UCSF
Would mHealth-based stress management be sufficient for family caregivers of those with dementia?
Presentation Type: Poster - Regular
Poster Number: 146
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, Outcomes Improvement and Equity
Primary Track: Advancing Wellness for Providers and Community with Consideration of Human Factors
Family caregivers (FCGs) of people living with dementia experience high stress and often cannot attend in-person mindfulness programs. Despite the widespread use of mindfulness apps, little is known about FCGs' needs and preferences. Because user-centered design is essential for engagement and effective stress management, this study evaluates FCGs’ preference and satisfaction in a RCT comparing two app interfaces, one featuring a virtual human assistant and one without, to guide the development of scalable, user-centered interventions.
Speaker(s):
Lingling Zhang, ScD, FAMIA
University of Massachusetts Boston
Author(s):
Jan Mutchler, PhD - University of Massachusetts Boston; Chengjie Zheng, MS - University of Massachusetts Boston;
Presentation Type: Poster - Regular
Poster Number: 146
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, Outcomes Improvement and Equity
Primary Track: Advancing Wellness for Providers and Community with Consideration of Human Factors
Family caregivers (FCGs) of people living with dementia experience high stress and often cannot attend in-person mindfulness programs. Despite the widespread use of mindfulness apps, little is known about FCGs' needs and preferences. Because user-centered design is essential for engagement and effective stress management, this study evaluates FCGs’ preference and satisfaction in a RCT comparing two app interfaces, one featuring a virtual human assistant and one without, to guide the development of scalable, user-centered interventions.
Speaker(s):
Lingling Zhang, ScD, FAMIA
University of Massachusetts Boston
Author(s):
Jan Mutchler, PhD - University of Massachusetts Boston; Chengjie Zheng, MS - University of Massachusetts Boston;
Lingling
Zhang,
ScD, FAMIA - University of Massachusetts Boston
Co-Designing an Outpatient Clinical Decision Support Tool for Managing Social Risks in Patients Living with Dementia
Presentation Type: Poster - Student
Poster Number: 147
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Human Factors and Usability, Social Determinants of Health (SDoH), Clinical Decision Support and Care Pathways, Clinician Well-Being
Primary Track: Advancing Wellness for Providers and Community with Consideration of Human Factors
We co-designed the iSMART clinical decision support tool with a representative sample of stakeholders. The tool integrates an AI-driven polysocial risk score into the EHR to summarize key social determinants for patients with dementia. It helps providers identify patients with elevated social risk for dementia-related adverse outcomes and act on priority risks through referrals and other workflow steps. Nine outpatient clinicians reviewed the tool, reported strong perceived usefulness, and suggested concrete workflow integration and refinements.
Speaker(s):
Philipp Haessner, BS, MBA, MS
University of Florida
Author(s):
Philipp Haessner, BS, MBA, MS - University of Florida; Miad Alfaqih, Phd - University of Florida; Jennifer LeLaurin, Ph.D. - University of Florida; Serena Jingchuan Guo, MD, PhD - Purdue University; Nicole Hammer, B.A. - University of Florida - Department of Health Outcomes and Biomedical Informatics; Onyekachi Ike-Okpe, MD - University Of Florida; Michael Pappa, MSHI - University of Florida; Xing He, Ph.D. - Indiana University; Ramzi Salloum, PhD; Jiang Bian, PhD - Indiana University/Regenstrief Institute; Megan Gregory, Ph.D. - University of Florida;
Presentation Type: Poster - Student
Poster Number: 147
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Human Factors and Usability, Social Determinants of Health (SDoH), Clinical Decision Support and Care Pathways, Clinician Well-Being
Primary Track: Advancing Wellness for Providers and Community with Consideration of Human Factors
We co-designed the iSMART clinical decision support tool with a representative sample of stakeholders. The tool integrates an AI-driven polysocial risk score into the EHR to summarize key social determinants for patients with dementia. It helps providers identify patients with elevated social risk for dementia-related adverse outcomes and act on priority risks through referrals and other workflow steps. Nine outpatient clinicians reviewed the tool, reported strong perceived usefulness, and suggested concrete workflow integration and refinements.
Speaker(s):
Philipp Haessner, BS, MBA, MS
University of Florida
Author(s):
Philipp Haessner, BS, MBA, MS - University of Florida; Miad Alfaqih, Phd - University of Florida; Jennifer LeLaurin, Ph.D. - University of Florida; Serena Jingchuan Guo, MD, PhD - Purdue University; Nicole Hammer, B.A. - University of Florida - Department of Health Outcomes and Biomedical Informatics; Onyekachi Ike-Okpe, MD - University Of Florida; Michael Pappa, MSHI - University of Florida; Xing He, Ph.D. - Indiana University; Ramzi Salloum, PhD; Jiang Bian, PhD - Indiana University/Regenstrief Institute; Megan Gregory, Ph.D. - University of Florida;
Philipp
Haessner,
BS, MBA, MS - University of Florida
Not Little Adults: Developing and Operationalizing Pediatric EHR Content for General Emergency Departments
Presentation Type: Poster - Regular
Poster Number: 148
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Human Factors and Usability, Outcomes Improvement and Equity, Clinical Decision Support and Care Pathways, Innovation Partnerships, Implementation Science, and Learning Health Systems
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Most children receive emergency care in general EDs lacking pediatric-specific EHR content, creating workflow challenges, poor EHR usability for pediatric patients, and risk that care provided may not consistently align with evidence-based standards. We partnered with two community EDs to design evidence-based pediatric order sets based on user-centered principles and national guidelines. Order sets include simple embedded decision support, enabling a portable design for broader adoption across EHR environments.
Speaker(s):
Emily Sentman, MD
Nationwide Children's Hospital
Author(s):
Emily Sentman, MD - Nationwide Children's Hospital; Anne Runkle, MD - Nationwide Children's Hospital; Charmaine Lo, PhD MPH - Nationwide Children's Hospital; Kate Remick, MD - Dell Medical School at the University of Texas at Austin;
Presentation Type: Poster - Regular
Poster Number: 148
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Human Factors and Usability, Outcomes Improvement and Equity, Clinical Decision Support and Care Pathways, Innovation Partnerships, Implementation Science, and Learning Health Systems
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Most children receive emergency care in general EDs lacking pediatric-specific EHR content, creating workflow challenges, poor EHR usability for pediatric patients, and risk that care provided may not consistently align with evidence-based standards. We partnered with two community EDs to design evidence-based pediatric order sets based on user-centered principles and national guidelines. Order sets include simple embedded decision support, enabling a portable design for broader adoption across EHR environments.
Speaker(s):
Emily Sentman, MD
Nationwide Children's Hospital
Author(s):
Emily Sentman, MD - Nationwide Children's Hospital; Anne Runkle, MD - Nationwide Children's Hospital; Charmaine Lo, PhD MPH - Nationwide Children's Hospital; Kate Remick, MD - Dell Medical School at the University of Texas at Austin;
Emily
Sentman,
MD - Nationwide Children's Hospital
From Concern to Conversation: EHR-Linked Surveys for Vaccine Education
Presentation Type: Poster - Regular
Poster Number: 149
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Infrastructure and Cloud Computing, Clinical Decision Support and Care Pathways, Population Health, Digital Therapeutics, Remote Patient Monitoring (RPM), and Digital Engagement, Public Surveillance and Reporting, Environmental Exposure, & Global Health
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
We developed an EHR-integrated text messaging system designed to elicit families’ vaccine concerns prior to well-child visits. Responses were transmitted into the EHR as practice advisories with tailored talking points for providers. The system was successfully implemented, demonstrating feasibility and establishing infrastructure for evaluating text-based survey strategies that enhance family-provider communication. Early results, however, revealed limited family engagement with the survey link text message, underscoring the need for strategies to increase participation.
Speaker(s):
Jonathan Beus, MD, MSCR, FAAP, FAMIA
Children's Healthcare of Atlanta
Author(s):
Jonathan Beus, MD, MSCR, FAAP, FAMIA - Children's Healthcare of Atlanta; Elease McLaurin, PhD - Emory University; Mallory Tidwell, BSN, RN - Children's Healthcare of Atlanta; Brad Cundiff, BS - Children's Healthcare of Atlanta; Terri Skipper, Senior Applications Analyst - Children's Healthcare of Atlanta; Yvette Reynolds, RN - Children's Healthcare of Atlanta; Michelle Goryn, MA - Johns Hopkins Bloomberg School of Public Health; Matthew Dudley, PhD - Johns Hopkins Bloomberg School of Public Health; Naveen Muthu, MD - Children's Healthcare of Atlanta; Daniel Salmon, PhD - Johns Hopkins Bloomberg School of Public Health;
Presentation Type: Poster - Regular
Poster Number: 149
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Infrastructure and Cloud Computing, Clinical Decision Support and Care Pathways, Population Health, Digital Therapeutics, Remote Patient Monitoring (RPM), and Digital Engagement, Public Surveillance and Reporting, Environmental Exposure, & Global Health
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
We developed an EHR-integrated text messaging system designed to elicit families’ vaccine concerns prior to well-child visits. Responses were transmitted into the EHR as practice advisories with tailored talking points for providers. The system was successfully implemented, demonstrating feasibility and establishing infrastructure for evaluating text-based survey strategies that enhance family-provider communication. Early results, however, revealed limited family engagement with the survey link text message, underscoring the need for strategies to increase participation.
Speaker(s):
Jonathan Beus, MD, MSCR, FAAP, FAMIA
Children's Healthcare of Atlanta
Author(s):
Jonathan Beus, MD, MSCR, FAAP, FAMIA - Children's Healthcare of Atlanta; Elease McLaurin, PhD - Emory University; Mallory Tidwell, BSN, RN - Children's Healthcare of Atlanta; Brad Cundiff, BS - Children's Healthcare of Atlanta; Terri Skipper, Senior Applications Analyst - Children's Healthcare of Atlanta; Yvette Reynolds, RN - Children's Healthcare of Atlanta; Michelle Goryn, MA - Johns Hopkins Bloomberg School of Public Health; Matthew Dudley, PhD - Johns Hopkins Bloomberg School of Public Health; Naveen Muthu, MD - Children's Healthcare of Atlanta; Daniel Salmon, PhD - Johns Hopkins Bloomberg School of Public Health;
Jonathan
Beus,
MD, MSCR, FAAP, FAMIA - Children's Healthcare of Atlanta
Clinician Perspectives on Aging in Place and Medication Safety Research
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, Telemedicine, Health at Home, and Virtual Care, Workforce Automation, Communication, and Workflow Efficiency
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
The majority of older Americans wish to age in their homes rather than a hospital or long-term residential care. Despite this, research on medication safety in the home health care settings (HHC) is relatively nascent, and data remains scarce. This study makes progress addressing these gaps by interviewing HHC workers for their perspectives on care gaps, challenges, and opportunities for medication-related intervention design.
Speaker(s):
Thomas Tam, MS
University of Pittsburgh
Author(s):
Sandra Kane-Gill, PharmD, MS, FCCM, FCCP - University of Pittsburgh; Steven Albert, PhD, MA, MS, FGSA, FAAN - University of Pittsburgh; Richard Boyce, PhD - University of Pittsburgh;
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, Telemedicine, Health at Home, and Virtual Care, Workforce Automation, Communication, and Workflow Efficiency
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
The majority of older Americans wish to age in their homes rather than a hospital or long-term residential care. Despite this, research on medication safety in the home health care settings (HHC) is relatively nascent, and data remains scarce. This study makes progress addressing these gaps by interviewing HHC workers for their perspectives on care gaps, challenges, and opportunities for medication-related intervention design.
Speaker(s):
Thomas Tam, MS
University of Pittsburgh
Author(s):
Sandra Kane-Gill, PharmD, MS, FCCM, FCCP - University of Pittsburgh; Steven Albert, PhD, MA, MS, FGSA, FAAN - University of Pittsburgh; Richard Boyce, PhD - University of Pittsburgh;
Thomas
Tam,
MS - University of Pittsburgh
Operationalizing AI Translation in a Learning Health System: Evolution of Mayo Clinic’s AI Translation Advisory Council (2022-Present) and its Best Practices Checklist
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, Clinical Decision Support and Care Pathways, Leadership and Strategy
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Digital health enables scalable clinical impact, but AI translation often fails at the “code-to-bedside” boundary where governance, workflow integration, and evidence expectations diverge. Established in 2022, Mayo Clinic’s AI Translation Advisory Council (AI TAC) has evolved into a multidisciplinary, consultative operating model that streamlines implementation of clinical AI into practice. A key culmination is an AI Translation Best Practices checklist, distilled from iterative translation evaluations, to standardize evidence, surface gaps, and support real-world deployment decisions and lifecycle monitoring.
Speaker(s):
Lauren Rost, PhD
Mayo Clinic
Author(s):
Lauren Rost, PhD - Mayo Clinic; Joshua Ohde, PhD - Mayo Clinic; Young Juhn, M.D., M.P.H. - Mayo Clinic; Chenyu Gai, Master - Mayo Clinic; Lu Zheng, Ph.D., M.S. - Mayo Clinic; Jason Greenwood, MD, MS - Mayo Clinic; Momin Malik, PhD in Societal Computing - Mayo Clinic; Chung Wi, MD - Mayo Clinic; Kevin Peterson, PhD - Mayo Clinic; Yongwen Wu, MS - Mayo Clinic; Tracey Brereton, Masters - Mayo Clinic;
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, Clinical Decision Support and Care Pathways, Leadership and Strategy
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Digital health enables scalable clinical impact, but AI translation often fails at the “code-to-bedside” boundary where governance, workflow integration, and evidence expectations diverge. Established in 2022, Mayo Clinic’s AI Translation Advisory Council (AI TAC) has evolved into a multidisciplinary, consultative operating model that streamlines implementation of clinical AI into practice. A key culmination is an AI Translation Best Practices checklist, distilled from iterative translation evaluations, to standardize evidence, surface gaps, and support real-world deployment decisions and lifecycle monitoring.
Speaker(s):
Lauren Rost, PhD
Mayo Clinic
Author(s):
Lauren Rost, PhD - Mayo Clinic; Joshua Ohde, PhD - Mayo Clinic; Young Juhn, M.D., M.P.H. - Mayo Clinic; Chenyu Gai, Master - Mayo Clinic; Lu Zheng, Ph.D., M.S. - Mayo Clinic; Jason Greenwood, MD, MS - Mayo Clinic; Momin Malik, PhD in Societal Computing - Mayo Clinic; Chung Wi, MD - Mayo Clinic; Kevin Peterson, PhD - Mayo Clinic; Yongwen Wu, MS - Mayo Clinic; Tracey Brereton, Masters - Mayo Clinic;
Lauren
Rost,
PhD - Mayo Clinic
Systematic Governance of CDS to Reduce Alert Fatigue and Improve Outcomes
Presentation Type: Poster Invite - Regular
Poster Number: 152
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Leadership and Strategy, Clinical Decision Support and Care Pathways, Change Management, Clinician Well-Being, Quality Informatics and Lean, Outcomes Improvement and Equity, Human Factors and Usability
Primary Track: Driving Change at Scale through Effective Leadership and Governance
This study examines the impact of a centralized Clinical Decision Support (CDS) governance model on reducing alert fatigue and improving clinical outcomes within a multi-hospital health system using an Oracle Health EMR. Alert fatigue—driven by high-volume, low-value alerts—can lead to clinician disengagement, missed critical warnings, and patient safety risks. Penn State Health established a multidisciplinary CDS Informatics Council to evaluate CDS performance, eliminate low-value alerts, and redesign tools using data analytics, provider feedback, and the Five Rights framework.
The council implemented two major strategies: suppressing low-utility alerts and deploying redesigned, high-value CDS interventions. This approach eliminated more than one million annual pop-up alerts, including a SIRS alert with <1% action rate. Targeted CDS tools yielded measurable improvements: telemetry usage decreased 15% system-wide; code status documentation improved from 79% to 94%; diphenhydramine prescribing in adults ≥65 dropped 90%; and a redesigned sepsis alert increased action rates to ~20%, reduced alert volume by 45%, and was associated with a 25% reduction in sepsis mortality and 16% shorter length of stay—projecting 135 lives saved and significant cost reduction annually.
Findings demonstrate that structured CDS governance can reduce cognitive burden without compromising safety, and may enhance clinical outcomes when alerts are user-centered and workflow-aligned. A centralized, data-driven model offers a scalable blueprint for optimizing CDS performance, improving provider engagement, and supporting institutional quality goals.
Speaker(s):
Prateek Grover, MD PhD MHA
Penn State
Author(s):
Shadi Hijjawi, MD - Penn State Health; Prateek Grover, MD PhD MHA - Penn State; Richard Schreiber, MD, FACP, FAMIA - Penn State Health; Douglas Morrissey, MD - Penn State Medical Group; Francis Quigley, DO - Pennstate health; Thomas Scott, MD - Penn State Medical Group; Christopher DeFlitch, MD - Penn State;
Presentation Type: Poster Invite - Regular
Poster Number: 152
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Leadership and Strategy, Clinical Decision Support and Care Pathways, Change Management, Clinician Well-Being, Quality Informatics and Lean, Outcomes Improvement and Equity, Human Factors and Usability
Primary Track: Driving Change at Scale through Effective Leadership and Governance
This study examines the impact of a centralized Clinical Decision Support (CDS) governance model on reducing alert fatigue and improving clinical outcomes within a multi-hospital health system using an Oracle Health EMR. Alert fatigue—driven by high-volume, low-value alerts—can lead to clinician disengagement, missed critical warnings, and patient safety risks. Penn State Health established a multidisciplinary CDS Informatics Council to evaluate CDS performance, eliminate low-value alerts, and redesign tools using data analytics, provider feedback, and the Five Rights framework.
The council implemented two major strategies: suppressing low-utility alerts and deploying redesigned, high-value CDS interventions. This approach eliminated more than one million annual pop-up alerts, including a SIRS alert with <1% action rate. Targeted CDS tools yielded measurable improvements: telemetry usage decreased 15% system-wide; code status documentation improved from 79% to 94%; diphenhydramine prescribing in adults ≥65 dropped 90%; and a redesigned sepsis alert increased action rates to ~20%, reduced alert volume by 45%, and was associated with a 25% reduction in sepsis mortality and 16% shorter length of stay—projecting 135 lives saved and significant cost reduction annually.
Findings demonstrate that structured CDS governance can reduce cognitive burden without compromising safety, and may enhance clinical outcomes when alerts are user-centered and workflow-aligned. A centralized, data-driven model offers a scalable blueprint for optimizing CDS performance, improving provider engagement, and supporting institutional quality goals.
Speaker(s):
Prateek Grover, MD PhD MHA
Penn State
Author(s):
Shadi Hijjawi, MD - Penn State Health; Prateek Grover, MD PhD MHA - Penn State; Richard Schreiber, MD, FACP, FAMIA - Penn State Health; Douglas Morrissey, MD - Penn State Medical Group; Francis Quigley, DO - Pennstate health; Thomas Scott, MD - Penn State Medical Group; Christopher DeFlitch, MD - Penn State;
Prateek
Grover,
MD PhD MHA - Penn State
Timely Notification of Veterans' Deaths Outside the VA Healthcare System
Presentation Type: Poster - Regular
Poster Number: 153
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Leadership and Strategy, Health Policy, Reimbursement and Affordability, and Sustainability, Public Surveillance and Reporting, Environmental Exposure, & Global Health, Innovation Partnerships, Implementation Science, and Learning Health Systems
Primary Track: Driving Change at Scale through Effective Leadership and Governance
Timely notification of Veterans’ deaths outside VA facilities is critical to preventing inappropriate prescriptions and uncancelled appointments. Currently, VA Salt Lake City lacks a consistent process to receive this information, leading to wasted provider time, resources, and distressing communications to families. Through a new data-sharing agreement with the Utah Department of Health and Human Services, we will receive twice-weekly death updates and promptly cancel appointments and prescriptions for deceased Veterans.
Speaker(s):
Andrea Bleak, BHA
Department of Veterans Affairs
Author(s):
Shardool Patel - VA Salt Lake City; Trevor Jones, MD - Salt Lake City VA;
Presentation Type: Poster - Regular
Poster Number: 153
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Leadership and Strategy, Health Policy, Reimbursement and Affordability, and Sustainability, Public Surveillance and Reporting, Environmental Exposure, & Global Health, Innovation Partnerships, Implementation Science, and Learning Health Systems
Primary Track: Driving Change at Scale through Effective Leadership and Governance
Timely notification of Veterans’ deaths outside VA facilities is critical to preventing inappropriate prescriptions and uncancelled appointments. Currently, VA Salt Lake City lacks a consistent process to receive this information, leading to wasted provider time, resources, and distressing communications to families. Through a new data-sharing agreement with the Utah Department of Health and Human Services, we will receive twice-weekly death updates and promptly cancel appointments and prescriptions for deceased Veterans.
Speaker(s):
Andrea Bleak, BHA
Department of Veterans Affairs
Author(s):
Shardool Patel - VA Salt Lake City; Trevor Jones, MD - Salt Lake City VA;
Andrea
Bleak,
BHA - Department of Veterans Affairs
Leveraging EHR Data to Identify Gaps in Osteoporosis Screening for Patients with Rheumatoid Arthritis and/or Chronic Glucocorticoids
Presentation Type: Poster - Regular
Poster Number: 154
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Outcomes Improvement and Equity, Social Determinants of Health (SDoH), Quality Informatics and Lean, Population Health, Digital Therapeutics, Remote Patient Monitoring (RPM), and Digital Engagement
Primary Track: Big Data for Health
Osteoporosis screening is recommended for patients with rheumatoid arthritis (RA) and for individuals receiving prolonged glucocorticoids, yet screening rates vary widely across healthcare settings. This study leveraged national Veterans Affairs (VA) electronic health record (EHR) data to determine the proportion of Veterans with RA or chronic oral prednisone exposure who completed dual-energy X-ray absorptiometry (DXA) screening and to identify demographic factors associated with testing.
Two high-risk cohorts were constructed. The RA cohort included patients with ≥2 encounters coded with ICD-10 M05* or M06* at least 30 days apart. The glucocorticoid cohort consisted of patients who received ≥90 non-overlapping days of oral prednisone within the prior year. DXA scans performed within the past 5 years were identified using CPT codes (77080, 77081, 77085, 77086) from both VA and Community Care sources.
The RA cohort included 89,703 Veterans (mean age 68 years; 79% male), of whom 19% had DXA screening.
The prednisone cohort included 39,909 Veterans (mean age 70.5 years; 92% male), with 32% receiving DXA screening.
These findings demonstrate substantial gaps in osteoporosis screening among high-risk Veterans and highlight opportunities for targeted improvement efforts. A national Power BI dashboard is currently under development to enable facilities to visualize screening performance and support data-driven quality improvement.
Speaker(s):
Sara Faghihi Kashani, MD, MPH
UCSF
Author(s):
Sara Faghihi Kashani, MD, MPH - UCSF; Karen Beltran, MD - UCSF; Sristi Sharma, M.D., M.P.H. - UCSF; Gary Tarasovsky, BS - SFVA; Cherish Wilson, BA - UCSF; Mary Whooley, MD - University of California, San Francisco; Gabriela Schmajuk, MD, MS - UCSF;
Presentation Type: Poster - Regular
Poster Number: 154
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Outcomes Improvement and Equity, Social Determinants of Health (SDoH), Quality Informatics and Lean, Population Health, Digital Therapeutics, Remote Patient Monitoring (RPM), and Digital Engagement
Primary Track: Big Data for Health
Osteoporosis screening is recommended for patients with rheumatoid arthritis (RA) and for individuals receiving prolonged glucocorticoids, yet screening rates vary widely across healthcare settings. This study leveraged national Veterans Affairs (VA) electronic health record (EHR) data to determine the proportion of Veterans with RA or chronic oral prednisone exposure who completed dual-energy X-ray absorptiometry (DXA) screening and to identify demographic factors associated with testing.
Two high-risk cohorts were constructed. The RA cohort included patients with ≥2 encounters coded with ICD-10 M05* or M06* at least 30 days apart. The glucocorticoid cohort consisted of patients who received ≥90 non-overlapping days of oral prednisone within the prior year. DXA scans performed within the past 5 years were identified using CPT codes (77080, 77081, 77085, 77086) from both VA and Community Care sources.
The RA cohort included 89,703 Veterans (mean age 68 years; 79% male), of whom 19% had DXA screening.
The prednisone cohort included 39,909 Veterans (mean age 70.5 years; 92% male), with 32% receiving DXA screening.
These findings demonstrate substantial gaps in osteoporosis screening among high-risk Veterans and highlight opportunities for targeted improvement efforts. A national Power BI dashboard is currently under development to enable facilities to visualize screening performance and support data-driven quality improvement.
Speaker(s):
Sara Faghihi Kashani, MD, MPH
UCSF
Author(s):
Sara Faghihi Kashani, MD, MPH - UCSF; Karen Beltran, MD - UCSF; Sristi Sharma, M.D., M.P.H. - UCSF; Gary Tarasovsky, BS - SFVA; Cherish Wilson, BA - UCSF; Mary Whooley, MD - University of California, San Francisco; Gabriela Schmajuk, MD, MS - UCSF;
Sara
Faghihi Kashani,
MD, MPH - UCSF
Factors Associated with Patient-Generated Blood Pressure Data in an Urban Safety-Net Health System
Presentation Type: Poster - Regular
Poster Number: 155
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Outcomes Improvement and Equity, Population Health, Digital Therapeutics, Remote Patient Monitoring (RPM), and Digital Engagement, Social Determinants of Health (SDoH)
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Remote blood pressure (BP) monitoring is underutilized in safety-net settings. We used hierarchical logistic regression with patient-level covariates and random effects for primary care providers and clinics to model variance in documentation of patient-reported BP. While patient factors accounted for 75% of variance, 19% was attributable to clinics and 6% to providers, indicating that strengthening clinic-level support and infrastructure for documenting patient-reported BP is essential for equitable and widespread adoption to improve hypertension control.
Speaker(s):
Elaine Khoong, MD, MS
University of California San Francisco
Author(s):
Isabel Luna, BA - University of California, San Francisco; Elaine Khoong, MD, MS - University of California San Francisco; Jaivind Grewal, BS, MS - California Health Sciences University; Raaga Karumanchi, N/A - Barnard College, Columbia University,;
Presentation Type: Poster - Regular
Poster Number: 155
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Outcomes Improvement and Equity, Population Health, Digital Therapeutics, Remote Patient Monitoring (RPM), and Digital Engagement, Social Determinants of Health (SDoH)
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Remote blood pressure (BP) monitoring is underutilized in safety-net settings. We used hierarchical logistic regression with patient-level covariates and random effects for primary care providers and clinics to model variance in documentation of patient-reported BP. While patient factors accounted for 75% of variance, 19% was attributable to clinics and 6% to providers, indicating that strengthening clinic-level support and infrastructure for documenting patient-reported BP is essential for equitable and widespread adoption to improve hypertension control.
Speaker(s):
Elaine Khoong, MD, MS
University of California San Francisco
Author(s):
Isabel Luna, BA - University of California, San Francisco; Elaine Khoong, MD, MS - University of California San Francisco; Jaivind Grewal, BS, MS - California Health Sciences University; Raaga Karumanchi, N/A - Barnard College, Columbia University,;
Elaine
Khoong,
MD, MS - University of California San Francisco
The Accuracy of Nutrition in Recipes Generated by Artificial Intelligence
Presentation Type: Poster - 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, Generative AI in Clinical Workflow: Ambient Listening, Chart Summarization, Automated Response with LLM, Telemedicine, Health at Home, and Virtual Care
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Generative Artificial intelligence (AI) offers new opportunities to deliver practical dietary guidance to individuals. We developed and pilot-tested the accuracy of DASH GENIE, an AI chatbot that uses large language models to generate recipes designed to meet DASH (Dietary Approaches to Stop Hypertension) nutrient targets. The results showed that our AI chatbot (without any tuning) can generate DASH guideline compliant recipes and accurate value estimates on some but not all key nutrients.
Speaker(s):
Ming-Yuan Chih, PhD
University of Kentucky
Author(s):
Ming-Yuan Chih, PhD - University of Kentucky; Kendra OoNorasak, PhD - University of Kentucky; Mansura Shahad Bawa, BS - University of Kentucky; Yongwook Song, MS - University of Kentucky; Vikram Gazula, MS - University of Kentucky; Brandi White, PhD - University of Kentucky;
Presentation Type: Poster - 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, Generative AI in Clinical Workflow: Ambient Listening, Chart Summarization, Automated Response with LLM, Telemedicine, Health at Home, and Virtual Care
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Generative Artificial intelligence (AI) offers new opportunities to deliver practical dietary guidance to individuals. We developed and pilot-tested the accuracy of DASH GENIE, an AI chatbot that uses large language models to generate recipes designed to meet DASH (Dietary Approaches to Stop Hypertension) nutrient targets. The results showed that our AI chatbot (without any tuning) can generate DASH guideline compliant recipes and accurate value estimates on some but not all key nutrients.
Speaker(s):
Ming-Yuan Chih, PhD
University of Kentucky
Author(s):
Ming-Yuan Chih, PhD - University of Kentucky; Kendra OoNorasak, PhD - University of Kentucky; Mansura Shahad Bawa, BS - University of Kentucky; Yongwook Song, MS - University of Kentucky; Vikram Gazula, MS - University of Kentucky; Brandi White, PhD - University of Kentucky;
Ming-Yuan
Chih,
PhD - University of Kentucky
Closing the Diagnostic Gap: Next Generation Sequencing for Enhanced Pathogen and Resistance Detection in Burn Wounds
Presentation Type: Poster - Student
Poster Number: 157
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Precision Medicine, Multi-Omics, and Pharmacology Integration, Diagnostics, Health Data Science
Primary Track: Big Data for Health
Infections are the leading cause of mortality in burn patients and are often complicated by antimicrobial resistance (AMR). This study compared standard culture with next-generation sequencing (NGS) for microbial and AMR detection in burn tissue. NGS identified substantially more bacteria and fungi than culture, including organisms missed by routine diagnostics. Among cultured isolates, most showed potential AMR. These findings highlight the diagnostic value of integrating NGS into burn infection management.
Speaker(s):
Riley Watson, B.S.
University of Texas Medical Branch at Galveston
Author(s):
Riley Watson, B.S. - University of Texas Medical Branch at Galveston; Vivian Tat, PhD - University of Texas Medical Branch at Galveston; Katherine Araya, BS - University of Texas Medical Branch at Galveston; Sarah Alnemrat, BDS - University of Texas Medical Branch at Galveston; Elizaveta Naydanova, M.D./Ph.D. Student - University of Texas Medical Branch; Juquan Song, MD - University of Texas Medical Branch at Galveston; Amina El Ayadi, PhD - University of Texas Medical Branch at Galveston; Jennifer Narvaez, BS - University of Texas Medical Branch at Galveston; Kamil Khanipov, PhD - University of Texas Medical Branch; Steven Wolf, MD - University of Texas Medical Branch at Galveston; George Golovko, PhD - University of Texas Medical Branch;
Presentation Type: Poster - Student
Poster Number: 157
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Precision Medicine, Multi-Omics, and Pharmacology Integration, Diagnostics, Health Data Science
Primary Track: Big Data for Health
Infections are the leading cause of mortality in burn patients and are often complicated by antimicrobial resistance (AMR). This study compared standard culture with next-generation sequencing (NGS) for microbial and AMR detection in burn tissue. NGS identified substantially more bacteria and fungi than culture, including organisms missed by routine diagnostics. Among cultured isolates, most showed potential AMR. These findings highlight the diagnostic value of integrating NGS into burn infection management.
Speaker(s):
Riley Watson, B.S.
University of Texas Medical Branch at Galveston
Author(s):
Riley Watson, B.S. - University of Texas Medical Branch at Galveston; Vivian Tat, PhD - University of Texas Medical Branch at Galveston; Katherine Araya, BS - University of Texas Medical Branch at Galveston; Sarah Alnemrat, BDS - University of Texas Medical Branch at Galveston; Elizaveta Naydanova, M.D./Ph.D. Student - University of Texas Medical Branch; Juquan Song, MD - University of Texas Medical Branch at Galveston; Amina El Ayadi, PhD - University of Texas Medical Branch at Galveston; Jennifer Narvaez, BS - University of Texas Medical Branch at Galveston; Kamil Khanipov, PhD - University of Texas Medical Branch; Steven Wolf, MD - University of Texas Medical Branch at Galveston; George Golovko, PhD - University of Texas Medical Branch;
Riley
Watson,
B.S. - University of Texas Medical Branch at Galveston
Utilizing Clinical Decision Support To Improve Recognition and Management of Glucocorticoid-Induced Adrenal Insufficiency in Pediatric Rheumatology Patients
Presentation Type: Poster - Regular
Poster Number: 158
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Quality Informatics and Lean, Clinical Decision Support and Care Pathways, Outcomes Improvement and Equity, Change Management, Education and Training
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
This quality improvement project successfully demonstrates that the combination of provider education along with the implementation of electronic health record-integrated clinical decision support tools, including emergency action plans and passive alerts, is associated with increased recognition and appropriate emergency management of adrenal insufficiency and adrenal crisis in a specific high-risk pediatric patient population.
Speaker(s):
Emily Schildt, MD
University of Washington
Author(s):
Emily Schildt, MD - University of Washington; Hayley Lynch, MD - Seattle Children's Hospital; Natalie Rosenwasser, MD - University of Washington; Susan Shenoi, MD - University of Washington; Lori Rutman, MD - University of Washington; Meenal Gupta, MD - University of Washington;
Presentation Type: Poster - Regular
Poster Number: 158
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Quality Informatics and Lean, Clinical Decision Support and Care Pathways, Outcomes Improvement and Equity, Change Management, Education and Training
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
This quality improvement project successfully demonstrates that the combination of provider education along with the implementation of electronic health record-integrated clinical decision support tools, including emergency action plans and passive alerts, is associated with increased recognition and appropriate emergency management of adrenal insufficiency and adrenal crisis in a specific high-risk pediatric patient population.
Speaker(s):
Emily Schildt, MD
University of Washington
Author(s):
Emily Schildt, MD - University of Washington; Hayley Lynch, MD - Seattle Children's Hospital; Natalie Rosenwasser, MD - University of Washington; Susan Shenoi, MD - University of Washington; Lori Rutman, MD - University of Washington; Meenal Gupta, MD - University of Washington;
Emily
Schildt,
MD - University of Washington
Experiences in Reaching Near Full eConsent Compliance
Presentation Type: Poster - Regular
Poster Number: 159
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Quality Informatics and Lean, Change Management, Leadership and Strategy
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
MetroHealth has pursued eConsent since 2014, but efforts were limited by technology and operational challenges. In 2024, less than 20% of consents were eConsents In 2025, a coordinated initiative, including electronic health record (EHR)functionality upgrade, deployment of mobile devices, and leadership and operational engagement enabled full implementation. By September of 2025 over 99% of all consents were eConsents, closing documentation gaps and improving patient safety.
Speaker(s):
Jui-En Lo, MD
Metrohealth Medical Center
Author(s):
Jui-En Lo, MD - Metrohealth Medical Center; David Kaelber, MD, PhD, MPH - Case Western Reserve University/The MetroHealth System;
Presentation Type: Poster - Regular
Poster Number: 159
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Quality Informatics and Lean, Change Management, Leadership and Strategy
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
MetroHealth has pursued eConsent since 2014, but efforts were limited by technology and operational challenges. In 2024, less than 20% of consents were eConsents In 2025, a coordinated initiative, including electronic health record (EHR)functionality upgrade, deployment of mobile devices, and leadership and operational engagement enabled full implementation. By September of 2025 over 99% of all consents were eConsents, closing documentation gaps and improving patient safety.
Speaker(s):
Jui-En Lo, MD
Metrohealth Medical Center
Author(s):
Jui-En Lo, MD - Metrohealth Medical Center; David Kaelber, MD, PhD, MPH - Case Western Reserve University/The MetroHealth System;
Jui-En
Lo,
MD - Metrohealth Medical Center
Successfully Leveraging LLMs to Audit Large-Scale and Complex ETL Mappings
Presentation Type: Poster - Regular
Poster Number: 160
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Quality Informatics and Lean, Workforce Automation, Communication, and Workflow Efficiency, Health Data Science
Working Group: Clinical Research Informatics Working Group
Primary Track: Big Data for Health
The Texas Childhood Trauma Research Network ETL process involves mapping thousands of variables, many with multiple versions and multilingual sources merged into single targets. To ensure accurate mapping, we tested GPT-based large language models to detect errors. We evaluated multiple models, configurations, and prompts. Many setups identified known errors and uncovered new ones. LLM-assisted auditing improved the accuracy and reliability of our data-warehouse pipeline and increased confidence in our processes.
Speaker(s):
Luke Klima, BBA
Dell Medical School, University of Texas at Austin
Author(s):
Luke Klima, BBA - Dell Medical School, University of Texas at Austin; Fei Teng, Ph.D. - Dell Medical School, University of Texas at Austin; Liza Hoke, MS - Dell Medical School, University of Texas at Austin; Alexis Larraga, MS - Dell Medical School, University of Texas at Austin; Vladislav Krendelev, MS - Texas Advanced Computing Center, University of Texas at Austin; Pat Scherer, MBA - Texas Advanced Computing Center, University of Texas at Austin; Tomislav Urban, MGIS - Texas Advanced Computing Center, University of Texas at Austin; D. Jeffrey Newport, MD - Dell Medical School, University of Texas at Austin; Karen Dineen Wagner, MD, PhD - University of Texas Medical Branch, Galveston; Charles B. Nemeroff, MD, PhD - Dell Medical School, University of Texas at Austin; Justin Rousseau, MD, MMSc - University of Texas Southwestern Medical Center;
Presentation Type: Poster - Regular
Poster Number: 160
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Quality Informatics and Lean, Workforce Automation, Communication, and Workflow Efficiency, Health Data Science
Working Group: Clinical Research Informatics Working Group
Primary Track: Big Data for Health
The Texas Childhood Trauma Research Network ETL process involves mapping thousands of variables, many with multiple versions and multilingual sources merged into single targets. To ensure accurate mapping, we tested GPT-based large language models to detect errors. We evaluated multiple models, configurations, and prompts. Many setups identified known errors and uncovered new ones. LLM-assisted auditing improved the accuracy and reliability of our data-warehouse pipeline and increased confidence in our processes.
Speaker(s):
Luke Klima, BBA
Dell Medical School, University of Texas at Austin
Author(s):
Luke Klima, BBA - Dell Medical School, University of Texas at Austin; Fei Teng, Ph.D. - Dell Medical School, University of Texas at Austin; Liza Hoke, MS - Dell Medical School, University of Texas at Austin; Alexis Larraga, MS - Dell Medical School, University of Texas at Austin; Vladislav Krendelev, MS - Texas Advanced Computing Center, University of Texas at Austin; Pat Scherer, MBA - Texas Advanced Computing Center, University of Texas at Austin; Tomislav Urban, MGIS - Texas Advanced Computing Center, University of Texas at Austin; D. Jeffrey Newport, MD - Dell Medical School, University of Texas at Austin; Karen Dineen Wagner, MD, PhD - University of Texas Medical Branch, Galveston; Charles B. Nemeroff, MD, PhD - Dell Medical School, University of Texas at Austin; Justin Rousseau, MD, MMSc - University of Texas Southwestern Medical Center;
Luke
Klima,
BBA - Dell Medical School, University of Texas at Austin
A Thematic Analysis of Portable Medical Summaries in Pediatric to Adult Healthcare Transition
Presentation Type: Poster - Student
Click to View Presentation
Poster Number: 161
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Standards, Terminology, and Interoperability, TEFCA, FHIR, Outcomes Improvement and Equity, Clinical Decision Support and Care Pathways
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Due to a lack of standardization, the transition from pediatric to adult healthcare has been associated with poor clinical outcomes, increased and redundant costs, and low patient and family satisfaction. This study aims to outline and categorize the core necessary data elements of a standard outpatient Portable Care Summary document for patients who are transitioning from pediatric to adult care through a thematic analysis, which will create the foundation for future health IT work.
Speaker(s):
Carly Noel, DO, MPH
Cincinnati Children's Hospital
Author(s):
Carly Noel, DO, MPH - Cincinnati Children's Hospital; Sean Dornbush, DO, MBA - University of California Irvine; Shannon Blair, MD - Providence Portland Medical Center; Matthew Molloy, MD, MPH - Cincinnati Children's Hospital Medical Center; Rachel Peterson, MD - Cincinnati Children's Hospital Medical Center;
Presentation Type: Poster - Student
Click to View Presentation
Poster Number: 161
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Standards, Terminology, and Interoperability, TEFCA, FHIR, Outcomes Improvement and Equity, Clinical Decision Support and Care Pathways
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
Due to a lack of standardization, the transition from pediatric to adult healthcare has been associated with poor clinical outcomes, increased and redundant costs, and low patient and family satisfaction. This study aims to outline and categorize the core necessary data elements of a standard outpatient Portable Care Summary document for patients who are transitioning from pediatric to adult care through a thematic analysis, which will create the foundation for future health IT work.
Speaker(s):
Carly Noel, DO, MPH
Cincinnati Children's Hospital
Author(s):
Carly Noel, DO, MPH - Cincinnati Children's Hospital; Sean Dornbush, DO, MBA - University of California Irvine; Shannon Blair, MD - Providence Portland Medical Center; Matthew Molloy, MD, MPH - Cincinnati Children's Hospital Medical Center; Rachel Peterson, MD - Cincinnati Children's Hospital Medical Center;
Carly
Noel,
DO, MPH - Cincinnati Children's Hospital
From Legacy HL7 v2.3 to NPHIES FHIR: A Real-World Implementation Framework for Medication Module Transformation
Presentation Type: Poster - Regular
Poster Number: 162
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Standards, Terminology, and Interoperability, TEFCA, FHIR, Standards, Terminology, and Interoperability, TEFCA, FHIR, Standards, Terminology, and Interoperability, TEFCA, FHIR
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
This project describes implementation of HL7 v2.3 to FHIR integration for the medication module with the National Platform for Health and Insurance Exchange Services (NPHIES) at Prince Sultan Military Medical City in Riyadh, Saudi Arabia, in collaboration with Lean Business Services. NPHIES is a national FHIR based health information exchange. The initial scope covers outpatient prescription and dispensing messages. Cerner generates HL7 v2.3 ORM and RDS messages that are enriched using Cerner Command Language (CCL) to add missing clinical and demographic fields, then completed in an integration layer through database lookup tables, national value set mappings, and calls to patient and practitioner registry services. An interface engine, Rhapsody 6, converts the enriched HL7 messages into FHIR resources for NPHIES medication services.
A pre go live criterion of 5 percent maximum error rate was used to move from staging to production. In production the error rate is now below this threshold and the team aims to keep it under 1 percent. On average the system processes about 17,000 prescription messages and 5,000 dispensing messages per day. Failed messages are logged to a monitoring database and summarized in automated email and spreadsheet reports reviewed with clinical and integration teams.
The poster will present the architecture, a summary of twelve key challenges and solutions, and a time series of production success and error rates. This experience shows how a medication first transformation layer can connect legacy HL7 environments to a national FHIR based infrastructure without replacing the core electronic health record.
Speaker(s):
Abdullah ALOMRAN, RPh, MHI, MBA
Prince Sultan Military Medical City
Author(s):
Nawaf Alkhayat, MD - PSMMC; Asma Aloqail, BSc (IT), MSc (SWE) - PSMMC; Ashique Khan, BTech Computer Science and Engineering - PSMMC; Taleb Alanazi, BSc - PSMMC; Owais Syed, Oracle and Microsoft Certified, PGPDS - PSMMC; Saleh Alharbi, PharmD - PSMMC; Eman Alhumaid, BSc (IT), MSc (IS) - PSMMC; Julnisar Pajiri, BS Computer Science - PSMMC; Norah Alhorishi, PharmD, MHI - PSMMC; Hanan Alotaibi, BSc (IT), MSc (IS) - PSMMC;
Presentation Type: Poster - Regular
Poster Number: 162
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Standards, Terminology, and Interoperability, TEFCA, FHIR, Standards, Terminology, and Interoperability, TEFCA, FHIR, Standards, Terminology, and Interoperability, TEFCA, FHIR
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
This project describes implementation of HL7 v2.3 to FHIR integration for the medication module with the National Platform for Health and Insurance Exchange Services (NPHIES) at Prince Sultan Military Medical City in Riyadh, Saudi Arabia, in collaboration with Lean Business Services. NPHIES is a national FHIR based health information exchange. The initial scope covers outpatient prescription and dispensing messages. Cerner generates HL7 v2.3 ORM and RDS messages that are enriched using Cerner Command Language (CCL) to add missing clinical and demographic fields, then completed in an integration layer through database lookup tables, national value set mappings, and calls to patient and practitioner registry services. An interface engine, Rhapsody 6, converts the enriched HL7 messages into FHIR resources for NPHIES medication services.
A pre go live criterion of 5 percent maximum error rate was used to move from staging to production. In production the error rate is now below this threshold and the team aims to keep it under 1 percent. On average the system processes about 17,000 prescription messages and 5,000 dispensing messages per day. Failed messages are logged to a monitoring database and summarized in automated email and spreadsheet reports reviewed with clinical and integration teams.
The poster will present the architecture, a summary of twelve key challenges and solutions, and a time series of production success and error rates. This experience shows how a medication first transformation layer can connect legacy HL7 environments to a national FHIR based infrastructure without replacing the core electronic health record.
Speaker(s):
Abdullah ALOMRAN, RPh, MHI, MBA
Prince Sultan Military Medical City
Author(s):
Nawaf Alkhayat, MD - PSMMC; Asma Aloqail, BSc (IT), MSc (SWE) - PSMMC; Ashique Khan, BTech Computer Science and Engineering - PSMMC; Taleb Alanazi, BSc - PSMMC; Owais Syed, Oracle and Microsoft Certified, PGPDS - PSMMC; Saleh Alharbi, PharmD - PSMMC; Eman Alhumaid, BSc (IT), MSc (IS) - PSMMC; Julnisar Pajiri, BS Computer Science - PSMMC; Norah Alhorishi, PharmD, MHI - PSMMC; Hanan Alotaibi, BSc (IT), MSc (IS) - PSMMC;
Abdullah
ALOMRAN,
RPh, MHI, MBA - Prince Sultan Military Medical City
Core Terminology Management Features for Non-Terminologists
Presentation Type: Poster - Regular
Poster Number: 164
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Standards, Terminology, and Interoperability, TEFCA, FHIR, Health Data Science, Human Factors and Usability, Infrastructure and Cloud Computing, Quality Informatics and Lean
Primary Track: Big Data for Health
Standard terminologies are essential for decision support, data exchange, and quality analytics, yet many content developers lack the expertise to use complex terminology services effectively. We identified seven capability areas needed for lightweight, intuitive tools that support unified terminology access, flexible value set assembly, and clarity-optimized user experiences. These capabilities reduce dependence on scarce experts and lower implementation risks. Future work will validate and prioritize these needs through prospective surveys and interviews with non-terminologist content developers.
Speaker(s):
Irene N. Zhuo, MD, MS
Semedy, Inc.
Author(s):
Charles Lagor, MD, PhD, MBA - Semedy; Aline Aronsky - Semedy AG; Roberto Rocha, MD, PhD, FACMI - Semedy, Inc.; Irene N. Zhuo, MD, MS - Semedy, Inc.;
Presentation Type: Poster - Regular
Poster Number: 164
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Standards, Terminology, and Interoperability, TEFCA, FHIR, Health Data Science, Human Factors and Usability, Infrastructure and Cloud Computing, Quality Informatics and Lean
Primary Track: Big Data for Health
Standard terminologies are essential for decision support, data exchange, and quality analytics, yet many content developers lack the expertise to use complex terminology services effectively. We identified seven capability areas needed for lightweight, intuitive tools that support unified terminology access, flexible value set assembly, and clarity-optimized user experiences. These capabilities reduce dependence on scarce experts and lower implementation risks. Future work will validate and prioritize these needs through prospective surveys and interviews with non-terminologist content developers.
Speaker(s):
Irene N. Zhuo, MD, MS
Semedy, Inc.
Author(s):
Charles Lagor, MD, PhD, MBA - Semedy; Aline Aronsky - Semedy AG; Roberto Rocha, MD, PhD, FACMI - Semedy, Inc.; Irene N. Zhuo, MD, MS - Semedy, Inc.;
Irene N.
Zhuo,
MD, MS - Semedy, Inc.
Identifying Barriers and Strategies to Optimize Primary Care Telemedicine for People with Dementia
Presentation Type: Poster Invite - Regular
Poster Number: 165
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Telemedicine, Health at Home, and Virtual Care, Outcomes Improvement and Equity, Human Factors and Usability
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
While telemedicine is now a routinely available delivery modality for primary care, little is known about how to tailor use for people with dementia (PWD) given their unique needs. To identify current barriers to optimal use and strategies to address them, we combined a literature review and structured set of expert panel sessions. Final results include six barriers and nine socio-technical strategies that would support more effective and equitable telemedicine use to meet the needs of PWD.
Speaker(s):
Jade Christey, BA
UCSF
Author(s):
Jade Christey, BA - UCSF; Aditi Sriram, MPH - UCSF; Anjali Gopalan, MD - Kaiser Permanente Northern California Division of Research; Mary Reed, DrPH - Kaiser Permanente Division of Research; Julia Adler-Milstein, PhD, FACMI - UCSF School of Medicine;
Presentation Type: Poster Invite - Regular
Poster Number: 165
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Telemedicine, Health at Home, and Virtual Care, Outcomes Improvement and Equity, Human Factors and Usability
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
While telemedicine is now a routinely available delivery modality for primary care, little is known about how to tailor use for people with dementia (PWD) given their unique needs. To identify current barriers to optimal use and strategies to address them, we combined a literature review and structured set of expert panel sessions. Final results include six barriers and nine socio-technical strategies that would support more effective and equitable telemedicine use to meet the needs of PWD.
Speaker(s):
Jade Christey, BA
UCSF
Author(s):
Jade Christey, BA - UCSF; Aditi Sriram, MPH - UCSF; Anjali Gopalan, MD - Kaiser Permanente Northern California Division of Research; Mary Reed, DrPH - Kaiser Permanente Division of Research; Julia Adler-Milstein, PhD, FACMI - UCSF School of Medicine;
Jade
Christey,
BA - UCSF
Building an Electronic Workflow for Minnesota Paid Leave Forms: Insights from Two Minnesota Metropolitan Health Systems
Presentation Type: Poster - Student
Poster Number: 166
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Workforce Automation, Communication, and Workflow Efficiency, Innovation Partnerships, Implementation Science, and Learning Health Systems, Change Management
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
The Minnesota Paid Leave Program will add additional administrative burdens due to increased medical certification requests. To address this issue, Hennepin Healthcare and M Health Fairview, in collaboration with the state of Minnesota, developed two workflows that use the state’s FHIR-based exchanges to streamline documentation, reduce errors, improve efficiency, decrease delays, and enhance patient and provider satisfaction. Planned evaluation of these two integrations will provide valuable insight into the performance of cross-sector FHIR integrations and into patients' and providers' perceptions.
Speaker(s):
Tony Huy Nguyen, MD
University of Minnesota Clinical Health Informatic Fellowship
Author(s):
Tony Huy Nguyen, MD - University of Minnesota Clinical Health Informatic Fellowship; Justine Mrosak, MD - Hennepin Healthcare; Bryan Jarabek, MD, PHD - Fairview;
Presentation Type: Poster - Student
Poster Number: 166
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Workforce Automation, Communication, and Workflow Efficiency, Innovation Partnerships, Implementation Science, and Learning Health Systems, Change Management
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
The Minnesota Paid Leave Program will add additional administrative burdens due to increased medical certification requests. To address this issue, Hennepin Healthcare and M Health Fairview, in collaboration with the state of Minnesota, developed two workflows that use the state’s FHIR-based exchanges to streamline documentation, reduce errors, improve efficiency, decrease delays, and enhance patient and provider satisfaction. Planned evaluation of these two integrations will provide valuable insight into the performance of cross-sector FHIR integrations and into patients' and providers' perceptions.
Speaker(s):
Tony Huy Nguyen, MD
University of Minnesota Clinical Health Informatic Fellowship
Author(s):
Tony Huy Nguyen, MD - University of Minnesota Clinical Health Informatic Fellowship; Justine Mrosak, MD - Hennepin Healthcare; Bryan Jarabek, MD, PHD - Fairview;
Tony Huy
Nguyen,
MD - University of Minnesota Clinical Health Informatic Fellowship
Optimizing Inpatient Secure Messaging Workflow via Native Mobile OS Analytics and Socio-Technical Intervention
Presentation Type: Poster - Regular
Poster Number: 167
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Workforce Automation, Communication, and Workflow Efficiency, Quality Informatics and Lean, Clinician Well-Being, Human Factors and Usability, Change Management
Primary Track: Advancing Wellness for Providers and Community with Consideration of Human Factors
Inpatient secure messaging contributes to notification burden and cognitive fragmentation. On a general medicine service, we used consumer mobile operating system analytics (iOS Screen Time) to quantify 58 interruptions per 12-hour shift in a vendor-independent, novel approach. We then implemented a socio-technical intervention using native Voalte app features (busy status, plan-of-care broadcast templates) to improve signal-to-noise. Future work will apply LLM-based message classification to analyze communication patterns and identify additional intervention targets.
Speaker(s):
Jonathan Colston, MD
Stanford University
Author(s):
Jonathan Colston, MD - Stanford University; Casey Ren, MD - Stanford Medicine; Oluseyi Fayanju, MD - Stanford University; Sam Sandhu, PA-C - Stanford Medicine; Chloe Planche, RN, BSN - Stanford Medicine; Tiffany Lee, MD - Stanford Medicine; Pamela Gallant, RN, BSN - Stanford Medicine; Angela Chen, RN, BSN, PHN - Stanford Medicine; Yingjie Weng, MHS - Stanford University School Of Medicine - - Stanford, CA; Stephen Ma, MD, PhD - Stanford University School of Medicine; April Liang, MD - Stanford University; Lisa Shieh, MD, PhD - Stanford Medicine; Christopher Sharp, MD - Stanford University School of Medicine;
Presentation Type: Poster - Regular
Poster Number: 167
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Workforce Automation, Communication, and Workflow Efficiency, Quality Informatics and Lean, Clinician Well-Being, Human Factors and Usability, Change Management
Primary Track: Advancing Wellness for Providers and Community with Consideration of Human Factors
Inpatient secure messaging contributes to notification burden and cognitive fragmentation. On a general medicine service, we used consumer mobile operating system analytics (iOS Screen Time) to quantify 58 interruptions per 12-hour shift in a vendor-independent, novel approach. We then implemented a socio-technical intervention using native Voalte app features (busy status, plan-of-care broadcast templates) to improve signal-to-noise. Future work will apply LLM-based message classification to analyze communication patterns and identify additional intervention targets.
Speaker(s):
Jonathan Colston, MD
Stanford University
Author(s):
Jonathan Colston, MD - Stanford University; Casey Ren, MD - Stanford Medicine; Oluseyi Fayanju, MD - Stanford University; Sam Sandhu, PA-C - Stanford Medicine; Chloe Planche, RN, BSN - Stanford Medicine; Tiffany Lee, MD - Stanford Medicine; Pamela Gallant, RN, BSN - Stanford Medicine; Angela Chen, RN, BSN, PHN - Stanford Medicine; Yingjie Weng, MHS - Stanford University School Of Medicine - - Stanford, CA; Stephen Ma, MD, PhD - Stanford University School of Medicine; April Liang, MD - Stanford University; Lisa Shieh, MD, PhD - Stanford Medicine; Christopher Sharp, MD - Stanford University School of Medicine;
Jonathan
Colston,
MD - Stanford University
ChatUCM: AI-Powered Hospital Staff Support Chatbot
Presentation Type: Poster Invite - Regular
Poster Number: 168
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Workforce Automation, Communication, and Workflow Efficiency, Education and Training, Innovation Partnerships, Implementation Science, and Learning Health Systems, Clinical Decision Support and Care Pathways
Primary Track: Innovation through Industry, Public Health, Non-profit and Commercial Partnerships
ChatUCM is an AI-powered conversational platform providing hospital staff immediate access to institutional policies, procedures, and safety protocols. Recently deployed, it addresses fragmented information workflows that divert clinicians from patient care. A standardized questionnaire evaluates satisfaction, efficiency, and perceived quality improvements. Early findings demonstrate feasibility of rapid, HIPAA-compliant AI deployment under institutional governance.
Speaker(s):
Samir Atiya, MD
University of Chicago Medical Ceneter
Author(s):
Samir Atiya, MD - University of Chicago Medical Ceneter; Cheng-Kai Kao, MD - University of Chicago;
Presentation Type: Poster Invite - Regular
Poster Number: 168
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Workforce Automation, Communication, and Workflow Efficiency, Education and Training, Innovation Partnerships, Implementation Science, and Learning Health Systems, Clinical Decision Support and Care Pathways
Primary Track: Innovation through Industry, Public Health, Non-profit and Commercial Partnerships
ChatUCM is an AI-powered conversational platform providing hospital staff immediate access to institutional policies, procedures, and safety protocols. Recently deployed, it addresses fragmented information workflows that divert clinicians from patient care. A standardized questionnaire evaluates satisfaction, efficiency, and perceived quality improvements. Early findings demonstrate feasibility of rapid, HIPAA-compliant AI deployment under institutional governance.
Speaker(s):
Samir Atiya, MD
University of Chicago Medical Ceneter
Author(s):
Samir Atiya, MD - University of Chicago Medical Ceneter; Cheng-Kai Kao, MD - University of Chicago;
Samir
Atiya,
MD - University of Chicago Medical Ceneter
Retrieval-Augmented Generation (RAG)-Powered Agent: A Scalable Framework for Modernizing Department-Wide and Hospital-Wide Clinical Operations Support in a Veterans Affairs Healthcare System
Presentation Type: Poster - Student
Poster Number: 169
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Workforce Automation, Communication, and Workflow Efficiency, Clinician Well-Being, Education and Training, Leadership and Strategy, Quality Informatics and Lean
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
At the VA Long Beach Healthcare System, fragmented clinical and EHR support across Teams, SharePoint, and policy repositories creates delays, inconsistent guidance, and manual burden. We developed a retrieval-augmented generation (RAG) AI agent that unifies knowledge retrieval and delivers real-time workflow and policy support. Early results show high response accuracy, successful ingestion of multi-year data, and strong user satisfaction. This proof of concept provides a scalable foundation for department- and enterprise-level operational AI assistants.
Speaker(s):
Roderick Eguilos, DO
University of California, Irvine
Author(s):
Peter Nguyen, MD - VA Long Beach Health Care System; Scott Rudkin, MD, MBA - University of California Irvine Medical Center; Department of Veterans Affairs; Jonathan Pham, DO - VA Long Beach Healthcare System; Vishnu Bharani, MD - University of California, Irvine School of Medicine;
Presentation Type: Poster - Student
Poster Number: 169
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Workforce Automation, Communication, and Workflow Efficiency, Clinician Well-Being, Education and Training, Leadership and Strategy, Quality Informatics and Lean
Primary Track: Implementing Real-World Change, Digital Engagement, and Connected Health
At the VA Long Beach Healthcare System, fragmented clinical and EHR support across Teams, SharePoint, and policy repositories creates delays, inconsistent guidance, and manual burden. We developed a retrieval-augmented generation (RAG) AI agent that unifies knowledge retrieval and delivers real-time workflow and policy support. Early results show high response accuracy, successful ingestion of multi-year data, and strong user satisfaction. This proof of concept provides a scalable foundation for department- and enterprise-level operational AI assistants.
Speaker(s):
Roderick Eguilos, DO
University of California, Irvine
Author(s):
Peter Nguyen, MD - VA Long Beach Health Care System; Scott Rudkin, MD, MBA - University of California Irvine Medical Center; Department of Veterans Affairs; Jonathan Pham, DO - VA Long Beach Healthcare System; Vishnu Bharani, MD - University of California, Irvine School of Medicine;
Roderick
Eguilos,
DO - University of California, Irvine
3D IntelliGenes: The next generation AI/ML application with multidimensional visualization for predictive medicine
Presentation Type: Poster Invite - Regular
Poster Number: 170
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
We present 3D IntelliGenes, a next generation, interactive, customizable, cross-platform, user-friendly, and multidimensional AI/ML application for multi-omics and clinical data exploration to discover biomarkers and predict diseases. The overall computational methodology of 3D IntelliGenes is based on a unique nexus of statistical techniques and cutting-edge AI/ML algorithms. Furthermore, 3D IntelliGenes produces intuitive, interactive and multidimensional visualization, which offers deeper insights, most importantly by capturing greater variability in the patient data by understanding both linear and non-linear structures, evaluating AI/ML model performance, and delineating the joint impact of biomarkers on the corresponding disease states. The friendly graphical user interface of 3D IntelliGenes supports users from diverse backgrounds in applying default and creating customized AI/ML pipelines to discover novel biomarkers and predict diseases. Furthermore, it allows users to generate and export analysis results in variable visual (image) and text formats.
Speaker(s):
Zeeshan Ahmed, PhD
Department of Medicine, Robert Wood Johnson Medical School. Institute for Health, Health Care Policy and Aging Research. Rutgers Health
Author(s):
Zeeshan Ahmed, PhD - Department of Medicine, Robert Wood Johnson Medical School. Institute for Health, Health Care Policy and Aging Research. Rutgers Health;
Presentation Type: Poster Invite - Regular
Poster Number: 170
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
We present 3D IntelliGenes, a next generation, interactive, customizable, cross-platform, user-friendly, and multidimensional AI/ML application for multi-omics and clinical data exploration to discover biomarkers and predict diseases. The overall computational methodology of 3D IntelliGenes is based on a unique nexus of statistical techniques and cutting-edge AI/ML algorithms. Furthermore, 3D IntelliGenes produces intuitive, interactive and multidimensional visualization, which offers deeper insights, most importantly by capturing greater variability in the patient data by understanding both linear and non-linear structures, evaluating AI/ML model performance, and delineating the joint impact of biomarkers on the corresponding disease states. The friendly graphical user interface of 3D IntelliGenes supports users from diverse backgrounds in applying default and creating customized AI/ML pipelines to discover novel biomarkers and predict diseases. Furthermore, it allows users to generate and export analysis results in variable visual (image) and text formats.
Speaker(s):
Zeeshan Ahmed, PhD
Department of Medicine, Robert Wood Johnson Medical School. Institute for Health, Health Care Policy and Aging Research. Rutgers Health
Author(s):
Zeeshan Ahmed, PhD - Department of Medicine, Robert Wood Johnson Medical School. Institute for Health, Health Care Policy and Aging Research. Rutgers Health;
Zeeshan
Ahmed,
PhD - Department of Medicine, Robert Wood Johnson Medical School. Institute for Health, Health Care Policy and Aging Research. Rutgers Health
Design Considerations for Translating Image Segmentation Foundation Models into Reliable and Scalable AI Medical-Imaging Applications in Healthcare
Presentation Type: Poster Invite - Regular
Poster Number: 171
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
The launch of the open-source Segment Anything Model 2 (SAM2) foundation model opens opportunities for pixel-precise, interactive image segmentations in the medical imaging field, enabling rapid generation of large, annotated datasets. However, integrating a SAM2 into clinical workflow is challenging and requires thoughtful design considerations. This study presents bottlenecks and implementation optimization steps for adapting the SAM2 to medical imaging using cone-beam computed tomography (CBCT) scans of the temporomandibular joint (TMJ). We systematically addressed computational constraints across the SAM2 training–testing pipeline, and implemented mixed-precision training, gradient accumulation, multi-threaded data loading, and GPU-based metric computation to improve efficiency and stability. The optimized workflow achieved reduced memory overhead, faster inference latency, and enhanced GPU utilization without compromising segmentation accuracy. Our results highlight the importance of architectural and computational optimization in adapting large foundation models, offering a best practice for reliable, scalable, and time-efficient SAM2 deployment in AI-medical imaging applications.
Speaker(s):
Toufeeq Syed, PhD, MS
UT Health Houston
Author(s):
Zulfiia Ditto, PhD, MS - The University of Texas Health Science Center at Houston; Deevakar Rogith, MBBS, PhD - The University of Texas Health Science Center at Houston (UTHealth) School of Biomedical Informatics; Katie Stinson, MLIS - The University of Texas Health Science Center at Houston; Toufeeq Syed, PhD, MS - UT Health Houston;
Presentation Type: Poster Invite - Regular
Poster Number: 171
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
The launch of the open-source Segment Anything Model 2 (SAM2) foundation model opens opportunities for pixel-precise, interactive image segmentations in the medical imaging field, enabling rapid generation of large, annotated datasets. However, integrating a SAM2 into clinical workflow is challenging and requires thoughtful design considerations. This study presents bottlenecks and implementation optimization steps for adapting the SAM2 to medical imaging using cone-beam computed tomography (CBCT) scans of the temporomandibular joint (TMJ). We systematically addressed computational constraints across the SAM2 training–testing pipeline, and implemented mixed-precision training, gradient accumulation, multi-threaded data loading, and GPU-based metric computation to improve efficiency and stability. The optimized workflow achieved reduced memory overhead, faster inference latency, and enhanced GPU utilization without compromising segmentation accuracy. Our results highlight the importance of architectural and computational optimization in adapting large foundation models, offering a best practice for reliable, scalable, and time-efficient SAM2 deployment in AI-medical imaging applications.
Speaker(s):
Toufeeq Syed, PhD, MS
UT Health Houston
Author(s):
Zulfiia Ditto, PhD, MS - The University of Texas Health Science Center at Houston; Deevakar Rogith, MBBS, PhD - The University of Texas Health Science Center at Houston (UTHealth) School of Biomedical Informatics; Katie Stinson, MLIS - The University of Texas Health Science Center at Houston; Toufeeq Syed, PhD, MS - UT Health Houston;
Toufeeq
Syed,
PhD, MS - UT Health Houston
Application of Machine Learning on Biomarker Discovery for CLN3 Disease
Presentation Type: Poster Invite - Regular
Poster Number: 172
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
CLN3 disease, also known as juvenile neuronal ceroid lipofuscinosis, is a rare and neurodegenerative disorder characterized by the accumulation of lipopigments in cells, progressive cognitive decline, seizures, and vision loss. Biomarkers are critical for elucidating disease mechanisms, monitoring disease progression, and evaluating therapeutic responses. In this study, we introduced a computational framework to programmatically identify fluid biomarker candidates for CLN3 disease using proteomics data and clinical data derived from prospective CLN3 observational clinical trials.
Speaker(s):
Shixue Sun, PhD
NCATS/NIH
Author(s):
Shixue Sun, PhD - NCATS/NIH; An Dang Do, MD, PhD - Eunice Kennedy Shriver National Institute of Child Health and Human Development; Audrey Thurm, Ph.D. - National Institute of Mental Health; Ariane Soldatos, M.D., M.P.H. - National Institute of Mental Health; Qian Zhu - National Institutes of Health;
Presentation Type: Poster Invite - Regular
Poster Number: 172
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
CLN3 disease, also known as juvenile neuronal ceroid lipofuscinosis, is a rare and neurodegenerative disorder characterized by the accumulation of lipopigments in cells, progressive cognitive decline, seizures, and vision loss. Biomarkers are critical for elucidating disease mechanisms, monitoring disease progression, and evaluating therapeutic responses. In this study, we introduced a computational framework to programmatically identify fluid biomarker candidates for CLN3 disease using proteomics data and clinical data derived from prospective CLN3 observational clinical trials.
Speaker(s):
Shixue Sun, PhD
NCATS/NIH
Author(s):
Shixue Sun, PhD - NCATS/NIH; An Dang Do, MD, PhD - Eunice Kennedy Shriver National Institute of Child Health and Human Development; Audrey Thurm, Ph.D. - National Institute of Mental Health; Ariane Soldatos, M.D., M.P.H. - National Institute of Mental Health; Qian Zhu - National Institutes of Health;
Shixue
Sun,
PhD - NCATS/NIH
Protein-Protein Interaction Analysis of Cervical Squamous Cell Carcinoma
Presentation Type: Poster - Student
Poster Number: 173
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Translation Bioinformatics/Precision Medicine
Protein-Protein Interactions (PPI) dictate physiological processes throughout the body and are disrupted in malignant cells, allowing invasion and metastatic spread. PPI networks provide further molecular information about pathogenic complexities, and identify unique subtypes of oncogenic processes. This study used Proteinarium, a tool for multi-sample PPI-network analysis, to identify consensus networks for patients with cervical squamous cell carcinoma. Five unique patient subgroups were identified within the TCGA cohort opening a new pathway for individualized targeted therapies.
Speaker(s):
Andrea Llamas Sanchez, B.A.
Warren Alpert Medical School of Brown University
Author(s):
Andrea Llamas Sanchez, B.A. - Warren Alpert Medical School of Brown University; Jessica Claus, M.D. - Brown University Health; Alper Uzun, PhD - Brown, University, Department of Pathology and Laboratory Medicine, Department of Pediatrics; Ece Uzun, PhD - Brown University Health/Brown University;
Presentation Type: Poster - Student
Poster Number: 173
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Translation Bioinformatics/Precision Medicine
Protein-Protein Interactions (PPI) dictate physiological processes throughout the body and are disrupted in malignant cells, allowing invasion and metastatic spread. PPI networks provide further molecular information about pathogenic complexities, and identify unique subtypes of oncogenic processes. This study used Proteinarium, a tool for multi-sample PPI-network analysis, to identify consensus networks for patients with cervical squamous cell carcinoma. Five unique patient subgroups were identified within the TCGA cohort opening a new pathway for individualized targeted therapies.
Speaker(s):
Andrea Llamas Sanchez, B.A.
Warren Alpert Medical School of Brown University
Author(s):
Andrea Llamas Sanchez, B.A. - Warren Alpert Medical School of Brown University; Jessica Claus, M.D. - Brown University Health; Alper Uzun, PhD - Brown, University, Department of Pathology and Laboratory Medicine, Department of Pediatrics; Ece Uzun, PhD - Brown University Health/Brown University;
Andrea
Llamas Sanchez,
B.A. - Warren Alpert Medical School of Brown University
Impact of Cigarette Smoke on Host Immune Response to Tuberculosis: A Transcriptomic Study
Presentation Type: Poster Invite - Regular
Poster Number: 174
Presentation Time: 05:00 PM - 06:30 PM
Working Group: Genomics and Translational Bioinformatics Working Group
Primary Track: Translation Bioinformatics/Precision Medicine
Tuberculosis (TB) remains a global health threat, with smoking recognized as a major modifiable risk factor. In this study, we used bulk RNA-seq to examine gene expression changes in macrophages exposed to cigarette smoke during M. tuberculosis infection. We observed significant downregulation of key immune pathways, including interleukin and interferon signaling. These findings highlight how smoking may impair host immune responses, contributing to worsened TB outcomes.
Speaker(s):
Sadia Akter, PhD, FAMIA
Marshall University Joan C. Edward School of Medicine
Author(s):
Hadi Ul Bashar, Graduate Student - Marshall University; Annu Devi, PhD - Texas Biomedical Research Institute; Mushtaq Ahmed, PhD - The University of Chicago; Larry S. Schlesinger, MS - Texas Biomedical Research Institute; Shabaana A. Khader, PhD - The University of Chicago; Deepak Kaushal, PhD - Texas Biomedical Research Institute; Smriti Mehra, PhD - Texas Biomedical Research Institute;
Presentation Type: Poster Invite - Regular
Poster Number: 174
Presentation Time: 05:00 PM - 06:30 PM
Working Group: Genomics and Translational Bioinformatics Working Group
Primary Track: Translation Bioinformatics/Precision Medicine
Tuberculosis (TB) remains a global health threat, with smoking recognized as a major modifiable risk factor. In this study, we used bulk RNA-seq to examine gene expression changes in macrophages exposed to cigarette smoke during M. tuberculosis infection. We observed significant downregulation of key immune pathways, including interleukin and interferon signaling. These findings highlight how smoking may impair host immune responses, contributing to worsened TB outcomes.
Speaker(s):
Sadia Akter, PhD, FAMIA
Marshall University Joan C. Edward School of Medicine
Author(s):
Hadi Ul Bashar, Graduate Student - Marshall University; Annu Devi, PhD - Texas Biomedical Research Institute; Mushtaq Ahmed, PhD - The University of Chicago; Larry S. Schlesinger, MS - Texas Biomedical Research Institute; Shabaana A. Khader, PhD - The University of Chicago; Deepak Kaushal, PhD - Texas Biomedical Research Institute; Smriti Mehra, PhD - Texas Biomedical Research Institute;
Sadia
Akter,
PhD, FAMIA - Marshall University Joan C. Edward School of Medicine
Bridging the Clinical Expertise Gap: Development of a Web-Based Platform for Accessible Time Series Forecasting and Analysis
Presentation Type: Poster Invite - Regular
Poster Number: 175
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Time series forecasting has applications across domains and industries, especially in healthcare, but the technical expertise required to analyze data, build models, and interpret results can be a barrier to using these techniques. This article presents a web platform that makes the process of analyzing and plotting data, training forecasting models, and interpreting and viewing results accessible to researchers and clinicians. Users can upload data and generate plots to showcase their variables and the relationships between them. The platform supports multiple forecasting models and training techniques which are highly customizable according to the user’s needs. Additionally, recommendations and explanations can be generated from a large language model that can help the user choose appropriate parameters for their data and understand the results for each model. The goal is to integrate this platform into learning health systems for continuous data collection and inference from clinical pipelines.
Speaker(s):
Aaron Mullen, M.S.
University of Kentucky
Author(s):
Daniel Harris, PhD - University of Kentucky; Svetla Slavova, PhD - University of Kentucky; Cody Bumgardner, PhD - University of Kentucky;
Presentation Type: Poster Invite - Regular
Poster Number: 175
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Time series forecasting has applications across domains and industries, especially in healthcare, but the technical expertise required to analyze data, build models, and interpret results can be a barrier to using these techniques. This article presents a web platform that makes the process of analyzing and plotting data, training forecasting models, and interpreting and viewing results accessible to researchers and clinicians. Users can upload data and generate plots to showcase their variables and the relationships between them. The platform supports multiple forecasting models and training techniques which are highly customizable according to the user’s needs. Additionally, recommendations and explanations can be generated from a large language model that can help the user choose appropriate parameters for their data and understand the results for each model. The goal is to integrate this platform into learning health systems for continuous data collection and inference from clinical pipelines.
Speaker(s):
Aaron Mullen, M.S.
University of Kentucky
Author(s):
Daniel Harris, PhD - University of Kentucky; Svetla Slavova, PhD - University of Kentucky; Cody Bumgardner, PhD - University of Kentucky;
Aaron
Mullen,
M.S. - University of Kentucky
POI-KB: Revolutionizing Perioperative Care Through AI-Powered Multi-Agent Evidence Synthesis and Real-Time Knowledge Discovery
Presentation Type: Poster - Regular
Poster Number: 176
Presentation Time: 05:00 PM - 06:30 PM
Working Group: Knowledge Discovery and Data Mining Working Group
Primary Track: Data Science/Artificial Intelligence
Background/Objectives: Perioperative organ injury (POI)—encompassing stroke, myocardial injury (MINS), ARDS, AKI, gut injury (AGI), SIRS, POCD, and IRS—affects 5-30% of surgical patients, elevating morbidity, mortality, and costs ($5-10B annually). Evidence synthesis lags due to publication volume and manual biases. We developed POI-KB, a multi-agent AI platform using BioMCP and Model Context Protocol (MCP) for automated literature mining, entity extraction, and real-time knowledge base (KB) construction to uncover POI mechanisms and support clinical translation.
Methods: Boolean searches ("perioperative" OR "postoperative" AND POI subtypes) queried PubMed, Embase, Cochrane, Web of Science, and ClinicalTrials.gov (2000–2025). Hybrid extraction integrated BioMCP for NER (diseases, biomarkers, genes), rule-based patterns, and ML (BioBERT) for relations. MCP-unified agents handled querying, retrieval, normalization (UMLS/MONDO), analysis, and summarization. Workflow: Retrieve records → Extract entities/relations → Build knowledge graphs → Ingest with provenance. QC combined auto-cross-references and manual curation (kappa >0.80).
Results: Across six POI subdomains, 28,500 records yielded 15,200 entities (e.g., 3,450 biomarkers) and 8,700 relations (e.g., IL-6 in ARDS-SIRS). Normalization: 95% ontology mapping; BioMCP F1-score >90%. Graphs revealed understudied links (e.g., gut-lung axis). Real-time updates processed 150 trials in <5 min; projected $2-5M savings/100,000 surgeries.
Conclusion: POI-KB accelerates unbiased POI synthesis, enabling hypothesis generation and ERAS personalization. Scalable for EHR integration, it promises equitable outcomes in high-risk cohorts.
Speaker(s):
jinlian wang, PhD
UTHealth
Author(s):
jinlian wang, PhD - UTHealth;
Presentation Type: Poster - Regular
Poster Number: 176
Presentation Time: 05:00 PM - 06:30 PM
Working Group: Knowledge Discovery and Data Mining Working Group
Primary Track: Data Science/Artificial Intelligence
Background/Objectives: Perioperative organ injury (POI)—encompassing stroke, myocardial injury (MINS), ARDS, AKI, gut injury (AGI), SIRS, POCD, and IRS—affects 5-30% of surgical patients, elevating morbidity, mortality, and costs ($5-10B annually). Evidence synthesis lags due to publication volume and manual biases. We developed POI-KB, a multi-agent AI platform using BioMCP and Model Context Protocol (MCP) for automated literature mining, entity extraction, and real-time knowledge base (KB) construction to uncover POI mechanisms and support clinical translation.
Methods: Boolean searches ("perioperative" OR "postoperative" AND POI subtypes) queried PubMed, Embase, Cochrane, Web of Science, and ClinicalTrials.gov (2000–2025). Hybrid extraction integrated BioMCP for NER (diseases, biomarkers, genes), rule-based patterns, and ML (BioBERT) for relations. MCP-unified agents handled querying, retrieval, normalization (UMLS/MONDO), analysis, and summarization. Workflow: Retrieve records → Extract entities/relations → Build knowledge graphs → Ingest with provenance. QC combined auto-cross-references and manual curation (kappa >0.80).
Results: Across six POI subdomains, 28,500 records yielded 15,200 entities (e.g., 3,450 biomarkers) and 8,700 relations (e.g., IL-6 in ARDS-SIRS). Normalization: 95% ontology mapping; BioMCP F1-score >90%. Graphs revealed understudied links (e.g., gut-lung axis). Real-time updates processed 150 trials in <5 min; projected $2-5M savings/100,000 surgeries.
Conclusion: POI-KB accelerates unbiased POI synthesis, enabling hypothesis generation and ERAS personalization. Scalable for EHR integration, it promises equitable outcomes in high-risk cohorts.
Speaker(s):
jinlian wang, PhD
UTHealth
Author(s):
jinlian wang, PhD - UTHealth;
jinlian
wang,
PhD - UTHealth
Workflow Implications of an AI-Driven Surgical Blood Ordering Tool
Presentation Type: Poster - Regular
Poster Number: 177
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
This qualitative study examined anticipated clinical and workflow implications of integrating a machine learning–based personalized Maximum Surgical Blood Order Schedule into the EHR. Through multidisciplinary focus groups analyzed using the SEIPS 101 framework, clinicians highlighted potential improvements in standardizing blood-ordering practices and workflow efficiency, while noting challenges such as verification burden and trust in predictive accuracy. Findings inform sociotechnical readiness for seamless clinical–research integration.
Speaker(s):
Ye-Eun Park, BA
Asan Medical Center
Author(s):
Ye-Eun Park, BA - Asan Medical Center; Minsu Ock, PhD - Ulsan University Hospital; Jae-Ho Lee, MD, PhD - Asan Medical Center; Dae-Hyun Ko, MD, PhD - Asan Medical Center; Hak-Jae Lee, MD, PhD - Asan Medical Center; Taezoon Park, PhD - Soongsil University; Junsang Yoo, RN, PhD - Samsung Advanced Institute for Health Sciences; Yura Lee, MD, PhD - Asan Medical Center;
Presentation Type: Poster - Regular
Poster Number: 177
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
This qualitative study examined anticipated clinical and workflow implications of integrating a machine learning–based personalized Maximum Surgical Blood Order Schedule into the EHR. Through multidisciplinary focus groups analyzed using the SEIPS 101 framework, clinicians highlighted potential improvements in standardizing blood-ordering practices and workflow efficiency, while noting challenges such as verification burden and trust in predictive accuracy. Findings inform sociotechnical readiness for seamless clinical–research integration.
Speaker(s):
Ye-Eun Park, BA
Asan Medical Center
Author(s):
Ye-Eun Park, BA - Asan Medical Center; Minsu Ock, PhD - Ulsan University Hospital; Jae-Ho Lee, MD, PhD - Asan Medical Center; Dae-Hyun Ko, MD, PhD - Asan Medical Center; Hak-Jae Lee, MD, PhD - Asan Medical Center; Taezoon Park, PhD - Soongsil University; Junsang Yoo, RN, PhD - Samsung Advanced Institute for Health Sciences; Yura Lee, MD, PhD - Asan Medical Center;
Ye-Eun
Park,
BA - Asan Medical Center
Understanding Clinician Perspectives on Predictive Models: A Pilot Study
Presentation Type: Poster Invite - Regular
Poster Number: 178
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
Machine learning (ML) models are increasingly proposed for clinical decision support, yet their impact on clinician judgment remains unclear. In this pilot study, 21 clinicians evaluated post-surgical VTE risk across three stages: clinical data alone, data plus ML risk scores, and data plus model explanations. Mixed-effects modeling showed that ML assistance improved differentiation between VTE-positive and VTE-negative patients, with larger adjustments among less-experienced clinicians. Findings highlight the importance of experience-specific strategies for ML implementation.
Speaker(s):
Smitha Edakalavan, PhD
University of Pittsburgh
Author(s):
Shyam Visweswaran, MD PhD - University of Pittsburgh; Rafael Ceschin, PhD - University of Pittsburgh; Gregory Cooper, MD - University of Pittsburgh; Ryan Zeh, MD - UPMC/University of Pittsburgh;
Presentation Type: Poster Invite - Regular
Poster Number: 178
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
Machine learning (ML) models are increasingly proposed for clinical decision support, yet their impact on clinician judgment remains unclear. In this pilot study, 21 clinicians evaluated post-surgical VTE risk across three stages: clinical data alone, data plus ML risk scores, and data plus model explanations. Mixed-effects modeling showed that ML assistance improved differentiation between VTE-positive and VTE-negative patients, with larger adjustments among less-experienced clinicians. Findings highlight the importance of experience-specific strategies for ML implementation.
Speaker(s):
Smitha Edakalavan, PhD
University of Pittsburgh
Author(s):
Shyam Visweswaran, MD PhD - University of Pittsburgh; Rafael Ceschin, PhD - University of Pittsburgh; Gregory Cooper, MD - University of Pittsburgh; Ryan Zeh, MD - UPMC/University of Pittsburgh;
Smitha
Edakalavan,
PhD - University of Pittsburgh
Bridging Repositories: Enabling Access to Sensitive Data through Federated Governance on the CLASSIC Portal
Presentation Type: Poster - Regular
Poster Number: 179
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Sage Bionetworks deployed a federated, CLASSIC Portal1 in a standards-aligned, controlled-access infrastructure, compliant with NIH data sharing requirements. These include authentication and authorization services that interoperate across data stores. This work aims to do the following:
- Improve data discoverability and accessibility by indexing data in a centralized portal and metadata catalog
- Support consistent policy enforcement for data access committees through modular and delegated governance controls
- Demonstrate a sustainable, extensible model for sharing deeply phenotyped, longitudinal social and behavioral data on aging
Speaker(s):
Ann Novakowski, MPH
Sage Bionetworks
Author(s):
Samia Ahmed, BS - Sage Bionetworks; Jessica Malenfant, MPH - Sage Bionetworks; Amelia Kallaher, MS - Sage Bionetworks; Christine Brugh, PhD - North Carolina State University; Emily Keller, BS - North Carolina State University; Shevaun Neupert, PhD - North Carolina State University; Stacey Scott, PhD - Stony Brook University; Ann Novakowski, MPH - Sage Bionetworks;
Presentation Type: Poster - Regular
Poster Number: 179
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Sage Bionetworks deployed a federated, CLASSIC Portal1 in a standards-aligned, controlled-access infrastructure, compliant with NIH data sharing requirements. These include authentication and authorization services that interoperate across data stores. This work aims to do the following:
- Improve data discoverability and accessibility by indexing data in a centralized portal and metadata catalog
- Support consistent policy enforcement for data access committees through modular and delegated governance controls
- Demonstrate a sustainable, extensible model for sharing deeply phenotyped, longitudinal social and behavioral data on aging
Speaker(s):
Ann Novakowski, MPH
Sage Bionetworks
Author(s):
Samia Ahmed, BS - Sage Bionetworks; Jessica Malenfant, MPH - Sage Bionetworks; Amelia Kallaher, MS - Sage Bionetworks; Christine Brugh, PhD - North Carolina State University; Emily Keller, BS - North Carolina State University; Shevaun Neupert, PhD - North Carolina State University; Stacey Scott, PhD - Stony Brook University; Ann Novakowski, MPH - Sage Bionetworks;
Ann
Novakowski,
MPH - Sage Bionetworks
Drug-COIN: Drug Co-prescription and Interaction Network
Presentation Type: Poster Invite - Regular
Click to View Presentation
Poster Number: 181
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Current approaches to knowledge graph (KG) construction are limited to generic techniques that are often not based on real-world data sources such as electronic health records (EHR). As EHR contains heterogeneous real-world information on the health status and history of patients, EHR-generated KGs are more capable of ensuring better results in subsequent Artificial Intelligence/Machine Learning (AI/ML) tasks like knowledge extraction and inference, diagnosis prediction, and medical decision support. To demonstrate how to construct a KG from real-world EHR data, we used the Department of Veteran Affairs’ (VA) Corporate Data Warehouse (CDW) and DrugBank to build a drug co-prescription and interaction network (Drug-COIN) as part of a foundational EHR-based knowledge graph. Drug-drug interactions (DDIs) play a key role in patient safety, as they can lead to unpredictable pharmacological effects and adverse drug reactions, which are significant causes of mortality and morbidity. We created the Drug-COIN KG by merging two graphs: i) a drug co-prescription graph that was built using the patient-level prescription data in 2024 for a subset of VA drug classes, and ii) a drug-drug interaction graph that was constructed from a set of DrugBank files. The resulting KG has >300,000 entities and ~1 million relationships. We demonstrated the strength of graph queries in identifying drug-drug interaction paths recursively. We also developed a web-based dashboard to support dynamic graph filtering and visualization. Our approach represents a major step in creating an enterprise-wide EHR-based knowledge graph to support AI/ML application development/enrichment and high-impact clinical use cases.
Speaker(s):
Kelson Zawack, PhD
Veterans Affairs Cooperative Studies Program Clinical Epidemiology Research Center (CSP-CERC)
Author(s):
Farah Kidwai-Khan, DEng - VA Connecticut/Yale University; Lei Yan, PhD - Yale University; Mihaela Aslan, PhD - VA Connecticut/Yale University; Kelson Zawack, PhD - VA Connecticut;
Presentation Type: Poster Invite - Regular
Click to View Presentation
Poster Number: 181
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Current approaches to knowledge graph (KG) construction are limited to generic techniques that are often not based on real-world data sources such as electronic health records (EHR). As EHR contains heterogeneous real-world information on the health status and history of patients, EHR-generated KGs are more capable of ensuring better results in subsequent Artificial Intelligence/Machine Learning (AI/ML) tasks like knowledge extraction and inference, diagnosis prediction, and medical decision support. To demonstrate how to construct a KG from real-world EHR data, we used the Department of Veteran Affairs’ (VA) Corporate Data Warehouse (CDW) and DrugBank to build a drug co-prescription and interaction network (Drug-COIN) as part of a foundational EHR-based knowledge graph. Drug-drug interactions (DDIs) play a key role in patient safety, as they can lead to unpredictable pharmacological effects and adverse drug reactions, which are significant causes of mortality and morbidity. We created the Drug-COIN KG by merging two graphs: i) a drug co-prescription graph that was built using the patient-level prescription data in 2024 for a subset of VA drug classes, and ii) a drug-drug interaction graph that was constructed from a set of DrugBank files. The resulting KG has >300,000 entities and ~1 million relationships. We demonstrated the strength of graph queries in identifying drug-drug interaction paths recursively. We also developed a web-based dashboard to support dynamic graph filtering and visualization. Our approach represents a major step in creating an enterprise-wide EHR-based knowledge graph to support AI/ML application development/enrichment and high-impact clinical use cases.
Speaker(s):
Kelson Zawack, PhD
Veterans Affairs Cooperative Studies Program Clinical Epidemiology Research Center (CSP-CERC)
Author(s):
Farah Kidwai-Khan, DEng - VA Connecticut/Yale University; Lei Yan, PhD - Yale University; Mihaela Aslan, PhD - VA Connecticut/Yale University; Kelson Zawack, PhD - VA Connecticut;
Kelson
Zawack,
PhD - Veterans Affairs Cooperative Studies Program Clinical Epidemiology Research Center (CSP-CERC)
Query-Optimized Integrable Ledger for Federated Analysis
Presentation Type: Poster Invite - Regular
Poster Number: 182
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
Collaborative biomedical analysis tools have become increasingly important, allowing collaborators to perform joint analyses of their genomic data without sharing patient-level data. In this study, we develop a blockchain-based ledger designed for collaborative biomedical analysis process recording, that adopts smart contracts to reduce query time, and is integrable into existing federated learning systems. Our results show our ledger improves querying speeds by 15–80 times over our baseline, without affecting speed of adding to ledger.
Speaker(s):
Juliet Lam, BS
Yale
Author(s):
Lana Kareem, Bachelor's of Computer Science & Engineering - Yale School of Medicine; Tsung-Ting Kuo, PhD, FAMIA - Yale University; Hyunghoon Cho, PhD - Yale University; Hua Xu, Ph.D - Yale University; Denis Loginov, M.Sc. - StratoLogics LLC / Yale University; Chi Wing Ng, BS - Yale University; Juliet Lam, BS - Yale University;
Presentation Type: Poster Invite - Regular
Poster Number: 182
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
Collaborative biomedical analysis tools have become increasingly important, allowing collaborators to perform joint analyses of their genomic data without sharing patient-level data. In this study, we develop a blockchain-based ledger designed for collaborative biomedical analysis process recording, that adopts smart contracts to reduce query time, and is integrable into existing federated learning systems. Our results show our ledger improves querying speeds by 15–80 times over our baseline, without affecting speed of adding to ledger.
Speaker(s):
Juliet Lam, BS
Yale
Author(s):
Lana Kareem, Bachelor's of Computer Science & Engineering - Yale School of Medicine; Tsung-Ting Kuo, PhD, FAMIA - Yale University; Hyunghoon Cho, PhD - Yale University; Hua Xu, Ph.D - Yale University; Denis Loginov, M.Sc. - StratoLogics LLC / Yale University; Chi Wing Ng, BS - Yale University; Juliet Lam, BS - Yale University;
Juliet
Lam,
BS - Yale
Downtime Preparedness
Presentation Type: Poster - Student
Poster Number: 183
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
As cybersecurity threats increase across healthcare, hospitals must be prepared for the inevitability of electronic health record (EHR) downtime. Such disruptions can jeopardize patient safety, delay clinical workflows, reduce productivity, and create financial strain. Building cybersecurity resilience and protecting patient information at critical touchpoints are essential strategies. A proactive approach engages frontline staff in the downtime processes. Activities such as regular downtime drills, managing designated downtime computers, and streamlining key workflows strengthen organizational readiness. Preparing in advance rather than reacting during an outage supports safe, efficient care delivery when EHR systems are unavailable.
Speaker(s):
Corey Smith, MS Healthcare Informatics
University of Colorado Anschutz
Author(s):
Corey Smith, MS Healthcare Informatics - University of Colorado Anschutz;
Presentation Type: Poster - Student
Poster Number: 183
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
As cybersecurity threats increase across healthcare, hospitals must be prepared for the inevitability of electronic health record (EHR) downtime. Such disruptions can jeopardize patient safety, delay clinical workflows, reduce productivity, and create financial strain. Building cybersecurity resilience and protecting patient information at critical touchpoints are essential strategies. A proactive approach engages frontline staff in the downtime processes. Activities such as regular downtime drills, managing designated downtime computers, and streamlining key workflows strengthen organizational readiness. Preparing in advance rather than reacting during an outage supports safe, efficient care delivery when EHR systems are unavailable.
Speaker(s):
Corey Smith, MS Healthcare Informatics
University of Colorado Anschutz
Author(s):
Corey Smith, MS Healthcare Informatics - University of Colorado Anschutz;
Corey
Smith,
MS Healthcare Informatics - University of Colorado Anschutz
Exploring Coded Laboratory Data in the NIH - All of Us Researcher Workbench Using Rare Disorder Exemplars
Presentation Type: Poster Invite - Regular
Poster Number: 184
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
The All of Us Research Workbench offers standardized data that can support research on the impact and natural history of rare diseases. To explore the value of these data, we looked at plausibility and completeness of laboratory results for two rare conditions (primary biliary cholangitis and Hashimoto's thyroiditis) with lab tests that in practice tend to be specific for each disease.
Speaker(s):
Rachel Richesson, PhD, MPH, FACMI, FAMIA
University of Michigan Medical School
Author(s):
Komal Maniar, MHI - University of Michigan; Emily Balczewski, BA - University of Michigan; David Hanauer, MD - University of Michigan;
Presentation Type: Poster Invite - Regular
Poster Number: 184
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
The All of Us Research Workbench offers standardized data that can support research on the impact and natural history of rare diseases. To explore the value of these data, we looked at plausibility and completeness of laboratory results for two rare conditions (primary biliary cholangitis and Hashimoto's thyroiditis) with lab tests that in practice tend to be specific for each disease.
Speaker(s):
Rachel Richesson, PhD, MPH, FACMI, FAMIA
University of Michigan Medical School
Author(s):
Komal Maniar, MHI - University of Michigan; Emily Balczewski, BA - University of Michigan; David Hanauer, MD - University of Michigan;
Rachel
Richesson,
PhD, MPH, FACMI, FAMIA - University of Michigan Medical School
Evaluating LLM-Assisted Programming for REDCap-to-i2b2 Data Conversion: A Comparative Case Study in Development Time and Effort
Presentation Type: Poster Invite - Regular
Poster Number: 185
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
We compared an R-based ETL pipeline with an LLM-assisted, prompt-driven approach for converting eMERGE REDCap data to the i2b2 Common Data Model. The LLM-assisted method reduced development time from an estimated 70 to 40 hours (43%), primarily through rapid feature generation via prompt engineering. However, two-thirds of effort involved debugging and validation of LLM-generated code. Both pipelines produced equivalent ontologies with identical patient counts. Results highlight trade-offs between accelerated feature creation and increased validation burden.
Speaker(s):
Jeffrey Klann, PhD
Massachusetts General Hospital
Author(s):
Jeffrey Klann, PhD - Massachusetts General Hospital; Victor Castro, MS - Mass General Brigham; Michael Mendis, BS - Mass General Brigham; Shawn Murphy, MD, Ph.D. - Massachusetts General Hospital;
Presentation Type: Poster Invite - Regular
Poster Number: 185
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
We compared an R-based ETL pipeline with an LLM-assisted, prompt-driven approach for converting eMERGE REDCap data to the i2b2 Common Data Model. The LLM-assisted method reduced development time from an estimated 70 to 40 hours (43%), primarily through rapid feature generation via prompt engineering. However, two-thirds of effort involved debugging and validation of LLM-generated code. Both pipelines produced equivalent ontologies with identical patient counts. Results highlight trade-offs between accelerated feature creation and increased validation burden.
Speaker(s):
Jeffrey Klann, PhD
Massachusetts General Hospital
Author(s):
Jeffrey Klann, PhD - Massachusetts General Hospital; Victor Castro, MS - Mass General Brigham; Michael Mendis, BS - Mass General Brigham; Shawn Murphy, MD, Ph.D. - Massachusetts General Hospital;
Jeffrey
Klann,
PhD - Massachusetts General Hospital
RWD-Driven Catchment Area Modeling to Inform Clinical Trial Site Selection and Enrollment Forecasting
Presentation Type: Poster - Regular
Poster Number: 186
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Clinical trial feasibility assessments often rely on historical site performance or subjective estimates that do not accurately reflect where eligible patients live or which sites they can realistically access. Real-world data (RWD) offers a powerful opportunity to ground feasibility decisions in observed care patterns, disease burden, and geographic accessibility. In this work, we present a scalable, informatics-driven framework that integrates claims data, census demographics, diagnostic timelines, and operational trial metadata to derive geospatially defined catchment areas and generate machine-learning–ready features for site selection. We assigned each patient a primary ZIP code and modeled road-network distances to identify the nearest accessible site. Catchment areas were constructed by linking ZIP populations to their most feasible site, producing realistic representations of patient flow. Within each catchment, we aggregated therapeutic-area–specific patient counts, socioeconomic context, utilization patterns, and diagnostic trajectories to create explainable features optimized for predictive modeling. Across therapeutic areas, the RWD-driven approach revealed differences between administrative site boundaries and the true distribution of eligible patients. These enriched features improved interpretability, enhanced ML-based feasibility predictions, and enabled more equitable identification of underserved regions. To demonstrate extensibility, we conducted a population-level analysis of prenatal testing and newborn spinal muscular atrophy (SMA) screening using aggregated ZIP-level claims, illustrating how this framework can surface broader care patterns while maintaining privacy. This work shows how integrating RWE with operational data can reimagine clinical trial feasibility and support more data-informed, equitable site strategies.
Speaker(s):
Saranya Duraisamy, Master of Science
Biogen
Author(s):
Cheyenne Solomon, MS - Biogen; David Clifford, Master of Science - Biogen; Ellen Tworkoski, MS - Biogen; Li Li, Master of Science - Biogen;
Presentation Type: Poster - Regular
Poster Number: 186
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Clinical trial feasibility assessments often rely on historical site performance or subjective estimates that do not accurately reflect where eligible patients live or which sites they can realistically access. Real-world data (RWD) offers a powerful opportunity to ground feasibility decisions in observed care patterns, disease burden, and geographic accessibility. In this work, we present a scalable, informatics-driven framework that integrates claims data, census demographics, diagnostic timelines, and operational trial metadata to derive geospatially defined catchment areas and generate machine-learning–ready features for site selection. We assigned each patient a primary ZIP code and modeled road-network distances to identify the nearest accessible site. Catchment areas were constructed by linking ZIP populations to their most feasible site, producing realistic representations of patient flow. Within each catchment, we aggregated therapeutic-area–specific patient counts, socioeconomic context, utilization patterns, and diagnostic trajectories to create explainable features optimized for predictive modeling. Across therapeutic areas, the RWD-driven approach revealed differences between administrative site boundaries and the true distribution of eligible patients. These enriched features improved interpretability, enhanced ML-based feasibility predictions, and enabled more equitable identification of underserved regions. To demonstrate extensibility, we conducted a population-level analysis of prenatal testing and newborn spinal muscular atrophy (SMA) screening using aggregated ZIP-level claims, illustrating how this framework can surface broader care patterns while maintaining privacy. This work shows how integrating RWE with operational data can reimagine clinical trial feasibility and support more data-informed, equitable site strategies.
Speaker(s):
Saranya Duraisamy, Master of Science
Biogen
Author(s):
Cheyenne Solomon, MS - Biogen; David Clifford, Master of Science - Biogen; Ellen Tworkoski, MS - Biogen; Li Li, Master of Science - Biogen;
Saranya
Duraisamy,
Master of Science - Biogen
SMART: A Disease-Agnostic, Translational Informatics Platform for Symptom Self-Management
Presentation Type: Poster Invite - Regular
Poster Number: 187
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
The Self-Management and Research Technology (SMART) platform is a disease-agnostic, patient-centered mobile app designed to support symptom self-management across diverse populations. Developed using human-centered design and stakeholder engagement, SMART features educational resources, symptom tracking, and peer support. Piloted in cancer, wellness, diabetes, and heart failure populations, SMART enables rapid adaptation, collaborative research, and improved outcomes, particularly for rural and underserved communities.
Speaker(s):
Heath Davis, MS, MLIS, FAMIA
University of Iowa, Carver College of Medicine
Author(s):
Stephanie Gilbertson-White, PhD, APRN-BC, FAAN - University of Iowa, College of Nursing and Internal Medicine; Heath Davis, MS, MLIS, FAMIA - University of Iowa, Carver College of Medicine;
Presentation Type: Poster Invite - Regular
Poster Number: 187
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
The Self-Management and Research Technology (SMART) platform is a disease-agnostic, patient-centered mobile app designed to support symptom self-management across diverse populations. Developed using human-centered design and stakeholder engagement, SMART features educational resources, symptom tracking, and peer support. Piloted in cancer, wellness, diabetes, and heart failure populations, SMART enables rapid adaptation, collaborative research, and improved outcomes, particularly for rural and underserved communities.
Speaker(s):
Heath Davis, MS, MLIS, FAMIA
University of Iowa, Carver College of Medicine
Author(s):
Stephanie Gilbertson-White, PhD, APRN-BC, FAAN - University of Iowa, College of Nursing and Internal Medicine; Heath Davis, MS, MLIS, FAMIA - University of Iowa, Carver College of Medicine;
Heath
Davis,
MS, MLIS, FAMIA - University of Iowa, Carver College of Medicine
Human-in-the-Loop in Healthcare AI: Measuring Involvement for Smarter Integration
Presentation Type: Poster Invite - Student
Poster Number: 188
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
As artificial intelligence (AI) becomes increasingly integrated into healthcare, human-in-the-loop (HITL) frameworks have emerged to overcome the limitations of full automation by embedding human expertise at various stages of the AI lifecycle. This systematic review analyzed 50 original studies from PubMed to assess the extent and stage of human involvement in healthcare AI applications. HITL engagement was most frequent during model development (68%) and data preparation (38%), with fewer studies addressing design (14%), feature engineering (2%), evaluation (16%), and deployment (16%). Most systems operated at Level 1 (AI suggests, human acts) or Level 2 (AI acts, human confirms), reflecting continued reliance on human oversight to ensure safety and reliability. A two-axis framework combining lifecycle stages and initiative levels was developed to guide standardized evaluation of HITL integration, emphasizing that meaningful human participation remains essential for transparent and trustworthy AI in healthcare.
Speaker(s):
Deevakar Rogith, MBBS, PhD
McWilliams School of Biomedical Informatics at The University of Texas Health Science Center at Houston (UTHealth)
Author(s):
Hien V. Tran, MS - The University of Texas Health Science Center at Houston; Deevakar Rogith, MBBS, PhD - The University of Texas Health Science Center at Houston (UTHealth) School of Biomedical Informatics; Zulfiia Ditto, PhD, MS - The University of Texas Health Science Center at Houston; Adib Shafipour, MS - The University of Texas Health Science Center at Houston; Katie Stinson, MLIS - The University of Texas Health Science Center at Houston; Toufeeq Syed, PhD, MS - UT Health Houston;
Presentation Type: Poster Invite - Student
Poster Number: 188
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
As artificial intelligence (AI) becomes increasingly integrated into healthcare, human-in-the-loop (HITL) frameworks have emerged to overcome the limitations of full automation by embedding human expertise at various stages of the AI lifecycle. This systematic review analyzed 50 original studies from PubMed to assess the extent and stage of human involvement in healthcare AI applications. HITL engagement was most frequent during model development (68%) and data preparation (38%), with fewer studies addressing design (14%), feature engineering (2%), evaluation (16%), and deployment (16%). Most systems operated at Level 1 (AI suggests, human acts) or Level 2 (AI acts, human confirms), reflecting continued reliance on human oversight to ensure safety and reliability. A two-axis framework combining lifecycle stages and initiative levels was developed to guide standardized evaluation of HITL integration, emphasizing that meaningful human participation remains essential for transparent and trustworthy AI in healthcare.
Speaker(s):
Deevakar Rogith, MBBS, PhD
McWilliams School of Biomedical Informatics at The University of Texas Health Science Center at Houston (UTHealth)
Author(s):
Hien V. Tran, MS - The University of Texas Health Science Center at Houston; Deevakar Rogith, MBBS, PhD - The University of Texas Health Science Center at Houston (UTHealth) School of Biomedical Informatics; Zulfiia Ditto, PhD, MS - The University of Texas Health Science Center at Houston; Adib Shafipour, MS - The University of Texas Health Science Center at Houston; Katie Stinson, MLIS - The University of Texas Health Science Center at Houston; Toufeeq Syed, PhD, MS - UT Health Houston;
Deevakar
Rogith,
MBBS, PhD - McWilliams School of Biomedical Informatics at The University of Texas Health Science Center at Houston (UTHealth)
‘Vibe Coding’ is Viable Option for Boosting Software Development Productivity and Enhancing Capabilities in Clinical Research Software
Presentation Type: Poster - Regular
Poster Number: 189
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
‘Vibe coding’ is an informal, iterative technique where users generate code with large language models (LLMs) by describing desired features. It offers a promising avenue for rapid software prototyping in clinical research. We employed vibe coding to develop an interactive timeline feature for EMERSE, an enterprise-level electronic medical record search engine. A non-developer iteratively prompted various LLMs (Claude 3.7 Sonnet, GPT-4o, DeepSeek V3, Grok 3), resulting in a locally testable web-based prototype. Despite reaching limits in LLM performance as code complexity increased, a senior developer successfully refined and finalized the feature from the model-generated code. Vibe coding facilitated brainstorming and UI experimentation without the need for initial developer input and proved more dynamic than static wireframing. While some coding knowledge aids the process and LLMs have scalability limitations, vibe coding demonstrated itself to be a cost-effective and efficient strategy for boosting development productivity and feature exploration.
Speaker(s):
David Hanauer, MD
University of Michigan
Author(s):
David Hanauer, MD - University of Michigan; Lisa Ferguson, MS - University of Michigan; Kellen McClain - University of Michigan; Guan Wang, MS - University of Michigan;
Presentation Type: Poster - Regular
Poster Number: 189
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
‘Vibe coding’ is an informal, iterative technique where users generate code with large language models (LLMs) by describing desired features. It offers a promising avenue for rapid software prototyping in clinical research. We employed vibe coding to develop an interactive timeline feature for EMERSE, an enterprise-level electronic medical record search engine. A non-developer iteratively prompted various LLMs (Claude 3.7 Sonnet, GPT-4o, DeepSeek V3, Grok 3), resulting in a locally testable web-based prototype. Despite reaching limits in LLM performance as code complexity increased, a senior developer successfully refined and finalized the feature from the model-generated code. Vibe coding facilitated brainstorming and UI experimentation without the need for initial developer input and proved more dynamic than static wireframing. While some coding knowledge aids the process and LLMs have scalability limitations, vibe coding demonstrated itself to be a cost-effective and efficient strategy for boosting development productivity and feature exploration.
Speaker(s):
David Hanauer, MD
University of Michigan
Author(s):
David Hanauer, MD - University of Michigan; Lisa Ferguson, MS - University of Michigan; Kellen McClain - University of Michigan; Guan Wang, MS - University of Michigan;
David
Hanauer,
MD - University of Michigan
Improved drug discovery through accurate prediction of adverse drug reactions within the CANDO platform
Presentation Type: Poster - Student
Poster Number: 190
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Translation Bioinformatics/Precision Medicine
Many adverse drug reactions (ADRs) are not fully captured by preclinical testing or structured labeling. Using the CANDO platform, we modeled compound–proteome interaction signatures to predict ADR risk. In a benchmark of 715 ADRs absent from DrugBank labels, CANDO retrieved at least one causative drug within the top 100 ranks for 76.6% of ADRs and identified risks such as keratitis and hyperlactatemia, supporting proteome-wide modeling for ADR safety profiling.
Speaker(s):
Vida Bodaghi-Namileh, PhD Student
University at Buffalo School of Medicine and Biomedical Sciences
Author(s):
Vida Bodaghi-Namileh, PhD Student - University at Buffalo School of Medicine and Biomedical Sciences; Zackary Falls, PhD - University at Buffalo, Jacobs School of Medicine and Biomedical Sciences; Ram Samudrala, PhD - University at Buffalo; William Mangione, PhD - University at Buffalo;
Presentation Type: Poster - Student
Poster Number: 190
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Translation Bioinformatics/Precision Medicine
Many adverse drug reactions (ADRs) are not fully captured by preclinical testing or structured labeling. Using the CANDO platform, we modeled compound–proteome interaction signatures to predict ADR risk. In a benchmark of 715 ADRs absent from DrugBank labels, CANDO retrieved at least one causative drug within the top 100 ranks for 76.6% of ADRs and identified risks such as keratitis and hyperlactatemia, supporting proteome-wide modeling for ADR safety profiling.
Speaker(s):
Vida Bodaghi-Namileh, PhD Student
University at Buffalo School of Medicine and Biomedical Sciences
Author(s):
Vida Bodaghi-Namileh, PhD Student - University at Buffalo School of Medicine and Biomedical Sciences; Zackary Falls, PhD - University at Buffalo, Jacobs School of Medicine and Biomedical Sciences; Ram Samudrala, PhD - University at Buffalo; William Mangione, PhD - University at Buffalo;
Vida
Bodaghi-Namileh,
PhD Student - University at Buffalo School of Medicine and Biomedical Sciences
Detection of Multiple Sclerosis Disease Activity in MRI Reports Using Fine-tuned LLM
Presentation Type: Poster - Regular
Poster Number: 191
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Electronic Health Records (EHRs) are crucial for clinical research but traditional approaches for extracting information – manual chart review, rule-based phenotyping, and supervised learning – face limitations such as labor intensity, inability to capture complex clinical patterns, or the request for high-quality labels. NLP improves efficiency yet struggles with clinical nuances such as negation and temporality. Large language models (LLMs), including BERT and GPT-4, offer stronger interpretive capacity. This study evaluates LLMs for automating MRI chart review, specifically detecting new or enhanced lesions.
Speaker(s):
Yunqing Han, MS
Brigham and Women's Hospital
Author(s):
Tianrun Cai, MD - Harvard Medical School; Tanuja Chitnis, MD - Mass General Brigham; Evan Madill, MD - Mass General Brigham;
Presentation Type: Poster - Regular
Poster Number: 191
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Electronic Health Records (EHRs) are crucial for clinical research but traditional approaches for extracting information – manual chart review, rule-based phenotyping, and supervised learning – face limitations such as labor intensity, inability to capture complex clinical patterns, or the request for high-quality labels. NLP improves efficiency yet struggles with clinical nuances such as negation and temporality. Large language models (LLMs), including BERT and GPT-4, offer stronger interpretive capacity. This study evaluates LLMs for automating MRI chart review, specifically detecting new or enhanced lesions.
Speaker(s):
Yunqing Han, MS
Brigham and Women's Hospital
Author(s):
Tianrun Cai, MD - Harvard Medical School; Tanuja Chitnis, MD - Mass General Brigham; Evan Madill, MD - Mass General Brigham;
Yunqing
Han,
MS - Brigham and Women's Hospital
Knowledge Management Resources for Computable Phenotype Definitions: A Scoping Review
Presentation Type: Poster - Regular
Poster Number: 192
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
Computable phenotype definitions are executable logic to identify cohorts from clinical data. Applying knowledge management processes can improve their creation and potential for reuse, yet there is no collective information on available resources. This work describes an interim analysis of 38 articles in a scoping review to describe knowledge management resources and phases for managing computable phenotype definitions. While various resources exist, future work is needed to support and evaluate computable phenotype definition reuse.
Speaker(s):
Luke Rasmussen, MS, FAMIA
Northwestern University
Author(s):
Luke Rasmussen, MS, FAMIA - Northwestern University; Liwei Wang, MD, PhD - UTHealth; Theresa Walunas, PhD - Northwestern University; Marisa Conte, MLIS - University of Michigan; Elina Guralnik, PhD(c), MPH - George Mason University; Laura Wiley, PhD - Washington University in St. Louis; Martin Chapman, PhD - King's College London; Fang Chen, Master - University of Texas Health Science Center at Houston; Hongfang Liu, PhD - University of Texas Health Science Center at Houston;
Presentation Type: Poster - Regular
Poster Number: 192
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
Computable phenotype definitions are executable logic to identify cohorts from clinical data. Applying knowledge management processes can improve their creation and potential for reuse, yet there is no collective information on available resources. This work describes an interim analysis of 38 articles in a scoping review to describe knowledge management resources and phases for managing computable phenotype definitions. While various resources exist, future work is needed to support and evaluate computable phenotype definition reuse.
Speaker(s):
Luke Rasmussen, MS, FAMIA
Northwestern University
Author(s):
Luke Rasmussen, MS, FAMIA - Northwestern University; Liwei Wang, MD, PhD - UTHealth; Theresa Walunas, PhD - Northwestern University; Marisa Conte, MLIS - University of Michigan; Elina Guralnik, PhD(c), MPH - George Mason University; Laura Wiley, PhD - Washington University in St. Louis; Martin Chapman, PhD - King's College London; Fang Chen, Master - University of Texas Health Science Center at Houston; Hongfang Liu, PhD - University of Texas Health Science Center at Houston;
Luke
Rasmussen,
MS, FAMIA - Northwestern University
Reporting of Sensors in Environmental and Exposure Health Research: A Scoping Review
Presentation Type: Poster - Student
Poster Number: 193
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
This scoping review examined reporting practices for sensors in exposure health research. After screening 673 articles published between 2019 and 2024, 121 studies were included. The reporting of the sensor model, manufacturer, measurement method, and deployment context was inconsistent. Only 78% identified a specific instrument, and approximately half described the measurement method. Such variability limits reproducibility and the synthesis of evidence. Standardized metadata reporting and automated extraction tools are needed to improve sensor-based exposure research.
Speaker(s):
Sunho Im, Doctoral Degree
University of Utah College of Nursing
Author(s):
Sunho Im, Doctoral Degree - University of Utah College of Nursing; Sukrut Shishupal, MS - University of Utah; Julio Facelli, PhD - University of Utah; Katherine Sward, PhD - University of Utah; Fatemeh Shah-Mohammadi, PhD - University of Utah; Ram Gouripeddi, MD - University of Utah; Mollie Cummins, PhD, RN, FAAN, FACMI - University of Utah;
Presentation Type: Poster - Student
Poster Number: 193
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
This scoping review examined reporting practices for sensors in exposure health research. After screening 673 articles published between 2019 and 2024, 121 studies were included. The reporting of the sensor model, manufacturer, measurement method, and deployment context was inconsistent. Only 78% identified a specific instrument, and approximately half described the measurement method. Such variability limits reproducibility and the synthesis of evidence. Standardized metadata reporting and automated extraction tools are needed to improve sensor-based exposure research.
Speaker(s):
Sunho Im, Doctoral Degree
University of Utah College of Nursing
Author(s):
Sunho Im, Doctoral Degree - University of Utah College of Nursing; Sukrut Shishupal, MS - University of Utah; Julio Facelli, PhD - University of Utah; Katherine Sward, PhD - University of Utah; Fatemeh Shah-Mohammadi, PhD - University of Utah; Ram Gouripeddi, MD - University of Utah; Mollie Cummins, PhD, RN, FAAN, FACMI - University of Utah;
Sunho
Im,
Doctoral Degree - University of Utah College of Nursing
Designing and Evaluating a Sensor Library for Translational Exposure Health: A User-Centered Approach
Presentation Type: Poster - Regular
Poster Number: 194
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
Environmental health research requires appropriate sensor selection, yet researchers lack resources to discover and compare sensor metadata. The SMARTER project developed a sensor library using user-centered design through qualitative interviews, expert consultation, and iterative testing. Results showed excellent usability (System Usability Score: 90.0) and high task completion rates (92-100%). Key insights included successful filter-based navigation, need for improved contribution workflows, and essential side-by-side comparison features. This work demonstrates how FAIRifying sensor metadata enhances exposure study design robustness and reproducibility.
Speaker(s):
Urvi Varma, MS
University of Utah
Author(s):
Rachel Nelson, BS - University of Utah; Mollie Cummins, PhD, RN, FAAN, FACMI - University of Utah; Julio Facelli, PhD - University of Utah; Ramkiran Gouripeddi, MS MBBS FAMIA - University of Utah; Pavan Motagi, MS - University of Utah; Urvi Varma, MS - University of Utah; Katherine Sward, PhD - University of Utah;
Presentation Type: Poster - Regular
Poster Number: 194
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
Environmental health research requires appropriate sensor selection, yet researchers lack resources to discover and compare sensor metadata. The SMARTER project developed a sensor library using user-centered design through qualitative interviews, expert consultation, and iterative testing. Results showed excellent usability (System Usability Score: 90.0) and high task completion rates (92-100%). Key insights included successful filter-based navigation, need for improved contribution workflows, and essential side-by-side comparison features. This work demonstrates how FAIRifying sensor metadata enhances exposure study design robustness and reproducibility.
Speaker(s):
Urvi Varma, MS
University of Utah
Author(s):
Rachel Nelson, BS - University of Utah; Mollie Cummins, PhD, RN, FAAN, FACMI - University of Utah; Julio Facelli, PhD - University of Utah; Ramkiran Gouripeddi, MS MBBS FAMIA - University of Utah; Pavan Motagi, MS - University of Utah; Urvi Varma, MS - University of Utah; Katherine Sward, PhD - University of Utah;
Urvi
Varma,
MS - University of Utah
Illuminating the FHIR Landscape: A Review of Public FHIR Datasets for Research
Presentation Type: Poster - Student
Poster Number: 196
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
FHIR has become foundational in health data exchange, yet its potential to advance clinical research and real-world evidence generation is limited by the scarcity of accessible datasets. Using a modified PRISMA approach, we identified 29 publicly available FHIR datasets, nearly all synthetic or narrowly scoped. This gap restricts developers’ ability to build and validate tools that support secondary use of clinical data.
Speaker(s):
Mike Enger, M.S.
Vanderbilt University
Author(s):
Mike Enger, M.S. - Vanderbilt University; Alex Cheng, PhD - Vanderbilt University Medical Center;
Presentation Type: Poster - Student
Poster Number: 196
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
FHIR has become foundational in health data exchange, yet its potential to advance clinical research and real-world evidence generation is limited by the scarcity of accessible datasets. Using a modified PRISMA approach, we identified 29 publicly available FHIR datasets, nearly all synthetic or narrowly scoped. This gap restricts developers’ ability to build and validate tools that support secondary use of clinical data.
Speaker(s):
Mike Enger, M.S.
Vanderbilt University
Author(s):
Mike Enger, M.S. - Vanderbilt University; Alex Cheng, PhD - Vanderbilt University Medical Center;
Mike
Enger,
M.S. - Vanderbilt University
Medications Influence Plasma Proteomic Profiles after Parkinson’s Disease Diagnosis
Presentation Type: Poster Invite - Regular
Poster Number: 197
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
We analyzed longitudinal 16,323 proteoforms in plasma from 348 Parkinson’s disease (PD) cases, integrating post-onset samples linked with medication history from real-world data. Medications, especially dopaminergic therapy, substantially influenced proteomic profiles: 9 of the 41 proteins most strongly modulated by PD were also associated with medication (p<0.001, |Cohen’s d| > 0.5). Enrichment analyses and correlations with known proteins modulated by PD medication support these findings, indicating that medication effects can confound proteomic biomarker discovery.
Speaker(s):
idit kosti, PhD
Alkahest Inc.
Author(s):
Balint File, Phd - Alkahest Inc.; Ying Wang, BSc - Alkahest Inc.; Rad Suchecki, PhD - Alkahest Inc.; Tibor Nanasi, MD, PhD - Alkahest Inc.; John Guo, PhD - Alkahest Inc.; Nhi Hin, PhD - Alkahest Inc.; Scott Lohr, BSc - Alkahest Inc.; Chunmiao Feng, PhD - Alkahest Inc.; idit kosti, PhD - Alkahest Inc.; Benoit Lehallier, PhD - Alkahest Inc.;
Presentation Type: Poster Invite - Regular
Poster Number: 197
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
We analyzed longitudinal 16,323 proteoforms in plasma from 348 Parkinson’s disease (PD) cases, integrating post-onset samples linked with medication history from real-world data. Medications, especially dopaminergic therapy, substantially influenced proteomic profiles: 9 of the 41 proteins most strongly modulated by PD were also associated with medication (p<0.001, |Cohen’s d| > 0.5). Enrichment analyses and correlations with known proteins modulated by PD medication support these findings, indicating that medication effects can confound proteomic biomarker discovery.
Speaker(s):
idit kosti, PhD
Alkahest Inc.
Author(s):
Balint File, Phd - Alkahest Inc.; Ying Wang, BSc - Alkahest Inc.; Rad Suchecki, PhD - Alkahest Inc.; Tibor Nanasi, MD, PhD - Alkahest Inc.; John Guo, PhD - Alkahest Inc.; Nhi Hin, PhD - Alkahest Inc.; Scott Lohr, BSc - Alkahest Inc.; Chunmiao Feng, PhD - Alkahest Inc.; idit kosti, PhD - Alkahest Inc.; Benoit Lehallier, PhD - Alkahest Inc.;
idit
kosti,
PhD - Alkahest Inc.
From Risk to Trajectories: How Alzheimer's Genetic Variants Differentially Affect Memory, Language, and Executive Function
Presentation Type: Poster - Student
Poster Number: 198
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Translation Bioinformatics/Precision Medicine
Using hierarchical mixed-effects models in 4,005 participants with 33,819 cognitive assessments across three domains, we characterized genetic effects on decline trajectories across memory, executive function, and language for 64 select variants. We identified three variants with pleiotropic effects across executive function and language (rs60755019/TREML2, rs10952097/ICA1, rs16941239/FOXF1, all p<5×10⁻⁸), plus executive-specific and memory-specific associations. This longitudinal framework demonstrates increased statistical power for detecting rate-of-change effects, advancing from identifying disease risk variants to characterizing their temporal cognitive effects.
Speaker(s):
Shraddha Dumawat, PhD Biomedical and Health Informatics
Case Western Reserve University
Author(s):
Shraddha Dumawat, PhD Biomedical and Health Informatics - Case Western Reserve University; William Bush - Case Western Reserve University;
Presentation Type: Poster - Student
Poster Number: 198
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Translation Bioinformatics/Precision Medicine
Using hierarchical mixed-effects models in 4,005 participants with 33,819 cognitive assessments across three domains, we characterized genetic effects on decline trajectories across memory, executive function, and language for 64 select variants. We identified three variants with pleiotropic effects across executive function and language (rs60755019/TREML2, rs10952097/ICA1, rs16941239/FOXF1, all p<5×10⁻⁸), plus executive-specific and memory-specific associations. This longitudinal framework demonstrates increased statistical power for detecting rate-of-change effects, advancing from identifying disease risk variants to characterizing their temporal cognitive effects.
Speaker(s):
Shraddha Dumawat, PhD Biomedical and Health Informatics
Case Western Reserve University
Author(s):
Shraddha Dumawat, PhD Biomedical and Health Informatics - Case Western Reserve University; William Bush - Case Western Reserve University;
Shraddha
Dumawat,
PhD Biomedical and Health Informatics - Case Western Reserve University
Predictive Net Promoter Score: A Comprehensive Approach to Measuring Member Experience Across a Healthcare Organization
Presentation Type: Poster - Regular
Poster Number: 199
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Key components and performance outcomes of a suite of predictive models designed to estimate member experience in Medicare Advantage members of a large, national payor are described. Rich foundational data sources with interoperability and technological capabilities enabled predictive models to improve the timing, scale, and quality of member experience insights, which can be used to proactively address or prevent suboptimal member experiences.
Speaker(s):
Eleanor Beltz, PhD, ATC
CVS Health
Author(s):
Eleanor Beltz, PhD, ATC - CVS Health; Kelly Jean Craig, PhD - CVS Health; Sara Okun, MS - CVS Health; Sarah Latto, BS - CVS Health; Anurag Agarwal, MS - CVS Health; Emily Hague, PhD - CVS Health; Elisea Avalos, PhD - CVS Health; Amanda Zaleski, PhD, MS, FACSM - Clinical Evidence Development, CVS Health; Srikant Narasimhan, MA, MBA - CVS Health; Adam Zambuto, BA - CVS Health; Andrew Croll, BS - CVS Health; Shawn Smith, MBA - CVS Health; Thomas Sargent, BS - CVS Health; Naman Bansal, BS - CVS Health; Lukas Hansen, MBA - CVS Health;
Presentation Type: Poster - Regular
Poster Number: 199
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Key components and performance outcomes of a suite of predictive models designed to estimate member experience in Medicare Advantage members of a large, national payor are described. Rich foundational data sources with interoperability and technological capabilities enabled predictive models to improve the timing, scale, and quality of member experience insights, which can be used to proactively address or prevent suboptimal member experiences.
Speaker(s):
Eleanor Beltz, PhD, ATC
CVS Health
Author(s):
Eleanor Beltz, PhD, ATC - CVS Health; Kelly Jean Craig, PhD - CVS Health; Sara Okun, MS - CVS Health; Sarah Latto, BS - CVS Health; Anurag Agarwal, MS - CVS Health; Emily Hague, PhD - CVS Health; Elisea Avalos, PhD - CVS Health; Amanda Zaleski, PhD, MS, FACSM - Clinical Evidence Development, CVS Health; Srikant Narasimhan, MA, MBA - CVS Health; Adam Zambuto, BA - CVS Health; Andrew Croll, BS - CVS Health; Shawn Smith, MBA - CVS Health; Thomas Sargent, BS - CVS Health; Naman Bansal, BS - CVS Health; Lukas Hansen, MBA - CVS Health;
Eleanor
Beltz,
PhD, ATC - CVS Health
Identifying irAEs Associated with Immune Checkpoint Inhibitors and Their Incidences in Lung Cancer Using Real-World Data
Presentation Type: Poster Invite - Regular
Poster Number: 200
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
The advent of immune checkpoint inhibitors (ICIs) significantly improves survival in lung cancer but can cause rare, severe immune-related adverse events (irAEs) not always captured in randomized controlled trials. This study aimed to identify irAEs associated with 8 ICIs and determine their incidences by integrating the FDA Adverse Event Reporting System (AERS/FAERS) for signal detection with the large-scale, EHR-based Cosmos platform for validation and quantitative characterization. AERS/FAERS data (2004–2024) was normalized and analyzed using disproportionality methods (PRR and ROR). Eleven clinically relevant irAEs (e.g., pneumonitis, hepatitis, hypothyroidism) were characterized in a lung cancer cohort of 7,051 patients from Cosmos. This study demonstrates a valuable informatics framework for integrated real-world data analysis.
Speaker(s):
Liwei Wang, MD, PhD
UTHealth
Author(s):
Rui Li, Phd - UT health; Shuyu Lu, Master - University of Texas Health Science Center at Houston; Xiaomeng Wang, Master of Science - University of Texas Health Science Center at Houston; MOUSUMI SINHA, MS in Data Analytics - UTHealth; Laila Rasmy, PhD, MSc, MBA, RPh. - UTHealth MSBMI; Xiaoyang Ruan, PhD - The University of Texas Health Science Center at Houston; Sunyang Fu, PhD, MHI - UTHealth Houston; Andrew Wen, MS - University of Texas Health Sciences Center at Houston; Heidi Dowst, MS - Baylor College of Medicine; Jun Jiang, Ph.D. - University of Texas Health Science Center at Houston; Rui Zhang, PhD, FACMI, FAMIA, FIAHSI - University of Minnesota, Twin Cities; Jacob New, MD, PhD - Scripps Cancer Center; Hongfang Liu, PhD - University of Texas Health Science Center at Houston;
Presentation Type: Poster Invite - Regular
Poster Number: 200
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
The advent of immune checkpoint inhibitors (ICIs) significantly improves survival in lung cancer but can cause rare, severe immune-related adverse events (irAEs) not always captured in randomized controlled trials. This study aimed to identify irAEs associated with 8 ICIs and determine their incidences by integrating the FDA Adverse Event Reporting System (AERS/FAERS) for signal detection with the large-scale, EHR-based Cosmos platform for validation and quantitative characterization. AERS/FAERS data (2004–2024) was normalized and analyzed using disproportionality methods (PRR and ROR). Eleven clinically relevant irAEs (e.g., pneumonitis, hepatitis, hypothyroidism) were characterized in a lung cancer cohort of 7,051 patients from Cosmos. This study demonstrates a valuable informatics framework for integrated real-world data analysis.
Speaker(s):
Liwei Wang, MD, PhD
UTHealth
Author(s):
Rui Li, Phd - UT health; Shuyu Lu, Master - University of Texas Health Science Center at Houston; Xiaomeng Wang, Master of Science - University of Texas Health Science Center at Houston; MOUSUMI SINHA, MS in Data Analytics - UTHealth; Laila Rasmy, PhD, MSc, MBA, RPh. - UTHealth MSBMI; Xiaoyang Ruan, PhD - The University of Texas Health Science Center at Houston; Sunyang Fu, PhD, MHI - UTHealth Houston; Andrew Wen, MS - University of Texas Health Sciences Center at Houston; Heidi Dowst, MS - Baylor College of Medicine; Jun Jiang, Ph.D. - University of Texas Health Science Center at Houston; Rui Zhang, PhD, FACMI, FAMIA, FIAHSI - University of Minnesota, Twin Cities; Jacob New, MD, PhD - Scripps Cancer Center; Hongfang Liu, PhD - University of Texas Health Science Center at Houston;
Liwei
Wang,
MD, PhD - UTHealth
Unified EHR Phenotyping of Adverse Pregnancy Outcomes for Postpartum CVD Risk
Presentation Type: Poster - Student
Poster Number: 202
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Adverse pregnancy outcomes reflect elevated future cardiovascular disease risk, yet many affected women remain overlooked when relying only on diagnosis codes in electronic health records. Using 23,833 deliveries from the Duke University Health System, we built a unified computable phenotype that integrates structured codes with natural language processing derived clinical concepts. Indirect validation through postpartum cardiovascular disease prediction demonstrated improved discrimination, reclassification, and clinical utility, especially among women without coded adverse pregnancy outcomes.
Speaker(s):
Yang Yang, BS
Duke University
Author(s):
Yang Yang, BS - Duke University; Yunqian Liu, Master - Duke University; Johanna Quist-Nelson, MD - University of North Carolina School of Medicine; Marie-Louise Meng, MD - Vanderbilt University Medical Center; Olga Demler, PhD - Harvard T.H. Chan School of Public Health; Kathryn Rexrode, MD, MPH - Harvard Medical School Internal Medicine, Preventive Medicine, Women's Health; Janet Rich-Edwards, ScD - Harvard T.H. Chan School of Public Health; Ricardo Henao, PhD - Duke University; Chuan Hong, PhD - Duke University;
Presentation Type: Poster - Student
Poster Number: 202
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Adverse pregnancy outcomes reflect elevated future cardiovascular disease risk, yet many affected women remain overlooked when relying only on diagnosis codes in electronic health records. Using 23,833 deliveries from the Duke University Health System, we built a unified computable phenotype that integrates structured codes with natural language processing derived clinical concepts. Indirect validation through postpartum cardiovascular disease prediction demonstrated improved discrimination, reclassification, and clinical utility, especially among women without coded adverse pregnancy outcomes.
Speaker(s):
Yang Yang, BS
Duke University
Author(s):
Yang Yang, BS - Duke University; Yunqian Liu, Master - Duke University; Johanna Quist-Nelson, MD - University of North Carolina School of Medicine; Marie-Louise Meng, MD - Vanderbilt University Medical Center; Olga Demler, PhD - Harvard T.H. Chan School of Public Health; Kathryn Rexrode, MD, MPH - Harvard Medical School Internal Medicine, Preventive Medicine, Women's Health; Janet Rich-Edwards, ScD - Harvard T.H. Chan School of Public Health; Ricardo Henao, PhD - Duke University; Chuan Hong, PhD - Duke University;
Yang
Yang,
BS - Duke University
Impact of Treatment Response Definitions on Effectiveness Estimates in Major Depression: Toward Treatment Selection Models to Predict Therapeutic Efficacy
Presentation Type: Poster - Student
Poster Number: 203
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Major depressive disorder is one of the leading causes of disability worldwide. Antidepressant selection is a challenge, as treatment decisions rely on trial-and-error. Real-world evidence comparing the effectiveness of existing treatments is lacking. This study intends to 1) estimate treatment effectiveness for common antidepressants from large, naturalistic healthcare records across multiple published definitions of treatment response. 2) predict treatment response using multimodal data, including information from electronic health record data and polygenic risk scores.
Speaker(s):
Sahit Menon, MD
Vanderbilt University Medical Center
Author(s):
Yogesh Barve, PhD - Vanderbilt Institute for Software Integration Systems; Anwar Said, PhD - Vanderbilt Institute for Software Integration Systems; Sahit Menon, MD - Vanderbilt University Medical Center; Forrest Laine, PhD - Vanderbilt Institute for Software Integration Systems; Janos Sztipanovits, PhD - Institute for Software Integration Systems; Colin Walsh, MD MA - Vanderbilt University;
Presentation Type: Poster - Student
Poster Number: 203
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Major depressive disorder is one of the leading causes of disability worldwide. Antidepressant selection is a challenge, as treatment decisions rely on trial-and-error. Real-world evidence comparing the effectiveness of existing treatments is lacking. This study intends to 1) estimate treatment effectiveness for common antidepressants from large, naturalistic healthcare records across multiple published definitions of treatment response. 2) predict treatment response using multimodal data, including information from electronic health record data and polygenic risk scores.
Speaker(s):
Sahit Menon, MD
Vanderbilt University Medical Center
Author(s):
Yogesh Barve, PhD - Vanderbilt Institute for Software Integration Systems; Anwar Said, PhD - Vanderbilt Institute for Software Integration Systems; Sahit Menon, MD - Vanderbilt University Medical Center; Forrest Laine, PhD - Vanderbilt Institute for Software Integration Systems; Janos Sztipanovits, PhD - Institute for Software Integration Systems; Colin Walsh, MD MA - Vanderbilt University;
Sahit
Menon,
MD - Vanderbilt University Medical Center
Methods for Assuring Quality Measures with the Claim Argument Evidence System
Presentation Type: Poster Invite - Regular
Poster Number: 204
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
The Claim-Argument-Evidence System (CAES) is an Artificial Intelligence powered assurance case generation and evaluation software system. Battelle developed the CAES to enhance evaluation of safety and trustworthiness in Clinical Quality Measures; however, the system is applicable across diverse domains, for example environmental sustainability. This system addresses challenges of proving safety and trustworthiness by employing a methodological approach and reducing cognitive bias. Motivated by the Partnership for Quality Measurement, wherein Battelle serves a role as a Consensus-based Entity for the Center for Medicare & Medicaid Services, the CAES accelerates Clinical Quality Measure assessment and improves results. The system leverages established Claim Argument Evidence and Assurance Case frameworks to analyze information, minimizing human errors such as confirmation bias. The CAES utilizes large language models for evidence extraction, claim generation, and evaluation. This paper details the CAES evaluation module’s methodology, and preliminary findings.
Speaker(s):
Gerrit Bryan, BS
Battelle
Author(s):
Jeffrey Geppert, EdM JD - Battelle Memorial Institute; Stephen Boxwell, Ph.D. - Battelle; Jeremy Bellay, Ph.D. - Battelle; Chun Lin Liu, B.S.E. - Battelle; Max Chambers, B.S.E. - Battelle;
Presentation Type: Poster Invite - Regular
Poster Number: 204
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
The Claim-Argument-Evidence System (CAES) is an Artificial Intelligence powered assurance case generation and evaluation software system. Battelle developed the CAES to enhance evaluation of safety and trustworthiness in Clinical Quality Measures; however, the system is applicable across diverse domains, for example environmental sustainability. This system addresses challenges of proving safety and trustworthiness by employing a methodological approach and reducing cognitive bias. Motivated by the Partnership for Quality Measurement, wherein Battelle serves a role as a Consensus-based Entity for the Center for Medicare & Medicaid Services, the CAES accelerates Clinical Quality Measure assessment and improves results. The system leverages established Claim Argument Evidence and Assurance Case frameworks to analyze information, minimizing human errors such as confirmation bias. The CAES utilizes large language models for evidence extraction, claim generation, and evaluation. This paper details the CAES evaluation module’s methodology, and preliminary findings.
Speaker(s):
Gerrit Bryan, BS
Battelle
Author(s):
Jeffrey Geppert, EdM JD - Battelle Memorial Institute; Stephen Boxwell, Ph.D. - Battelle; Jeremy Bellay, Ph.D. - Battelle; Chun Lin Liu, B.S.E. - Battelle; Max Chambers, B.S.E. - Battelle;
Gerrit
Bryan,
BS - Battelle
Leveraging Large Language Models to Integrate Patient Sentiment and Adverse Drug Reaction Pattern Detection from Social Media in Atopic Dermatitis Therapies
Presentation Type: Poster - Student
Poster Number: 205
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
This study applies a large language model (GPT-4o) to analyze Reddit posts describing systemic atopic dermatitis therapies. We first classify patient sentiment, then extract adverse drug reactions from 300 comments per drug. Results reveal drug-specific patterns, including higher positive sentiment for upadacitinib and dupilumab and distinct adverse reactions. This approach demonstrates the value of LLMs for real-world pharmacovigilance.
Speaker(s):
Jack Cummins, undergrad
Princeton University
Author(s):
Jack Cummins, undergrad - Princeton University; JiaDe Yu, M.D. - Virginia Commonwealth University School of Medicine;
Presentation Type: Poster - Student
Poster Number: 205
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
This study applies a large language model (GPT-4o) to analyze Reddit posts describing systemic atopic dermatitis therapies. We first classify patient sentiment, then extract adverse drug reactions from 300 comments per drug. Results reveal drug-specific patterns, including higher positive sentiment for upadacitinib and dupilumab and distinct adverse reactions. This approach demonstrates the value of LLMs for real-world pharmacovigilance.
Speaker(s):
Jack Cummins, undergrad
Princeton University
Author(s):
Jack Cummins, undergrad - Princeton University; JiaDe Yu, M.D. - Virginia Commonwealth University School of Medicine;
Jack
Cummins,
undergrad - Princeton University
Improving Prospective Weight Entry Error Detection using Deep Learning Methods
Presentation Type: Poster Invite - Student
Poster Number: 206
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Weight entry errors can cause significant patient harm in pediatrics due to pervasive weight-based dosing practices. Existing studies primarily utilize rule-based methods, achieving an Area Under the Receiver Operating Characteristic Curve (AUROC) ranging from 0.546 to 0.620. In this study, a two-module method employing bi-directional Long Short-Term Memory (bi-LSTM) with the Multi-Head Attention Mechanism for the prospective detection of anomalous weight entries was implemented to improve prospective error detection. The proposed method consists of a predictor and a classifier module. The predictor module learns the normal pattern of weight changes, and the classifier module identifies anomalous weight entries. The performance of both modules was evaluated, exhibiting a clear superiority over other methods in distinguishing normal from anomalous data points. The proposed approach achieved an AUROC of 0.977 and a precision of 8.98% when calibrated for 90% sensitivity, close to double the previous results, significantly outperforming other methods.
Speaker(s):
Han-Nieh Shih, M.Sc.
University of Cincinnati
Author(s):
Raj Bhatnagar, PhD - University of Cincinnati; Vikram Ravindra, PhD - University of Cincinnati; Justin Zhan, PhD - University of Cincinnati; Stephen Spooner, MD, FAAP - Cincinnati Children's Hospital; Danny Wu, PhD - University of North Carolina at Chapel Hill;
Presentation Type: Poster Invite - Student
Poster Number: 206
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Weight entry errors can cause significant patient harm in pediatrics due to pervasive weight-based dosing practices. Existing studies primarily utilize rule-based methods, achieving an Area Under the Receiver Operating Characteristic Curve (AUROC) ranging from 0.546 to 0.620. In this study, a two-module method employing bi-directional Long Short-Term Memory (bi-LSTM) with the Multi-Head Attention Mechanism for the prospective detection of anomalous weight entries was implemented to improve prospective error detection. The proposed method consists of a predictor and a classifier module. The predictor module learns the normal pattern of weight changes, and the classifier module identifies anomalous weight entries. The performance of both modules was evaluated, exhibiting a clear superiority over other methods in distinguishing normal from anomalous data points. The proposed approach achieved an AUROC of 0.977 and a precision of 8.98% when calibrated for 90% sensitivity, close to double the previous results, significantly outperforming other methods.
Speaker(s):
Han-Nieh Shih, M.Sc.
University of Cincinnati
Author(s):
Raj Bhatnagar, PhD - University of Cincinnati; Vikram Ravindra, PhD - University of Cincinnati; Justin Zhan, PhD - University of Cincinnati; Stephen Spooner, MD, FAAP - Cincinnati Children's Hospital; Danny Wu, PhD - University of North Carolina at Chapel Hill;
Han-Nieh
Shih,
M.Sc. - University of Cincinnati
Large Language Models Improve Identification of Memory Clinic Patients Diagnosed with Alzheimer Disease Using Electronic Health Records Data
Presentation Type: Poster Invite - Regular
Poster Number: 207
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Alzheimer disease is a neurodegenerative disorder marked by memory and cognitive decline. This study used electronic health records (EHR), large language models (LLMs), and machine learning to predict symptomatic AD in patients with cognitive concerns. By analyzing demographics, comorbidities, biomarkers, and clinical notes, the models predicted AD diagnosis a year later (AUROC 0.858). Key predictors included age, hypertension, and repetitive statements, demonstrating early identification and intervention for AD through LLM-derived phenotypes from EHR data.
Speaker(s):
William Powell, MS
Washington University in St. Louis
Author(s):
William Powell, MS - Washington University in St. Louis; Inez Oh, PhD - Institute for Informatics at Washington University in St. Louis, School of Medicine; Suzanne Schindler, MD/PhD - Department of Neurology, Washington University School of Medicine; Anna Hofmann, MD - Department of Neurology, Washington University School of Medicine; Barbara Joy Snider, MD, PhD - Department of Neurology, Washington University School of Medicine; Madeline Paczynski, PA-C - Department of Neurology, Washington University School of Medicine; Nupur Ghoshal, MD, PhD - Department of Neurology, Washington University School of Medicine; Philip Payne, PhD, FACMI, FAMIA - WashU Medicine and BJC Healthcare; Albert Lai, PhD, FACMI, FAMIA - Washington University; Mackenzie Hofford, MD - Washington University; Aditi Gupta - Washington University in St. Louis;
Presentation Type: Poster Invite - Regular
Poster Number: 207
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Alzheimer disease is a neurodegenerative disorder marked by memory and cognitive decline. This study used electronic health records (EHR), large language models (LLMs), and machine learning to predict symptomatic AD in patients with cognitive concerns. By analyzing demographics, comorbidities, biomarkers, and clinical notes, the models predicted AD diagnosis a year later (AUROC 0.858). Key predictors included age, hypertension, and repetitive statements, demonstrating early identification and intervention for AD through LLM-derived phenotypes from EHR data.
Speaker(s):
William Powell, MS
Washington University in St. Louis
Author(s):
William Powell, MS - Washington University in St. Louis; Inez Oh, PhD - Institute for Informatics at Washington University in St. Louis, School of Medicine; Suzanne Schindler, MD/PhD - Department of Neurology, Washington University School of Medicine; Anna Hofmann, MD - Department of Neurology, Washington University School of Medicine; Barbara Joy Snider, MD, PhD - Department of Neurology, Washington University School of Medicine; Madeline Paczynski, PA-C - Department of Neurology, Washington University School of Medicine; Nupur Ghoshal, MD, PhD - Department of Neurology, Washington University School of Medicine; Philip Payne, PhD, FACMI, FAMIA - WashU Medicine and BJC Healthcare; Albert Lai, PhD, FACMI, FAMIA - Washington University; Mackenzie Hofford, MD - Washington University; Aditi Gupta - Washington University in St. Louis;
William
Powell,
MS - Washington University in St. Louis
Interpretable Machine Learning for Rare Chest Pain CCSR Label Detection with Clinical Reasoning
Presentation Type: Poster Invite - Regular
Poster Number: 208
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
In this study, we propose an automated, interpretable machine learning (ML) framework to detect chest pain–related Clinical Classification Software Refined (CCSR) labels directly from Electronic Health Record (EHR) data without relying on resource-intensive expert labeling. ICD-10-CM codes from patient encounters were mapped to 53 CCSR labels, most of which were rare. Using only the first 6 hours of clinical data, we extracted 745 features and developed a binary classification pipeline using Random Forest, XGBoost, Easy Ensemble, Logistic Regression, and Neural Networks. A grid search was performed to optimize hyperparameters for each of the 53 models. The best models were selected based on TPR and AUROC, followed by interpretability analysis to identify key predictive features using LIME. Our approach achieved an overall weighted accuracy of 88%, demonstrating its potential to support clinicians by reducing early diagnostic uncertainty in chest pain evaluation.
Speaker(s):
Sourav Kumar Ghosh, MS in Electrical and Computer Engineering
University of Minnesota
Author(s):
Tonushree Dutta, BSc in Electrical and Electronic Engineering - University of Minnesota; Nicholas E. Ingraham, MD, MS - University of Minnesota; Michael A. Puskarich, MD, MS - University of Minnesota; Sarah KS. Knack, MD, MS - University of Minnesota; John Sartori, PhD - University of Minnesota;
Presentation Type: Poster Invite - Regular
Poster Number: 208
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
In this study, we propose an automated, interpretable machine learning (ML) framework to detect chest pain–related Clinical Classification Software Refined (CCSR) labels directly from Electronic Health Record (EHR) data without relying on resource-intensive expert labeling. ICD-10-CM codes from patient encounters were mapped to 53 CCSR labels, most of which were rare. Using only the first 6 hours of clinical data, we extracted 745 features and developed a binary classification pipeline using Random Forest, XGBoost, Easy Ensemble, Logistic Regression, and Neural Networks. A grid search was performed to optimize hyperparameters for each of the 53 models. The best models were selected based on TPR and AUROC, followed by interpretability analysis to identify key predictive features using LIME. Our approach achieved an overall weighted accuracy of 88%, demonstrating its potential to support clinicians by reducing early diagnostic uncertainty in chest pain evaluation.
Speaker(s):
Sourav Kumar Ghosh, MS in Electrical and Computer Engineering
University of Minnesota
Author(s):
Tonushree Dutta, BSc in Electrical and Electronic Engineering - University of Minnesota; Nicholas E. Ingraham, MD, MS - University of Minnesota; Michael A. Puskarich, MD, MS - University of Minnesota; Sarah KS. Knack, MD, MS - University of Minnesota; John Sartori, PhD - University of Minnesota;
Sourav Kumar
Ghosh,
MS in Electrical and Computer Engineering - University of Minnesota
Real-World Data on ADHD Medication Consumption: Moving Beyond Prescription Data with Large Language Models
Presentation Type: Poster - Regular
Click to View Presentation
Poster Number: 209
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Electronic health record studies often rely on prescription orders, which indicate treatment intent but not whether patients took medications, limiting real-world assessment of treatment. We evaluated the accuracy of a large language model (OpenAI-o1) in identifying methylphenidate consumption documented in clinical notes of pediatric patients with ADHD. OpenAI-o1 demonstrated high accuracy, particularly in notes with consumption documentation, suggesting LLM-based analysis of clinical notes could improve the precision of real-world studies of medication adherence and effectiveness.
Speaker(s):
Tracy Huang, MSPH
Stanford University School of Medicine
Author(s):
Tracy Huang, MSPH - Stanford University School of Medicine; Fatma Gunturkun, PhD - Stanford University; Yair Bannett, MD, MS - Stanford University School of Medicine;
Presentation Type: Poster - Regular
Click to View Presentation
Poster Number: 209
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Electronic health record studies often rely on prescription orders, which indicate treatment intent but not whether patients took medications, limiting real-world assessment of treatment. We evaluated the accuracy of a large language model (OpenAI-o1) in identifying methylphenidate consumption documented in clinical notes of pediatric patients with ADHD. OpenAI-o1 demonstrated high accuracy, particularly in notes with consumption documentation, suggesting LLM-based analysis of clinical notes could improve the precision of real-world studies of medication adherence and effectiveness.
Speaker(s):
Tracy Huang, MSPH
Stanford University School of Medicine
Author(s):
Tracy Huang, MSPH - Stanford University School of Medicine; Fatma Gunturkun, PhD - Stanford University; Yair Bannett, MD, MS - Stanford University School of Medicine;
Tracy
Huang,
MSPH - Stanford University School of Medicine
Less Is More: Evaluating Knowledge Sources for AI-Powered SCI Patient Education Chatbot
Presentation Type: Poster Invite - Regular
Poster Number: 210
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
This study evaluated how different knowledge sources—expert‑curated websites, journal articles, their combination, and a vanilla LLM models—affect a RAG‑powered SCI patient education chatbot. Expert‑curated websites, though limited in number, produced the most understandable, actionable, and clinician‑aligned responses, whereas journal articles were more technical and less actionable. Combining sources achieved comparable performance, while vanilla outputs were inconsistent. These findings demonstrate that focused, high‑quality sources deliver stronger patient-education impact with less development effort.
Speaker(s):
Maria Yuliana, Bachelor
University of Pittsburgh
Author(s):
Bayu Aryoyudanta, Master - University of Pittsburgh; Wilbert Soekinto, B.S - University of Pittsburgh; I Made Agus Setiawan, PhD - University of Pittsburgh; Brad Dicianno, MD - University of Pittsburgh School of Medicine; Andi Saptono, PhD - University of Pittsburgh School of Health and Rehabilitation Services; Bambang Parmanto, PhD - University of Pittsburgh School of Health and Rehabilitation Sciences;
Presentation Type: Poster Invite - Regular
Poster Number: 210
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
This study evaluated how different knowledge sources—expert‑curated websites, journal articles, their combination, and a vanilla LLM models—affect a RAG‑powered SCI patient education chatbot. Expert‑curated websites, though limited in number, produced the most understandable, actionable, and clinician‑aligned responses, whereas journal articles were more technical and less actionable. Combining sources achieved comparable performance, while vanilla outputs were inconsistent. These findings demonstrate that focused, high‑quality sources deliver stronger patient-education impact with less development effort.
Speaker(s):
Maria Yuliana, Bachelor
University of Pittsburgh
Author(s):
Bayu Aryoyudanta, Master - University of Pittsburgh; Wilbert Soekinto, B.S - University of Pittsburgh; I Made Agus Setiawan, PhD - University of Pittsburgh; Brad Dicianno, MD - University of Pittsburgh School of Medicine; Andi Saptono, PhD - University of Pittsburgh School of Health and Rehabilitation Services; Bambang Parmanto, PhD - University of Pittsburgh School of Health and Rehabilitation Sciences;
Maria
Yuliana,
Bachelor - University of Pittsburgh
Predicting Breast Cancer Recurrence with a Multimodal Deep Learning Model Integrating Whole-Slide Image, Clinical, Genomic Data
Presentation Type: Poster - Student
Poster Number: 211
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Translation Bioinformatics/Precision Medicine
Breast cancer recurrence prediction remains limited by single-modality models. We developed a novel multimodal deep learning model integrating whole-slide images, clinical, gene expression, and somatic mutation data from 539 breast cancer patients (TCGA PanCancer Atlas). Using modality-specific encoders and multi-stage attention, our model achieved superior performance versus four traditional machine-learning baselines (AUROC 0.809; accuracy 0.875). These findings demonstrate the clinical promise of tetramodal frameworks for breast cancer prognostication.
Speaker(s):
Justin Kim, A.B. '21 M.D. '28
Brown University
Author(s):
Justin Kim, A.B. '21 M.D. '28 - Brown University; Zhicheng Jiao, PhD - Brown University; Sean Hacking, MB,BCh - NYU Langone; Ece Uzun, PhD - Brown University Health/Brown University;
Presentation Type: Poster - Student
Poster Number: 211
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Translation Bioinformatics/Precision Medicine
Breast cancer recurrence prediction remains limited by single-modality models. We developed a novel multimodal deep learning model integrating whole-slide images, clinical, gene expression, and somatic mutation data from 539 breast cancer patients (TCGA PanCancer Atlas). Using modality-specific encoders and multi-stage attention, our model achieved superior performance versus four traditional machine-learning baselines (AUROC 0.809; accuracy 0.875). These findings demonstrate the clinical promise of tetramodal frameworks for breast cancer prognostication.
Speaker(s):
Justin Kim, A.B. '21 M.D. '28
Brown University
Author(s):
Justin Kim, A.B. '21 M.D. '28 - Brown University; Zhicheng Jiao, PhD - Brown University; Sean Hacking, MB,BCh - NYU Langone; Ece Uzun, PhD - Brown University Health/Brown University;
Justin
Kim,
A.B. '21 M.D. '28 - Brown University
Using a Computer Vision Ensemble and Weight Data to Achieve Reliable Syringe Exchange
Presentation Type: Poster - Regular
Poster Number: 212
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
In this work, we address the challenges of accurately counting needles and detecting foreign objects in a needle exchange vending machine using computer vision techniques. Various methods were explored, with the best results achieved using DINOv3 embeddings. The developed framework leverages both visual data and the weight of all the contents within the vending machine, enabling accurate needle counting while effectively detecting foreign objects.
Speaker(s):
Evan Damron, B.S.
University of Kentucky
Author(s):
Evan Damron, B.S. - University of Kentucky; Samuel Armstrong, MS - University of Kentucky;
Presentation Type: Poster - Regular
Poster Number: 212
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
In this work, we address the challenges of accurately counting needles and detecting foreign objects in a needle exchange vending machine using computer vision techniques. Various methods were explored, with the best results achieved using DINOv3 embeddings. The developed framework leverages both visual data and the weight of all the contents within the vending machine, enabling accurate needle counting while effectively detecting foreign objects.
Speaker(s):
Evan Damron, B.S.
University of Kentucky
Author(s):
Evan Damron, B.S. - University of Kentucky; Samuel Armstrong, MS - University of Kentucky;
Evan
Damron,
B.S. - University of Kentucky
Generating Clinically Relevant Questions from EHR Data Using Large Language Models: An Iterative Evaluation Study
Presentation Type: Poster Invite - Regular
Poster Number: 213
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Large language models (LLMs) offer new opportunities to support patient engagement by generating tailored, comprehensible question prompts that help patients prepare for clinician visits. Effective patient–clinician communication is particularly critical when interpreting laboratory test results, yet many patients struggle to identify what to ask or how to relate lab values to their diagnoses, medications, and overall health. This study evaluates the feasibility of using LLMs to generate clinically relevant, patient-friendly questions grounded in electronic health record (EHR) data. Using nine de-identified clinical profiles of adults with Type 2 diabetes and chronic kidney disease, we extracted the most recent laboratory values, medications, and diagnoses to construct structured textual profiles. These profiles informed an iterative LLM prompt-engineering process involving GPT-4o and LLaMA 3.2. Across three rounds of clinician-in-the-loop refinement, three board-certified family physicians evaluated over 300 generated questions for clarity, clinical appropriateness, primary-care relevance, and actionability. Prompt refinements emphasizing abnormal results, alignment with diagnoses, and patient-centered language produced substantial improvements: in Round 2, clinicians rated 100% of questions as clear and clinically appropriate, with increases across all Likert-based usefulness metrics. Model comparisons indicated complementary strengths—GPT-4o generated more readable questions, while LLaMA 3.2 produced more clinically detailed prompts. A subsequent evaluation with 134 patients demonstrated high understandability and moderate-to-high usefulness for most questions, especially those focused on lifestyle guidance and actionable next steps. Findings demonstrate that LLMs, when guided by structured EHR data and iterative expert feedback, can reliably generate questions that enhance patient preparedness, support shared decision-making, and improve comprehension of lab results.
Speaker(s):
Zhe He, PhD, FIAHSI, FAMIA
Florida State University
Author(s):
Zhe He, PhD, FIAHSI, FAMIA - Florida State University; Balu Bhasuran, Ph.D - Florida State University; Mia Lustria, PhD - Florida State University; Karim Hanna, MD - University of South Florida Health; Michael Killian, PhD - Florida State University; Cindy Shavor, MD - University of South Florida; Mandy Dailey, MD - University of South Florida; Xiao Luo, PhD - Southern Methodist University;
Presentation Type: Poster Invite - Regular
Poster Number: 213
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Large language models (LLMs) offer new opportunities to support patient engagement by generating tailored, comprehensible question prompts that help patients prepare for clinician visits. Effective patient–clinician communication is particularly critical when interpreting laboratory test results, yet many patients struggle to identify what to ask or how to relate lab values to their diagnoses, medications, and overall health. This study evaluates the feasibility of using LLMs to generate clinically relevant, patient-friendly questions grounded in electronic health record (EHR) data. Using nine de-identified clinical profiles of adults with Type 2 diabetes and chronic kidney disease, we extracted the most recent laboratory values, medications, and diagnoses to construct structured textual profiles. These profiles informed an iterative LLM prompt-engineering process involving GPT-4o and LLaMA 3.2. Across three rounds of clinician-in-the-loop refinement, three board-certified family physicians evaluated over 300 generated questions for clarity, clinical appropriateness, primary-care relevance, and actionability. Prompt refinements emphasizing abnormal results, alignment with diagnoses, and patient-centered language produced substantial improvements: in Round 2, clinicians rated 100% of questions as clear and clinically appropriate, with increases across all Likert-based usefulness metrics. Model comparisons indicated complementary strengths—GPT-4o generated more readable questions, while LLaMA 3.2 produced more clinically detailed prompts. A subsequent evaluation with 134 patients demonstrated high understandability and moderate-to-high usefulness for most questions, especially those focused on lifestyle guidance and actionable next steps. Findings demonstrate that LLMs, when guided by structured EHR data and iterative expert feedback, can reliably generate questions that enhance patient preparedness, support shared decision-making, and improve comprehension of lab results.
Speaker(s):
Zhe He, PhD, FIAHSI, FAMIA
Florida State University
Author(s):
Zhe He, PhD, FIAHSI, FAMIA - Florida State University; Balu Bhasuran, Ph.D - Florida State University; Mia Lustria, PhD - Florida State University; Karim Hanna, MD - University of South Florida Health; Michael Killian, PhD - Florida State University; Cindy Shavor, MD - University of South Florida; Mandy Dailey, MD - University of South Florida; Xiao Luo, PhD - Southern Methodist University;
Zhe
He,
PhD, FIAHSI, FAMIA - Florida State University
Enhancing Dietary Supplement Question Answering using an Agent Built on the eDISK Knowledge Graph
Presentation Type: Poster - Regular
Poster Number: 214
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
The eDISK Agent Chatbot integrates a large dietary supplement knowledge graph with LLM-guided retrieval and reasoning to support evidence-based question answering. The agent processes natural language or image queries, maps them to eDISK entities, and orchestrates KG-based grounding, contextual traversal, and graph-driven inference. Module-level outputs are integrated into a citation-aware response, reducing hallucination and improving trustworthiness for research, consumer education, and future decision-support use.
Speaker(s):
Yu Hou, PhD
University of Minnesota
Author(s):
Yu Hou, PhD - University of Minnesota; Rui Zhang, PhD, FACMI, FAMIA, FIAHSI - University of Minnesota, Twin Cities;
Presentation Type: Poster - Regular
Poster Number: 214
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
The eDISK Agent Chatbot integrates a large dietary supplement knowledge graph with LLM-guided retrieval and reasoning to support evidence-based question answering. The agent processes natural language or image queries, maps them to eDISK entities, and orchestrates KG-based grounding, contextual traversal, and graph-driven inference. Module-level outputs are integrated into a citation-aware response, reducing hallucination and improving trustworthiness for research, consumer education, and future decision-support use.
Speaker(s):
Yu Hou, PhD
University of Minnesota
Author(s):
Yu Hou, PhD - University of Minnesota; Rui Zhang, PhD, FACMI, FAMIA, FIAHSI - University of Minnesota, Twin Cities;
Yu
Hou,
PhD - University of Minnesota
Developing an Adaptive Self-Corrective Question Answering Framework for Rare Disease Patients and Families
Presentation Type: Poster Invite - Regular
Poster Number: 215
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
We developed an adaptive, self-corrective LLM framework for rare disease question answering that integrates a dedicated evaluator model and dynamic context refinement. Using locally deployed open-weight LLMs, the system generates patient-facing answers, evaluates their accuracy, and iteratively improves them through evaluator-guided feedback. Findings demonstrate substantial gains in answer quality, highlighting the promise of self-corrective LLM pipelines for trustworthy, privacy-preserving clinical informatics applications.
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; Inez Oh, PhD - Institute for Informatics at Washington University in St. Louis, School of Medicine; Aditi Gupta - Washington University in St. Louis; Bryan Sisk, MD, MSCI - Washington University in St. Louis, School of Medicine; Albert Lai, PhD, FACMI, FAMIA - Washington University;
Presentation Type: Poster Invite - Regular
Poster Number: 215
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
We developed an adaptive, self-corrective LLM framework for rare disease question answering that integrates a dedicated evaluator model and dynamic context refinement. Using locally deployed open-weight LLMs, the system generates patient-facing answers, evaluates their accuracy, and iteratively improves them through evaluator-guided feedback. Findings demonstrate substantial gains in answer quality, highlighting the promise of self-corrective LLM pipelines for trustworthy, privacy-preserving clinical informatics applications.
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; Inez Oh, PhD - Institute for Informatics at Washington University in St. Louis, School of Medicine; Aditi Gupta - Washington University in St. Louis; Bryan Sisk, MD, MSCI - Washington University in St. Louis, School of Medicine; Albert Lai, PhD, FACMI, FAMIA - Washington University;
Min
Zhao,
MS - Washington University in St. Louis, School of Medicine
Using Clinical Ontologies and Interpretable Machine Learning Models to Scalably, Transparently, and Accurately Automate Prior Authorization
Presentation Type: Poster - Regular
Poster Number: 216
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
To automatically approve prior authorization (PA) requests, one must verify that required evidence is present in unstructured clinical documentation. We have developed a system that does so transparently, scalably, and accurately. First, clinicians translate coverage determination policies into standardized clinical concepts from medical ontologies. These concepts are then programmatically found in clinical documents using NLP. The prevalence of relevant concepts found can predict approval status through neural classification models, which routinely achieve AUC scores ≥0.8.
Speaker(s):
Samuel Zorowitz, PhD
Cohere Health
Author(s):
David Coar, MA - Cohere Health; Christopher de Freitas, MA - Cohere Health; Parth Jawale, MA - Cohere Health; Paulo Pinho, MD - Lumeris; Gigi Yuen-Reed, PHD;
Presentation Type: Poster - Regular
Poster Number: 216
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
To automatically approve prior authorization (PA) requests, one must verify that required evidence is present in unstructured clinical documentation. We have developed a system that does so transparently, scalably, and accurately. First, clinicians translate coverage determination policies into standardized clinical concepts from medical ontologies. These concepts are then programmatically found in clinical documents using NLP. The prevalence of relevant concepts found can predict approval status through neural classification models, which routinely achieve AUC scores ≥0.8.
Speaker(s):
Samuel Zorowitz, PhD
Cohere Health
Author(s):
David Coar, MA - Cohere Health; Christopher de Freitas, MA - Cohere Health; Parth Jawale, MA - Cohere Health; Paulo Pinho, MD - Lumeris; Gigi Yuen-Reed, PHD;
Samuel
Zorowitz,
PhD - Cohere Health
Fairness-Aware Fine-Tuning of Vision-Language Models for Medical Glaucoma Diagnosis
Presentation Type: Poster Invite - Regular
Poster Number: 217
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Vision-language models achieve expert-level performance on medical imaging tasks but exhibit significant diagnostic accuracy disparities across demographic groups. We introduce fairness-aware Low-Rank Adaptation for medical VLMs, combining parameter efficiency with explicit fairness optimization.
Our key algorithmic contribution is a differentiable MaxAccGap loss that enables end-to-end optimization of accuracy parity across demographic groups. We propose three methods: FR-LoRA integrates MaxAccGap regularization into the training objective, GR-LoRA applies inverse frequency weighting to balance gradient contributions, and Hybrid-LoRA combines both mechanisms.
Evaluated on 10,000 glaucoma fundus images, GR-LoRA reduces diagnostic accuracy disparities by 69% while maintaining 53.15% overall accuracy. Ablation studies reveal that strong regularization strength achieves optimal fairness with minimal accuracy trade-off, and race-specific optimization yields 60% disparity reduction.
Our approach requires only 0.24% trainable parameters, enabling practical deployment of fair medical AI in resource-constrained healthcare settings.
Speaker(s):
Zijian Gu, Master
University of Rochester
Author(s):
Zijian Gu, Master - University of Rochester; YUXI LIU, PHD - Indiana University School of Medicine; Zhenhao Zhang, master - Indiana University; Song Wang, Ph.D. - University of Central Florida;
Presentation Type: Poster Invite - Regular
Poster Number: 217
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Vision-language models achieve expert-level performance on medical imaging tasks but exhibit significant diagnostic accuracy disparities across demographic groups. We introduce fairness-aware Low-Rank Adaptation for medical VLMs, combining parameter efficiency with explicit fairness optimization.
Our key algorithmic contribution is a differentiable MaxAccGap loss that enables end-to-end optimization of accuracy parity across demographic groups. We propose three methods: FR-LoRA integrates MaxAccGap regularization into the training objective, GR-LoRA applies inverse frequency weighting to balance gradient contributions, and Hybrid-LoRA combines both mechanisms.
Evaluated on 10,000 glaucoma fundus images, GR-LoRA reduces diagnostic accuracy disparities by 69% while maintaining 53.15% overall accuracy. Ablation studies reveal that strong regularization strength achieves optimal fairness with minimal accuracy trade-off, and race-specific optimization yields 60% disparity reduction.
Our approach requires only 0.24% trainable parameters, enabling practical deployment of fair medical AI in resource-constrained healthcare settings.
Speaker(s):
Zijian Gu, Master
University of Rochester
Author(s):
Zijian Gu, Master - University of Rochester; YUXI LIU, PHD - Indiana University School of Medicine; Zhenhao Zhang, master - Indiana University; Song Wang, Ph.D. - University of Central Florida;
Zijian
Gu,
Master - University of Rochester
Predicting Early-Onset Colorectal Cancer Risk in Routine Care
Presentation Type: Poster - Regular
Poster Number: 218
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Introduction: Colorectal cancer (CRC) remains a leading cause of cancer-related mortality in the United States, with rising incidence among younger adults aged 18–49 years, referred to as early-onset CRC (EOCRC). Existing screening strategies are predominantly age-based and may miss high-risk individuals who routinely visit outpatient clinics. We aimed to develop a predictive model using routine electronic health record (EHR) data to identify patients at elevated 2-year risk of EOCRC.
Methods: This retrospective cohort study used EHRs from Washington University and BJC Health System. We included adults aged 18–49 years with at least one outpatient visit in 2022 and at least 365 days of prior observation, excluding those with previous cancer diagnoses (except non-melanoma skin cancer). EOCRC was defined as ≥2 CRC diagnosis codes within 2 years after the index date. Candidate predictors from the prior year included demographics, medical conditions, medications, procedures, observations, and laboratory data. Features occurring in fewer than 0.04% of patients were removed, yielding 3,856 predictors. We trained an L1-regularized logistic regression model using nested 5-fold cross-validation.
Results: Among 251,097 eligible patients, 113 (0.045%) developed EOCRC. The model achieved a precision of 0.56 and recall of 0.27. This identified group exhibited an EOCRC incidence roughly 1,200-fold higher than the cohort baseline. Top predictors included anemia-related laboratory values (RDW, MCH), Crohn’s disease, and age.
Conclusion: We developed an EHR-based high-dimensional model that enriches for individuals at markedly elevated 2-year EOCRC risk. Future work includes temporal and external validation using EHR data from independent health systems.
Speaker(s):
Linying Zhang, PhD
Washington University in St. Louis
Author(s):
Ruochong Fan, MA - Washington University in St. Louis; Sina Azadnajafabad, MD, MPH - Washington University School of Medicine in St. Louis; Benjamin Bowe, PhD - Washington University School of Medicine in St. Louis; Yin Cao, ScD, MPH - Washington University School of Medicine in St. Louis;
Presentation Type: Poster - Regular
Poster Number: 218
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Introduction: Colorectal cancer (CRC) remains a leading cause of cancer-related mortality in the United States, with rising incidence among younger adults aged 18–49 years, referred to as early-onset CRC (EOCRC). Existing screening strategies are predominantly age-based and may miss high-risk individuals who routinely visit outpatient clinics. We aimed to develop a predictive model using routine electronic health record (EHR) data to identify patients at elevated 2-year risk of EOCRC.
Methods: This retrospective cohort study used EHRs from Washington University and BJC Health System. We included adults aged 18–49 years with at least one outpatient visit in 2022 and at least 365 days of prior observation, excluding those with previous cancer diagnoses (except non-melanoma skin cancer). EOCRC was defined as ≥2 CRC diagnosis codes within 2 years after the index date. Candidate predictors from the prior year included demographics, medical conditions, medications, procedures, observations, and laboratory data. Features occurring in fewer than 0.04% of patients were removed, yielding 3,856 predictors. We trained an L1-regularized logistic regression model using nested 5-fold cross-validation.
Results: Among 251,097 eligible patients, 113 (0.045%) developed EOCRC. The model achieved a precision of 0.56 and recall of 0.27. This identified group exhibited an EOCRC incidence roughly 1,200-fold higher than the cohort baseline. Top predictors included anemia-related laboratory values (RDW, MCH), Crohn’s disease, and age.
Conclusion: We developed an EHR-based high-dimensional model that enriches for individuals at markedly elevated 2-year EOCRC risk. Future work includes temporal and external validation using EHR data from independent health systems.
Speaker(s):
Linying Zhang, PhD
Washington University in St. Louis
Author(s):
Ruochong Fan, MA - Washington University in St. Louis; Sina Azadnajafabad, MD, MPH - Washington University School of Medicine in St. Louis; Benjamin Bowe, PhD - Washington University School of Medicine in St. Louis; Yin Cao, ScD, MPH - Washington University School of Medicine in St. Louis;
Linying
Zhang,
PhD - Washington University in St. Louis
Accurate Skin Lesion Classification Using Multimodal Learning on the HAM10000 and ISIC 2017 Datasets
Presentation Type: Poster Invite - Regular
Poster Number: 219
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
We propose a novel use of multimodal deep learning for the classification of the HAM10000 and ISIC 2017 skin lesion images. We combined the images with patients’ data (e.g., sex, age, lesion location) for training and evaluating a multimodal deep learning classification model. The dataset was split into 70% for training the model, 20% for the validation set, and 10% for the testing set. We compared the multimodal model’s performance to well-known deep learning models that only use images for classification. Our multimodal model outperformed the competitors and achieved the best results. Our model’s accuracy and AUCROC was 0.9411 and 0.9426, respectively, on HAM10000. On ISIC 2017, our model’s accuracy and AUCROC was 0.7971 and 0.8253, respectively. Our study showed that a multimodal deep learning model can outperform traditional deep learning models for skin lesion classification on the HAM10000 and ISIC 2017 datasets.
Speaker(s):
Abdulmateen Adebiyi, MSc
University of Missouri-Columbia
Author(s):
Abdulmateen Adebiyi, MSc - University of Missouri-Columbia; Nader Abdalnabi, MBA/MSB - University of Missouri; Mirna Becevic, PhD - University of Missouri Department of Dermatology; Praveen Rao, PhD - University of Missouri; Eduardo Simoes - University of Missouri; Emily Hoffman Smith, MD - Department of Dermatology, Saint Louis University; Jesse Hirner, MD - University of Missouri;
Presentation Type: Poster Invite - Regular
Poster Number: 219
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
We propose a novel use of multimodal deep learning for the classification of the HAM10000 and ISIC 2017 skin lesion images. We combined the images with patients’ data (e.g., sex, age, lesion location) for training and evaluating a multimodal deep learning classification model. The dataset was split into 70% for training the model, 20% for the validation set, and 10% for the testing set. We compared the multimodal model’s performance to well-known deep learning models that only use images for classification. Our multimodal model outperformed the competitors and achieved the best results. Our model’s accuracy and AUCROC was 0.9411 and 0.9426, respectively, on HAM10000. On ISIC 2017, our model’s accuracy and AUCROC was 0.7971 and 0.8253, respectively. Our study showed that a multimodal deep learning model can outperform traditional deep learning models for skin lesion classification on the HAM10000 and ISIC 2017 datasets.
Speaker(s):
Abdulmateen Adebiyi, MSc
University of Missouri-Columbia
Author(s):
Abdulmateen Adebiyi, MSc - University of Missouri-Columbia; Nader Abdalnabi, MBA/MSB - University of Missouri; Mirna Becevic, PhD - University of Missouri Department of Dermatology; Praveen Rao, PhD - University of Missouri; Eduardo Simoes - University of Missouri; Emily Hoffman Smith, MD - Department of Dermatology, Saint Louis University; Jesse Hirner, MD - University of Missouri;
Abdulmateen
Adebiyi,
MSc - University of Missouri-Columbia
Early Detection of ePRO Non-Adherence in Remote Oncology Monitoring
Presentation Type: Poster Invite - Regular
Poster Number: 220
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
Electronic patient-reported outcomes (ePROs) improve symptom management in oncology, but daily engagement can falter. We frame near-term non-adherence as a next-window prediction task: given the most recent 𝑇 T days of symptom reports (overlapping 𝑇 ∈ { 3 , 5 , 7 } T∈{3,5,7}), forecast whether the following 𝑇 T-day window contains any missed entries. We curate one record per patient-day, reconstruct calendar gaps, apply inclusion criteria (≥35 observed days and ≤30% overall non-adherence), and impute within-window partial missingness by row-wise means (all-missing set to 0). Class imbalance is addressed with class-weighted loss; features are standardized. We compare LSTM, CNN–LSTM, CNN–LSTM + Attention, CNN–LSTM + Multi-Head Attention, GRU, a non-deep baseline (XGBoost), and an imbalance-aware variant (LSTM + SMOTE). Decision thresholds are chosen on validation sweeps (with optional precision floors) and held fixed for test evaluation. On 7-day windows, LSTM and CNN–LSTM with attention deliver the strongest ranking (ROC-AUC ≈ 0.755; accuracy ≈ 0.71–0.73), while GRU attains a balanced profile (accuracy 0.706; precision/recall/F1: 0.595/0.556/0.575). At 5 days, GRU achieves high case-finding sensitivity (recall 0.891; F1 0.469) at the cost of more false positives; attention-augmented CNN–LSTM variants offer higher accuracy/AUC with modest recall, and recall-oriented baselines (e.g., XGBoost or LSTM + SMOTE) emphasize sensitivity over precision. With 3 days, all models show limited minority-class F1. These results support two deployment modes: a 5-day, recall-oriented model for proactive outreach and a 7-day, balanced model for risk triage, with thresholding as a practical lever to match clinical capacity and risk tolerance.
Speaker(s):
AREF SMILEY, Associate Professor/PhD
University of Arizona college of medicine, T-Health Institute
Author(s):
AREF SMILEY, Associate Professor/PhD - University of Arizona college of medicine, T-Health Institute; Kathi Mooney, PhD, RN, FAAN - College of Nursing, The University of Utah, Salt Lake City, UT, USA; Christina Echeverria, MS - College of Nursing, The University of Utah, Salt Lake City, UT, USA; Joseph Finkelstein, MD, PhD - University of Arizona;
Presentation Type: Poster Invite - Regular
Poster Number: 220
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
Electronic patient-reported outcomes (ePROs) improve symptom management in oncology, but daily engagement can falter. We frame near-term non-adherence as a next-window prediction task: given the most recent 𝑇 T days of symptom reports (overlapping 𝑇 ∈ { 3 , 5 , 7 } T∈{3,5,7}), forecast whether the following 𝑇 T-day window contains any missed entries. We curate one record per patient-day, reconstruct calendar gaps, apply inclusion criteria (≥35 observed days and ≤30% overall non-adherence), and impute within-window partial missingness by row-wise means (all-missing set to 0). Class imbalance is addressed with class-weighted loss; features are standardized. We compare LSTM, CNN–LSTM, CNN–LSTM + Attention, CNN–LSTM + Multi-Head Attention, GRU, a non-deep baseline (XGBoost), and an imbalance-aware variant (LSTM + SMOTE). Decision thresholds are chosen on validation sweeps (with optional precision floors) and held fixed for test evaluation. On 7-day windows, LSTM and CNN–LSTM with attention deliver the strongest ranking (ROC-AUC ≈ 0.755; accuracy ≈ 0.71–0.73), while GRU attains a balanced profile (accuracy 0.706; precision/recall/F1: 0.595/0.556/0.575). At 5 days, GRU achieves high case-finding sensitivity (recall 0.891; F1 0.469) at the cost of more false positives; attention-augmented CNN–LSTM variants offer higher accuracy/AUC with modest recall, and recall-oriented baselines (e.g., XGBoost or LSTM + SMOTE) emphasize sensitivity over precision. With 3 days, all models show limited minority-class F1. These results support two deployment modes: a 5-day, recall-oriented model for proactive outreach and a 7-day, balanced model for risk triage, with thresholding as a practical lever to match clinical capacity and risk tolerance.
Speaker(s):
AREF SMILEY, Associate Professor/PhD
University of Arizona college of medicine, T-Health Institute
Author(s):
AREF SMILEY, Associate Professor/PhD - University of Arizona college of medicine, T-Health Institute; Kathi Mooney, PhD, RN, FAAN - College of Nursing, The University of Utah, Salt Lake City, UT, USA; Christina Echeverria, MS - College of Nursing, The University of Utah, Salt Lake City, UT, USA; Joseph Finkelstein, MD, PhD - University of Arizona;
AREF
SMILEY,
Associate Professor/PhD - University of Arizona college of medicine, T-Health Institute
Using meta-data to measure participant engagement with a cancer symptom self-management mobile app
Presentation Type: Poster Invite - Regular
Poster Number: 221
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
Effective symptom self-management includes access to patient education materials, developing self-monitoring skills, and/or receipt of social support. The purposes of this study are to describe how we mapped each component of a cancer symptom self-management mobile app to the theoretical underpinning of the intervention and to describe results from our pilot study to designed to measure participant engagement with each app component.
Speaker(s):
Stephanie Gilbertson-White, PhD, APRN-BC, FAAN
University of Iowa, College of Nursing and Internal Medicine
Author(s):
Marvin Nukunu-Attachey, MS - University of Iowa; W. Nick Street, PhD - University of Iowa; Heath Davis, MS, MLIS, FAMIA - University of Iowa, Carver College of Medicine; Nayung Youn, Research assistance / MSN - University of Iowa College of Nursing;
Presentation Type: Poster Invite - Regular
Poster Number: 221
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
Effective symptom self-management includes access to patient education materials, developing self-monitoring skills, and/or receipt of social support. The purposes of this study are to describe how we mapped each component of a cancer symptom self-management mobile app to the theoretical underpinning of the intervention and to describe results from our pilot study to designed to measure participant engagement with each app component.
Speaker(s):
Stephanie Gilbertson-White, PhD, APRN-BC, FAAN
University of Iowa, College of Nursing and Internal Medicine
Author(s):
Marvin Nukunu-Attachey, MS - University of Iowa; W. Nick Street, PhD - University of Iowa; Heath Davis, MS, MLIS, FAMIA - University of Iowa, Carver College of Medicine; Nayung Youn, Research assistance / MSN - University of Iowa College of Nursing;
Stephanie
Gilbertson-White,
PhD, APRN-BC, FAAN - University of Iowa, College of Nursing and Internal Medicine
AI Residency 2.0: Tracking LLM Progress in Medical Note Summarization
Presentation Type: Poster - Regular
Poster Number: 222
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Imagine a physician with only minutes to review a complex patient's history. EHRs contribute to burnout rather than ease it, yet LLMs offer a promising solution for summarizing notes. The challenge is evaluation: standard NLP metrics don't capture clinical value. In this work, an expert physician (40+ years) assessed LLM progress: earlier models achieved ~90% accuracy, while today's reasoning-enhanced models (Gemini 2.5 Pro, o3, Sonnet 3.7) reached 100% acceptability, eliminating prior failures. Zero harm events occurred.
Speaker(s):
Rene Ahlsdorf, Master of Science
RWTH Aachen University
Author(s):
Rene Ahlsdorf, Master of Science - RWTH Aachen University; Rene Ahlsdorf, MS - RWTH Aachen University; Joseph Terdiman, MD, PhD - Independent Researcher; Daniel Gruhl, PhD - Google LLC;
Presentation Type: Poster - Regular
Poster Number: 222
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Imagine a physician with only minutes to review a complex patient's history. EHRs contribute to burnout rather than ease it, yet LLMs offer a promising solution for summarizing notes. The challenge is evaluation: standard NLP metrics don't capture clinical value. In this work, an expert physician (40+ years) assessed LLM progress: earlier models achieved ~90% accuracy, while today's reasoning-enhanced models (Gemini 2.5 Pro, o3, Sonnet 3.7) reached 100% acceptability, eliminating prior failures. Zero harm events occurred.
Speaker(s):
Rene Ahlsdorf, Master of Science
RWTH Aachen University
Author(s):
Rene Ahlsdorf, Master of Science - RWTH Aachen University; Rene Ahlsdorf, MS - RWTH Aachen University; Joseph Terdiman, MD, PhD - Independent Researcher; Daniel Gruhl, PhD - Google LLC;
Rene
Ahlsdorf,
Master of Science - RWTH Aachen University
Domain-Adapted Small Language Models for Reliable Clinical Triage
Presentation Type: Poster Invite - Regular
Poster Number: 223
Presentation Time: 05:00 PM - 06:30 PM
Working Group: Natural Language Processing Working Group
Primary Track: Clinical Research Informatics
Accurate Emergency Severity Index (ESI) assignment is challenging due to noisy free-text triage documentation, contributing to mistriage and inefficiencies. We evaluate open-source small language models (SLMs) for pediatric ESI prediction across six prompting pipelines and show that vignette-style summaries outperform raw notes and structured-only inputs. QLoRA fine-tuning of Qwen2.5-7B on institution-specific silver and expert data substantially reduces discordance and significant errors while maintaining real-time inference speed, enabling privacy-preserving decision support.
Speaker(s):
Xuan Wang, PhD
Virginia Tech
Author(s):
Manar Aljohani, Phd - Virginia Polytechnic Institute and State University; Brandon Ho, MD - Seattle Children’s Hospital; Kenneth McKinley, MD - Children's National; Dennis Ren, MD - Children’s National; Xuan Wang, Ph.D. in Computer Science - Virginia Polytechnic Institute and State University;
Presentation Type: Poster Invite - Regular
Poster Number: 223
Presentation Time: 05:00 PM - 06:30 PM
Working Group: Natural Language Processing Working Group
Primary Track: Clinical Research Informatics
Accurate Emergency Severity Index (ESI) assignment is challenging due to noisy free-text triage documentation, contributing to mistriage and inefficiencies. We evaluate open-source small language models (SLMs) for pediatric ESI prediction across six prompting pipelines and show that vignette-style summaries outperform raw notes and structured-only inputs. QLoRA fine-tuning of Qwen2.5-7B on institution-specific silver and expert data substantially reduces discordance and significant errors while maintaining real-time inference speed, enabling privacy-preserving decision support.
Speaker(s):
Xuan Wang, PhD
Virginia Tech
Author(s):
Manar Aljohani, Phd - Virginia Polytechnic Institute and State University; Brandon Ho, MD - Seattle Children’s Hospital; Kenneth McKinley, MD - Children's National; Dennis Ren, MD - Children’s National; Xuan Wang, Ph.D. in Computer Science - Virginia Polytechnic Institute and State University;
Xuan
Wang,
PhD - Virginia Tech
Scalable NLP Pipeline for Automated ICF Classification from Clinical Notes
Presentation Type: Poster - Regular
Poster Number: 224
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Traditional approach to manual coding of the WHO's International Classification of Functioning, Disability and Health (ICF) for documenting patient function can be very labor-intensive. This bottleneck hinders the large-scale implementation of standardized disability classification, limiting our ability to understand population-level functional mobility patterns and inform evidence-based rehabilitation practices. We implemented an NLP pipeline to extract ICF concepts from rehabilitation clinical notes of patients with chronic low back pain, leveraging patterns learned from expert annotations. The transformer-based NLP model ClinicalBERT was fine-tuned on 390 expertly annotated notes, achieving 87% precision and 75% recall, and transforming narrative text into structured, machine-readable disability data.
We deployed the model on 3,762 clinical notes from patients with chronic pain, processing the entire dataset in 34 minutes. The system identified 40 unique ICF codes, with mobility impairments accounting for the majority (79% of extractions). To validate clinical relevance, we linked the extracted codes to the Patient-Reported Outcomes Measurement Information Systems (PROMIS) measure. Patients with mobility limitations showed significantly lower physical function T-scores compared to national norms (39-43 vs. 50), confirming that our model captures meaningful functional impairments.
Our work demonstrates that AI can help identify patterns of disability hidden in unstructured clinical narratives on a larger scale. By automating ICF extraction, there is potential to study physical function and disability at the population health level, support data-driven intervention planning, and create pathways for real-time functional profiling across rehabilitation settings.
Speaker(s):
Sang Pak, PT, DPT, ACHIP
UCSF
Author(s):
Yuxi Jiang, M.S. - UCSF; Sara Temple, PT, DPT - UCSF; Sang Pak, PT, DPT, ACHIP - UCSF;
Presentation Type: Poster - Regular
Poster Number: 224
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Traditional approach to manual coding of the WHO's International Classification of Functioning, Disability and Health (ICF) for documenting patient function can be very labor-intensive. This bottleneck hinders the large-scale implementation of standardized disability classification, limiting our ability to understand population-level functional mobility patterns and inform evidence-based rehabilitation practices. We implemented an NLP pipeline to extract ICF concepts from rehabilitation clinical notes of patients with chronic low back pain, leveraging patterns learned from expert annotations. The transformer-based NLP model ClinicalBERT was fine-tuned on 390 expertly annotated notes, achieving 87% precision and 75% recall, and transforming narrative text into structured, machine-readable disability data.
We deployed the model on 3,762 clinical notes from patients with chronic pain, processing the entire dataset in 34 minutes. The system identified 40 unique ICF codes, with mobility impairments accounting for the majority (79% of extractions). To validate clinical relevance, we linked the extracted codes to the Patient-Reported Outcomes Measurement Information Systems (PROMIS) measure. Patients with mobility limitations showed significantly lower physical function T-scores compared to national norms (39-43 vs. 50), confirming that our model captures meaningful functional impairments.
Our work demonstrates that AI can help identify patterns of disability hidden in unstructured clinical narratives on a larger scale. By automating ICF extraction, there is potential to study physical function and disability at the population health level, support data-driven intervention planning, and create pathways for real-time functional profiling across rehabilitation settings.
Speaker(s):
Sang Pak, PT, DPT, ACHIP
UCSF
Author(s):
Yuxi Jiang, M.S. - UCSF; Sara Temple, PT, DPT - UCSF; Sang Pak, PT, DPT, ACHIP - UCSF;
Sang
Pak,
PT, DPT, ACHIP - UCSF
Leveraging LLMs for Structured Data Extraction from Unstructured Patient Records
Presentation Type: Poster Invite - Regular
Poster Number: 225
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Manual chart review remains an extremely time-consuming and resource-intensive component of clinical research, requiring experts to extract often complex information from unstructured electronic health record (EHR) narratives. We present a secure, modular framework for automated structured feature extraction from clinical notes leveraging locally deployed large language models (LLMs) on institutionally approved, Health Insurance Portability and Accountability Act (HIPPA)-compliant compute infrastructure. This system integrates retrieval augmented generation (RAG) and structured response methods of LLMs into a widely deployable and scalable container to provide feature extraction for diverse clinical domains. In evaluation, the framework achieved high accuracy across multiple medical characteristics present in large bodies of patient notes when compared against an expert-annotated dataset and identified several annotation errors missed in manual review. This framework demonstrates the potential of LLM systems to reduce the burden of manual chart review through automated extraction and increase consistency in data capture, accelerating clinical research.
Speaker(s):
Elizabeth Solie, Undergraduate
University of Kentucky
Author(s):
Mitchell Klusty, B.S. Computer Science - University of Kentucky; Elizabeth Solie, Undergraduate - University of Kentucky; Caroline Leach, B.S. in Physics - University of Kentucky; William Logan, B.S. in Computer Engineering - UKY; Lynnet Richey, BS - University of Kentucky; John Gensel, PhD - University of Kentucky; David Szczykutowicz, BA - University of Kentucky; Bryan McLellan, BS - University of Kentucky; Emily Collier, MSLS - University of Kentucky; Samuel Armstrong, MS - University of Kentucky; Cody Bumgardner, PhD - University of Kentucky;
Presentation Type: Poster Invite - Regular
Poster Number: 225
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Manual chart review remains an extremely time-consuming and resource-intensive component of clinical research, requiring experts to extract often complex information from unstructured electronic health record (EHR) narratives. We present a secure, modular framework for automated structured feature extraction from clinical notes leveraging locally deployed large language models (LLMs) on institutionally approved, Health Insurance Portability and Accountability Act (HIPPA)-compliant compute infrastructure. This system integrates retrieval augmented generation (RAG) and structured response methods of LLMs into a widely deployable and scalable container to provide feature extraction for diverse clinical domains. In evaluation, the framework achieved high accuracy across multiple medical characteristics present in large bodies of patient notes when compared against an expert-annotated dataset and identified several annotation errors missed in manual review. This framework demonstrates the potential of LLM systems to reduce the burden of manual chart review through automated extraction and increase consistency in data capture, accelerating clinical research.
Speaker(s):
Elizabeth Solie, Undergraduate
University of Kentucky
Author(s):
Mitchell Klusty, B.S. Computer Science - University of Kentucky; Elizabeth Solie, Undergraduate - University of Kentucky; Caroline Leach, B.S. in Physics - University of Kentucky; William Logan, B.S. in Computer Engineering - UKY; Lynnet Richey, BS - University of Kentucky; John Gensel, PhD - University of Kentucky; David Szczykutowicz, BA - University of Kentucky; Bryan McLellan, BS - University of Kentucky; Emily Collier, MSLS - University of Kentucky; Samuel Armstrong, MS - University of Kentucky; Cody Bumgardner, PhD - University of Kentucky;
Elizabeth
Solie,
Undergraduate - University of Kentucky
Multi-Agent Applications for Family History Extraction
Presentation Type: Poster Invite - Student
Poster Number: 226
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Much of a patient’s family history (FH) information is embedded in clinical narratives, limiting its utility for analytics and decision support. Natural language processing (NLP) systems have been explored for automatic extraction, but development remains resource-intensive. Advances in large language models (LLMs) offer opportunities to tackle the task, but their tendency to hallucinate requires domain-specific training and expert validation. Here, we adopt a multi-agent framework for FH information extraction. Multi-agent systems (MAS) integrate multiple tools and machine learning models that work collaboratively to complete complex tasks. We hypothesize that by dividing the task into multiple sub-tasks and assigning them to specialized agents, the framework can mitigate the’ limitations of LLMs in handling long context chains and improve the overall performance of the model. Our results demonstrate that this multi-agent approach achieves strong performance in FH extraction, surpassing state-of-the-art models and establishing new benchmarks for the task.
Speaker(s):
Zacharia Husain, BS
University of Toronto
Author(s):
Nan Wang, Graduate Student - UTH; Sunyang Fu, PhD, MHI - UTHealth Houston; Liwei Wang, MD, PhD - UTHealth; Qiuhao Lu, PhD - Ensemble Health Partners; Andrew Wen, MS - University of Texas Health Sciences Center at Houston; Hongfang Liu, PhD - University of Texas Health Science Center at Houston;
Presentation Type: Poster Invite - Student
Poster Number: 226
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Much of a patient’s family history (FH) information is embedded in clinical narratives, limiting its utility for analytics and decision support. Natural language processing (NLP) systems have been explored for automatic extraction, but development remains resource-intensive. Advances in large language models (LLMs) offer opportunities to tackle the task, but their tendency to hallucinate requires domain-specific training and expert validation. Here, we adopt a multi-agent framework for FH information extraction. Multi-agent systems (MAS) integrate multiple tools and machine learning models that work collaboratively to complete complex tasks. We hypothesize that by dividing the task into multiple sub-tasks and assigning them to specialized agents, the framework can mitigate the’ limitations of LLMs in handling long context chains and improve the overall performance of the model. Our results demonstrate that this multi-agent approach achieves strong performance in FH extraction, surpassing state-of-the-art models and establishing new benchmarks for the task.
Speaker(s):
Zacharia Husain, BS
University of Toronto
Author(s):
Nan Wang, Graduate Student - UTH; Sunyang Fu, PhD, MHI - UTHealth Houston; Liwei Wang, MD, PhD - UTHealth; Qiuhao Lu, PhD - Ensemble Health Partners; Andrew Wen, MS - University of Texas Health Sciences Center at Houston; Hongfang Liu, PhD - University of Texas Health Science Center at Houston;
Zacharia
Husain,
BS - University of Toronto
Extensional Value Set Curation with Knowledge Graph Augmented LLMs
Presentation Type: Poster - Regular
Poster Number: 227
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Extensional value sets require substantial manual effort to construct and validate. This study presents a two-stage framework that combines transformer-based retrieval with a lightweight LLM augmented with Knowledge Graph (KG) context to automate value set generation. Across three SNOMED CT value sets, supplying KG derived hierarchical and relational context improved precision, recall, and F1 over a non-augmented baseline. Findings suggest that KG enhanced LLM reasoning can improve concordance with curated extensional value sets.
Speaker(s):
Ali Daowd, MD, PhD
Semedy, Inc.
Author(s):
Marcelo Fiszman, MD, Ph.D. FACMI, DipIBLM - Semedy, Inc; Irene Zhuo, MS, MS - Semedy, inc.; Roberto Rocha, MD, PhD, FACMI - Semedy, Inc.;
Presentation Type: Poster - Regular
Poster Number: 227
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Extensional value sets require substantial manual effort to construct and validate. This study presents a two-stage framework that combines transformer-based retrieval with a lightweight LLM augmented with Knowledge Graph (KG) context to automate value set generation. Across three SNOMED CT value sets, supplying KG derived hierarchical and relational context improved precision, recall, and F1 over a non-augmented baseline. Findings suggest that KG enhanced LLM reasoning can improve concordance with curated extensional value sets.
Speaker(s):
Ali Daowd, MD, PhD
Semedy, Inc.
Author(s):
Marcelo Fiszman, MD, Ph.D. FACMI, DipIBLM - Semedy, Inc; Irene Zhuo, MS, MS - Semedy, inc.; Roberto Rocha, MD, PhD, FACMI - Semedy, Inc.;
Ali
Daowd,
MD, PhD - Semedy, Inc.
Common Laboratory Test Results-Based Biological Age, Aging, and its Clinical Associations
Presentation Type: Poster Invite - Regular
Poster Number: 229
Presentation Time: 05:00 PM - 06:30 PM
Working Group: Genomics and Translational Bioinformatics Working Group
Primary Track: Data Science/Artificial Intelligence
Background:
Biological age (BA), derived from physiological markers, may better reflect health status than chronological age (CA). Laboratory tests offer a scalable approach for estimating BA, yet their clinical implications remain underexplored.
Objectives:
To improve a laboratory-based BA prediction model and evaluate its associations with health outcomes in two large cohorts across a broad age range.
Methods:
We trained XGBoost models on common laboratory tests from the UK Biobank (UKBB) and Leumit Health Services (LHS). We examined associations between BA–CA deviation and 21 diseases, and assessed aging rate using repeated measurements.
Results:
Individuals with lower BA than CA showed reduced disease prevalence, while those with higher BA exhibited increased risk. Findings were consistent across UKBB and LHS. Aging rate analyses further revealed disease-specific associations with accelerated or decelerated biological aging.
Conclusions:
Lab-derived BA is reproducibly associated with health status and aging dynamics in real-world populations.
Speaker(s):
Nadav Rappoport, Ph.D.
Ben-Gurion University of the Negev
Author(s):
Presentation Type: Poster Invite - Regular
Poster Number: 229
Presentation Time: 05:00 PM - 06:30 PM
Working Group: Genomics and Translational Bioinformatics Working Group
Primary Track: Data Science/Artificial Intelligence
Background:
Biological age (BA), derived from physiological markers, may better reflect health status than chronological age (CA). Laboratory tests offer a scalable approach for estimating BA, yet their clinical implications remain underexplored.
Objectives:
To improve a laboratory-based BA prediction model and evaluate its associations with health outcomes in two large cohorts across a broad age range.
Methods:
We trained XGBoost models on common laboratory tests from the UK Biobank (UKBB) and Leumit Health Services (LHS). We examined associations between BA–CA deviation and 21 diseases, and assessed aging rate using repeated measurements.
Results:
Individuals with lower BA than CA showed reduced disease prevalence, while those with higher BA exhibited increased risk. Findings were consistent across UKBB and LHS. Aging rate analyses further revealed disease-specific associations with accelerated or decelerated biological aging.
Conclusions:
Lab-derived BA is reproducibly associated with health status and aging dynamics in real-world populations.
Speaker(s):
Nadav Rappoport, Ph.D.
Ben-Gurion University of the Negev
Author(s):
Nadav
Rappoport,
Ph.D. - Ben-Gurion University of the Negev
MentalNet: A Individual Level Mental Disease Detection Framework Demonstrated with Users’ Social Media Posts
Presentation Type: Poster Invite - Regular
Poster Number: 230
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
In this study, we propose a novel individual level mental disease detection framework, MentalNet, for identifying social media users with mental health conditions with their social media posts for potential digital intervention. The current detection framework includes and implemented data preprocessing, sematic embeddings generated from pre-trained language models, and classifiers based on Convolutional Neural Networks (CNN). Our results show that the MentalNet framework outperforms existing deep learning methods such as GPT-4 and BERT-based models, showing its feasibility and reliability for the effective detection of mental health conditions in the user level. The detection framework is a highly flexible architecture and enables the incorporation of different health documents, embedding techniques, and classification methods. The framework also allows the applications beyond the mental health field, although it is demonstrated for mental disease detection.
Speaker(s):
Ming Huang, PhD
UTHealth Houston
Author(s):
Tzu-Hao Mo, MS - University of Pennsylvania; Jialiang Zhou, MS - University of Pennsylvania; Salih Selek, MD - UTHealth Houston; Hongfang Liu, PhD - University of Texas Health Science Center at Houston; Ming Huang, PhD - UTHealth Houston;
Presentation Type: Poster Invite - Regular
Poster Number: 230
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
In this study, we propose a novel individual level mental disease detection framework, MentalNet, for identifying social media users with mental health conditions with their social media posts for potential digital intervention. The current detection framework includes and implemented data preprocessing, sematic embeddings generated from pre-trained language models, and classifiers based on Convolutional Neural Networks (CNN). Our results show that the MentalNet framework outperforms existing deep learning methods such as GPT-4 and BERT-based models, showing its feasibility and reliability for the effective detection of mental health conditions in the user level. The detection framework is a highly flexible architecture and enables the incorporation of different health documents, embedding techniques, and classification methods. The framework also allows the applications beyond the mental health field, although it is demonstrated for mental disease detection.
Speaker(s):
Ming Huang, PhD
UTHealth Houston
Author(s):
Tzu-Hao Mo, MS - University of Pennsylvania; Jialiang Zhou, MS - University of Pennsylvania; Salih Selek, MD - UTHealth Houston; Hongfang Liu, PhD - University of Texas Health Science Center at Houston; Ming Huang, PhD - UTHealth Houston;
Ming
Huang,
PhD - UTHealth Houston
Towards A Trustworthy Policy: Conservative Q Learning for Treatment Rule Optimization in Perioperative Blood Glucose Monitoring and Management
Presentation Type: Poster - Student
Poster Number: 231
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Perioperative blood glucose (BG) management is highly variable and linked to adverse outcomes. Using data of 97,343 surgical cases from UCSF, we developed an offline reinforcement learning framework with Conservative Q-Learning to optimize BG treatment decisions. Our learned policy consistently outperformed clinician behavior policy in off-policy evaluation. Our work demonstrates the potential for trustworthy AI to support safer perioperative glycemic management.
Speaker(s):
Sylvia Cheng, PhD candidate
University of California, Berkeley
Author(s):
Romain Pirracchio, MD MPH PhD - UCSF; Nicholas Fong, B.A. - University of California, San Francisco;
Presentation Type: Poster - Student
Poster Number: 231
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Data Science/Artificial Intelligence
Perioperative blood glucose (BG) management is highly variable and linked to adverse outcomes. Using data of 97,343 surgical cases from UCSF, we developed an offline reinforcement learning framework with Conservative Q-Learning to optimize BG treatment decisions. Our learned policy consistently outperformed clinician behavior policy in off-policy evaluation. Our work demonstrates the potential for trustworthy AI to support safer perioperative glycemic management.
Speaker(s):
Sylvia Cheng, PhD candidate
University of California, Berkeley
Author(s):
Romain Pirracchio, MD MPH PhD - UCSF; Nicholas Fong, B.A. - University of California, San Francisco;
Sylvia
Cheng,
PhD candidate - University of California, Berkeley
Branching Out: Fifteen Years of Electronic Laboratory Reporting Expansion
Presentation Type: Poster - Regular
Poster Number: 232
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
CDC’s Epidemiology and Laboratory Capacity Cooperative Agreement has supported public health department health information systems since 2010, including advancement of electronic laboratory reporting (ELR) to replace paper-based reporting methods. By 2019, over 93% of laboratory reports sent to public health departments were via ELR. Preexisting ELR infrastructure provided critical foundation during the COVID-19 pandemic, supporting response activities within jurisdictions and nationally. ELR continues to support public health action through faster, cleaner, and more complete data.
Speaker(s):
Caitlin Duffy, MPH
CDC
Author(s):
Caitlin Duffy, MPH - CDC; Megan Mueller, MPH - CDC; Teresa Jue, MPH - Centers for Disease Control and Prevention; Tricia Aden, MT(ASCP) - CDC; Alexandra Ganim, MPH - CDC;
Presentation Type: Poster - Regular
Poster Number: 232
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Clinical Research Informatics
CDC’s Epidemiology and Laboratory Capacity Cooperative Agreement has supported public health department health information systems since 2010, including advancement of electronic laboratory reporting (ELR) to replace paper-based reporting methods. By 2019, over 93% of laboratory reports sent to public health departments were via ELR. Preexisting ELR infrastructure provided critical foundation during the COVID-19 pandemic, supporting response activities within jurisdictions and nationally. ELR continues to support public health action through faster, cleaner, and more complete data.
Speaker(s):
Caitlin Duffy, MPH
CDC
Author(s):
Caitlin Duffy, MPH - CDC; Megan Mueller, MPH - CDC; Teresa Jue, MPH - Centers for Disease Control and Prevention; Tricia Aden, MT(ASCP) - CDC; Alexandra Ganim, MPH - CDC;
Caitlin
Duffy,
MPH - CDC
Arkansas Maternal Health Scorecard: An Empirically-Based Data Portal Developed to Support the Maternal Health Stakeholders in Arkansas
Presentation Type: Poster - Regular
Poster Number: 233
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Translation Bioinformatics/Precision Medicine
The Arkansas Maternal Health Scorecard, the state’s first comprehensive public data portal on maternal health, addresses persistent maternal health challenges by integrating data from multiple sources. It presents data on births, maternal mortality, severe maternal morbidity, behavioral risk factors, and obstetric service closures. To maximize usefulness, usability, and trustworthiness, we identified requirements through a systematic review of comparable portals and key stakeholder interviews. We automated data workflows and developed a storyboard-driven portal with interactive visualizations.
Speaker(s):
Güneş Koru, PhD, FAMIA
University of Arkansas for Medical Sciences, Northwest Regional Campus
Author(s):
Güneş Koru, PhD, FAMIA - University of Arkansas for Medical Sciences, Northwest Regional Campus; Tanvangi Tiwari, MPP; Altay Genç, BS - 3Faktoriyel Technologies; Türkay Palancı, MS - 3Faktoriyel Technologies; Jennifer Callaghan-Koru, PhD;
Presentation Type: Poster - Regular
Poster Number: 233
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Translation Bioinformatics/Precision Medicine
The Arkansas Maternal Health Scorecard, the state’s first comprehensive public data portal on maternal health, addresses persistent maternal health challenges by integrating data from multiple sources. It presents data on births, maternal mortality, severe maternal morbidity, behavioral risk factors, and obstetric service closures. To maximize usefulness, usability, and trustworthiness, we identified requirements through a systematic review of comparable portals and key stakeholder interviews. We automated data workflows and developed a storyboard-driven portal with interactive visualizations.
Speaker(s):
Güneş Koru, PhD, FAMIA
University of Arkansas for Medical Sciences, Northwest Regional Campus
Author(s):
Güneş Koru, PhD, FAMIA - University of Arkansas for Medical Sciences, Northwest Regional Campus; Tanvangi Tiwari, MPP; Altay Genç, BS - 3Faktoriyel Technologies; Türkay Palancı, MS - 3Faktoriyel Technologies; Jennifer Callaghan-Koru, PhD;
Güneş
Koru,
PhD, FAMIA - University of Arkansas for Medical Sciences, Northwest Regional Campus
aoutools: A Python Package for Efficient Polygenic Risk Score Calculation in the All of Us Researcher Workbench
Presentation Type: Poster Invite - Regular
Poster Number: 234
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Translation Bioinformatics/Precision Medicine
Polygenic risk scores (PRS) are central to modern genomics, yet their use in the All of Us Researcher Workbench is limited by computational and workflow challenges. To address this, we developed aoutools, a lightweight Python package that enables efficient PRS computation directly on the native Hail VariantDataset (VDS). It ensures accurate variant matching through minimal representation, resolving multi-allelic sites inherent to the VDS, an issue often overlooked by other tools. aoutools further improves performance with a batch-processing framework that computes genotype dosages once for multiple scores, substantially reducing runtime and resource usage. The package also provides an automated workflow to convert PGS Catalog identifiers directly into final scores, lowering technical barriers for users. Benchmarking against established methods confirms its accuracy and efficiency. aoutools is open-source (MIT), available on the Python Package Index (PyPI), and documented at https://aoutools.readthedocs.io.
Speaker(s):
Jaehyun Joo, Ph.D.
University of Pennsylvania School of Medicine
Author(s):
Jaehyun Joo, Ph.D. - University of Pennsylvania School of Medicine; Dokyoon Kim, PhD - Institute for Biomedical Informatics, University of Pennsylvania;
Presentation Type: Poster Invite - Regular
Poster Number: 234
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Translation Bioinformatics/Precision Medicine
Polygenic risk scores (PRS) are central to modern genomics, yet their use in the All of Us Researcher Workbench is limited by computational and workflow challenges. To address this, we developed aoutools, a lightweight Python package that enables efficient PRS computation directly on the native Hail VariantDataset (VDS). It ensures accurate variant matching through minimal representation, resolving multi-allelic sites inherent to the VDS, an issue often overlooked by other tools. aoutools further improves performance with a batch-processing framework that computes genotype dosages once for multiple scores, substantially reducing runtime and resource usage. The package also provides an automated workflow to convert PGS Catalog identifiers directly into final scores, lowering technical barriers for users. Benchmarking against established methods confirms its accuracy and efficiency. aoutools is open-source (MIT), available on the Python Package Index (PyPI), and documented at https://aoutools.readthedocs.io.
Speaker(s):
Jaehyun Joo, Ph.D.
University of Pennsylvania School of Medicine
Author(s):
Jaehyun Joo, Ph.D. - University of Pennsylvania School of Medicine; Dokyoon Kim, PhD - Institute for Biomedical Informatics, University of Pennsylvania;
Jaehyun
Joo,
Ph.D. - University of Pennsylvania School of Medicine
Accelerating open science in deeply phenotyped longitudinal studies of aging
Presentation Type: Poster - Regular
Poster Number: 236
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Translation Bioinformatics/Precision Medicine
We describe OpenDRaWeR, an open-source, web-based metadata catalog developed within the CLASSIC consortium to accelerate open science in deeply phenotyped longitudinal studies of aging. By capturing harmonized, study-level metadata on design, measures, and codebooks, and by providing governance and data request workflows, OpenDRaWeR reduces administrative burden, increases findability of underused datasets, and enables secondary and coordinated data analyses across diverse aging studies.
Speaker(s):
Jessica Malenfant, MPH
Sage Bionetworks
Author(s):
Shevaun Neupert, PhD - Department of Psychology, North Carolina State University; Stacey Scott, PhD - Department of Psychology, Stony Brook University; Christine Brugh, PhD - Office of Research and Innovation, North Carolina State University; Emily Keller, TBD - Office of Research and Innovation, North Carolina State University; Ann Novakowski, MPH - Sage Bionetworks; Amelia Kallaher, MS - Sage Bionetworks; Samia Ahmed, BS - Sage Bionetworks; Jessica Malenfant, MPH - Sage Bionetworks;
Presentation Type: Poster - Regular
Poster Number: 236
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Translation Bioinformatics/Precision Medicine
We describe OpenDRaWeR, an open-source, web-based metadata catalog developed within the CLASSIC consortium to accelerate open science in deeply phenotyped longitudinal studies of aging. By capturing harmonized, study-level metadata on design, measures, and codebooks, and by providing governance and data request workflows, OpenDRaWeR reduces administrative burden, increases findability of underused datasets, and enables secondary and coordinated data analyses across diverse aging studies.
Speaker(s):
Jessica Malenfant, MPH
Sage Bionetworks
Author(s):
Shevaun Neupert, PhD - Department of Psychology, North Carolina State University; Stacey Scott, PhD - Department of Psychology, Stony Brook University; Christine Brugh, PhD - Office of Research and Innovation, North Carolina State University; Emily Keller, TBD - Office of Research and Innovation, North Carolina State University; Ann Novakowski, MPH - Sage Bionetworks; Amelia Kallaher, MS - Sage Bionetworks; Samia Ahmed, BS - Sage Bionetworks; Jessica Malenfant, MPH - Sage Bionetworks;
Jessica
Malenfant,
MPH - Sage Bionetworks
A Pan-Cell Transcriptomic Aging Clock from Mouse Brain Single-Cell Data
Presentation Type: Poster Invite - Regular
Poster Number: 237
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Translation Bioinformatics/Precision Medicine
In this paper, we develop a pan-cell transcriptomic aging clock with single-cell RNA-seq data from the subventricular zone (SVZ) of the mouse brain. Using RNA-seq of six different cell-types in the SVZ, we present a method of integrating cell-type-specific gene expression information into a single pan-cell aging clock. We discuss the development of our model and compare the performance with cell-type-specific transcriptomic aging clocks. Our research introduces a novel approach to the development of aging clock models that incorporates cell-type-specific transcriptomic patterns of aging into age predictions. Evaluation of our pan-cell model demonstrates improved performance compared to cell-type-specific models, highlighting the importance of the cell-type context of molecular aging features.
Speaker(s):
Kayla Xu, BS
University of Pennsylvania
Author(s):
Kayla Xu, BS - University of Pennsylvania; Travyse Edwards, PhD - University of Pennsylvania; Peiming Li, BS - University of Pennsylvania; Tianhua Zhai, Ph.D. - University of Pennsylvania; Yijing Su, Ph.D. - University of Pennsylvania; Li Shen, Ph.D., FAIMBE, FACMI, FAMIA - University of Pennsylvania;
Presentation Type: Poster Invite - Regular
Poster Number: 237
Presentation Time: 05:00 PM - 06:30 PM
Primary Track: Translation Bioinformatics/Precision Medicine
In this paper, we develop a pan-cell transcriptomic aging clock with single-cell RNA-seq data from the subventricular zone (SVZ) of the mouse brain. Using RNA-seq of six different cell-types in the SVZ, we present a method of integrating cell-type-specific gene expression information into a single pan-cell aging clock. We discuss the development of our model and compare the performance with cell-type-specific transcriptomic aging clocks. Our research introduces a novel approach to the development of aging clock models that incorporates cell-type-specific transcriptomic patterns of aging into age predictions. Evaluation of our pan-cell model demonstrates improved performance compared to cell-type-specific models, highlighting the importance of the cell-type context of molecular aging features.
Speaker(s):
Kayla Xu, BS
University of Pennsylvania
Author(s):
Kayla Xu, BS - University of Pennsylvania; Travyse Edwards, PhD - University of Pennsylvania; Peiming Li, BS - University of Pennsylvania; Tianhua Zhai, Ph.D. - University of Pennsylvania; Yijing Su, Ph.D. - University of Pennsylvania; Li Shen, Ph.D., FAIMBE, FACMI, FAMIA - University of Pennsylvania;
Kayla
Xu,
BS - University of Pennsylvania
Leveraging EHR Data to Identify Gaps in Osteoporosis Screening for Patients with Rheumatoid Arthritis and/or Chronic Glucocorticoids
Category
Clinical Informatics Conference > Poster - Regular
Description
Custom CSS
double-click to edit, do not edit in source