Times are displayed in (UTC-05:00) Central Time (US & Canada) Change
Custom CSS
double-click to edit, do not edit in source
11/9/2026 |
5:00 PM – 6:30 PM |
Trinity Exhibit Hall (Posters)
Poster Session 2
Presentation Type: Poster
This activity is non-accredited.
Description
Join us for an engaging Poster Session where ideas come to life through one-on-one conversations with presenters. Explore a diverse range of topics, learn directly from the researchers behind the work, and dive deeper into the studies that spark your interest. This is your opportunity to connect with others who share your passions, exchange perspectives, and build new professional relationships. Whether you’re looking to gain insights, ask questions, or network with peers, the Poster Session offers a dynamic, interactive environment to expand your knowledge and your professional circle.
Optimizing the Admissions Process for Health Informatics Graduate Programs
Poster Number: 100
2026 Symposium LIEAF Presentation
2026 Symposium 25x5 Presentation
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Administrative Systems, Data transformation/ETL, Documentation Burden, Teaching Innovation
Programmatic Theme: Academic Informatics / LIEAF
We replaced a manual, labor-intensive graduate admissions workflow with an automated Microsoft Office 365 solution. Utilizing Microsoft Bookings for scheduling, Power Automate for Salesforce data ingestion, and Power Apps for a centralized evaluation rubric, this system significantly reduced administrative bottlenecks. The scalable, low-code architecture eliminated redundant data entry, automatically calculated weighted evaluation scores, and improved data integrity, providing a generalizable framework for health informatics programs to streamline admissions.
Speaker(s):
Bunyamin Ozaydin, PhD
University of Alabama at Birmingham
Author(s):
Bunyamin Ozaydin, PhD - University of Alabama at Birmingham; Sue Feldman, RN, MEd, PhD, FACMI - University of Alabama at Birmingham; Garrett Smith, M.A. - University of Alabama at Birmingham;
Bunyamin Ozaydin, PhD - University of Alabama at Birmingham
A Multi-Agent LLM Pipeline for Automated Rationale Generation and Evidence-Ranked Reference Selection in Medical Board Review Items
Poster Number: 101
2026 Symposium LIEAF Presentation
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Information Retrieval, Large Language Models (LLMs)
Programmatic Theme: Academic Informatics / LIEAF
Creating rationales and selecting evidence-based references for medical board examination items is a labor-intensive process that requires clinical expertise and familiarity with current literature. We present a multi-agent pipeline that automates both tasks using a five-node LangGraph workflow orchestrating large language model (LLM) generation, PubMed retrieval with ten progressive search strategies, cross-encoder semantic screening, a two-tier composite ranking algorithm that encodes a six-level evidence hierarchy (H1–H6) derived from institutional reference guidance, and an LLM fallback guarantee ensuring every item receives exactly two AMA-formatted citations. A novel composite scoring function weights semantic relevance, evidence level, recency, journal accessibility, and lexical overlap, with tier-adaptive weighting that prevents off-topic guidelines from outranking on-topic studies. We evaluated the system on pediatric board review items and demonstrated that the pipeline produces rationales and references suitable for expert review, with built-in quality flags that reduce reviewer burden. The architecture is modular, stateless per item, and scales linearly across batch sizes.
Speaker(s):
Ting Wang, PhD
American Board of Pediatrics
Author(s):
Ting Wang, PhD - American Board of Pediatrics;
Ting Wang, PhD - American Board of Pediatrics
Topic Modeling of Social Media Short Texts Incorporating Opinion and Sentiment Contexts in Public Health Discussions
Poster Number: 102
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Natural Language Processing, Large Language Models (LLMs), Data Mining, Deep Learning, Information Extraction, Knowledge Representation & Information Modeling, Public Health
Programmatic Theme: Public Health Informatics
Turning unstructured public responses from large-scale online discourse into actionable public health intelligence remains challenging due to the short and noisy nature of social media text. Traditional topic modeling approaches often overlook emotional and attitudinal contexts that are essential for understanding public reactions. To address this limitation, we propose COSiR, an LLM-based framework that incorporates opinion and sentiment contexts into the topic discovery process. COSiR adopts a two-stage design: micro-level structured reasoning first extracts the subject, opinion, and sentiment of each reply, followed by macro-level global inference to derive coherent discussion topics. Experiments on a dataset of public replies to U.S. CDC posts demonstrate that COSiR consistently outperforms traditional probabilistic and neural topic modeling baselines. The framework achieves higher topic coherence, maintains strong topic diversity, and reduces off-topic noise. By aligning thematic discovery with affective signals, COSiR provides an effective approach for large-scale public health infoveillance and intelligence.
Speaker(s):
Wangjiaxuan Xin, M.S.
University of North Carolina at Charlotte
Author(s):
Wangjiaxuan Xin, M.S. - University of North Carolina at Charlotte; Shuhua Yin, Ph.D. - Department of Epidemiology and Community Health, University of North Carolina at Charlotte; Yaorong Ge, PhD - University of North Carolina at Charlotte; Shi Chen, PhD - UNC Charlotte;
Wangjiaxuan Xin, M.S. - University of North Carolina at Charlotte
Successes and Challenges with Ambient Scribe Deployment in a Federally Qualified Health Center
Poster Number: 103
2026 Symposium Health Equity Presentation
2026 Symposium 25x5 Presentation
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Documentation Burden, Health Equity
Programmatic Theme: Clinical Informatics
Evidence on ambient scribe use is largely concentrated in large academic medical centers and may not reflect the experiences of federally qualified health centers. We analyzed clinician satisfaction and documentation during an ambient scribe pilot at an FQHC using mixed methods. We found no reduction in time per note but observed improvements in satisfaction, 48 hour note closure, and after-hours documentation with rich qualitative insights on workflow adjustments and priorities.
Speaker(s):
Paige Nong, PhD
University of Minnesota
Author(s):
Anna Schulte, MPH - University of Minnesota; Miranda Considine, DO - University of Minnesota; Dori Cross, PhD - University of Minnesota School of Public Health; Eric Maurer, MHA, MHI - Community-University Health Care Center; Michelle Oden, RN, MHI - Community University Health Care Center; Roli Dwivedi, MD - Community university health care center;
Paige Nong, PhD - University of Minnesota
Impact of Individual- and Community-Level Social Determinants of Health on AI Model Performance and Fairness for Glaucoma Progression Prediction
Poster Number: 104
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Machine Learning, Fairness and elimination of bias
Programmatic Theme: Clinical Informatics
This study evaluates how integrating individual- and community-level Social Determinants of Health (SDoH) as input features impacts AI models predicting glaucoma progression. Using XGBoost model on multicenter electronic health record data, we found that individual-level SDoH, particularly substance use history, significantly improves predictive performance and algorithmic fairness. Conversely, relying solely on community-level SDoH fails to improve predictions and potentially exacerbates disparities in fairness across diverse patient populations.
Speaker(s):
Sean Lin, MPH
Byers Eye Institute, Stanford University
Author(s):
Yu Tseng Lin, DVM, MPH - Byers Eye Institute, Stanford University; Joshua Stein, MD, MS - Kellogg Eye Center, University of Michigan; Tina Hernandez-Boussard, PhD - Stanford University; Sophia Wang, MD, MS - Stanford University;
Sean Lin, MPH - Byers Eye Institute, Stanford University
Predicting Communication Effectiveness in Proactive Maternity Care
Poster Number: 105
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Causal Inference, Data Mining, Large Language Models (LLMs), Healthcare Quality, Real-World Evidence Generation
Programmatic Theme: Clinical Informatics
Traditional maternity care relies on episodic visits. To bridge gaps in care, we built a virtual maternity care model that utilizes rapid asynchronous messaging to maintain a continuous patient connection. We present a retrospective causal inference framework using LLMs to evaluate clinician messaging strategies across 56,290 patient-week snapshots. This proof-of-concept "digital twin" approach could act as a communication co-pilot, helping care teams select conversational styles that maximize patient connection and ultimately improve clinical outcomes.
Speaker(s):
Samuel Weiss, Mps
pomelo care
Author(s):
Samuel Weiss, Mps - pomelo care; Robert Martorano, BS - pomelo care; Ian Blakley, BS - pomelo care; Ian Hooley, BS - pomelo care;
Samuel Weiss, Mps - pomelo care
Nursing Informatics 2.0: Nurses’ AI Use in Practice, Perceived Roles in AI Adoption, and Educational Needs for Integrating AI into Clinical Workflows
Poster Number: 106
2026 Symposium LIEAF Presentation
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Workflow, Workforce Development, Teaching Innovation
Working Group: Nursing Informatics Working Group
Programmatic Theme: Academic Informatics / LIEAF
The rapid growth of data science and artificial intelligence (DSAI) is transforming healthcare; however, nursing adoption remains limited. This study examined DSAI use among nurses and nursing informaticians, their perceived roles in AI adoption, and priority education topics. Results indicate that clinical settings are increasingly adopting DSAI tools, yet nurses lack experience with them. Participants emphasized nurses’ leadership in integrating DSAI into workflows and the need for education, including DSAI literacy, evaluation, and safe application.
Speaker(s):
Eun-Shim Nahm, PhD
University of Maryland School of Nursing
Author(s):
Stephen Prouse; Brooke Gaskins, DNP, RN, NI-BC - University of Maryland Medical System; Gyasi Moscou-Jackson, PhD, MHS, RN - University of Maryland Medical System; Kristin L. Seidl, PhD, RN - University of Maryland Medical Center; Jiwon Park, MSN - UMSON; Dahae Choi, MSN - UMSON;
Eun-Shim Nahm, PhD - University of Maryland School of Nursing
Continuous-Value Tokenization Improves Medical Foundation Model Performance
Poster Number: 107
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Large Language Models (LLMs), Deep Learning
Programmatic Theme: Clinical Informatics
Current medical event foundation models typically discretize numeric data into bins, resulting in information loss and bin boundaries that are misaligned with clinical thresholds. Higher-fidelity representation of numeric data through continuous-value tokenization improves the efficiency and predictive performance of medical foundation models by preserving numerical precision and compressing sequence length. This has the potential to improve the efficiency of medical foundation models and address clinical scenarios defined by specific numeric thresholds.
Speaker(s):
Ikgyu Shin, Master of Science in Health Informatics
Yale University
Author(s):
Kent McCann, MD - Yale University; Huan Li, PhD - Yale; Edward Melnick, MD - Yale University, School of Medicine; Mark Iscoe, MD, MHS; Andrew Loza, MD PhD - Yale University School of Medicine | VA Connecticut Healthcare System, US Department of Veterans Affairs;
Ikgyu Shin, Master of Science in Health Informatics - Yale University
Extraction of HER2 IHC Scoring Parameters with Large Language Models
Poster Number: 108
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Information Extraction, Large Language Models (LLMs)
Programmatic Theme: Clinical Informatics
Manual extraction of HER2 immunohistochemistry parameters from unstructured pathology reports is labor-intensive. We adopted large language models to extract granular HER2 attributes (e.g., status, intensity, percent expression, and scoring system) from pathology reports with embedded, complex boilerplate text. Initial evaluation yielded F1 scores above 0.93 for most parameters. Our work demonstrates the utility of LLMs for granular HER2 information extraction and forms the foundation for a pan-tumor clinical trial alert system to identify HER2-expressing patients.
Speaker(s):
Enshuo Hsu, MS
University of Texas MD Anderson Cancer Center
Author(s):
Ana GalanCobo, PhD - The University of Texas MD Anderson Cancer Center; Michelle Elders, BS - University of Texas MD Anderson Cancer Center; Hung Le, PhD - UT MD Anderson Cancer Center; Jiaqi Gu, MD - UTHealth Houston; Hui Chen, MD, PhD - University of Texas MD Anderson Cancer Center; Donna Hansel, MD, PhD - University of Texas MD Anderson Cancer Center; Kirk Roberts, PhD - University of Texas Health Science Center at Houston; Funda Meric-Bernstam, MD - MD Anderson Cancer Center, University of Texas;
Enshuo Hsu, MS - University of Texas MD Anderson Cancer Center
Optimization Instability in Autonomous Agentic Workflows for Clinical Symptom Detection
Poster Number: 109
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Large Language Models (LLMs), Workflow
Programmatic Theme: Clinical Research Informatics
Autonomous prompt optimization frameworks' behavior under class imbalance remains poorly characterized. We evaluated Pythia, a prompt optimization system, on the detection of three clinical symptoms, where we observed prevalence-dependent optimization instability. The instability was most severe for brain fog, where sensitivity oscillated between 0 and 1.0 across iterations. These findings characterize a critical failure mode of autonomous agentic systems under class imbalance and demonstrate stabilization solutions for low-prevalence classification tasks.
Speaker(s):
Cameron Cagan, Bachelors of science
Mass General Brigham
Author(s):
Cameron Cagan, Bachelors of science - Mass General Brigham; Pedram Fard, PhD; Jiazi Tian, Master of Biomedical Informatics - Massachusettes General Hospital; Jingya Cheng, Master - Massachusetts General Hospital,; Shawn Murphy, MD, Ph.D. - Massachusetts General Hospital; Hossein Estiri, PhD - Harvard Medical School;
Cameron Cagan, Bachelors of science - Mass General Brigham
When Not Submitting is Hard: Challenges of Error Detection in Agent-Driven Genetic Test Pre-Authorization Workflows
Poster Number: 110
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Large Language Models (LLMs), Evaluation, Information Retrieval, Administrative Systems
Programmatic Theme: Clinical Research Informatics
Browser-based agents powered by large language models (LLMs) may help automate insurance pre-authorization workflows, but their reliability in clinical settings remains unclear. We developed a sandbox environment with synthetic patient profiles to benchmark agent performance in genetic test pre-authorization tasks. While larger models achieved higher task completion rates, all tested models struggled to detect problematic cases and terminate the ongoing process, highlighting significant limitations in using general-purpose LLM reasoning for complex clinical workflows.
Speaker(s):
Mengshu Nie, MA
Boston Children's Hospital
Author(s):
Mengshu Nie, MA - Boston Children's Hospital; Cong Liu, PhD - Boston Children's Hospital;
Mengshu Nie, MA - Boston Children's Hospital
IA-Coach: Personalized feedback to Improve Clinical Insight and Collaborative Care
Poster Number: 111
2026 Symposium LIEAF Presentation
2026 Symposium 25x5 Presentation
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Teaching Innovation, Healthcare Quality, Large Language Models (LLMs)
Programmatic Theme: Academic Informatics / LIEAF
Generative AI is increasingly used by medical students for summarization, raising concerns about skill development. We created a GenAI-based platform that provides personalized feedback on clinical summaries using expert-validated vignettes. Multiple language models were evaluated on relevance, structure, and language through iterative testing. Google Gemini 2.5 Pro performed best, delivering consistently accurate feedback. The platform shows strong educational potential, though further research is needed to assess its impact on learners’ summarization skills.
Speaker(s):
Katherine Blondon, MD, PhD
Hopitaux Universitaires Geneve
Author(s):
Assiya Sirbal, MD student - University of Geneva; David Vicente Alvarez, MSc - University of Geneva; Douglas Teodoro, PhD - University of Geneva; Katherine Blondon, MD, PhD - Hopitaux Universitaires Geneve;
Katherine Blondon, MD, PhD - Hopitaux Universitaires Geneve
Comparing Multi-Agent Large Language Model Framework Architectures for Detecting Checkpoint Inhibitor Pneumonitis from Longitudinal Electronic Health Records
Poster Number: 112
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Large Language Models (LLMs), Patient Safety, Clinical Decision Support
Working Group: Clinical Information Systems Working Group
Programmatic Theme: Clinical Informatics
This study evaluated three multi-agent LLM frameworks for detecting checkpoint inhibitor pneumonitis (CIP) and inferring onset dates from longitudinal EHR data. All methods utilized a two-level architecture for note-level evidence extraction and patient-level assessment. Results demonstrate that a unified "all-in-one" agent architecture consistently outperforms separate agents for patient-level assessment. Furthermore, note-level summarization proved superior to snippet extraction, with the summary-based all-in-one approach achieving the highest F1-score of 0.81.
Speaker(s):
Yuansheng Zhang, PhD
Harvard Medical School
Author(s):
Yuansheng Zhang, PhD - Harvard Medical School; Danielle Bitterman, MD - Harvard Medical School; Danielle Bitterman, MD - Marvard Medical School;
Yuansheng Zhang, PhD - Harvard Medical School
Clinician experience with EHR-based chart summarization tools in inpatient care: A mixed-methods study
Poster Number: 113
2026 Symposium 25x5 Presentation
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Documentation Burden, Workflow
Programmatic Theme: Clinical Informatics
In a mixed-methods pre–post study of inpatient clinicians, EHR-based chart summarization tools were associated with reduced cognitive workload and improved efficiency for key tasks, including patient orientation and discharge documentation. Documentation-focused summarization showed the strongest perceived value, while orientation-focused summarization depended on workflow fit and content relevance. Qualitative findings highlighted the importance of accuracy, human verification, and actionable synthesis. These results inform generalizable design principles for clinical chart summarization.
Speaker(s):
Lincoln Abbott, MD, FACEP
Hartford Healthcare
Author(s):
Tony Kin Wai Hung, MD, MBA, MSCR, FACP, FAMIA - Hartford HeatlhCare; Candace Carlson, BA - Hartford HealthCare; Daniel Kombert, MD, MBA - Hartford Healthcare; Lincoln Abbott, MD - Hartford HealthCare;
Lincoln Abbott, MD, FACEP - Hartford Healthcare
Forecasting Rheumatoid Arthritis Flares Up to Four Weeks in Advance Using Wearable-Derived Digital Biomarkers and Deep Learning
Poster Number: 114
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Chronic Care Management, Deep Learning, Patient-/Person-Generated Health Data, Mobile Health, Real-World Evidence Generation
Programmatic Theme: Clinical Research Informatics
Continuous wearable data analyzed with transformer-based deep learning models predicted symptomatic and inflammatory Rheumatoid Arthritis flares up to four weeks in advance (7-day ROCAUC/AUPRC 0.95/0.92 and 0.95/0.91, respectively, on out-of-time data). Models were robust to real-world missing data and outperformed classical machine learning approaches. Incorporating daily patient reported outcomes enhanced per-participant sensitivity for inflammatory flare prediction (0.67 to 0.87). Integrating wearable-derived digital biomarkers with patient-reported data enables scalable and proactive RA disease monitoring.
Speaker(s):
Pragya Sharma, PhD
Icahn School of Medicine at Mount Sinai
Author(s):
Pragya Sharma, PhD - Icahn School of Medicine at Mount Sinai; Matteo Danieletto, PhD - Icahn School of Medicine at Mount Sinai; Jessica Whang, MS - Icahn School of Medicine at Mount Sinai; Kyle Landell, BS - Icahn School of Medicine at Mount Sinai; Drew Helmus, PhD - Icahn School of Medicine at Mount Sinai; Mayte Suarez-Farinas, PhD - Icahn School of Medicine at Mount Sinai; Percio Gulko, MD - Icahn School of Medicine at Mount Sinai; Robert Hirten, MD - Icahn School of Medicine at Mount Sinai;
Pragya Sharma, PhD - Icahn School of Medicine at Mount Sinai
The Impact of Quality Improvement on Evaluation of AI: Application to Neonatal Respiratory Outcomes
Poster Number: 115
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Evaluation, Governance, Machine Learning, Deep Learning
Programmatic Theme: Clinical Research Informatics
Quality improvement (QI) initiatives alter clinical workflows, potentially introducing temporal data drift. We evaluated a chest radiograph and clinical variable-based artificial intelligence (AI) model to predict outcomes in neonates with transient tachypnea of the newborn (TTN) during an ongoing QI initiative (N=218). Cross-validation showed moderate discrimination (AUROC 0.77 for CPAP ≥24h), but temporal hold-out testing revealed quarter-to-quarter variability, with performance stabilizing as clinical workflows matured, highlighting the need for continual AI monitoring in dynamic environments.
Speaker(s):
Pragya Sharma, PhD
Icahn School of Medicine at Mount Sinai
Author(s):
Pragya Sharma, PhD - Icahn School of Medicine at Mount Sinai; Felix Richter, MD, PhD - Children's Hospital of Philadelphia; Julia Maslinski, BS - Icahn School of Medicine at Mount Sinai; Richelle Reinhart, MD - Icahn School of Medicine at Mount Sinai; Sushma Krishna, MD - Weill Cornell Medicine; Benjamin Glicksberg, PhD - Icahn School of Medicine at Mount Sinai;
Pragya Sharma, PhD - Icahn School of Medicine at Mount Sinai
Scoring Rubric to Assess Ethical Issues in Clinical Artificial Intelligence Tools
Poster Number: 116
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Clinical Decision Support, Delivering Health Information and Knowledge to the Public, Evaluation, Fairness and elimination of bias, Governance, Health Equity, Policy
Programmatic Theme: Clinical Informatics
A multidisciplinary team of clinical and AI experts developed and tested a scoring rubric to evaluate AI tools in clinical settings from an ethical perspective, considering potential issues that may arise during their use. The rubric provides a replicable scoring system comprising five content areas and four scoring categories. The content areas include evaluation population, accessibility, clinical bias, user control, monitoring and updates. We tested the rubric by applying it to three AI tools designed for clinical decision-making, documentation, and image analysis. The evaluation relied solely on publicly available information to simulate typical clinical settings. The rubric addresses a critical gap in the evaluation and governance of clinical AI tools by providing a structured framework to assess their ethical aspects. This approach enables healthcare systems to address and monitor the ethical impact of AI tools, fostering greater trust among clinicians and patients and guiding improvements in clinical practice.
Speaker(s):
Olga Kravchenko, PhD
University of Pittsbugh
Author(s):
Olga Kravchenko, PhD - University of Pittsbugh; Kevin Kindler, MD - UPMC; Shyam Visweswaran, MD PhD - University of Pittsburgh; Yanshan Wang, PhD - University of Pittsburgh; John Maier, MD - University of Pittsbugh; Jose Abad, MD - University of Pittsburgh;
Olga Kravchenko, PhD - University of Pittsbugh
A RPM-driven Digital Twin of hypertension in an underserved population
Poster Number: 117
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Patient-/Person-Generated Health Data, Large Language Models (LLMs), Health Equity, Machine Learning
Programmatic Theme: Public Health Informatics
We developed a digital twin for systolic blood pressure (SBP) trajectory forecasting in an underserved, primarily Hispanic/Latinx hypertensive population using a recurrent variational autoencoder (rVAE) coupled with LightGBM. Trained on quarterly EHR data from 1,350 patients, the model achieved R²=0.58–0.73 for predicting SBP 15–33 months out, with rVAE latent features reducing RMSE by ~6 mmHg over clinical covariates alone. ACTIVATE remote monitoring patients were distributionally distinct in latent space, with a hypothesis-generating −5.1 mmHg signal at 33 months.
Speaker(s):
Katherine Kim, PhD, MPH, MBA, FAMIA, FIAHSI
University of California Davis
Author(s):
Vinay Pai, PhD, MBA - Health Tequity, LLC.; Joshua Weand, BS - Health Tequity, LLC.; Tyler Kim-Hayes, BS - Health Tequity, LLC; Katherine Kim, PhD, MPH, MBA, FAMIA - University of California Davis;
Katherine Kim, PhD, MPH, MBA, FAMIA, FIAHSI - University of California Davis
AI Health Coaching and Clinical Outcomes in GLP-1 Weight Management: A Large-Scale Deployment Study
Poster Number: 118
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Patient Engagement and Preferences, Causal Inference
Programmatic Theme: Clinical Informatics
We studied 150,000+ patients taking GLP-1 weight-loss medications who had access to a 24/7 AI coaching agent, analyzing 500,000+ conversations. Patients who engaged more frequently lost weight 3.2x faster than non-engagers, and within-person analyses confirmed this was not simply due to more motivated patients engaging more. Higher engagement also predicted greater patient trust, earlier identification of at-risk individuals, broader reach than traditional coaching, and 23% fewer medication orders to achieve equivalent results.
Speaker(s):
John Xing, MEng
Concurrence
Author(s):
Connie Xu, BSE - Concurrence; John Alderete, PhD - Concurrence; Stella Lee, BS - Concurrence; John Xing, MEng - Concurrence;
John Xing, MEng - Concurrence
Digital-Clinical Adversarial Stress Testing (D-CAST): A Multi-Layer Security Framework for Algorithmic Vigilance in Patient-Centered Healthcare
Poster Number: 119
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Patient Safety, Privacy and Security, Large Language Models (LLMs)
Programmatic Theme: Clinical Research Informatics
As AI-driven digital health tools proliferate, the risk of adversarial hallucinations and security breaches poses a direct threat to patient safety. While existing models demonstrate high benchmark accuracy, their resilience against malicious manipulation remains untested. We present the Digital-Clinical Adversarial Stress Testing (D-CAST) framework, a five-layer methodology synthesized from 2022–2026 safety benchmarks. D-CAST evaluates AI systems across sensor stress, security integrity, clinical logic, demographic equity, and operational resilience. Our evaluation reveals that authority impersonation and targeted confabulation are primary failure modes in clinical LLMs. This framework provides a standardized protocol for algorithmic vigilance, ensuring that digital health innovations prioritize Safety-by-Design over Accuracy-by-Default.
Speaker(s):
L. Raymond Guo, PhD.
West Virginia University
Author(s):
Shuimei Liu, S.J.D. - China University of Political Science and Law;
L. Raymond Guo, PhD. - West Virginia University
Semantic and Thematic Characterization of a National Registry of Disrupted National Institutes of Health Grants: A Mixed Text-Mining and Network Analysis
Poster Number: 121
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Data Mining, Information Extraction, Machine Learning, Knowledge Representation & Information Modeling, Natural Language Processing, Quantitative Methods, Policy
Programmatic Theme: Clinical Research Informatics
During 2024-2025, thousands of National Institutes of Health (NIH) grants were subject to administrative disruptions, including termination, freezing, and reinstatement actions. We applied multi-method natural language processing, including TF-IDF, topic modeling, sentence embeddings, and named entity recognition with co-occurrence networks, to characterize a curated registry of 5,474 affected grants (~$17.4 billion) assembled from the Tracking Accountability in Government Grants System (TAGGS). The corpus spans broad biomedical domains; flagged terminology concentrates on equity and disparities contexts with substantial polysemy. Administrative status categories are not separable by abstract semantics alone.
Speaker(s):
Fares Alahdab, MD, MS, MSc, FAHA
University of Missouri
Author(s):
Fares Alahdab, MD, MS, MSc, FAHA - University of Missouri;
Fares Alahdab, MD, MS, MSc, FAHA - University of Missouri
Interpretable Voice Biomarkers for Hearing Loss Using Structured Acoustic Feature Families
Poster Number: 122
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Clinical Decision Support, Machine Learning, Natural Language Processing
Programmatic Theme: Clinical Informatics
Hearing loss is common, underdiagnosed, and associated with cognitive and psychosocial morbidity. Current screening depends on resource-intensive audiometry. Voice production offers a non-invasive and scalable source of biomarkers because degraded auditory feedback in hearing loss alters multiple acoustic subsystems (e.g., speech timing). Using the multi-site Bridge2AI Voice dataset, we developed machine learning models using ML methods (gradient boosting, LASSO, XGBoost) to detect hearing loss from voice recordings with an AUROC of 0.74.
Speaker(s):
Peter Dixon, MD
Medical University of South Carolina
Author(s):
Carl Ehrett, PhD - Clemson University; Paul Heider, PhD - Medical University of South Carolina; Peter Dixon, MD - Medical University of South Carolina;
Peter Dixon, MD - Medical University of South Carolina
Assessing Clinical and Privacy Inference Abilities of Generalist Medical Vision-Language Models
Poster Number: 123
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Large Language Models (LLMs), Privacy and Security
Programmatic Theme: Clinical Research Informatics
Current vision-language models are capable of providing advanced cross-modality inference abilities that could advance medical research and improve patient care. At the same time, adopting these techniques in practice poses novel utility and privacy issues. In this on-going study, we systematically analyze the inference ability of a state-of-the-art medical vision language model to predict clinical outcomes from limited samples and assess potential privacy risks in inferring sensitive information from de-identified medical images.
Speaker(s):
Luca Bonomi, PhD
Vanderbilt University Department of Biomedical Informatics
Author(s):
Luca Bonomi, PhD - Vanderbilt University Department of Biomedical Informatics; Liyue Fan, PhD - University of North Carolina at Charlotte;
Luca Bonomi, PhD - Vanderbilt University Department of Biomedical Informatics
Clinician-in-the-Loop: A Workflow-Oriented Comparative Evaluation of SAM2 for Prompt-Aware Medical Image Segmentation
Poster Number: 124
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, User-centered Design Methods, Clinical Decision Support
Programmatic Theme: Clinical Informatics
Medical image segmentation is critical for diagnosis and treatment planning, yet it remains resource-intensive and difficult to evaluate. AI has been increasingly used to support segmentation with mixed results in effectiveness. This study compares the segmentation performance of three architectures: The foundational SAM2, a fine-tuned SAM2 model, and the U-Net architecture. Additionally, we evaluate seven prompt modes for both SAM2 variants to assess how prompt selection influences segmentation quality. A publicly-available TMJ CBCT dataset was utilized for the research. Fine-tuning employed both center-point and tight-box prompts, and evaluation was performed under multiple prompt conditions, including single-point, multi-point, and box-based modes. Model performance was assessed using Dice score, Intersection-over-Union, Boundary F1-score, Hausdorff distance, and inference time. Our findings demonstrate that model choice and prompt strategy require careful alignment with clinical tasks and workflow requirements, balancing segmentation accuracy, inference time, and available computational resources.
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; Jennifer Chang, DDS, MSD - The University of Texas Health Science Center at Houston; Ritu Tiwari, BDS, MDS - The University of Texas Health Science Center at Houston; Ryan Vahdani, DDS, MHA, MAGD - The University of Texas Health Science Center at Houston; Toufeeq Syed, PhD, MS - UT Health Houston;
Toufeeq Syed, PhD, MS - UT Health Houston
Generative AI–Enhanced Therapeutic Recreation for Behavioral Well-Being in People Living with Disabilities: Findings from the ArtAI Pilot
Poster Number: 125
2026 Symposium Health Equity Presentation
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Disability, Accessibility, and Human Function, Population Health
Programmatic Theme: Clinical Research Informatics
Care-AI is a generative artificial intelligence (GenAI)–enabled therapeutic recreation program designed to support creative expression and behavioral well-being among people living with disabilities. The program builds on the ArtAI pilot, in which 28 participants completed 98 sessions of GenAI-assisted creative activities and reported improvements in engagement, mood, relaxation, emotional expression, and autonomy. Care-AI expands these approaches to include art, music, and storytelling and evaluates behavioral health outcomes across diverse disability populations.
Speaker(s):
Senait Tekle, Phd
George Washington University
Author(s):
Debi Alexander, JD - Specially Adaptive Resource Centers (SPARC); Tadas Sean Vasaitis, Phd - University of Maryland Eastern Shore; Linda Zanin, EdD - George Washington University; Elena Libin, Phd - George Washington University; Jordan Potash, Phd - George Washington University; Kathryn Rader, MAT - Specially Adaptive Resource Centers (SPARC); Qing Zeng, PhD - George Washington University; Melissa M Goldstein, JD - George Washington University;
Senait Tekle, Phd - George Washington University
From Validation to Simulation to Operationalization: The Degradation of AI Performance as Workflow and Human Interactions are Incorporated
Poster Number: 126
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Evaluation, Human-computer Interaction, Clinical Decision Support, Workflow
Programmatic Theme: Clinical Informatics
Our investigation follows the change in artificial intelligence performance metrics through the implementation of a clinical based, vendor created predictive model, from validation, simulation, and operationalization. We witnessed a degradation in metrics which could be explained by unforeseen human behaviors and reactions to alert warnings. The ability to account for these interactions in future predictive model implementations could lead to an improved experience for already overburdened clinicians.
Speaker(s):
Lauren Cooper, MS
ECU Health
Author(s):
Lauren Cooper, MS - ECU Health; Christopher Dennis, PharmD, MMCi, BCPS, CPHIMS - ECU Health; john Hanna, MD - ECU Health;
Lauren Cooper, MS - ECU Health
From Frontline Problems to Embedded AI Tools: The EMBED-AI Framework for Learning Health Systems
Poster Number: 127
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Informatics Implementation, Real-World Evidence Generation
Programmatic Theme: Translational Bioinformatics
Generative artificial intelligence (AI) is expanding rapidly across health-related domains, yet many promising ideas fail to become trusted, sustainable tools integrated into real workflows. Address this gap, EMBED-AI, a seven-phase lifecycle framework, moves from frontline problems to embedded AI tools within Learning Health Systems (LHS). The framework was inductively formalized from implementation work led through the Center for Health Innovation and Technology Advancement (CHITA), including the Nursing-AI Challenge, Nursing AI Studio (NAIS), collaboration with institutional technology leadership, and the development of AI tools for nursing education, research support, and caregiving support. EMBED-AI comprises seven phases: Elicit, Map, Bridge, Engineer, Deploy, Appraise, and Institutionalize. The framework connects frontline problem identification, structured intake, technical translation, prototype development, governance alignment, evaluation, and sustainability planning. EMBED-AI offers a practical informatics framework for organizing AI implementation as a continuous learning cycle rather than a one-time build.
Speaker(s):
Bo Xie, PhD
University of Texas at Austin
Author(s):
Bo Xie, PhD - University of Texas at Austin;
Bo Xie, PhD - University of Texas at Austin
Evaluation and Mitigation of Bias in Large Language Models for Real-World Clinical Applications
Poster Number: 128
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Healthcare Quality, Human-computer Interaction
Programmatic Theme: Clinical Informatics
LLMs have the potential to assist physicians and improve the healthcare industry, but their outputs can be biased. This systematic review aims to identify the methods used, types of bias studied, and findings within the last three years of research in this area: demographic bias in LLMs applied in the medical.
Speaker(s):
Jie Yang, PhD, FACMI, FAMIA
Harvard Medical School
Author(s):
Joshua Zhang, Bachelor of Arts - Brigham and Women's Hospital; Jiageng Wu, PhD - Brigham and Women's Hospital and Harvard Medical School; Isabella Andrade, BS - Harvard University; Enqi Liu, Master of Science - Harvard University; Alice Yang, MS - Harvard University; Bowen Gu, MS - Brigham and Women's Hospital; Richard Wyss, PhD - Brigham and Women's Hospital; Kueiyu Joshua Lin, MD - Brigham and Women's Hospital; Jie Yang, PhD, FACMI, FAMIA - Harvard Medical School;
Jie Yang, PhD, FACMI, FAMIA - Harvard Medical School
Artificial Intelligence to Predict Systemic Mastocytosis in Patients from Electronic Health Records
Poster Number: 129
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Large Language Models (LLMs), Machine Learning, Information Retrieval, Information Extraction, Natural Language Processing, Real-World Evidence Generation, Workflow
Programmatic Theme: Clinical Informatics
Non-specific symptom presentation in systemic mastocytosis often leads to diagnostic delays, exposing patients to life-threatening anaphylaxis and progressive organ damage. Current diagnostic pathways rely on molecular testing and bone marrow biopsy after clinical suspicion arises, yet no workflows exist to identify at-risk patients using routine electronic health record data. We developed an informatics pipeline integrating large language model-assisted and rule-based feature extraction with machine learning models to predict systemic mastocytosis risk 12 months before diagnosis.
Speaker(s):
Muhammad Ali Khan, M.B.B.S.
Mayo Clinic
Author(s):
Muhammad Ali Khan, M.B.B.S. - Mayo Clinic; Richard Yim, PhD - University of California San Francisco; Hunter Mills - University of California San Francisco; Umair Ayub, PhD - Mayo Clinic; Syed Arsalan Ahmed Naqvi, MD - Mayo Clinic; Gundolf Schenk - University of California, San Francisco; Habibeh Ashouri Choshali, PhD - University of California San Francisco; Irbaz Riaz, MD, MS, MBI, PhD - Mayo Clinic; Cecilia Arana Yi, MD - Mayo Clinic; Vivek Rudrapatna, MD, PhD - University of California, San Francisco;
Muhammad Ali Khan, M.B.B.S. - Mayo Clinic
Human-Centered Requirements for Ubiquitous AI Health Risk Assessment
Poster Number: 132
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Human-computer Interaction, Usability
Programmatic Theme: Consumer Health Informatics
This narrative review examines requirements for AI-based health risk assessment technologies (e.g., AI-enhanced wearables and chatbots) to function effectively as ubiquitous computing systems. Health risk factors such as hypertension are highly prevalent, and their monitoring and management are often integral parts of everyday life. Thus, this review identifies human-centered requirements for ubiquitous deployment, including interpretability, privacy, fairness, context-awareness, trust, personalization, and construct validity, while highlighting gaps in the literature, including a lack of patient-centered perspectives.
Speaker(s):
Bianka Onwumbiko, B.S.
University of Maryland, Baltimore County
Author(s):
Tera Reynolds, PhD, MPH, MS - University of Maryland Baltimore County; Andrea Kleinsmith - University of Maryland Baltimore County;
Bianka Onwumbiko, B.S. - University of Maryland, Baltimore County
The Missing Link in Healthcare AI Fairness: From Principles to Practice
Poster Number: 133
2026 Symposium Health Equity Presentation
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Diversity, Equity, Inclusion, and Accessibility, Fairness and elimination of bias, Machine Learning, Health Equity, Governance, Policy, Usability
Programmatic Theme: Public Health Informatics
Healthcare AI increasingly informs clinical decisions, while existing fairness guidance remains largely principle oriented. We integrated 3 established guidelines to develop a 47-item checklist structured by AI lifecycle, fairness type, and objective categories, defined 82 vulnerability patterns; and implemented a web prototype. By translating fairness outputs into structured risks and follow-up actions, this framework supports more systematic planning, prioritization, and cross-stakeholder coordination across the healthcare AI lifecycle.
Speaker(s):
MyeongHo Shin, Undergraduate Student
Hallym Univesity
Author(s):
MyeongHo Shin, Undergraduate Student - Hallym Univesity; Adrian Zai, MD, PhD, MPH - University of Massachusetts Chan Medical School; JaeYong Yu, PhD - Hallym University;
MyeongHo Shin, Undergraduate Student - Hallym Univesity
AI-Simulated Interviews as a Controlled Reference Baseline for Quantifying Context-Dependent Diagnostic Reasoning Dynamics
Poster Number: 134
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Clinical Decision Support, Quantitative Methods, Human-computer Interaction
Programmatic Theme: Clinical Informatics
AI-simulated interviews were evaluated as a controlled reference condition for modeling context-dependent diagnostic reasoning dynamics. In a between-subject study of 60 final-year clinical trainees, effect-size analyses revealed substantial differences in hypothesis expansion, switching, persistence, and engagement timing between AI-simulated and emergency department encounters. These findings support computational characterization of context-associated reasoning variability for future AI-assisted calibration.
Speaker(s):
Ming-Yuan Huang, MD, MSc
MacKay Memorial Hospital
Author(s):
Ming-Yuan Huang, MD, MSc - MacKay Memorial Hospital;
Ming-Yuan Huang, MD, MSc - MacKay Memorial Hospital
Generation of Synthetic Data in Health Surveys Using Large Language Models
Poster Number: 135
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Large Language Models (LLMs), Global Health, Quantitative Methods, Surveys and Needs Analysis
Programmatic Theme: Clinical Informatics
To evaluate the consistency and psychometric plausibility of synthetic data generated by an LLM to simulate the responses of survey participants (user personas) in a national health survey. The synthetic PHQ-9, PHQ-8, and GAD-7 instruments showed optimal factor fit and high internal consistency. Synthetic measures were positively and significantly correlated with the real PHQ-2 and negatively correlated with EQ-5D, with moderate to high correlations, particularly for PHQ-8/PHQ-9 and GAD-7.
Speaker(s):
David Villarreal-Zegarra
Universidad Cesa Vallejo
Author(s):
David Villarreal-Zegarra, MPH - Universidad Científica del Sur; Luciana Bellido-Boza, MSc - Facultad de Ciencias de la Salud, Universidad Peruana de Ciencias Aplicadas;
David Villarreal-Zegarra - Universidad Cesa Vallejo
AT Navigator: An AI-Powered Chatbot for Assistive Technology Discovery and Funding in Pennsylvania
Poster Number: 136
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Large Language Models (LLMs), Population Health
Programmatic Theme: Public Health Informatics
AT Navigator is an AI-powered web application that was developed in partnership with the Pennsylvania Assistive Technology Foundation. This application streamlines assistive technology (AT) discovery and funding for individuals with disabilities in Pennsylvania. Using a Retrieval-Augmented Generation (RAG) architecture with a multi-stage pipeline including dynamic profile extraction, semantic query expansion, dual-index retrieval, and LLM re-ranking the system delivers personalized response.
Speaker(s):
Yanshan Wang, PhD
University of Pittsburgh
Author(s):
Mariano De Leon, MS - University of Pittsburgh; Dan Ding, PhD - University of Pittsburgh; Adit Bharat Shah, MS - University of Pittsburgh; Gina Novario, OTD, OTR/L - University of Pittsburgh; Yanshan Wang, PhD - University of Pittsburgh;
Yanshan Wang, PhD - University of Pittsburgh
From Trial Gaps to Guided Support: A Spanish-Language Agentic AI-Enabled Tool for Patient-Centered TB Support
Poster Number: 137
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Mobile Health, Usability
Programmatic Theme: Public Health Informatics
To address after hours and mental health gaps in digital health TB care, we developed a Spanish language conversational agent integrating TB education, an emotional-support “coach” with DBT-informed skills, and provider escalation. An agentic RAG architecture outperformed a single prompt model across iterative internal tests, improving relevance, empathy, and context tracking. Next steps include expert evaluation and patient facing pilot integration within TB TST to validate safety workflows and assess feasibility for scalable, patient centered deployment.
Speaker(s):
Sarah Iribarren, PhD
University of Washington
Author(s):
Javier Roberti, PhD - Centre for Research on Epidemiology and Public Health (CIESP), CONICET Buenos Aires, Argentina; Yvette Rodriguez, RN - University of Washington; Priscilla Carmiol-Rodriquez, RN - University of Washington; Alfie Aguilar Vidrio, BS - University of Washington; Priyanshi Garg, BS - University of Washington; Sikha Pentyala, PhD - University of Washington; Charles Kwanin, RN, MS - University of Washington; Jennifer Sprecher, MS - University of Washington; Martine De Cock, PhD - University of Washington;
Sarah Iribarren, PhD - University of Washington
Accuracy & Safety of an AI Agent for Hypertension Management
Poster Number: 138
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Chronic Care Management, Large Language Models (LLMs)
Programmatic Theme: Clinical Informatics
We evaluated the safety of an AI conversational agent for hypertension management over 4 weekly calls among 13 patients. Twelve calls were identified in which the AI provided inaccurate information. Most inaccuracies were minor and frequently involved misunderstandings of medication names. One call was flagged for the potential to cause patient harm. The agent advised holding a medication after an out-of-range blood pressure instead of connecting the patient with a nurse. Model refinement is ongoing.
Speaker(s):
Brian Wells, MD, PhD
Wake Forest University School of Medicine
Author(s):
Brian Wells, MD, PhD - Wake Forest University School of Medicine; Richard Lord, MD - Wake Forest University School of Medicine; Jennifer Roper, MD - Wake Forest University School of Medicine; Christopher Zagar, MD - One Health; Yhenneko Taylor, PhD - Center for Health System Sciences - Advocate Health; Andrew McWilliams, MD, MPH - Advocate Health; Justin Kramer, PhD, MAT - Wake Forest University School of Medicine;
Brian Wells, MD, PhD - Wake Forest University School of Medicine
AI-Based Facial Expression Analysis in Heart Failure Consultations
Poster Number: 139
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Machine Learning, Patient Engagement and Preferences, Patient-/Person-Generated Health Data, Clinical Decision Support, Chronic Care Management, Human-computer Interaction, Quantitative Methods
Programmatic Theme: Clinical Informatics
Using AI-based facial expression analysis, this study characterized emotional responses among 61 patients with heart failure during patient-reported outcome (PRO)-informed clinical consultations. Despite happiness being the most frequently expressed discrete emotion, overall emotional tone was slightly negative, with the majority of patients distributed across low-valence quadrants of the circumplex model. These findings highlight the emotional complexity of PRO-based consultations and the potential of AI-enabled tools to support emotionally responsive, patient-centered care.
Speaker(s):
So Hyeon Bang, Postdoctoral Research Scientist/Ph.D.
Columbia University School of Nursing
Author(s):
So Hyeon Bang, Postdoctoral Research Scientist/Ph.D. - Columbia University School of Nursing; Stacey Dai, MPH - Columbia University School of Nursing; Janejira Chaiyasit, MSN, BSN, DNP, AGNP-C - Columbia University; Yelitza Castanos, MSN, BSN, DNP, AGPCNP-BC - Columbia University; Kimia Mirabbaszadeh, DNP, FNP-BC - Columbia University; Alicia Bacchus, RN, MSN, FNP-BC - Columbia University; Annie Goldberg, NP - Columbia University; Jordan Lindekens, MSN, AGACNP-BC - Columbia University; Melani Ellison, MPH - Columbia University School of Nursing; Nickisha Mortimer, DNP - Columbia University; Parag Goyal, M.D., M.Sc. - Weill Cornell Medicine; Mary Beth Happ, PhD, RN - The Ohio State University; David Russell, Ph.D - Appalachian State University; Ruth Masterson Creber, PhD, MSc, RN - Columbia University;
So Hyeon Bang, Postdoctoral Research Scientist/Ph.D. - Columbia University School of Nursing
seeBias: A Comprehensive Tool for Assessing and Visualizing AI Fairness
Poster Number: 140
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Clinical Decision Support, Diversity, Equity, Inclusion, and Accessibility
Programmatic Theme: Clinical Informatics
Ensuring fairness of AI prediction models is critical for responsible use in healthcare. We present seeBias, an R package for comprehensive and clinically interpretable fairness evaluation with rich visualizations. When evaluating a clinical prediction model, seeBias reveals disparities in calibration and predictive values that may remain hidden when fairness assessment relies only on conventional parity metrics, and translates predictive values into numbers needed for true positive or true negative predictions to support clinical interpretation.
Speaker(s):
Yian Ma, MSc
Duke-NUS Medical School
Nan Liu, PhD
National University of Singapore
Author(s):
Yilin Ning, PhD - Duke-NUS Medical School; Yian Ma, MSc - Duke-NUS Medical School; Mingxuan Liu, PhD student - Duke-NUS Medical School; Xin Li, Master of science - UTHealth Houston; Nan Liu, PhD - National University of Singapore;
Yian Ma, MSc - Duke-NUS Medical School
Nan Liu, PhD - National University of Singapore
Nan Liu, PhD - National University of Singapore
AI-Powered Clinical Decision Support System Integrated with Remote Spirometry for Pediatric Asthma: Clinician and Caregiver Perspectives from a Pragmatic Feasibility Study
Poster Number: 141
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Healthcare Quality, Clinical Decision Support, Chronic Care Management, Patient-/Person-Generated Health Data
Programmatic Theme: Clinical Informatics
We evaluated user feedback from caregivers and clinicians regarding AsthmaTuner, an integrated, real-time home spirometry application, and A-GPS, an AI-powered clinical decision-making system embedded in EHR, in terms of clinical monitoring of childhood asthma, clinical patient management, adaptation to daily life and workflow, patient self-management and safety, accessibility to healthcare services, and many other standpoints like future adoption, based on the thematic analysis of the interviews conducted with them.
Speaker(s):
Elif Polat, MD
MAYO CLINIC
Author(s):
Momin Malik, PhD in Societal Computing - Mayo Clinic; Young Juhn, M.D., M.P.H. - Mayo Clinic; Shauna Overgaard, Ph.D. - Mayo Clinic; DAN KELLEHER, PHD,MS - MAYO CLINIC; CHUNG WI, MD - MAYO CLINIC; LYENNA MYERS, PHD - KAROLINSKA INSTITUTET; TRACEY BRERETON, MS - MAYO CLINIC; JASON GREENWOOD, MD,MS - MAYO CLINIC; LU ZHENG, PHD,MS - MAYO CLINIC; KATHY IHRKE, RN - MAYO CLINIC; MADISON BEENKEN, MS - MAYO CLINIC; JOSH BUBLITZ, . - MAYO CLINIC; TAYLOR GALLOWAY, MS,MBA - MAYO CLINIC; QUANTIA WILKES, MS - MAYO CLINIC; DANIELLE SHRADER, . - MAYO CLINIC; MIGUEL A. PARK, MD - MAYO CLINIC; DEEPAK SHARMA, PHD,MS - MAYO CLINIC; BJORN NORDLUND, PHD - KAROLINSKA INSTITUTET;
Elif Polat, MD - MAYO CLINIC
Evaluating Clinician Trust in AI-Generated Contours During Radiation Oncology Peer Review: A Qualitative Study Using AI Trust Theory
Poster Number: 142
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Clinical Decision Support, Qualitative Methods
Programmatic Theme: Clinical Research Informatics
Artificial intelligence (AI) contouring tools are increasingly used in radiation therapy planning, yet clinician adoption depends on trust in AI-generated outputs. Using the Theory of Trust and Acceptance of Artificial Intelligence Technology (TrAAIT), we examine how radiation oncology clinicians evaluate AI-generated clinical target volume (CTV) contours during simulated peer review. Qualitative transcript analysis reveals insights into the credibility, perceived value, and reliability of AI-assisted contours, highlighting the need for further research on trust in AI.
Speaker(s):
Samuel Isaac, Health Informatics
University of North Carolina - Chapel Hill
Author(s):
Samuel Isaac, Health Informatics - University of North Carolina - Chapel Hill; Elizabeth Byrd, Information Science - UNC-Chapel Hill; Yuan Gao, PhD - UNC; Amy Cole, PhD - UNC School of Medicine; Lukasz Mazur, PhD - University of North Carolina;
Samuel Isaac, Health Informatics - University of North Carolina - Chapel Hill
Can we predict users of AI?: A closer look at Pieces AI implementation in a safety net health system
Poster Number: 143
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Informatics Implementation, Evaluation
Programmatic Theme: Clinical Research Informatics
The study examined the implementation of Pieces AI, a generative artificial intelligence (AI) tool that creates discharge summaries using data from the electronic health record. We surveyed end users on their general trust in AI and perceived workload for creating discharge summaries pre- and post- Pieces AI implementation. Findings highlight reduced workload post-Pieces AI implementation and a need to further understand how trust in AI may play a role in specific AI tool implementations.
Speaker(s):
Ashley Hughes, PhD, MS
MetroHealth System/Case Western Reserve University
Author(s):
Ashley Hughes, PhD, MS - MetroHealth System/Case Western Reserve University; Kevin Driscoll, DO - MetroHealth/Case Western Reserve University; Jui-En Lo, MD - Metrohealth Medical Center; Yasir Tarabichi, MD, MSCR - MetroHealth;
Ashley Hughes, PhD, MS - MetroHealth System/Case Western Reserve University
A Multi-Agent Retrieval-Grounded AI System for Institutional Research Dataset Discovery
Poster Number: 144
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Large Language Models (LLMs), Information Retrieval, Knowledge Representation & Information Modeling, Information Extraction, Data Sharing, Informatics Implementation, Natural Language Processing
Programmatic Theme: Clinical Research Informatics
ARIN-Data is an agentic AI system for institutional research dataset discovery within the AI Research Infrastructure Navigator (ARIN) framework. The system combines a curated inventory of institutional dataset records with a retrieval-augmented generation (RAG) architecture and a multi-agent workflow. ARIN-Data performs query clarification, hybrid semantic retrieval, and grounded recommendation generation to identify relevant datasets and stewardship pathways, enabling transparent and traceable dataset discovery to support collaboration and accelerate clinical and translational research.
Speaker(s):
syandan jagadam, Masters
UAB
Author(s):
Abu Mosa, PhD, MS, FAMIA - University of Alabama at Birmingham;
syandan jagadam, Masters - UAB
Syntactic and Semantic Normalization of >2.5M Cancer Staging Records
Poster Number: 145
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Data transformation/ETL, Usability, Data Standards, Controlled Terminologies, Ontologies, Vocabularies, Large Language Models (LLMs)
Programmatic Theme: Clinical Research Informatics
Syntactic and semantic normalization are essential for aggregating and analyzing cancer staging data for research purposes. Using large-scale, real-world cancer staging data extracted from clinical notes by a large language model, we demonstrate our process for making these data available for health system, academic, government, and life science researchers in a consistent and research-ready format.
Speaker(s):
Jamalynne Deckard, MS
Truveta, Inc.
Author(s):
Jamalynne Deckard, MS - Truveta, Inc.; Swapna Abhyankar, MD - Truveta, Inc.; Sujatha Sagiraju, MS, MBA - Truveta; Jennifer Liang, MD - Truveta, Inc.;
Jamalynne Deckard, MS - Truveta, Inc.
Patient use of an AI chatbot for heart failure self-management: A feasibility study
Poster Number: 146
2026 Symposium Health Equity Presentation
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Chronic Care Management, Health Equity, Usability
Programmatic Theme: Clinical Informatics
Heart failure (HF) patients struggle to adhere to self-management behaviors, highlighting the need for scalable support tools. Artificial intelligence (AI) conversational agents may address this gap. This feasibility study evaluated patient engagement and usability of an AI chatbot based on evidence-based HF guidelines and tailored to users’ psychosocial phenotypes. Fifteen HF patients completed surveys and interviews. The chatbot was perceived as natural and easy to use, though some responses were generic, indicating areas for improvement.
Speaker(s):
Liya Wang, BSN, PhD student
the University of Texas at Austin, School of Nursing
Author(s):
Liya Wang, BSN, PhD student - the University of Texas at Austin, School of Nursing; Yuko Hamajima, MSN - The University of Texas at Austin; School of Nursing; Shaked Gorin, BS - The University of Texas at Austin, School of Information; Rachel Tunis, PhD Student - University of Texas at Austin; Ying Ding, PhD - University of Texas at Austin; Kavita Radhakrishnan, PhD - University of Texas - Austin;
Liya Wang, BSN, PhD student - the University of Texas at Austin, School of Nursing
Patient Preferences for Using Artificial Intelligence for Chronic Pain Management
Poster Number: 147
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Chronic Care Management, Patient Engagement and Preferences
Programmatic Theme: Clinical Informatics
Artificial intelligence (AI) tools that summarize electronic health record (EHR) data may support clinical decision-making, but patient preferences are essential for adoption. We surveyed 104 adults with chronic non-cancer pain at a Midwestern safety-net health system about AI use in their care. About half supported AI for information retrieval, treatment selection, and communication improvement. Most emphasized the need for clear explanation and transparent disclosure of AI involvement.
Speaker(s):
MaKenzie Gee, MS
Regenstrief Institute
Author(s):
Randall Grout, MD - Indiana University; Jennifer Hillstrom, N/A - Eskenazi Health; Lindsey Sanner, MPH - Indiana University Richard M. Fairbanks School of Public Health; Sarah Roth, MHA, MPH, PMP, CCRP - Regenstrief Institute; Christopher Harle, PhD - Indiana University; Olena Mazurenko, MD, PhD - Indiana University Fairbanks School of Public Health;
MaKenzie Gee, MS - Regenstrief Institute
A Machine Learning-Based Simulated Synthetic Head-to-Head Comparison of Panitumumab-FOLFIRI and Bevacizumab-FOLFIRI in Wild-Type KRAS Metastatic Colorectal Cancer
Poster Number: 148
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Machine Learning, Clinical Decision Support, Evaluation, Informatics Implementation
Programmatic Theme: Clinical Research Informatics
We developed machine learning models to predict treatment response, adverse events, and serious adverse events for patients receiving panitumumab-FOLFIRI for wild-type KRAS metastatic colorectal cancer. The models were trained on retrospective clinical trial data from 470 patients to generate outcome predictions. We then simulated a synthetic treatment comparison by predicting outcomes under panitumumab-FOLFIRI for 42 patients from a separate trial treated with bevacizumab-FOLFIRI.
Speaker(s):
Vinodh Balaraman, MBA
KolateAI
Author(s):
Achilles Saxby, MS - KolateAI; Karthik Chandrakant, BS - KolateAI; Dahlia Musa, PhD - KolateAI; Devesh Sharma, MTech - KolateAI; Chandra Ranganathan, MBA - KolateAI; Vinodh Balaraman, MBA - KolateAI;
Vinodh Balaraman, MBA - KolateAI
Cybersecurity and Privacy Threats in Generative-AI Mental-Health Chatbots: A Scoping Review
Poster Number: 149
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Public Health, Privacy and Security, Large Language Models (LLMs), Surveys and Needs Analysis
Programmatic Theme: Consumer Health Informatics
Generative-AI mental-health chatbots process highly sensitive disclosures yet face poorly characterized cybersecurity and privacy threats. Through a scoping review of 48 studies, we identify data leakage, adversarial attacks, and prompt-interface exploitation as the dominant cybersecurity threats, alongside profiling/inference and PII exposure as leading privacy risks. Chat logs, crisis narratives, and private data are most implicated. Findings support evidence-based governance and health informatics training standards.
Speaker(s):
Bushra Rashrash, Masters
GVSU
Author(s):
Rajashekhar Reddy Shagamreddy, Masters in Bioinformatics - grand valley state university; Hemalatha sabbineni, Masters - Grand valley State University; Bushra Rashrash, Masters - GVSU; Suhila Sawesi, PhD - GVSU;
Bushra Rashrash, Masters - GVSU
CEHR-BERT for Alzheimer's Disease Prediction: Harnessing Longitudinal Real-World EHR Data
Poster Number: 150
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Deep Learning, Machine Learning, Clinical Decision Support
Programmatic Theme: Clinical Research Informatics
This study employs CEHR-BERT, a transformer-based model designed for OMOP-structured longitudinal EHR data, to assess Alzheimer's disease (AD) risk in a real-world cohort of 17,300 patients from the TriNetX network. By analyzing complete temporal sequences of clinical encounters, including diagnoses, medications, procedures, and vitals, CEHR-BERT outperforms baseline models such as logistic regression and XGBoost on AUROC and AUPRC metrics. Additionally, the attention weights reveal prodromal AD markers that align with findings in clinical literature.
Speaker(s):
Kostiantyn Botnar, PhD
Author(s):
George Golovko, PhD - University of Texas Medical Branch; Kamil Khanipov, PhD - University of Texas Medical Branch;
Kostiantyn Botnar, PhD -
Probabilistic Forecasting of Longitudinal Clinical Event Trajectories in Hypertensive Patients Using Medical Transformer-Based Generative Adversarial Networks
Poster Number: 151
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Data Mining, Clinical Decision Support, Machine Learning, Deep Learning
Programmatic Theme: Clinical Research Informatics
Hypertension represents a major chronic disease characterized by longitudinal clinical patterns. We designed a Medical Transformer-Based Generative Adversarial Networks (GANs) to model and simulate future clinical trajectories in hypertensive patients using Electronic Health Records (EHR). The model generates probabilistic sequences of diagnoses and medications and demonstrates improved performance over baseline methods for trajectory forecasting and next-visit prediction while providing interpretable risk estimates for comorbidities and treatment outcomes.
Speaker(s):
Xenia Liu, Doctoral Degree
Rutgers University
Author(s):
Rahul Mittal, MD - Rutgers University; Martin Pavlovski, Phd - Samsung Electronics America; Branimir Ljubic, MD, PhD - Rutgers University, NJ, USA;
Xenia Liu, Doctoral Degree - Rutgers University
Building a Voice-First Front Door: Standardized Patient Evaluation of the GENIE Autonomous Call Center Agent
Poster Number: 152
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Large Language Models (LLMs), Evaluation
Programmatic Theme: Clinical Research Informatics
As voice-first conversational AI transforms care, there is a critical need to validate systems. We developed GENIE 24/7, a proprietary inbound voice AI agent built to automate routine call center workflows. We used novel voice-centric Voice Usability Scale (VUS) and HumanELY instrument to rigorously and holistically assess the semantic intelligence, safety and patient experience with standardized patients. Our analysis shows high user satisfaction and opportunities for improvement especially on routing failures and unanswered questions.
Speaker(s):
Ashish Atreja, MD, MPH
Stanford University
Author(s):
Morish Shah, Precision Health - Genserve.AI;
Ashish Atreja, MD, MPH - Stanford University
Trajectory-Aware AI System for Lead Optimization in Drug Discovery
Poster Number: 153
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Machine Learning, Large Language Models (LLMs)
Programmatic Theme: Translational Bioinformatics
We present TRACE, a trajectory-aware AI system for molecular lead optimization in drug discovery. The framework coordinates multiple molecular editing tools through an AI planning module and reuses prior optimization trajectories to guide multi-step refinement. Experiments across several ADMET-related optimization tasks demonstrate improved optimization success rates and larger property improvements compared with standalone molecular optimization models.
Speaker(s):
Lingxia Li, MS Student
University of Michigan
Author(s):
Lingxiao Li, Master - University of Michigan;
Lingxia Li, MS Student - University of Michigan
Simulation-Based Human Factors Evaluation of AI-Enabled Clinical Technologies: A Framework for Pre-Deployment Assessment
Poster Number: 154
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Human-computer Interaction, Evaluation, Informatics Implementation, Governance
Programmatic Theme: Clinical Informatics
Pre-deployment evaluation of AI-enabled medical devices often emphasizes model and software performance while underassessing clinician use in real workflows. We describe a simulation-based human factors framework integrating risk stratification, realistic clinical scenarios, standardized task and usability measures, and AI-specific assurance testing. Applied to a digital stethoscope, the framework detected workflow and setup friction despite successful task completion, demonstrating its value for generating decision-relevant evidence to support Learning Health System implementation.
Speaker(s):
Shauna Overgaard, Ph.D.
Mayo Clinic
Author(s):
Joshua Ohde, PhD - Mayo Clinic; Lu Zheng, Ph.D., M.S. - Mayo Clinic; Tracey Brereton, Masters - Mayo Clinic; Young Juhn, M.D., M.P.H. - Mayo Clinic; Chung-Il Wi, MD - Mayo Clinic Minnesota; Momin Malik, PhD - Mayo Clinic; Deepak Sharma, MS - Mayo Clinic;
Shauna Overgaard, Ph.D. - Mayo Clinic
Physicians’ Public Discourse on Generative Artificial Intelligence: Implications for Trust, Governance, and Clinical Implementation
Poster Number: 155
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Large Language Models (LLMs), Natural Language Processing
Programmatic Theme: Clinical Informatics
Generative AI is rapidly entering health care, yet less is known about how physicians publicly discuss its potential and risks. We analyzed 20,704 physician-authored posts on X (2022–2025) using BERT topic modeling. Five prominent topic groups emerged: specialty-specific clinical applications, trust and governance concerns, research and knowledge support, workflow integration, and diagnostic support. The findings suggest that physician discourse reflects both enthusiasm for clinical uses and ongoing concerns about safety, bias, and reliability.
Speaker(s):
Samaneh Borji, Master in Information Science
University of South Carolina
Author(s):
Samaneh Borji, Master in Information Science - University of South Carolina; Ehsan Mohammadi, PhD - University of South Carolina; Amirreza Kalantari, MLIS - University of south carolins;
Samaneh Borji, Master in Information Science - University of South Carolina
SurvTEDVAE: A Disentangled Variational Autoencoder for Heterogeneous Treatment Effect Estimation with Time-to-Event Outcomes
Poster Number: 156
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Causal Inference, Real-World Evidence Generation, Artificial Intelligence, Deep Learning
Programmatic Theme: Clinical Research Informatics
Estimating heterogeneous treatment effects (HTE) from observational health data is essential for precision medicine, yet existing methods often struggle with high-dimensional covariates and time-to-event outcomes common in electronic health records (EHRs). We propose SurvTEDVAE, a disentangled variational autoencoder designed for causal survival analysis. The model learns latent representations corresponding to instrumental factors, confounders, and outcome-dependent risk factors, and integrates a survival likelihood to model time-to-event outcomes with censoring. The learned representations are used to estimate conditional average treatment effects using downstream causal estimators. We evaluated SurvTEDVAE using a semi-synthetic ACTG dataset and a high-dimensional EHR-based hypertension cohort with over 20,000 covariates. Across both datasets, SurvTEDVAE achieved lower estimation error for heterogeneous treatment effects compared with meta-learning and causal survival forest approaches. These results demonstrate that disentangled representation learning can improve causal effect estimation for survival outcomes in high-dimensional real-world health data.
Speaker(s):
William Powell, MS
Washington University in St. Louis
Author(s):
William Powell, MS - Washington University in St. Louis; Linying Zhang, PhD - Washington University in St. Louis;
William Powell, MS - Washington University in St. Louis
Operationalizing Structural Equation Modeling to Evaluate Poverty and Related Law and Policy on Preemptive Kidney Transplant Waitlist Times by Jurisdictions (2010-2013): A Work in Progress
Poster Number: 157
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Causal Inference, Clinical Decision Support, Population Health, Health Equity
Programmatic Theme: Public Health Informatics
We operationalized a causal inference model to evaluate the variation in state-level poverty-related legal index scores associated with differences in kidney transplant waitlist times across the U.S. Preliminary results show promise for a novel legal epidemiologic analysis that can be paired with traditional epidemiologic methods to refine algorithms for clinical decision support tools and lay the groundwork for future analysis of a state-level poverty-related legal index on kidney transplant waitlist times for health equity research.
Speaker(s):
Holly Tomlin, JM, MPH
Meharry School of Global Health
Author(s):
Holly Tomlin, JM, MPH - Meharry School of Global Health; Arindam RoyChoudhury, PhD - Weill Cornell Medicine; Sophia Kostelanetz, MD, MPH - Maimonides Medical Center; Samantha Weber, JD - Morehouse Medical College; Nadine Jawad, MD, MS - Weill Cornell Medicine; Wisdom Dayok, BS - Consumer Health Informatics Lab at Yale University School of Public Health; Shannon Dhindsa, B.Kin, MPH - Consumer Health Informatics Lab at Yale University School of Public Health; Monika Safford, MD - Weill Cornell Medicine; Fazal Khan, MD, JD - Meharry School of Global Health; Terika McCall, PhD, MPH, MBA - Yale School of Public Health;
Holly Tomlin, JM, MPH - Meharry School of Global Health
Evaluation of Causal Reasoning for Large Language Models in Contextualized Clinical Scenarios of Laboratory Test Interpretation
Poster Number: 158
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Causal Inference, Large Language Models (LLMs), Clinical Decision Support, Evaluation
Programmatic Theme: Clinical Research Informatics
We evaluated the causal reasoning of LLMs in laboratory test interpretation across Pearl’s Ladder of Causation. Using 99 clinically grounded scenarios, we assessed associational, interventional, and counterfactual reasoning. GPT-o1 outperformed Llama 3.2, showing higher precision and consistency in complex reasoning tasks. While models handle well-established interventions effectively, performance remains uneven in counterfactual scenarios. These findings emphasize the need for robust reasoning guardrails in clinical AI.
Speaker(s):
Balu Bhasuran, Ph.D
Florida State University
Author(s):
Balu Bhasuran, Ph.D - Florida State University; Zhe He, PhD, FIAHSI, FAMIA - Florida State University; Mattia Prosperi, PhD, FAMIA - University of Florida; Karim Hanna, MD - University of South Florida Health; Caretia JeLayne Washington, BS - University of Florida, Gainesville, FL; John Petrilli, MD - University of Florida, Gainesville, FL;
Balu Bhasuran, Ph.D - Florida State University
Probabilistic Methods in Large Language Models for Suicide-Related Outcome Detection
Poster Number: 159
2026 Symposium LIEAF Presentation
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Causal Inference, Quantitative Methods, Public Health
Programmatic Theme: Academic Informatics / LIEAF
Large language models (LLMs) are increasingly applied to suicide-related outcome prediction, yet their probabilistic outputs lack methodological standardization. This scoping review of studies (2022–2025) introduces a taxonomy distinguishing explicit approaches, where models are prompted to output numeric probabilities, from non-explicit approaches, where risk estimates are derived indirectly. Results reveal substantial heterogeneity and absent calibration standards, posing patient safety risks for learning health systems deploying generative AI in clinical suicide risk workflows.
Speaker(s):
Selen Bozkurt, PhD, MS
Emory University
Author(s):
Selen Bozkurt, PhD, MS - Emory University; Abeed Sarker, PhD - Emory University School of Medicine; Swati Rajwal, MS - Emory University; Yuhe Gao, Master in Information Science - Emory University; Yuhe Gao, PhD Student - Emory University;
Selen Bozkurt, PhD, MS - Emory University
Health Decision Patterns Embedded in Daily Life: Informatics Opportunities to Support Youth
Poster Number: 160
2026 Symposium Health Equity Presentation
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Chronic Care Management, Patient Engagement and Preferences, Diversity, Equity, Inclusion, and Accessibility, Health Equity, Disability, Accessibility, and Human Function, Qualitative Methods, Real-World Evidence Generation, Clinical Decision Support
Working Group: Consumer Health Informatics Working Group
Programmatic Theme: Consumer Health Informatics
Patients living with chronic conditions must constantly navigate complex health-related decisions in daily life. Youth with chronic kidney disease (CKD) exemplify this challenge, as managing their condition requires balancing medication regimens and life goals. Despite the increasing emphasis on patient-centered decision-making, most decision-support tools focus on clinician–patient interactions in clinical settings, neglecting the real-world contexts in which many health decisions actually unfold. To address this gap, we examined health decision-making patterns that emerge in the everyday lives of youth. Our analysis revealed four distinct patterns characterized by varying contextual conditions, engagement styles, and levels of goal alignment. These patterns suggest that the support needs of youth may vary by the specific pattern they fall under for a given decision. Recognizing these distinctions provides a foundation for identifying targeted opportunities for informatics interventions and underscores the importance of providing context-sensitive support that adapts to patients' changing decision-making needs and circumstances.
Speaker(s):
ZIXUAN XU, Master's
University of Washington School of Medicine
Author(s):
Wanda Pratt, PhD, FACMI - University of Washington; Ari Pollack, MD - Seattle Children's Hosp & University of Washington; Hyeyoung Ryu, PhD - Vanderbilt University Medical Center; Aaron Wightman, MD - University of Washington School of Medicine; Chelsea Ng, BS - University of Washington School of Medicine; Cozumel Pruette, MD - Johns Hopkins University; Ryan Hutson, BS - Johns Hopkins University; Jaime Snyder, PhD - University of Washington;
ZIXUAN XU, Master's - University of Washington School of Medicine
Does Machine Learning Outperform Traditional Fibrosis Scores in Predicting Liver Cirrhosis Risk? A Longitudinal EHR-Based Study
Poster Number: 161
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Chronic Care Management, Machine Learning, Clinical Decision Support
Programmatic Theme: Clinical Research Informatics
Early detection of liver cirrhosis is critical for timely intervention to prevent severe progression. Using longitudinal electronic health record (EHR) data, we developed machine learning models to predict incident cirrhosis one to three years before diagnosis among patients with fatty liver disease. Gradient-boosted tree models (XGBoost) leveraging demographic, laboratory, and comorbidity data were evaluated against the Fibrosis-4 score, demonstrating improved predictive performance and potential for earlier risk stratification in clinical practice.
Speaker(s):
Zhuqi Miao, Ph.D.
Oklahoma State University
Author(s):
Zhuqi Miao, Ph.D. - Oklahoma State University; Sujan Ravi, MD - University of Alabama at Birmingham; Abdulaziz Ahmed, PhD - University of Alabama at Birmingham;
Zhuqi Miao, Ph.D. - Oklahoma State University
Examining the validity and fairness of an electronic medical record-integrated pharmacy risk score in patients with diabetes and/or hypertension
Poster Number: 162
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Chronic Care Management, Clinical Decision Support, Informatics Implementation
Programmatic Theme: Clinical Informatics
There is a high need for valid risk scoring algorithms for prioritization of clinical pharmacy services in patients with diabetes and hypertension. We evaluated the predictive ability of an electronic medical record-integrated pharmacy risk score developed at Atrium Health. The score appeared to be fair but did not demonstrate adequate performance in predicting uncontrolled HbA1c and BP. Our study provides useful foundational information for future research on developing similar algorithms with optimal predictive performance.
Speaker(s):
Rohan Mahabaleshwarkar, PhD
Atrium Health
Author(s):
Hieu Nyugen, MS - Atrium Health; Nick Wilkins, PharmD, BCPS, CDCES, CPP, DPLA - Advocate Health; Molly Hinely, PharmD, BCPS - Atrium Health Wake Forest Baptist; Danielle Baker, PharmD, MS, BCPS - Atrium Health Wake Forest Baptist; Nicholas Pajewski, PhD - Wake Forest University School of Medicine;
Rohan Mahabaleshwarkar, PhD - Atrium Health
Characterizing Treatment Continuity in Relapsing Cohorts: Scalable Data Transformation to Expedite Research in a Learning Health System
Poster Number: 163
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Chronic Care Management, Data Modernization, Data transformation/ETL
Programmatic Theme: Clinical Research Informatics
Chronic conditions often require recurring episodic treatment, particularly for individuals with substance use disorders. Longitudinal retention data has historically been challenging to analyze across multiple treatment episodes. We developed a clinician-guided, learner-accessible framework to harmonize electronic health record data for >10,000 patients with relapsing opioid use disorder undergoing multiple buprenorphine treatment episodes. This approach modernizes data preparation within Learning Health Systems and supports reproducible, iterative analyses for conditions marked by irregular care continuity.
Speaker(s):
Donielle Beiler, BS
Geisinger Health System
Author(s):
Vanessa Troiani, PhD - Geisinger; Donielle Beiler, BS - Geisinger Health System; Joseph Valdez, MD - Geisinger Health System;
Donielle Beiler, BS - Geisinger Health System
Predicting 90-day Odontogenic Emergency Escalation from Routine Dental Visits Using Structured Electronic Health Record Data
Poster Number: 164
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support, Patient Safety, Population Health
Programmatic Theme: Clinical Research Informatics
We developed a machine learning model to predict 90-day odontogenic emergency escalation using 22 structured features from dental EHR data. Across 26,770 visits (7,518 patients), elastic net logistic regression achieved a temporally validated AUROC of 0.638 (95% CI: 0.601–0.676). The highest-risk tertile experienced 2.59 times the emergency rate of the lowest. Prior emergency history, endodontic procedures, and tooth-level clinical findings were among the strongest predictors; preventive care engagement was protective.
Speaker(s):
Oscar Peng, BS
University of California, San Francisco
Author(s):
Marianna Elia, PhD - UCSF; Julian Hong, M.D., M.S. - UCSF;
Oscar Peng, BS - University of California, San Francisco
Refining Clinical Decision Support Training to Improve Workflow Integration in a Multisite Pragmatic Trial
Poster Number: 165
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support, Informatics Implementation, Workflow, Human-computer Interaction
Programmatic Theme: Clinical Informatics
Clinical decision support (CDS) effectiveness depends on successful integration into clinical workflows. During the multisite rollout of the OPTIONS CDS program in skilled nursing facilities, training was refined to include hands-on practice. Early adoption exceeded 40%, and residents in the refined training cohort completed CDS workflows significantly faster (HR = 2.68). These findings suggest that training strategies aligned with CDS characteristics, organizational context, and clinical workflow may improve CDS implementation in pragmatic multisite trials.
Speaker(s):
Min Jeoung Kang, PhD
Brigham and Women's Hospital/ Harvard Medical School
Author(s):
Rosa Recio-García, MSN, RN - Brigham and Women's Hospital; Pamela Garabedian, MS - Mass General Brigham; Veysel Baris, PhD, RN - Brigham and Women's Hospital; Alice Kim, MS, RD - Brigham and Women’s Hospital; Kumiko Schnock, RN, PH.D - Brigham and Women's Hospital/ Harvard Medical School; Ling Tang, MS - University of Maryland School of Medicine; Stuart Lipsitz - Brigham and Women's Hospital; Nancy Latham, PhD PT - Brigham and Women's Hosptial; Rodrigo Valderrábano, MD - Brigham and Women's Hospital; Elizabeth Dennis, PhD, RD - University of Maryland School of Medicine; Jason Falvey, PhD - University of Maryland School of Medicine; Jay Magaziner, PhD - University of Maryland School of Medicine; Richard White, BSc - PointClickCare; Tyler Saumur, PhD - PointClickCare; Dana Weisenfeld, MS - Hebrew SeniorLife; Denise Orwig, PhD - University of Maryland School of Medicine; PATRICIA C DYKES, PhD, MA, RN - Emory University;
Min Jeoung Kang, PhD - Brigham and Women's Hospital/ Harvard Medical School
Deploying Scalable CDS: Technical Development of an EHR-Agnostic, SMART on FHIR Pulmonary Embolism Risk Prediction Tool
Poster Number: 166
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support, Interoperability and Health Information Exchange, Informatics Implementation, Workflow
Programmatic Theme: Clinical Informatics
Most clinical decision support tools are tightly coupled to local EHR builds, limiting scalability and portability. To address this, we developed an interoperable, EHR-agnostic web-based CDS platform for pulmonary embolism risk prediction, embedded within the CTPA ordering workflow. Using SMART on FHIR and native EHR frameworks, the tool automates Wells' scoring and delivers real-time recommendations to ED providers. Iterative co-development with institutional IT teams supported technical feasibility and deployment readiness across diverse health systems.
Speaker(s):
Lynn Xu, MPH
NYU Grossman School of Medicine
Author(s):
Lynn Xu, MPH - NYU Grossman School of Medicine; Angela Mastrianni, PhD - NYU Grossman School of Medicine; Natalie Henning; Yuhan Cui, MS - NYU Langone; Praveen Ranganath, MS - NYU Langone Health; Adam Kiefer, BA - NYU Langone Health; Christopher Larson, BSBA - NYU Langone Health; Edwin Pineda, MSN, RN, CEN - NYU Langone Health; Susan Moore, RN - NYU Langone Health; Adam Rubens, BS - NYU Langone Health; Eduardo Iturrate, MD - NYU School of Medicine; Vincent Xiao, MD - NYU Langone Health; Nicholas Genes, MD, PhD - NYU Grossman School of Medicine; Gregory Simon, MD - NYU Langone Health; Safiya Richardson, MD, MPH - NYU Langone;
Lynn Xu, MPH - NYU Grossman School of Medicine
Primary Care Provider Perspectives to Inform Development of Clinical Decision Support for Venous Thromboembolism Diagnosis
Poster Number: 167
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support, Patient Safety, User-centered Design Methods
Programmatic Theme: Clinical Research Informatics
Diagnosing venous thromboembolism (VTE) in primary care is challenging due to nonspecific symptoms and time-constrained clinical environments. Structured interviews with eight primary care providers from two U.S. health systems explored perspectives on clinical decision support (CDS) for VTE diagnosis. 88 coded comments were identified, with most related to CDS design considerations(n=31) and desirable features(n=28). Providers emphasized electronic health record integration, actionable recommendations, and minimal workflow disruption while raising concerns about alert fatigue and medico-legal implications.
Speaker(s):
Veysel Baris, PhD, RN
Brigham and Women's Hospital
Author(s):
Alexandra Frost, Bachelor's of Science (BS) - Brigham and Women's Hospital; Michael Sainlaire - Brigham and Women's Health; Jin Chen, PhD - University of Alabama at Birmingham; Minakshi V. Shukla, MD - University of Alabama at Birmingham; MD Shifatul Apurba, PhD Candidate - University of Alabama at Birmingham; Ann Hurley, DNSc - BWH; PATRICIA C DYKES, PhD, MA, RN - Emory University;
Veysel Baris, PhD, RN - Brigham and Women's Hospital
A Cross-Institutional Qualitative Analysis of CDS Alert Design Guides
Poster Number: 168
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support, Usability, User-centered Design Methods
Programmatic Theme: Clinical Informatics
We analyzed CDS alert design guides from six pediatric institutions and surveyed 19 clinical informaticists to identify priority domains for a consensus CDS Alert Design Guide. Thematic analysis showed consistent emphasis on minimizing alert fatigue, ensuring clinical relevance, and governance, with variable specificity in visual design and lifecycle management. Survey results demonstrated broad support for a core set of clinical, usability, and governance domains to inform standardized, sharable CDS alert design guidance.
Speaker(s):
Bayley Bennett, MD
Emory University/Children's Healthcare of Atlanta
Author(s):
Swaminathan Kandaswamy, PhD - Emory University School of Medicine; Bayley Bennett, MD - Emory University/Children's Healthcare of Atlanta;
Bayley Bennett, MD - Emory University/Children's Healthcare of Atlanta
Prioritizing What to Measure in CDS: Consensus-Based Domains for a Practical Evaluation Framework
Poster Number: 169
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support, Evaluation, Governance
Programmatic Theme: Clinical Research Informatics
Clinical decision support (CDS) evaluation is inconsistent across institutions and often focuses on narrow metrics. We conducted a modified Delphi study to prioritize CDS evaluation domains based on essentiality and feasibility. Eighteen operational CDS experts rated 12 domains; CDS Design Elements, Fidelity of Use, Burden, Adoption, Process Adherence, Information Quality, and Reach ranked highest. Round 2 (ongoing) is defining a minimal core, secondary, and “nice-to-have” tiered framework for CDS evaluation.
Speaker(s):
Swaminathan Kandaswamy, PhD
Emory University School of Medicine
Author(s):
Swaminathan Kandaswamy, PhD - Emory University School of Medicine; Evan Orenstein, MD - Children's Healthcare of Atlanta; Adam Moses, MHA, PMP, FAMIA - Advocate Health; Eric Kirkendall, MD, MBI - Wake Forest Baptist School of Medicine/Advocate Health; Naveen Muthu, MD - Children's Healthcare of Atlanta;
Swaminathan Kandaswamy, PhD - Emory University School of Medicine
Unsupervised Deep Clustering to Identify High‑Risk Emergency Department Chest Pain Subgroups Using Early EHR Data
Poster Number: 170
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support, Deep Learning, Machine Learning
Programmatic Theme: Clinical Research Informatics
Chest pain is a common reason for emergency department (ED) visits, yet only a small fraction represents life‑threatening conditions or require hospitalization, making rapid risk stratification difficult. Manual labeling of patient risk from EHR is time-consuming and resource intensive. To support ED providers in detection of high-risk groups, we applied unsupervised and deep clustering methods to 56,343 ED chest pain encounters using 58 structured features including demographics, symptoms, laboratory biomarkers, and early vital signs. Hierarchical clustering and deep hierarchical clustering were evaluated using outcomes such as mortality, hospital admission, ICU stay, and biomarker abnormalities. Both methods produced four clusters with two large lower risk subgroups and two smaller high-risk subgroups. Smaller clusters identified by Deep HC showed markedly higher mortality and admission rates, while smaller subgroups from HC highlighted abnormalities in glucose and platelet levels. These findings show that unsupervised clustering can reveal clinically meaningful patient subgroups.
Speaker(s):
Sourav Kumar Ghosh, MS in Electrical and Computer Engineering
University of Minnesota
Author(s):
Tonushree Dutta, M.S. - University of Minnesota Twin Cities; John Sartori, PhD - 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;
Sourav Kumar Ghosh, MS in Electrical and Computer Engineering - University of Minnesota
From Ambiguity to Accuracy: Enhancing Ordering Accuracy for Elemental Medications
Poster Number: 171
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support, Workflow, Usability, Qualitative Methods
Programmatic Theme: Clinical Informatics
Medications that can be dosed by either their elemental component or by the salt component vary in whether the strength displayed in mg in the EMR is the elemental or salt form of the product, resulting in inconsistent ordering and potential for medication errors. This project implemented discrete dosing units of “mg elemental” and “mg salt form” to offer clarity on the prescribed dose.
Speaker(s):
Leanne Murray, PharmD
Nationwide Children's Hospital
Author(s):
Leanne Murray, PharmD - Nationwide Children's Hospital;
Leanne Murray, PharmD - Nationwide Children's Hospital
Implementation of an Asthma Discharge SmartSet is Associated with Improved ED Prescription Practices
Poster Number: 172
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support, Administrative Systems, Healthcare Quality, Patient Safety, Workflow, Usability
Programmatic Theme: Clinical Informatics
Emergency departments (EDs) under-utilize Global Initiative for Asthma (GINA)-recommended inhaled corticosteroid (ICS)–containing therapy for asthma exacerbation. We evaluated whether implementing a discharge SmartSet, which offered recommended ICS-based regimens, was associated with increased guideline-concordant prescribing. After implementation, budesonide-formoterol prescriptions increased from 1.14% to 2.59% of inhaler prescriptions (RR 2.28, p<0.001). Interrupted time series analysis confirmed a 2.34-fold increase. Our results suggest that SmartSets are a practical solution for encouraging guideline-concordant ED prescribing.
Speaker(s):
Gregory Han, MD
New York Presbyterian Hospital
Author(s):
Juliet Jacobson, MD - Columbia University Medical Center, Weill Cornell Medicine; Judah Brown, PharmD - New York Presbyterian Hospital; Michael Alfonzo, MD, MS - Weill Cornell Medicine; Sarah Hanif, PhD - Weill Cornell Medicine; Jason Chu, MD - Columbia University Medical Center; Elizabeth Hartofilis, MD - Columbia University Medical Center; Dana Sacco, MD - Columbia University Medical Center; Eugene Kim, MD - Columbia University Medical Center; Edward Suh, MD - Columbia University Medical Center;
Gregory Han, MD - New York Presbyterian Hospital
Clinician Perspectives and Experiences with OPAs for Heart Failure Management
Poster Number: 173
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support, Chronic Care Management, Qualitative Methods
Programmatic Theme: Clinical Informatics
Guideline-directed medical therapy (GDMT) for heart failure management is underutilized, despite its ability to significantly reduce mortality. This study examined clinician perspectives and experiences with our practice advisories (OPAs) designed to assist cardiology clinicians in prescribing GDMT. We surveyed these clinicians and analyzed their provided acknowledgement reasons, which revealed that while they enjoy some elements of the OPAs, over half do not utilize them to order GDMT and prefer to order GDMT on their own.
Speaker(s):
Katharine Kleinbrahm, Medical Student
Geisinger College of Health Sciences
Author(s):
Katharine Kleinbrahm, BS - Geisinger College of Health Sciences; Cherie Shah, BS - Geisinger College of Health Sciences; Eric Mole, BS - Geisinger College of Health Sciences; Steven Voyce, MD - Geisinger College of Health Sciences; Amir Goren, PhD - Geisinger Program Evaluation; Apoorva Pradhan, BAMS, MPH - Geisinger; Vanessa Hayduk, BS, PMP - Geisinger; Gail Rosenbaum, PhD - Geisinger Behavioral Insights Team; Eric Wright, PharmD, MPH - Geisinger;
Katharine Kleinbrahm, Medical Student - Geisinger College of Health Sciences
Integrating AI-Enabled Clinical Decision Support for Proactive Workplace Violence Risk Identification
Poster Number: 174
2026 Symposium 25x5 Presentation
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support, Artificial Intelligence, Informatics Implementation, Patient Safety, Documentation Burden
Programmatic Theme: Clinical Informatics
Mayo Clinic Arizona and the Kern Center implemented an AI-enabled clinical decision support tool to enhance proactive workplace violence prevention. Nearly 67,000 manual screenings yielded 2% admission rate to a multidisciplinary support program. Using transfer learning to adapt an existing model, hourly inpatient risk scores prioritized review. During a two-month pilot, 18% of AI-flagged patients were admitted, improving review efficiency while preserving human oversight and supporting iterative model refinement consistent with Learning Health System principles.
Speaker(s):
Patrick Wilson, MPH
Mayo Clinic
Author(s):
Ricardo Rojas, BA - Mayo Clinic;
Patrick Wilson, MPH - Mayo Clinic
Caregivers’ Experiences Communicating Clinical Concerns in Pediatric Hospitalization
Poster Number: 175
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support, Patient Safety, Qualitative Methods, Patient-/Person-Generated Health Data
Working Group: Clinical Decision Support Working Group
Programmatic Theme: Clinical Research Informatics
The qualitative study used online focus groups with caregivers to explore experiences communicating clinical concerns during children’s hospitalization and identify strategies to incorporate their input into AI-enabled clinical decision support systems. Interim findings indicate perceived dismissal of caregiver concerns, differences in caregiver and clinician recognition of early risk, and nurses as trusted advocates who facilitate concerns escalation.
Speaker(s):
YU-CHING YANG, Ph.D.
Columbia University Department of Biomedical Informatics
Author(s):
YU-CHING YANG, Ph.D. - Columbia University Department of Biomedical Informatics; Sarah Rossetti, RN, PhD, FAAN, FACMI, FAMIA, FIAHSI - Columbia University Irving Medical Center; Brian Douthit, PhD, RN, NI-BC, FAMIA - Department of Biomedical Informatics, Vanderbilt University; Kenrick Cato, PhD, RN, CPHIMS, FAAN, FACMI - University of Pennsylvania/ Children's Hospital of Philadelphia; David Albers, PhD - University of Colorado, Department of Biomedical Informatics; Catherine Ivory, PhD, NI-BC, NEA-BC, FAAN, FAMIA - Vanderbilt University Medical Center; Alvin Jeffery, PhD, RN - Vanderbilt University Medical Center;
YU-CHING YANG, Ph.D. - Columbia University Department of Biomedical Informatics
Managing Dementia Social Risks: Toward a Clinical Decision Support Tool
Poster Number: 176
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support, User-centered Design Methods, Human-computer Interaction, Informatics Implementation
Programmatic Theme: Clinical Informatics
We used a qualitative user-centered design approach to refine iSMART, a clinical decision support tool for dementia care. Outpatient providers and staff identified caregiver support, financial strain, and transportation as key social risks. Participants found iSMART useful for identifying and prioritizing social risks and connecting patients and caregivers to resources. Findings will guide tool refinement and future implementation.
Speaker(s):
Miad Alfaqih, Phd
University of Florida
Author(s):
Miad Alfaqih, Phd - University of Florida; Philipp Haessner, BS, MBA, MS - University of Florida; Jennifer LeLaurin, PhD - 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;
Miad Alfaqih, Phd - University of Florida
EHR-Integrated Predictive Modeling in the ED: The Epic Deterioration Index Moderately Identifies Boarded Patients at Risk of Clinical Deterioration
Poster Number: 177
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support, Evaluation, Machine Learning, Informatics Implementation, Critical Care, Healthcare Quality, Artificial Intelligence
Programmatic Theme: Clinical Informatics
Emergency department (ED) boarding increases the risk of delayed care and clinical deterioration. We evaluated the Epic Deterioration Index (EDI), an EHR-integrated predictive model, among ED boarders across six hospitals at Emory Healthcare. In over 22,000 boarded encounters, the EDI demonstrated moderate discrimination for clinical deterioration (AUC 0.63). While performance varied by outcome, the model efficiently identified higher-risk patients, suggesting potential value for prioritizing monitoring and guiding early intervention in crowded ED settings.
Speaker(s):
Megan Schwinne, MPH
Emory University
Author(s):
Megan Schwinne, MPH - Emory University; Shea Sparks, MD - Emory University; Sharaf Khan, MD - Emory Healthcare; Sivasubramanium Bhavani, MD - Emory University;
Megan Schwinne, MPH - Emory University
A Mathematical Algorithm for Antibiotic Selection Using a Risk-Aware Weighted Decision Model
Poster Number: 178
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support, Patient Safety, Public Health, Infectious Diseases & Epidemiology
Programmatic Theme: Clinical Informatics
Appropriate antibiotic selection requires balancing multiple competing clinical factors, including infection risk, antimicrobial effectiveness, spectrum breadth, and patient-specific characteristics such as prior resistance patterns. However, clinical decision-making in this context remains heterogeneous and highly dependent on physician experience.
We propose an interpretable clinical decision support approach based on a Weighted Sum Model (WSM) that integrates multiple clinically relevant variables into a structured decision framework. The model produces a transparent score for candidate antibiotics, enabling second-opinion recommendations that support physician judgment.
Unlike black-box machine learning approaches, this method prioritizes interpretability, traceability of reasoning, and reduced algorithmic bias, making it particularly suitable for clinical environments where transparency and accountability are essential.
Speaker(s):
Ye Jun Won, Bachelors in Science
Sungkyunkwan university
Author(s):
Ye Jun Won, Bachelor in science - Sungkyunkwan university; Taerim Kim, Ph.D - Samsung Medical Center; Chae yeon Park, Bachelor in Science - Sungkyunkwan university;
Ye Jun Won, Bachelors in Science - Sungkyunkwan university
Federated ENACT EHR Data as Source for Generating Real-World Evidence in Comparative Safety Analysis of Three Classes of Anti-Diabetic Medications
Poster Number: 179
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support, Usability, Real-World Evidence Generation
Programmatic Theme: Clinical Research Informatics
Federated EHR networks enable the timely generation of real-world evidence while preserving privacy. Using aggregated data from 11 sites in the ENACT network (2012–2022), we evaluated the feasibility of pharmacovigilance analyses comparing SGLT-2 inhibitors with GLP-1 receptor agonists and DPP-4 inhibitors. Multivariable negative binomial regression showed lower risks of several adverse outcomes among SGLT-2i users, consistent with prior studies. These findings demonstrate the feasibility of using federated EHR networks for large-scale safety signal detection.
Speaker(s):
Malarkodi Jebathilagam Samayamuthu, MD
University of Pittsburgh
Author(s):
Malarkodi Jebathilagam Samayamuthu, MD - University of Pittsburgh; Michele Morris, BA - University of Pittsburgh; Chung-Chou H Chang, PhD - University of Pittsburgh; Wei-Hsuan Lo-Ciganic, MS, PhD - University of Pittsburgh; Shyam Visweswaran, MD PhD - University of Pittsburgh;
Malarkodi Jebathilagam Samayamuthu, MD - University of Pittsburgh
Machine Learning–Based Prediction of Major Adverse Cardiovascular Events in Cardiovascular-kidney-metabolic syndrome Stages 2–3 Using Nationwide Claims Data from South Korea and Japan
Poster Number: 180
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support, Machine Learning, Public Health
Programmatic Theme: Clinical Research Informatics
Patients with cardiometabolic conditions are at high risk of major adverse cardiovascular events (MACE). The American Heart Association recently introduced the cardiovascular–kidney–metabolic (CKM) syndrome framework to integrate cardiovascular, kidney, and metabolic risk. However, prediction models for MACE based on the CKM staging paradigm remain limited. We therefore developed a machine learning–based prediction model for MACE among patients with CKM stages 2–3 using nationwide claims data from South Korea and Japan..
Speaker(s):
Sunyeup Kim, BS
Sungkyunkwan University
Author(s):
Sunyeup Kim, BS - Sungkyunkwan University; Jaewon Cheon, BSc - Sungkyunkwan University; Beomseok Jeong, BSc - Sungkyunkwan University; Masao Iwagami, MD, PhD, MPH - Tsukuba University; Seungwon Lee, MD, PhD - Sungkyunkwan University;
Sunyeup Kim, BS - Sungkyunkwan University
Transcriptomic Stratification of Glucocorticoid Receptor Resistance in Multiple Myeloma: Toward Pre-Treatment Precision Therapy Decisions
Poster Number: 181
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support, Artificial Intelligence, Data Mining, Machine Learning, Patient-/Person-Generated Health Data, Chronic Care Management
Programmatic Theme: Translational Bioinformatics
Dexamethasone resistance is pre-existing and mechanistically embedded in multiple myeloma (MM) at diagnosis, yet no pre-treatment biomarker exists. Analyzing 997 MM transcriptomes (GSE276561), we trained 56 machine learning models across four gene panels. MM detection achieved AUC=0.985; glucocorticoid receptor (GR) resistance classification AUC=0.843; GR score regression R²=0.925. Immune deconvolution identified a pre-established immunosuppressive niche in resistant patients. This framework enables biomarker guided therapy decisions before dexamethasone exposure.
Speaker(s):
ashi keshariya, BDS, MPH
Public Health Informatics Institute and Jackson county public health (University Health)
Author(s):
Ashi Keshariya, BDS, MPH - Saint Louis University; Rohan Agarwal, MD - Saint Louis University;
ashi keshariya, BDS, MPH - Public Health Informatics Institute and Jackson county public health (University Health)
Early Adoption of AI-Enabled Medication Clinical Decision Support and Medication Safety Performance: An Analysis of Leapfrog Survey Data
Poster Number: 182
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support, Artificial Intelligence, Evaluation
Programmatic Theme: Clinical Informatics
A cross-sectional analysis of 1,399 hospitals participating in the 2025 Leapfrog CPOE Evaluation Tool and Annual Hospital Survey was performed to examine early adoption of AI-enabled medication clinical decision support and its association with medication safety performance. Only 9.3% of hospitals reported AI-enabled CDS. AI use was not significantly associated with CPOE safety scores after adjustment, suggesting AI-enabled CDS adoption remains limited and its impact on medication safety performance is not yet clear.
Speaker(s):
Zoe Co, BS
University of Michigan
Author(s):
David Classen, MD - University of Utah School of Medicine; David Bates, MD - Mass General Brigham; Harvard University;
Zoe Co, BS - University of Michigan
Selective Cross-Ontology Graph Traversal Improves Concept Expansion for OMOP Query Formulation
Poster Number: 183
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Controlled Terminologies, Ontologies, and Vocabularies, Knowledge Representation and Information Modeling, Information Retrieval, Interoperability and Health Information Exchange, Data Standards, Data Mining, Real-World Evidence Generation, Evaluation
Programmatic Theme: Clinical Research Informatics
Accurate cohort identification requires retrieving clinically meaningful concepts from EHRs encoded in heterogeneous medical terminologies. Semantic heterogeneity and ambiguous cross-ontology mappings complicate reliable query expansion. We developed an RDF cross-ontology terminology graph integrating OMOP Vocabulary and UMLS (~482M triples), defining relationship archetypes and linking concepts via shared vocabularies to enable traversal across ontologies. Using this graph, we implemented multiple traversal strategies based on selective edge-type inclusion. We evaluated query expansion from seed concepts representing common clinical terms and two CIPHER phenotypes. We found that naïve text-matching produced lexical false positives while missing clinically relevant variants. Expansion quality depended strongly on the types of relationships traversed. A strategy combining UMLS mappings with OMOP hierarchical edges achieved the best balance of coverage and semantic precision. Thus, graph traversal supports tunable accuracy-completeness tradeoffs. This framework provides a scalable, ontology-aware approach for tunable query expansion in phenotyping and cohort definition.
Speaker(s):
Ryan Birmingham, MS
Emory University
Author(s):
Ryan Birmingham, MS - Emory University; Tony Pan, PhD - Emory University;
Ryan Birmingham, MS - Emory University
Knowledge Graph Embedding-based Dual-Head Learning with Structure- Regularization for Relationship Classification in SNOMED CT
Poster Number: 184
2026 Symposium LIEAF Presentation
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Controlled Terminologies, Ontologies, and Vocabularies, Machine Learning, Knowledge Representation & Information Modeling
Programmatic Theme: Academic Informatics / LIEAF
SNOMED CT is a large medical ontology, encoding hierarchical and semantic relationships among medical concepts. Ensuring the correctness of these relationships is a difficult curation task, but it is essential for reliable clinical decision support and healthcare analytics. One task of ontology curation of relationships consists of identifying candidate concept pairs and assigning relationship types between them. Here, we focus on the latter task and assume that candidate pairs are already given. We formulate relationship classification in SNOMED CT as a supervised multi-class classification problem using existing SNOMED CT triples ⟨ concept₁, relation, concept₂⟩ as gold-standard. Utilizing millions of triples spanning over one-hundred relationship types, we train a structure-regularized two-head neural architecture that jointly optimizes relationship classification and knowledge graph embedding objectives. The proposed approach achieves a macro-F1 of 0.84 and an accuracy of 0.95, on the relationship classification task, outperforming strong BERT-based and KGE baselines with identical data sets.
Speaker(s):
T M Rubaith Bashar, B.Sc.
New Jersey Institute of Technology
Author(s):
T M Rubaith Bashar, B.Sc. - New Jersey Institute of Technology; Mengjia Xu, Ph.D. - New Jersey Institute of Technology; Christian De Guzman, High School Diploma - New Jersey Institute of Technology; James Geller, PhD - NJIT;
T M Rubaith Bashar, B.Sc. - New Jersey Institute of Technology
MMJ-O: An Ontology for Representing Cannabis Exposure and Clinical Context in Electronic Health Records
Poster Number: 185
2026 Symposium LIEAF Presentation
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Controlled Terminologies, Ontologies, and Vocabularies, Knowledge Representation and Information Modeling, Data Standards
Programmatic Theme: Academic Informatics / LIEAF
Cannabis use is increasingly documented in electronic health records (EHRs), yet information about cannabis exposure, including product type, route, dose form, and clinical context, is inconsistently represented across structured fields and clinical notes. Existing terminologies support concept identification but cannot model exposure events, temporal relationships, or documentation context. We are designing the Medical Marijuana Ontology (MMJ-O) to represent cannabis exposure and related clinical events for EHR-based phenotyping and Learning Health System workflows.
Speaker(s):
Wei Wei, PhD
Geisinger
Author(s):
Matthew Woll, PhD - Geisinger; Antoinette DiCriscio, PhD - Geisinger; Vanessa Troiani, PhD - Geisinger;
Wei Wei, PhD - Geisinger
Evaluating an AI-based Clinical Decision Support System for Sepsis and ARDS: An Interim Analysis of a Clinician Turing Test
Poster Number: 186
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Critical Care, Clinical Decision Support, Artificial Intelligence, Evaluation, Human-computer Interaction, Qualitative Methods
Programmatic Theme: Clinical Informatics
The first Clinician Turing Test, a novel Phase 1b approach for evaluating AI-CDSSs, is being conducted in a multi-site randomized vignette study testing the AI Ventilator Assistant (AVA). Interim analyses showed that clinicians were unable to distinguish AVA-generated treatment recommendations from those of human clinicians. Clinical appropriateness and safety ratings were lower for recommendations perceived as AI-generated. This method may support early validation of AI systems prior to clinical studies and deployment.
Speaker(s):
Antonia Angeli Gazola, MD
University of Pennsylvania
Author(s):
Antonia Angeli Gazola, MD - University of Pennsylvania; Benjamin Schmid, MS - University of Pennsylvania; Nicholas Bishop, BA - University of Pennsylvania; Alexander Ortiz, MD, MS - University of Pennsylvania; Dustin Krutsinger, MD, MSCE, ATSF - University of Nebraska Medical Center; Danielle Miltz, PA-C - Emory University; Heather Giannini, MD - Virtua Health Inc; Lyle Ungar - University of Pennsylvania; Meeta Prasad Kerlin, MD, MSCE - University of Pennsylvania; Megan Carr-Lettieri, MSN CRNP - University of Pennsylvania; Nuala Meyer, MD, MS - University of Pennsylvania; Romain Pirracchio, MD MPH PhD - UCSF; Sivasubramanium Bhavani, MD - Emory University; Thomas Valley, MD, MSc - University of Colorado; Yingying Lu, PhD, MBA - University of Pennsylvania; GARY WEISSMAN, MD, MSHP - University of Pennsylvania;
Antonia Angeli Gazola, MD - University of Pennsylvania
AI-PACE: From Competency Framework to Curriculum Implementation at an Academic Medical Center
Poster Number: 187
2026 Symposium LIEAF Presentation
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Curriculum Development, Artificial Intelligence, Large Language Models (LLMs), Machine Learning, Teaching Innovation
Programmatic Theme: Academic Informatics / LIEAF
As artificial intelligence transforms healthcare, medical education has struggled to keep pace. Current AI curricula remain fragmented, siloed within specialties or delivered as workshops that fail to produce lasting competency. We present the AI-PACE Framework, which adapts Bloom's Taxonomy to integrate Psychomotor, Affective, and Cognitive domains with longitudinal Embedding across the training continuum. We then report on its operationalization into a 4-week clinical elective for medical students at UC Davis School of Medicine.
Speaker(s):
Scott McGrath, PhD FAMIA
CITRIS Health UC Berkeley
Author(s):
Katherine Kim, PhD, MPH, MBA, FAMIA - University of California Davis; Karnjit Johl, MD - UC Davis Health; Nicholas Anderson, PhD - University of California, Davis;
Scott McGrath, PhD FAMIA - CITRIS Health UC Berkeley
Analytics in the Wild: A First-of-Its-Kind Graduate Curriculum Inside Epic
Poster Number: 188
2026 Symposium LIEAF Presentation
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Curriculum Development, Information Visualization, Teaching Innovation, Informatics Implementation, Knowledge Representation & Information Modeling
Programmatic Theme: Academic Informatics / LIEAF
What if graduate students learned analytics inside Epic, not a simulation, not a sandbox, but Epic? This poster presents the design of the first-ever graduate course built around Epic's SlicerDicer as its primary instructional platform. Students don't just learn to query data; they learn to defend visualization choices, identify risks of misinterpretation, and communicate analytic limitations to clinical and executive audiences. We also propose future licensing to make this curriculum available to other institutions.
Speaker(s):
Angie Hayes, DHI, MS
UTHealth MSBMI
Author(s):
Angie Hayes, DHI, MS - UTHealth MSBMI; Angela Ross, DNP, RN, MPH, PMP, DASM, PHCNS-BC - University of Texas Health McWilliams School of Biomedical Informatics; Camelia Addison, MEd - UTHealth Science Center Houston;
Angie Hayes, DHI, MS - UTHealth MSBMI
Identifying drug repurposing candidate for hidradenitis suppurativa using an integrated phenotype-to-mechanism framework
Poster Number: 189
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Data Mining, Population Health, Knowledge Representation & Information Modeling
Programmatic Theme: Translational Bioinformatics
We aim to resolve the clinical heterogeneity of hidradenitis suppurativa (HS) to enable mechanism-guided drug repurposing. We developed PiMInfer, an integrated phenotype-to-mechanism pipeline that combines real-world clinical data with network-based inference to identify phenotype-informed molecular modules. PiMInfer identified a coherent HS gene module comprising 69 genes and 9 functional endotypes, supported by multimodal evidence. Network-based drug repurposing further prioritized potential therapeutic candidates, highlighting PiMInfer as a scalable framework for precision medicine in complex diseases.
Speaker(s):
Wei-Ting Wang, MS
Weill Cornell Medicine
Author(s):
Wei-Ting Wang, MS - Weill Cornell Medicine; Manqi Zhou, PhD - Weill Cornell Medicine; Jie Tong, PhD - New York University Grossman School of Medicine; Meng-Ju Lin, PhD - New York University Grossman School of Medicine; Fei Wang, PhD - Weill Cornell Medicine; Catherine P. Lu, PhD - New York University Grossman School of Medicine; Chang Su, PhD - Weill Cornell Medicine;
Wei-Ting Wang, MS - Weill Cornell Medicine
MALAT1 Silencing Reduces EDB-Fibronectin Expression and Invasive Morphology in Triple-Negative Breast Cancer
Poster Number: 190
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Data Mining, Knowledge Representation and Information Modeling, Quantitative Methods
Programmatic Theme: Translational Bioinformatics
We investigate the regulatory link between lncRNA MALAT1 and EDB-Fibronectin (EDB-FN) in Triple-Negative Breast Cancer (TNBC). Using siRNA-mediated silencing in 3D spheroids, we demonstrated 91–95% MALAT1 knockdown across three independent sequences, leading to 92–94% EDB-FN reduction and phenotypic reversion from invasive to dormant morphology. TCGA analysis (n = 1,100+) confirms TNBC-selective MALAT1–FN1 correlation (R = 0.52), establishing a framework for imaging-accessible theranostic platforms.
Speaker(s):
Nikila Swaminathan, High School Student
Allen High School
Author(s):
Nikila Swaminathan, High School Student - Allen High School;
Nikila Swaminathan, High School Student - Allen High School
Architecting a Prototype Phenotyping Pipeline for Clinical Notes
Poster Number: 191
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Data Modernization, Natural Language Processing, Information Retrieval
Programmatic Theme: Clinical Research Informatics
We developed a pipeline to annotate History of Present Illness sections from MIMIC-IV discharge summaries with Human Phenotype Ontology (HPO) terms. Discharge summaries for five conditions — Multiple Myeloma, Multiple Sclerosis, Frontotemporal Dementia, Klinefelter— were retrieved via Elasticsearch and validated using a locally deployed LLM. Term rankings were clinically plausible and the pipeline is deployable in containerized environments.
Speaker(s):
Janos Hajagos, Ph.D.
Stony Brook University
Author(s):
Janos Hajagos, Ph.D. - Stony Brook University;
Janos Hajagos, Ph.D. - Stony Brook University
The ALS Knowledge Portal: A Federated Data Ecosystem for Governed Discovery and Access for AMP ALS
Poster Number: 192
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Data Sharing, Governance, Interoperability and Health Information Exchange, Data Standards, Informatics Implementation, Knowledge Representation & Information Modeling, Data Modernization, Privacy and Security
Programmatic Theme: Translational Bioinformatics
We describe the ALS Knowledge Portal, a governed, federated data ecosystem within the Accelerating Medicines Partnership for ALS (AMP ALS). Spanning Sage Bionetworks and Critical Path Institute platforms with indexed external resources, the portal supports metadata-driven discovery and governed access across 24+ datasets and 1,000+ samples covering transcriptomics, epigenomics, genomics, clinical and biomarker data types. We report on architecture, governance workflows, federation implementation and early adoption.
Speaker(s):
Jessica Malenfant, MPH
Sage Bionetworks
Author(s):
Jessica Malenfant, MPH - Sage Bionetworks; Ram Ayyala, PhD - Sage Bionetworks; Jonathan Liaw-Gray, MS - Sage Bionetworks; Vanessa Barone, PhD - Sage Bionetworks; Jay Hodgson, MSc - Sage Bionetworks; Bruce Hoff, MEng - Sage Bionetworks; Ann Novakowski, MPH - Sage Bionetworks; Robert Allaway, PhD - Sage Bionetworks; Christine Suver - Sage Bionetworks; Susheel Varma, PhD MBA FBCS - Sage Bionetworks;
Jessica Malenfant, MPH - Sage Bionetworks
Willingness to Participate in Brain Health Research: The Influence of Data Sharing and Use Practices
Poster Number: 193
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Data Sharing, Patient Engagement and Preferences, Patient-/Person-Generated Health Data
Programmatic Theme: Consumer Health Informatics
This study examined how data sharing practices influence willingness to participate in brain health research among U.S. adults. Participants rated their baseline willingness and then across 10 different scenarios. Willingness varied significantly across scenarios, with race and education moderating these effects. Willingness to participate in brain health research is highly dependent on specific data practices, highlighting the need for transparent data governance to foster trust and inclusive participation.
Speaker(s):
Stacey Dai, MPH
Columbia University School of Nursing
Author(s):
Stephanie Nino de Rivera, BA - Weill Cornell Medicine; Sarah Eslami, Bachelors of Science - Columbia University; Yihong Zhao, PhD, MPhil - Columbia School of Nursing; Meghan Reading Turchioe, PhD, MPH, RN - Columbia University School of Nursing; Ruth Masterson Creber, PhD, MSc, RN - Columbia University;
Stacey Dai, MPH - Columbia University School of Nursing
Modifications to a Pair of Care Coordination FHIR Applications for Behavioral Health
Poster Number: 194
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Data Sharing, Clinical Decision Support, Chronic Care Management, Transitions of Care
Programmatic Theme: Clinical Research Informatics
Independent behavioral health (BH) clinics need to exchange key health information with outside care sites, but barriers to sharing need to be addressed. Our team adapted a pair of extant light-weight applications, eCarePlan (Clinician facing) and MyCarePlanner (patient/client facing), that collate key data elements (diagnoses, medications, etc.) into one place, for behavioral health (BH) clinic needs. Adaptations were prioritized based on patient use cases, needs from behavioral health partner sites, and patient advisors’ feedback.
Speaker(s):
Kate Peak, MPH
Oregon Health & Science University
Author(s):
Kate Peak, MPH - Oregon Health & Science University; Matthew Storer, BS - Oregon Health & Science University; Amy Yates, MS - OHSU; LeAnn Michaels, BS - Oregon Health & Science University; Kari Stephens, PhD - University of Washington; David Dorr, MD, MS, FACMI, FAMIA, FIAHSI - Oregon Health & Science University;
Kate Peak, MPH - Oregon Health & Science University
Development of the National Library of Medicine Center for Clinical Observational Investigations (CCOI)
Poster Number: 195
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Data Sharing, Knowledge Representation and Information Modeling, Data Mining, Information Extraction
Programmatic Theme: Clinical Research Informatics
NLM developed the Center for Clinical Observation Investigations (CCOI) to support data reuse of large-scale clinical datasets by creating a knowledgebase of curated metadata. The purpose of CCOI is to help researchers find, understand and use datasets by displaying information about the administration of the dataset as well, as its population characteristics and clinical concept counts. CCOI currently includes 13 datasets from EHRs, claims, surveys, observational studies and registries and continues to expand.
Speaker(s):
Craig Mayer, MS in Bioinformatics
National Library of Medicine
Author(s):
Euny Lee, PhD, MS - NIH/NLM; Xin Gao, PhD - BCC;
Craig Mayer, MS in Bioinformatics - National Library of Medicine
Enhanced radiology services (Plain X-Ray) effectiveness by utilization of self-service appointment booking kiosk
Poster Number: 196
2026 Symposium 25x5 Presentation
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Data Sharing, Administrative Systems, Documentation Burden
Programmatic Theme: Clinical Informatics
The growing demand for diagnostic imaging, particularly X-ray services, highlights the need to enhance patient experience in radiology departments. Traditional booking methods often result in long wait times, patient dissatisfaction, and an uneven distribution of attendance, contributing to a stressful environment and potential complaints between patients and staff.
Speaker(s):
Lok Yann Stephanie Chan, Master Degree
Hospital Authority, Hong Kong
Author(s):
Yin Ling Esther Wong, Master Degree - Hospital Authority; Peter Wan, BSc., MBBS, MSc. - Hong Kong Hospital Authority - Information Tech & Health Informatics Division;
Lok Yann Stephanie Chan, Master Degree - Hospital Authority, Hong Kong
Standardizing ECMO Data for Clinical Care, Quality Improvement, and Research: Development of a Specialty-Specific Data Dictionary
Poster Number: 197
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Data Standards, Workflow, Informatics Implementation, Critical Care, Controlled Terminologies, Ontologies, Vocabularies, User-centered Design Methods
Programmatic Theme: Clinical Informatics
Learning Health Systems rely on standardized clinical data to support continuous improvement, research, and operational decision-making. ECMO programs generate complex clinical data across multiple systems, often resulting in fragmented documentation and inefficient manual abstraction. This project develops a standardized ECMO data dictionary aligned with ELSO registry requirements and internal clinical workflows. The methodology integrates multi-source data elements into a unified structure supporting clinical care, quality improvement, and research, providing a replicable framework for critical care learning health systems seeking to standardize high-acuity clinical data.
Speaker(s):
Ramya Govindarajan, Director
Emory University
Author(s):
Ramya Govindarajan, Director - Emory University;
Ramya Govindarajan, Director - Emory University
Keep the FHIR ® Burning: A Forward-Looking FHIR Capacity Building Initiative with Genomics Training Modules
Poster Number: 198
2026 Symposium LIEAF Presentation
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Data Standards, Interoperability and Health Information Exchange, Workforce Development
Programmatic Theme: Academic Informatics / LIEAF
Following the 2019 NIH guidance promoting FHIR adoption in clinical research, NIH ODSS and NIMHD assessed FHIR readiness across the NIMHD clinical research network. Surveys and interviews revealed varying expertise and a need for advanced genomics-focused training. In response, NIH developed a didactic genomics module and a hands-on FHIR, clinical decision support, and genomics workshop to strengthen capacity and support effective FHIR implementation in research.
Speaker(s):
Snipta Mallick, BS
The National Institutes of Health
Author(s):
Hsinyi Tsang, PhD - National Institutes of Health; Tomasz Adamusiak, MD PhD - MITRE Corporation; Alex Mays, MD - MITRE Corporation; Max Masnick, PhD; Belinda Seto, PhD - National Institutes of Health;
Snipta Mallick, BS - The National Institutes of Health
An Informatics Pipeline for Structuring Rare Disease Registry Data for Multi-System Phenotype Profiling in Wiedemann-Steiner Syndrome
Poster Number: 199
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Data Transformation/ETL, Global Health, Delivering Health Information and Knowledge to the Public, Governance
Programmatic Theme: Clinical Research Informatics
Rare disease registries often capture clinical variables in heterogeneous formats that limit phenotypic analysis. We
developed an informatics pipeline to transform non-standardized multi-select registry variables into structured
phenotype observations and applied it to 137 genetically confirmed Wiedemann-Steiner Syndrome patients. The
pipeline enabled reproducible multi-system phenotype profiling and demonstrates a scalable framework for
structuring registry data to support rare disease research and Learning Health Systems.
Speaker(s):
Sarika Vemana, Masters
Grand Valley State University
Author(s):
Sarika Vemana, Masters - Grand Valley State University; Nadia Falah, M.D., M.S., FACMG - MD Genetic Clinic; Bushra Rashrash, Masters - GVSU; Suhila Sawesi, PhD - GVSU;
Sarika Vemana, Masters - Grand Valley State University
Cohort Assessment of Stanford Medicine EHR Datasets for Machine Learning Studies in Polycystic Ovary Syndrome (PCOS)
Poster Number: 200
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Data Transformation/ETL, Machine Learning, Information Visualization, Knowledge Representation & Information Modeling
Working Group: Clinical Information Systems Working Group
Programmatic Theme: Clinical Research Informatics
Electronic Health Record (EHR) data are increasingly being used for predictive modeling. While they provide
detailed insights into patients’ interactions with the healthcare system, they must be interpreted within the appropriate context, considering data representation and variability. Through this study we assess EHR data from EHRSHOT and INSPECT EHR with heterogeneous care trajectories and sparse data of PCOS patients to build ML models.
Speaker(s):
Sravya Vunnam, MS in Health Informatics
George Mason University
Author(s):
Janusz Wojtusiak, PhD - George Mason University;
Sravya Vunnam, MS in Health Informatics - George Mason University
GEESE: Genotype-aware End-to-End Spatio-temporal Embedding for Behavioral Phenotyping
Poster Number: 201
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Deep Learning, Artificial Intelligence, Machine Learning
Programmatic Theme: Translational Bioinformatics
Behavioral phenotyping of genetic animal models currently requires labor-intensive manual feature engineering that limits reproducibility and scalability. We present GEESE, an end-to-end deep learning framework that learns behavioral representations directly from 3D pose dynamics without hand-crafted features. Using a pretrained time series foundation model, we encode movement sequences into a behavioral manifold that supports both behavior classification and genotype prediction. Evaluated across three autism-associated genetic models (CNTNAP2, CHD8, FMR1), our deep learning approach surpasses hand-crafted feature baselines in both tasks, revealing that learned representations capture genotype-specific behavioral signatures. The framework generalizes across genetic backgrounds, and an all-cohort model identifies both genetic background and genotype from movement patterns alone. We further provide HONK, an interactive intelligent tool enabling researchers without programming expertise to perform behavioral phenotyping from pose data through natural language interaction.
Speaker(s):
Yiran Ding, Sc. M.
Michigan State University
Author(s):
Yiran Ding, Sc. M. - Michigan State University; Yuen Gao, PhD - Michigan State University; Chunqi Qian, PhD - Michigan State University; Zijun Cui, PhD - Michigan State University;
Yiran Ding, Sc. M. - Michigan State University
Multimodal Deep Learning for Pressure Injury Staging Using Images and Nursing Records
Poster Number: 202
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Deep Learning, Natural Language Processing, Artificial Intelligence, Clinical Decision Support, Machine Learning, Information Extraction
Programmatic Theme: Clinical Informatics
This study developed an explainable multimodal pressure injury staging model integrating wound images with KLUE-BERT-embedded nursing records. Compared with the image-only baseline, the image-plus-nursing-records model outperformed the baseline across all classification metrics. Explainable AI analysis revealed that the model leveraged both visual wound features and systemic patient indicators such as 'physical restraint’, 'invasive', and 'neurologic'. By approximating holistic expert reasoning, the model enhances transparency and offers a reliable, interpretable decision-support tool for clinical nursing practice.
Speaker(s):
Yeonwoo Kim, BSN, RN
Yonsei University College of Nursing
Author(s):
Yeonwoo Kim, BSN, RN - Yonsei University College of Nursing; Jungmin You, PhD - Yonsei University College of Nursing; Yeonju Kim, MPH, RN - Yonsei University College of Nursing; Jiin Kim, RN, MPH - Yonsei University; Jong Chan Yeom, MS - Gachon University; Kwang Gi Kim, PhD - Gachon University; Mona Choi, PhD - Yonsei University College of Nursing;
Yeonwoo Kim, BSN, RN - Yonsei University College of Nursing
Prediction of Challenging Behaviors in Profound Autism in Classrooms Using Multimodal Wearable Sensors
Poster Number: 203
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Deep Learning, Patient Safety, Machine Learning, Clinical Decision Support
Programmatic Theme: Clinical Informatics
We developed a hierarchical multimodal deep learning model to predict challenging behaviors in nine students with profound autism using wearable physiological signals (EDA, accelerometry, skin temperature) in real-world classroom settings. Pre-trained foundation models extracted modality-specific features, which were fused for behavior prediction. The model achieved AUROC of 0.79 at a 10-minute horizon and 0.76 at 30 minutes, demonstrating clinically meaningful anticipatory prediction feasibility.
Speaker(s):
Hyeokhyen Kwon, Ph.D.
Emory University
Author(s):
Yadhu Kartha, Undergraduate - Georgia Institute of Technology; Hyeokhyen Kwon, Ph.D. - Emory University; Conor Anderson, MS - The Center for Discovery; Jenny Foster, MS - The Center for Discovery; Theresa Hamlin, Ed.D - The Center for Discovery; Johanna Lantz, PhD - The Center for Discovery; Gari Clifford, DPhil - Emory University - BMI; Juergen Hahn, PhD - Rensselaer Polytechnic Institute;
Hyeokhyen Kwon, Ph.D. - Emory University
Current status of anti-obesity medications and performance, an EHR based survey
Poster Number: 204
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Delivering Health Information and Knowledge to the Public, Information Visualization, Population Health
Programmatic Theme: Clinical Research Informatics
Recently, both FDA-approved AOMs (F-AOMs) and Off-label AOMs (O-AOMs) have gained popularity and demonstrated promising results in randomized clinical trials (RCTs). However, their effectiveness in real-world settings remains less understood. We evaluated population-level responses to AOMs and individual variability, leveraging Electronic Health Records (EHRs) and pharmacy dispense as real-world data sources.
Speaker(s):
Xiaoyang Ruan, PhD
The University of Texas Health Science Center at Houston
Author(s):
Xiaoyang Ruan, PhD - The University of Texas Health Science Center at Houston; Rui Li, Phd - UT health; Liwei Wang, MD, PhD - UTHealth; Shuyu Lu, Master - University of Texas Health Science Center at Houston; Andrew Wen, MS - University of Texas Health Sciences Center at Houston; Murali Sameer, PhD - UTHealth Housotn; Hongfang Liu, PhD - University of Texas Health Science Center at Houston;
Xiaoyang Ruan, PhD - The University of Texas Health Science Center at Houston
Monitoring novel psychoactive substance trends on social media: Analysis of discussions and dashboard implementation
Poster Number: 205
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Delivering Health Information and Knowledge to the Public, Information Visualization, Information Extraction, Data Mining
Programmatic Theme: Public Health Informatics
Novel psychoactive substances (NPSs) pose significant public health challenges due to rapid emergence and evolving use patterns. We mined 60,601 subreddits (2015–2025) for seven NPSs, comparing Reddit post volumes with National Forensic Laboratory Information System data. Moderate-to-strong correlations were observed across substances, with Reddit discussions preceding law enforcement reports by one year for several NPSs, suggesting early warning potential. A public dashboard provides interactive visualizations of these trends. https://sarkerlab.org/emerging-psychoactives-reddit/
Speaker(s):
Abeed Sarker, PhD
Emory University School of Medicine
Author(s):
Sahithi Lakamana, M.S. - Emory University; Sudeshna Das, PhD - Emory University; Jeanmarie Perrone, MD - University of Pennyslvania; Anthony Spadaro, MD - University of Pennsylvania; Rachel Wightman, MD - Brown University; Abeed Sarker, PhD - Emory University School of Medicine;
Abeed Sarker, PhD - Emory University School of Medicine
Reliability and Diagnostic Performance of Large Language Models in Identifying Literary Spin in Neurosurgical Press Releases
Poster Number: 206
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Delivering Health Information and Knowledge to the Public, Human-computer Interaction, Large Language Models (LLMs), Patient Safety, Artificial Intelligence
Programmatic Theme: Consumer Health Informatics
While Large Language Models (LLMs) may help automate the detection of subjective distortions in medical reporting, their ability to identify rhetorical bias and contextual gaps remains untested. Three human reviewers used a validated spin classification model to assess neurosurgical press releases. ChatGPT-5.1 and Gemini-3 evaluated the same text using both “complex” and “simple” prompts. LLMs performed poorly, especially with simple prompts, indicating LLMs may inappropriately validate articles and contribute to the spread of misinformation.
Speaker(s):
Jacob Sternberg-Sher, MBE
Warren Alpert Medical School
Author(s):
Lauren Sugarmann, BS - Department of Neurosurgery, Warren Alpert Medical School of Brown University; Joshua Feler, MD - Department of Neurosurgery, Warren Alpert Medical School of Brown University; Jack Klein, BA - Warren Alpert Medical School of Brown University; Alan Mach, BS - Warren Alpert Medical School of Brown University; Chris Nguyen, BS - Warren Alpert Medical School of Brown University; Andre Metzger, BS - Icahn School of Medicine at Mt. Sinai; Christian Schroeder, ScM - Department of Neurosurgery, Warren Alpert Medical School of Brown University; Madison Michaels, MS - Department of Neurosurgery, Warren Alpert Medical School of Brown University; Ziya Gokaslan, MD - Department of Neurosurgery, Warren Alpert Medical School of Brown University; Patricia Zadnik Sullivan, MD - Department of Neurosurgery, Warren Alpert Medical School of Brown University;
Jacob Sternberg-Sher, MBE - Warren Alpert Medical School
Phenotyping Temporal Patterns of EHR Use Among Primary Care Physicians
Poster Number: 207
2026 Symposium 25x5 Presentation
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Documentation Burden, Workflow, Data Mining, Quantitative Methods, Machine Learning
Programmatic Theme: Clinical Informatics
Primary care physicians spend substantial time interacting with the electronic health record, contributing to documentation burden. We applied k-means clustering to 15-minute Epic UAL Lite data from 197 primary care physicians and identified five temporal EHR-use phenotypes. Clusters differed in daytime intensity, midday breaks, and after-hours work; mean daily EHR time and physician sex differed significantly across clusters. This research helps characterize variation in EHR work patterns and may inform future efforts to reduce burden.
Speaker(s):
Jiarun Dai, PhD Student
University of Wisconsin - Madison
Author(s):
Jiarun Dai, PhD Student - University of Wisconsin - Madison; Mark Micek, MD - University of Wisconsin - Madison; Jeffrey Baltus, MS - University of Wisconsin - Madison; Brian Arndt, Dr. - University of Wisconsin - Madison; Adam Rule, PhD - University of Wisconsin - Madison;
Jiarun Dai, PhD Student - University of Wisconsin - Madison
Impact of Ambient AI on documentation burden in oncology: findings from a high-volume cancer center institution-wide implementation
Poster Number: 208
2026 Symposium 25x5 Presentation
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Documentation Burden, Healthcare Quality, Artificial Intelligence
Programmatic Theme: Clinical Informatics
While there is a growing body of evidence showing that ambient AI reduces clinician documentation burden and is perceived positively by clinicians, our understanding of ambient AI solutions in oncology is not well understood. Based on our assessment of an institution-wide rollout of ambient AI at a high-volume cancer center, despite somewhat low uptake, there were self-reported improvements in documentation time and overall satisfaction, suggesting the need for additional, targeted research in oncology.
Speaker(s):
Allison Lipitz-Snyderman, PhD
Memorial Sloan Kettering Cancer Center
Author(s):
SUSAN CHIMONAS, PhD - Memorial Sloan Kettering Cancer Center; Adam Watson, PhD - Memorial Sloan Kettering Cancer Center; Gilad Kuperman, MD, PhD - Columbia University; Maclain Damstrom, BS - Memorial Sloan Kettering Cancer Center; Suryan Goyal, BS - Memorial Sloan Kettering Cancer Center; Christina Ward, MBA - Memorial Sloan Kettering Cancer Center; Fariha Ahsan, BS - Memorial Sloan Kettering Cancer Center; Benjamin Roman, MD MBA - Memorial Sloan Kettering Cancer Center; Fernanda Polubriaginof, MD PhD - Memorial Sloan Kettering Cancer Center; Abigail Baldwin, MBA MSN RN - Memorial Sloan Kettering Cancer Center;
Allison Lipitz-Snyderman, PhD - Memorial Sloan Kettering Cancer Center
Implementation and Preliminary Outcomes of a Heat‑Responsive Care Management Initiative for Medicare Advantage Members
Poster Number: 209
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Environmental Health and Climate Informatics, Real-World Evidence Generation, Chronic Care Management, Population Health, Workflow
Working Group: Climate, Health and Informatics Working Group
Programmatic Theme: Clinical Informatics
This retrospective evaluation of a payor‑led quality improvement initiative examined the feasibility, implementation, and early utilization signals of integrating climate‑responsive outreach into routine care management for 3,978 Medicare Advantage members during extreme heat events. Nurse‑led delivery of heat‑specific education was feasible with high fidelity and engagement. Modest between‑group differences in 30‑day acute care utilization were observed. Findings provide early operational benchmarks for climate‑responsive initiatives within a learning health system.
Speaker(s):
Amanda Zaleski, PhD, MS, FACSM
Clinical Evidence Development, CVS Health
Author(s):
Amanda Zaleski, PhD, MS, FACSM - Clinical Evidence Development, CVS Health; Omar Soliman, MMS - CVS Health; Kelly Jean Craig, PhD - CVS Health; Annie Laurie Hines, PhD - CVS Health; Matthew Anderson, MS - CVS Health; Eric Hamilton, MS - CVS Health; Amanda Cheng, MPH - CVS Health; Eric Simoni, MBA - CVS Health; Daniel Knecht, MD - CVS Health; Paige Wickner, MD - CVS Health; Sreekanth Chaguturu, MD - CVS Health;
Amanda Zaleski, PhD, MS, FACSM - Clinical Evidence Development, CVS Health
Extracting Comprehensive Community Care-Driven Social Factors from Clinical Notes in a National Community Health Network
Poster Number: 210
2026 Symposium Health Equity Presentation
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Environmental Health and Climate Informatics, Natural Language Processing, Diversity, Equity, Inclusion, and Accessibility
Programmatic Theme: Clinical Research Informatics
This study examined 1,723 clinical notes from 540 patients in a national community health network to identify 12 social determinants of health domains using NLP. Compared with ClinicalBERT, a TF-IDF-based XGBoost model achieved better overall performance for detecting both SDOH-related information and social risk. The findings show that practice-driven SDOH extraction from community health notes is feasible and can support future screening, intervention, and research efforts.
Speaker(s):
Pedro Rivera, PhD
OCHIN
Author(s):
Pedro Rivera, PhD - OCHIN; Biao Yin, PhD - UMass Chan Medical School; Robert Voss, MS - OCHIN, Inc.; Joanna Georgescu, PhD - OCHIN; Ben Gerber, MD, MPH - University of Massachusetts Chan Medical School; Huanmei Wu, FAMIA, PhD - Temple University; Omar Martinez, JD - University of Central Florida; Teresa Schmidt, PhD - OCHIN; Feifan Liu, PhD - University of Massachusetts Chan Medical School;
Pedro Rivera, PhD - OCHIN
Toward Trustworthy Chatbots: A Red Teaming Protocol for Health-Related Conversations
Poster Number: 211
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Evaluation, Large Language Models (LLMs), Privacy and Security, Human-computer Interaction, Workflow, Artificial Intelligence
Programmatic Theme: Public Health Informatics
We propose a generalizable red teaming protocol for patient-facing health chatbots built on three pillars: error stratification (Document vs. Instruction Adherence), dual single- and multi-turn testing, and vulnerability-informed mitigation. Demonstrating this on an HRSN chatbot, we find a primary risk is behavioral noncompliance, rather than just factual inaccuracy. Instruction adherence failures escalate under multi-turn pressure, producing exclusively high-severity errors. Layered prompt and document augmentation eliminated all high-severity failures by establishing safe conversational failure states.
Speaker(s):
Syed-Amad Hussain, BSE
The Ohio State University
Author(s):
Syed-Amad Hussain, BSE - The Ohio State University; Daniel Jackson, B.Sc. - Nationwide Childrens Hospital at Abigail Wexner Research Institute; Ashley Lewis, BA - Ohio State University; Eric Fosler-Lussier, PhD - Ohio State University; Emre Sezgin, PhD - Nationwide Children's Hospital / The Ohio State University College of Medicine;
Syed-Amad Hussain, BSE - The Ohio State University
The Patient Experience and Acoustic Care Enhancement (PEACE) pilot: A Mixed-Methods Observational Study of White Noise and Biophilic Soundscaping in a Cancer Hospital
Poster Number: 212
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Evaluation, Healthcare Quality, Qualitative Methods, Quantitative Methods
Programmatic Theme: Clinical Informatics
This pilot study evaluated white noise and adaptive biophilic soundscaping compared to usual care in a post-anesthesia care unit at Memorial Sloan Kettering Cancer Center in New York. In this three-arm observational study of 1,758 patients, neither intervention improved pain, medication use, or discharge time in an already quiet setting. However, patients and staff reported largely positive experiences, emphasizing the importance of assessing experiential alongside clinical outcomes in environmental health research in hospitals.
Speaker(s):
Fernanda Polubriaginof, MD PhD
Memorial Sloan Kettering Cancer Center
Author(s):
SUSAN CHIMONAS, PhD - Memorial Sloan Kettering Cancer Center; Chris Brause, BA - Memorial Sloan Kettering Cancer Center; Brenden Bready, MA - Memorial Sloan Kettering Cancer Center; Kenneth Seier, MA - Memorial Sloan Kettering Cancer Center; Fernanda Polubriaginof, MD PhD - Memorial Sloan Kettering Cancer Center; Daniel Stein, MD, PhD - Memorial Sloan Kettering Cancer Center;
Fernanda Polubriaginof, MD PhD - Memorial Sloan Kettering Cancer Center
Calibration Analysis of an AI Model for Breast Cancer Risk Prediction
Poster Number: 213
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Evaluation, Artificial Intelligence, Clinical Decision Support, Deep Learning, Health Equity, Population Health, Machine Learning, Public Health
Working Group: Genomics and Translational Bioinformatics Working Group
Programmatic Theme: Clinical Informatics
The Mirai open-source deep learning algorithm for mammography-based risk prediction has demonstrated the potential to drive a paradigm shift toward personalized breast cancer screening. We conducted a large-scale external validation study to assess Mirai’s calibration performance, examining the numerical accuracy of its risk scores in determining a woman’s eligibility for risk-reduction interventions and supplemental screening. Our results suggest that Mirai may benefit from population-specific recalibration prior to clinical adoption.
Speaker(s):
Ojas Ankurbhai Ramwala, PhD
University of Wisconsin–Madison
Author(s):
Daniel Hippe, MS - Fred Hutchinson Cancer Center; Christoph Lee, MD, MS, MBA - University of Wisconsin-Madison; Kathryn Lowry, MD - University of Washington;
Ojas Ankurbhai Ramwala, PhD - University of Wisconsin–Madison
Fairness Evaluation of Machine Learning Models for Predicting Severe Chronic Kidney Disease Progression Using Electronic Health Records
Poster Number: 214
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Fairness and Elimination of Bias, Machine Learning, Artificial Intelligence, Health Equity
Programmatic Theme: Clinical Research Informatics
Machine learning (ML) models are increasingly used to predict chronic kidney disease (CKD) progression using electronic health records (EHRs). However, their fairness across demographic subgroups remains insufficiently studied. This study evaluates the predictive performance and demographic fairness of multiple ML models for severe CKD progression prediction. Using a retrospective EHR dataset, we assessed the fairness of random forest, support vector machines, neural network, and extreme gradient boosting decision tree models across age and sex groups using equality of opportunity difference, average odds difference, and disparate impact, including threshold-aware and intersectional analyses. While several models achieved strong overall discrimination (AUC up to 0.953), consistent age-related disparities were observed, with younger patients exhibiting lower sensitivity. Intersectional analyses further amplified subgroup differences. Threshold selection substantially influenced fairness outcomes, revealing performance–fairness trade-offs. These findings highlight the necessity of threshold-aware and intersectional fairness evaluation prior to deployment of ML-based CKD risk prediction models.
Speaker(s):
Feiyang Zhou, BEng
ShanghaiTech University
Author(s):
Feiyang Zhou, BEng - ShanghaiTech University; Huicong Chen, BEng - ShanghaiTech University; Zhiyu Wan, PhD - ShanghaiTech University;
Feiyang Zhou, BEng - ShanghaiTech University
Analytic Approaches to Estimating Effects and Associations of the Area Deprivation Index on Health Outcomes: A Methodological Scoping Review
Poster Number: 215
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Geospatial (GIS) Data/Analysis, Quantitative Methods, Population Health, Policy, Public Health
Programmatic Theme: Public Health Informatics
This scoping review characterizes analytic methods in 231 U.S. studies using the Area Deprivation Index (ADI) as a primary exposure during the policy implementation era (2021–present). Two-thirds employed no dependence handling for clustered (non-independent) observations, and 97.4% used no spatial method or diagnostic. Nearly half had neither. As CMS transitions to successor indices, these findings highlight an opportunity to strengthen analytic standards for area-level deprivation research.
Speaker(s):
Jacob Mackie, M.S.
Northwestern University
Author(s):
Jacob Mackie, M.S. - Northwestern University;
Jacob Mackie, M.S. - Northwestern University
Evaluating Kenya’s Influenza Sentinel Surveillance to Inform Modernization: A Mixed-Methods Study Using the CDC System Modernization Toolkit
Poster Number: 216
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Global Health, Data Modernization, Evaluation, Interoperability and Health Information Exchange, Infectious Diseases & Epidemiology, Public Health, Informatics Implementation, Usability
Programmatic Theme: Public Health Informatics
We evaluated Kenya’s influenza sentinel surveillance information systems using the CDC System Modernization Toolkit across nine sites. The SARI systems met 77% of assessed elements, and LIMS met 72%, with the lowest scores in usability, security, and documentation. Findings informed a prioritized modernization roadmap to harmonize identifiers, formalize synchronization, automate data-quality checks, and strengthen security controls to support detection, reporting, and response.
Speaker(s):
Clara Oguji, MPH
Centers for Disease Control and Prevention (CDC)
Author(s):
Clara Oguji, MPH - Centers for Disease Control and Prevention (CDC); Eric-Jan Manders, PhD - US Centers for Disease Control and Prevention; Carrie Peterson, AD - Centers for Disease Control and Prevention; Daniel Owusu, Phd - Centers for Disease Control and Prevention; Jorim Ayugi, Msc - Kenya Medical Research Institute (KEMRI), Nairobi, Kenya; Ian Orege, Msc - Kenya Medical Research Institute (KEMRI), Nairobi, Kenya; Peter Kinyanjui, Msc - Henry M. Jackson Foundation for the Advancement of Military Medicine Research Institute, Nairobi, Kenya; Irene Nachiro, Msc - National Influenza Center (NIC), Nairobi, Kenya; Rufus Nyaga, MPH - Global Health International, Association of Public Health Laboratories, Nairobi, Kenya; Rosalia Kalani, Msc - Ministry of Health, Kenya; Nzisa Liku, MPH - Centers for Disease Control and Prevention;
Clara Oguji, MPH - Centers for Disease Control and Prevention (CDC)
Machine Learning Based Prediction of Low Birthweight Using Maternal and Perinatal Factors in Eastern Jackson County
Poster Number: 217
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Global Health, Healthcare Quality, Delivering Health Information and Knowledge to the Public, Public Health, Surveys and Needs Analysis, Data Modernization, Informatics Implementation, Health Equity
Programmatic Theme: Public Health Informatics
We developed and externally validated four machine learning models for low birthweight (LBW <2,500 g) prediction using Missouri vital birth records from eastern Jackson County (training: 20,370 births, 2016–2022; validation: 5,315 births, 2023–2024). All models achieved AUC >0.92. An AUC-weighted ensemble achieved the highest balanced accuracy (0.856). SHAP analysis identified gestational age as the dominant predictor. Population attributable risk (PAR) analysis of actual observed LBW counts informs geospatial-targeted surveillance and intervention planning.
Speaker(s):
ashi keshariya, BDS, MPH
Public Health Informatics Institute and Jackson county public health (University Health)
Author(s):
ashi keshariya, BDS, MPH - Public Health Informatics Institute and Jackson county public health (University Health); Tolulope Awolusi, MPH, DrPH - Jackson County Public Health (University Health); Emily Olack, MPH - Jackson County Public Health;
ashi keshariya, BDS, MPH - Public Health Informatics Institute and Jackson county public health (University Health)
Residuals with JSONPlus Framework: Leveraging Pharmacovigilance Temporality
Click to View Handout
Poster Number: 218
2026 Symposium Health Equity Presentation
2026 Symposium 25x5 Presentation
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Health Equity, Data Modernization, Deep Learning, Real-World Evidence Generation, Usability, Population Health, Causal Inference, Informatics Implementation
Working Group: Consumer Health Informatics Working Group
Programmatic Theme: Clinical Research Informatics
𝘐𝘯𝘥𝘦𝘹𝘪𝘤𝘢𝘭 𝘦𝘭𝘦𝘤𝘵𝘳𝘰𝘯𝘪𝘤 𝘩𝘦𝘢𝘭𝘵𝘩 𝘳𝘦𝘤𝘰𝘳𝘥 (𝘌𝘏𝘙) 𝘥𝘦𝘱𝘭𝘰𝘺𝘮𝘦𝘯𝘵 𝘳𝘦𝘮𝘢𝘪𝘯𝘴 𝘶𝘯𝘦𝘷𝘦𝘯 𝘢𝘤𝘳𝘰𝘴𝘴 𝘥𝘦𝘷𝘦𝘭𝘰𝘱𝘪𝘯𝘨 𝘩𝘦𝘢𝘭𝘵𝘩 𝘴𝘺𝘴𝘵𝘦𝘮𝘴, 𝘤𝘰𝘯𝘴𝘵𝘳𝘢𝘪𝘯𝘪𝘯𝘨 𝘧𝘦𝘥𝘦𝘳𝘢𝘵𝘦𝘥 𝘱𝘩𝘢𝘳𝘮𝘢𝘤𝘰𝘷𝘪𝘨𝘪𝘭𝘢𝘯𝘤𝘦. 𝘕𝘰𝘷𝘦𝘭 𝘤𝘢𝘶𝘴𝘢𝘭 𝘪𝘯𝘧𝘰𝘳𝘮𝘢𝘵𝘪𝘤𝘴 𝘢𝘱𝘱𝘳𝘰𝘢𝘤𝘩𝘦𝘴 𝘦𝘷𝘢𝘭𝘶𝘢𝘵𝘦 𝘩𝘰𝘸 𝘴𝘤𝘢𝘭𝘢𝘣𝘭𝘦 𝘢𝘶𝘵𝘰𝘮𝘢𝘵𝘦𝘥 𝘪𝘯𝘧𝘳𝘢𝘴𝘵𝘳𝘶𝘤𝘵𝘶𝘳𝘦𝘴 𝘪𝘯𝘧𝘭𝘶𝘦𝘯𝘤𝘦 𝘢𝘥𝘷𝘦𝘳𝘴𝘦 𝘦𝘷𝘦𝘯𝘵 𝘳𝘦𝘱𝘰𝘳𝘵𝘪𝘯𝘨 (𝘈𝘌𝘙𝘴) 𝘢𝘯𝘥 𝘤𝘭𝘪𝘯𝘪𝘤𝘢𝘭 𝘵𝘳𝘪𝘢𝘭 𝘳𝘦𝘤𝘳𝘶𝘪𝘵𝘮𝘦𝘯𝘵 (𝘊𝘛𝘙) 𝘦𝘧𝘧𝘪𝘤𝘪𝘦𝘯𝘤𝘺. 𝘍𝘰𝘤𝘶𝘴𝘪𝘯𝘨 𝘰𝘯 𝘢 𝘴𝘤𝘰𝘱𝘪𝘯𝘨 𝘳𝘦𝘷𝘪𝘦𝘸 𝘸𝘪𝘵𝘩 𝘳𝘦𝘨𝘶𝘭𝘢𝘵𝘰𝘳𝘺 𝘳𝘦𝘨𝘪𝘴𝘵𝘳𝘪𝘦𝘴 𝘴𝘱𝘢𝘯𝘯𝘪𝘯𝘨 𝘵𝘩𝘦 𝘱𝘳𝘦𝘤𝘦𝘥𝘪𝘯𝘨 𝘵𝘦𝘯 𝘳𝘦𝘵𝘳𝘰𝘴𝘱𝘦𝘤𝘵𝘪𝘷𝘦 𝘺𝘦𝘢𝘳𝘴. 𝘖𝘣𝘴𝘦𝘳𝘷𝘢𝘵𝘪𝘰𝘯𝘢𝘭 𝘥𝘢𝘵𝘢𝘴𝘦𝘵𝘴, 𝘭𝘦𝘷𝘦𝘳𝘢𝘨𝘪𝘯𝘨 𝘵𝘳𝘢𝘯𝘴𝘱𝘰𝘳𝘵𝘢𝘣𝘪𝘭𝘪𝘵𝘺 𝘢𝘯𝘥 𝘨𝘦𝘯𝘦𝘳𝘢𝘭𝘪𝘻𝘢𝘣𝘪𝘭𝘪𝘵𝘺 𝘱𝘳𝘪𝘯𝘤𝘪𝘱𝘭𝘦𝘴, 𝘦𝘴𝘵𝘪𝘮𝘢𝘵𝘦𝘥 𝘤𝘰𝘶𝘯𝘵𝘦𝘳𝘧𝘢𝘤𝘵𝘶𝘢𝘭 𝘰𝘶𝘵𝘤𝘰𝘮𝘦𝘴 𝘶𝘯𝘥𝘦𝘳 𝘷𝘢𝘳𝘺𝘪𝘯𝘨 𝘌𝘏𝘙 𝘤𝘰𝘷𝘦𝘳𝘢𝘨𝘦. 𝘙𝘰𝘣𝘶𝘴𝘵 𝘤𝘢𝘶𝘴𝘢𝘭 𝘪𝘯𝘵𝘦𝘯𝘵𝘪𝘰𝘯-𝘵𝘰-𝘵𝘳𝘦𝘢𝘵 𝘢𝘯𝘥 𝘳𝘦𝘨𝘳𝘦𝘴𝘴𝘪𝘰𝘯 𝘥𝘪𝘴𝘤𝘰𝘯𝘵𝘪𝘯𝘶𝘪𝘵𝘺 𝘥𝘦𝘴𝘪𝘨𝘯𝘴 𝘲𝘶𝘢𝘯𝘵𝘪𝘧𝘪𝘦𝘥 𝘊𝘛𝘙 𝘮𝘢𝘯𝘢𝘨𝘦𝘮𝘦𝘯𝘵 𝘥𝘺𝘯𝘢𝘮𝘪𝘤𝘴. 𝘔𝘢𝘤𝘩𝘪𝘯𝘦 𝘭𝘦𝘢𝘳𝘯𝘪𝘯𝘨 𝘴𝘪𝘮𝘶𝘭𝘢𝘵𝘪𝘰𝘯𝘴 𝘶𝘴𝘪𝘯𝘨 𝘨𝘦𝘯𝘦𝘳𝘢𝘵𝘪𝘷𝘦 𝘢𝘥𝘷𝘦𝘳𝘴𝘢𝘳𝘪𝘢𝘭 𝘯𝘦𝘵𝘸𝘰𝘳𝘬𝘴 𝘮𝘰𝘥𝘦𝘭𝘦𝘥 𝘈𝘌𝘙 𝘵𝘳𝘢𝘫𝘦𝘤𝘵𝘰𝘳𝘪𝘦𝘴 𝘱𝘦𝘳 𝘪𝘯𝘤𝘳𝘦𝘮𝘦𝘯𝘵𝘢𝘭 𝘌𝘏𝘙 𝘴𝘢𝘵𝘶𝘳𝘢𝘵𝘪𝘰𝘯. 𝘈𝘥𝘷𝘢𝘯𝘤𝘦𝘥 𝘴𝘶𝘱𝘱𝘰𝘳𝘵 𝘷𝘦𝘤𝘵𝘰𝘳 𝘭𝘰𝘨𝘪𝘴𝘵𝘪𝘤 𝘧𝘰𝘳𝘮𝘶𝘭𝘢𝘵𝘪𝘰𝘯𝘴 𝘤𝘢𝘱𝘵𝘶𝘳𝘦𝘥 𝘯𝘰𝘯𝘭𝘪𝘯𝘦𝘢𝘳 𝘈𝘌𝘙 𝘵𝘳𝘦𝘯𝘥𝘴 𝘢𝘴𝘴𝘰𝘤𝘪𝘢𝘵𝘦𝘥 𝘸𝘪𝘵𝘩 𝘱𝘳𝘰𝘨𝘳𝘦𝘴𝘴𝘪𝘷𝘦 𝘥𝘪𝘨𝘪𝘵𝘢𝘭𝘪𝘻𝘢𝘵𝘪𝘰𝘯. 𝘛𝘳𝘢𝘯𝘴𝘱𝘰𝘳𝘵𝘢𝘣𝘭𝘦 𝘈𝘌𝘙 𝘮𝘰𝘥𝘦𝘭𝘪𝘯𝘨 𝘥𝘦𝘮𝘰𝘯𝘴𝘵𝘳𝘢𝘵𝘦𝘥 𝘢 2.15-𝘧𝘰𝘭𝘥 𝘱𝘰𝘰𝘭𝘦𝘥 𝘨𝘳𝘰𝘸𝘵𝘩 𝘢𝘯𝘥 𝘢 𝘧𝘪𝘷𝘦𝘧𝘰𝘭𝘥 𝘪𝘯𝘤𝘳𝘦𝘢𝘴𝘦 𝘪𝘯 𝘱𝘩𝘺𝘴𝘪𝘤𝘪𝘢𝘯-𝘭𝘦𝘷𝘦𝘭 𝘱𝘳𝘰𝘥𝘶𝘤𝘵𝘪𝘷𝘪𝘵𝘺. 𝘐𝘵𝘦𝘳𝘢𝘵𝘪𝘷𝘦 𝘥𝘪𝘢𝘨𝘯𝘰𝘴𝘵𝘪𝘤𝘴 𝘤𝘰𝘯𝘧𝘪𝘳𝘮𝘦𝘥 𝘴𝘵𝘳𝘰𝘯𝘨 𝘮𝘰𝘥𝘦𝘭 𝘧𝘪𝘵, 𝘴𝘵𝘢𝘣𝘭𝘦 𝘷𝘢𝘳𝘪𝘢𝘯𝘤𝘦, 𝘢𝘯𝘥 𝘮𝘪𝘯𝘪𝘮𝘢𝘭 𝘦𝘴𝘵𝘪𝘮𝘢𝘵𝘰𝘳 𝘰𝘷𝘦𝘳𝘧𝘪𝘵𝘵𝘪𝘯𝘨. 𝘊𝘢𝘶𝘴𝘢𝘭 𝘱𝘳𝘰𝘫𝘦𝘤𝘵𝘪𝘰𝘯𝘴 𝘪𝘯𝘥𝘪𝘤𝘢𝘵𝘦𝘥 𝘮𝘰𝘥𝘦𝘳𝘢𝘵𝘦𝘥 𝘊𝘛𝘙 𝘦𝘴𝘤𝘢𝘭𝘢𝘵𝘪𝘰𝘯 𝘶𝘯𝘥𝘦𝘳 𝘴𝘵𝘳𝘶𝘤𝘵𝘶𝘳𝘦𝘥 𝘌𝘏𝘙 𝘴𝘤𝘦𝘯𝘢𝘳𝘪𝘰𝘴. 𝘚𝘤𝘢𝘭𝘢𝘣𝘭𝘦 𝘪𝘯𝘵𝘦𝘨𝘳𝘢𝘵𝘪𝘰𝘯 𝘱𝘳𝘰𝘷𝘪𝘥𝘦𝘴 𝘱𝘰𝘭𝘪𝘤𝘺-𝘳𝘦𝘭𝘦𝘷𝘢𝘯𝘵 𝘧𝘳𝘢𝘮𝘦𝘸𝘰𝘳𝘬𝘴 𝘧𝘰𝘳 𝘦𝘲𝘶𝘪𝘵𝘢𝘣𝘭𝘦 𝘌𝘏𝘙 𝘦𝘹𝘱𝘢𝘯𝘴𝘪𝘰𝘯 𝘢𝘯𝘥 𝘦𝘯𝘩𝘢𝘯𝘤𝘦𝘥 𝘮𝘰𝘯𝘪𝘵𝘰𝘳𝘪𝘯𝘨.
Speaker(s):
Ogheneovo Uviase, Faculty
FUHSI
Author(s):
Ogheneovo Uviase, Faculty - FUHSI;
Ogheneovo Uviase, Faculty - FUHSI
Fractured logic? The Role of Race in Osteoporosis Risk Assessment
Poster Number: 219
2026 Symposium Health Equity Presentation
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Health Equity, Racial disparities, Fairness and elimination of bias, Evaluation
Programmatic Theme: Public Health Informatics
We examine how race-adjusted osteoporosis risk scoring contribute to screening disparities among women. We evaluated the impact of the race adjusted Simple Calculated Osteoporosis Risk Estimate (SCORE) using electronic health records, claims data, and linked datasets. Preliminary results show approximately 48% women are medium risk and 18% are high risk. Comparing performance across data sources and evaluating race-adjusted vs race-free models, we aim to identify data gaps and inform development of more equitable screening tools.
Speaker(s):
Elyse Lasser, DrPH
Johns Hopkins Bloomberg School of Public Health
Author(s):
Marika Cusick, PhD - Johns Hopkins Bloomberg School of Public Health; Grace Lyons, MS Applied Health Informatics - Deloitte; Kadija Ferryman, Ph.D. - New York University;
Elyse Lasser, DrPH - Johns Hopkins Bloomberg School of Public Health
Enabling OMOP in Safety-Net Health Systems Broadens Observational Research Populations
Poster Number: 220
2026 Symposium Health Equity Presentation
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Health Equity, Informatics Implementation, Real-World Evidence Generation
Programmatic Theme: Clinical Research Informatics
Clinical research often lacks population diversity, limiting generalizability. Large-scale observational research collaboratives offer a solution; however, Parkland Health and other diverse safety-net systems face barriers to participation. Parkland implemented the OMOP common data model (CDM) using University of Texas Southwestern Medical Center’s portable and efficient approach. Parkland’s 2.7 million patients (23% Black and/or African American and 46% Hispanic) represented in the CDM will diversify large-scale observational research populations locally and across the collaborative network.
Speaker(s):
Natalie Provenzale, MPH, MS, BSN, RN, CPHQ
Parkland Health
Author(s):
AAMIRAH VADSARIYA, RN MSN - UTSW; Mereeja Varghese;
Natalie Provenzale, MPH, MS, BSN, RN, CPHQ - Parkland Health
The Importance of Transparency and Functional Language in Designing Digital Health Tools for Reentry
Poster Number: 221
2026 Symposium Health Equity Presentation
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Health Equity, Human-computer Interaction, Mobile Health, Public Health
Programmatic Theme: Consumer Health Informatics
Justice-impacted women may view digital technologies as tools of surveillance rather than support, which can hinder access to mental health care during reentry. This qualitative study used four focus groups (N=14) with formerly incarcerated Black American women to explore design preferences for digital mental health tools. Participants favored transparent access to provider notes and functional, everyday language over clinical terminology. These features may reduce mistrust, stigma, and barriers to engagement with mental health services.
Speaker(s):
Kudrat Kadelu, Psychology
Yale University
Author(s):
Kudrat Kadelu, Psychology - Yale University;
Kudrat Kadelu, Psychology - Yale University
NLP-Based Tool Reduces Time Spent On Medication Reconciliation
Poster Number: 222
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Healthcare Quality, Natural Language Processing, Patient Safety
Programmatic Theme: Clinical Informatics
Medication reconciliation during hospital-to–skilled nursing facility transitions is time-consuming and error-prone due to incompatible electronic health records. We developed MedRex, an NLP-based web application that normalizes medications using RxNorm and highlights discrepancies between medication lists. In a pilot study with five physicians reconciling simulated cases, MedRex reduced reconciliation time by 75% while maintaining 100% discrepancy identification accuracy, suggesting potential to improve efficiency during care transitions.
Speaker(s):
li zhou, MD
VA
Author(s):
li zhou, MD - VA; Richard Korf, PhD - UCLA; Cindy Liu, PhD - UCLA; Jonathon Xu, BS - UCLA; Edmond Wen, BS - UCLA; Advaith Gowrishetty, BS - UCLA; Nicole Ju, BS - UCLA; Emily Sun, BS - UCLA; Elizabeth Kim, BS - UCLA;
li zhou, MD - VA
Glioblastoma Evaluation and Real-time Tracking (GLIOALERT): An EHR-Based Workflow Tool for Identifying Glioblastoma Patients and Coordinating Care Across VA and Community Health Systems
Poster Number: 223
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Healthcare Quality, Transitions of Care, Workflow, Patient Safety
Programmatic Theme: Clinical Informatics
We developed GLIOALERT, an electronic health records (EHR)-based workflow tool designed to identify patients with suspected glioblastoma (GBM) and to support care coordination across the Veterans Affairs (VA) healthcare system. We retrospectively evaluated GLIOALERT within the VA New England Healthcare System to characterize care pathways and identify coordination gaps. This approach can inform strategies to improve timeliness and integration of care across VA and community healthcare systems.
Speaker(s):
John Bihn, BA
VA Boston Healthcare System
Author(s):
John Bihn, BA - VA Boston Healthcare System; Leslie Brown, BA - VA Boston Healthcare System; Kaitlin Swinnerton, MIDS - VA Boston Healthcare System; Nathanael Fillmore, PhD - VA Boston Healthcare System; Colleen Shannon, MPH - VA Boston Healthcare System; Kelly Turner, MPH - VA Boston Healthcare System; Ryan Ferguson, ScD, MPH - VA Boston Healthcare System; Lauren Carter, Geography PhD - VA Boston; Svitlana Dipietro, BS - VA Boston Healthcare System; Linden Huhmann, PhD - VA Boston Healthcare System; Stephen La, BS - VA Boston Healthcare System; George Schneeloch, BS - VA Boston Healthcare System; Amelia Tarren, MS - VA Boston Healthcare System; Mary Brophy, MD; Brian Andersen, MD, PhD - VA Boston Healthcare System; Jacob Rachlin, MD, PhD - VA Boston Healthcare System; Genaro Villa, MD,PhD - VA Boston Healthcare System; Nhan Do, MD, MS, Clinical Informatics Diplomate - Boston VA HCS;
John Bihn, BA - VA Boston Healthcare System
User Interaction Patterns and Usability of a Global HIV Epidemiology Data Dashboard for Research and Decision-Making
Poster Number: 224
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Human-computer Interaction, User-centered Design Methods, Usability, Infectious Diseases & Epidemiology, Global Health, Evaluation, Qualitative Methods, Quantitative Methods
Programmatic Theme: Public Health Informatics
Dashboards can improve interpretation of health data, but poor design limit decision-making. This study evaluated
usability of the IeDEA Multi-Use Dataset (MUD) Dashboard, which integrates harmonized HIV data for researchers
and public health professionals. Using think-aloud interviews, the System Usability Scale, and NASA TLX (Task
Load Index) with 14 participants, we found strong usability, low task load, and five themes. Findings support the
dashboard’s value for HIV data analysis and suggest improvements in feature discoverability.
Speaker(s):
Creea Shannon, PhD, M.S., M.S.D.S.
Vanderbilt University
Author(s):
Stephany Duda, PhD - Vanderbilt University Medical Center; Judith Lewis, PhD - Vanderbilt University Medical Center; Peter Rebeiro, PhD, MHS - Vanderbilt University Medical Center; Jessica Ancker, MPH, PhD, FACMI - Vanderbilt University Medical Center;
Creea Shannon, PhD, M.S., M.S.D.S. - Vanderbilt University
Zero-Trust Claims: Integrating Homomorphic Encryption and Blockchain to Neutralize Clearinghouse Vulnerabilities
Poster Number: 225
2026 Symposium Health Equity Presentation
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Human-computer Interaction, Health Equity, Interoperability and Health Information Exchange, Machine Learning, Informatics Implementation
Programmatic Theme: Consumer Health Informatics
The research focuses on explaining the need of zero-trust claim architecture in better processing of medical claims at the recent time. The research proposed the data flow diagram for this architecture to combine homomorphic encryption and permissioned blockchain to eliminate clearinghouse-based attack surface. The study reflected on encrypted competition enabling validation without exposure of data, while immutable ledgers ensuring integrity, decentralized verification, auditability, breach impact and single-point failure vulnerability within the modern medical ecosystem.
Speaker(s):
Ruicha Gupta, Master in Health Informatics
New England college/ Claritev
Author(s):
Ruicha Gupta, Master in Health Informatics - New England college/ Claritev
One Model Card Does Not Fit All: A Layered Framework for Healthcare AI Model Cards from Empirical Stakeholder Analysis
Poster Number: 226
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Human-computer Interaction, User-centered Design Methods, Usability, Governance, Artificial Intelligence, Clinical Decision Support
Programmatic Theme: Clinical Informatics
We empirically examined stakeholder information priorities for healthcare AI Model Cards and derived a stakeholder-tailored, layered framework. Through content analysis of three Model Card templates and interviews with nine stakeholders (patients, clinicians, and developers), we prioritized 54 candidate items and identified a universal minimum Core set alongside stakeholder-specific Core needs. These findings were operationalized through a Core-Extended two-tier structure and six-section cognitive-flow ordering.
Speaker(s):
HyunSeok Lee, Master’s Student
lhs3789@gmail.com
Author(s):
Hyun Seok Lee, Master's student - Research Institute for Data Science and AI, Hallym University, Chuncheon, Republic of Korea; Adrian Zai, MD, PhD, MPH - University of Massachusetts Chan Medical School; Jae Yong Yu, PhD - Research Institute for Data Science and AI, Hallym University, Chuncheon, Republic of Korea;
HyunSeok Lee, Master’s Student - lhs3789@gmail.com
Augmentation Over Automation: Designing Human-AI Collaboration for Digital Emotional Intelligence
Poster Number: 227
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Human-computer Interaction, Artificial Intelligence, Natural Language Processing, User-centered Design Methods, Usability
Programmatic Theme: Consumer Health Informatics
Peer mentors in digital youth mental health support face significant cognitive load interpreting multimodal emotional expressions without nonverbal cues. We present an AI-augmented copilot system developed with a Singapore non-profit, evaluated with five peer mentors. Findings reveal an "assistance paradox" where real-time AI suggestions overwhelm while meta-analysis enhances, and a hierarchical trust model where foundational UX quality precedes acceptance of advanced AI features. Five design principles for responsible AI integration in emotional support are proposed.
Speaker(s):
ARIF WOOZEER, INFORMATION SYSTEMS
SINGAPORE MANAGEMENT UNIVERISITY
Author(s):
ARIF WOOZEER, INFORMATION SYSTEMS - SINGAPORE MANAGEMENT UNIVERISITY;
ARIF WOOZEER, INFORMATION SYSTEMS - SINGAPORE MANAGEMENT UNIVERISITY
Psychometric and Thematic Features of EHR During Infectious Outbreaks: Qualitative Review, Discriminant Analysis, and Biosurveillance Implications
Poster Number: 228
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Infectious Diseases and Epidemiology, Natural Language Processing, Qualitative Methods, Population Health, Machine Learning
Programmatic Theme: Public Health Informatics
Thematic and psycholinguistic themes in electronic health records (EHR) reveal clinicians’ cognitive processes and may be an informative and interpretable source of information in the detection of novel or poorly characterized infectious disease outbreaks. Frequencies of twenty-six dictionary themes related to biosurveillance were computed in laboratory confirmed, syndromically similar, and random case notes for early and late time periods of four infectious disease outbreaks in the VA hospital system. Means were computed and structural coefficients were derived from class-specific discriminant analysis (CSDA) investigating the early confirmed case notes. Increased references to laboratory studies and results, patient isolation, physician ambiguity or uncertainty, and public health authorities correlated with separation between confirmed cases and syndromically similar cases in the initial stages of an outbreak. Allusion to public health authority, precaution, activities leading to exposure, and isolation were most highly correlated with separation between early and late cases of disease.
Speaker(s):
Christian Barker, BS
University of Utah Health
Author(s):
Christian Barker, BS - University of Utah Health; Andrea Kalvesmaki, PhD - U.S. Department of Veterans Affairs; Alec Chapman, MS - University of Utah; Senthil Nachimuthu, MD, PhD, FAMIA - University of Utah School of Medicine; Kelly Peterson - University of Utah, Department of Epidemiology; Makoto Jones, MD, MS - University of Utah/Salt Lake City VA Healthcare System;
Christian Barker, BS - University of Utah Health
Integrating Sequencing-Based Pathogen and Resistance Detection into Burn Care
Poster Number: 229
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Infectious Diseases and Epidemiology, Surgery, Clinical Decision Support
Programmatic Theme: Translational Bioinformatics
This study compared the results of culture, targeted NGS (tNGS), and metagenomic NGS (mNGS) in burn wound samples. Sequencing identified significantly more bacterial and fungal taxa than culture while maintaining high agreement for key pathogens. Concordant detections between tNGS and mNGS supported the biological plausibility of additional organisms not recovered by culture. These findings highlight the potential of NGS to enhance burn infection diagnostics while complementing traditional culture methods.
Speaker(s):
Riley Watson, B.S.
University of Texas Medical Branch at Galveston
Author(s):
Vivian Tat, PhD - University of Texas Medical Branch; Katherine Araya, BS - University of Texas Medical Branch; Sarah Alnemrat, BDS - University of Texas Medical Branch; Elizaveta Naydanova, M.D./Ph.D. Student - University of Texas Medical Branch; Juquan Song, MD - University of Texas Medical Branch; Amina El Ayadi, PhD - University of Texas Medical Branch; Jennifer Narvaez, BS - University of Texas Medical Branch; Kamil Khanipov, PhD - University of Texas Medical Branch; Steven Wolf, MD - University of Texas Medical Branch; George Golovko, PhD - University of Texas Medical Branch;
Riley Watson, B.S. - University of Texas Medical Branch at Galveston
Integrating CSF Proteomics and Clinical Data to Improve Disease Progression Prediction in Niemann-Pick Disease Type C1
Poster Number: 231
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Informatics Implementation, Data Mining, Machine Learning
Programmatic Theme: Clinical Research Informatics
We evaluated whether cerebrospinal fluid (CSF) proteomic biomarkers improve the prediction of disease progression in Niemann–Pick disease type C1 (NPC1) beyond the age of neurological onset (ANO). Using cross-validated linear modeling, a two-protein panel outperformed ANO alone (LOOCV R² = 0.238 vs. 0.185) and demonstrated incremental predictive value when combined with ANO (LOOCV R² = 0.288). These findings support integrative molecular–clinical prognostic modeling in rare neurodegenerative disease.
Speaker(s):
Shixue Sun, PhD
NCATS/NIH
Author(s):
Shixue Sun, PhD - NCATS/NIH; Hajwa Kim, M.S. - Biostatistics Core, Center for Clinical and Translational Science, University of Illinois; Roshan Javanshad, PhD - Department of Chemistry, University of Illinois; Derek Alexander, MPH - Eunice Kennedy Shriver National Institute of Child Health and Human Development, National Institutes of Health; Stephanie Cologna, PhD - Department of Chemistry, University of Illinois; Forbes Porter, M.D.,Ph.D. - Eunice Kennedy Shriver National Institute of Child Health and Human Development, National Institutes of Health; Qian Zhu - National Institutes of Health;
Shixue Sun, PhD - NCATS/NIH
In Basket Patient Message Reduction Through Retrospective Text Analysis and Filter
Poster Number: 232
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Informatics Implementation, Patient-/Person-Generated Health Data, Natural Language Processing, Real-World Evidence Generation
Programmatic Theme: Clinical Informatics
Online patient portal messaging has become increasingly common and contributes to clinician workload. We evaluated the impact of a simple string-matching filter for suppressing brief gratitude messages, and assessed additional filter enhancements using vendor-suggested and locally derived phrase variants. The expanded filter was estimated to retrospectively suppress up to 7% of monthly patient messages. These findings demonstrate that simple text‑matching approaches can serve as effective, low‑resource strategies for reducing clinician in-basket burden.
Speaker(s):
Wendy Yin, MD
UT Southwestern
Author(s):
Samuel McDonald, MD - UT Southwestern; Robert Turer, MD - UT Southwestern Medical Center; Ling Chu, MD - UT Southwestern; DuWayne Willett, MD - University of Texas Southwestern Medical Center; Whitney Jones, MPA - UT Southwestern;
Wendy Yin, MD - UT Southwestern
Generating Aggregate Data of Diagnostic Concepts for an Effective Visualization Dashboard: An NLM Center for Clinical Observational Investigations (CCOI) Product
Poster Number: 233
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Informatics Implementation, Knowledge Representation and Information Modeling, Data transformation/ETL, Controlled Terminologies, Ontologies, Vocabularies
Programmatic Theme: Clinical Research Informatics
The NLM Center for Clinical Observational Investigations generated an aggregate data model for five large clinical datasets to deploy in a data visualization dashboard to better allow for understanding contents of datasets without directly accessing them. The aggregate data model calculated counts by diagnosis code, sex, race, and 10-year age groupings. This allowed for comparing across disparate codings used (four different code systems) and ranging amounts of used diagnosis codes (516 to 19,370).
Speaker(s):
Craig Mayer, MS in Bioinformatics
National Library of Medicine
Author(s):
Xin Gao, PhD - BCC; Euny Lee, PhD, MS - NIH/NLM;
Craig Mayer, MS in Bioinformatics - National Library of Medicine
MPath Central Notification Service: A Scalable Communication Infrastructure for Clinical Pathology Systems
Poster Number: 234
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Informatics Implementation, Workflow, Data Sharing
Programmatic Theme: Clinical Informatics
Molecular diagnostic laboratories are a critical part of patient care and require robust, bidirectional communication infrastructure to support real-time data exchange, and clinical decision-making. Traditional notification systems often operate in silos with redundant infrastructure and inconsistent messaging across clinical workflows. MPath CNS addresses these challenges by providing a unified, multitiered notification platform that streamlines communication between clinicians and laboratory staff while maintaining reliability, scalability, and compliance that is essential for modern day-to-day healthcare operations.
Speaker(s):
Sean Lachhander, D.Eng.
Memorial Sloan Kettering Cancer Center
Author(s):
Sean Lachhander, D.Eng. - Memorial Sloan Kettering Cancer Center; Eakrum Awwal, M.Eng. - Capital One; Aijazuddin Syed, Masters in Computer Science - Memorial Sloan Kettering Cancer Center;
Sean Lachhander, D.Eng. - Memorial Sloan Kettering Cancer Center
Patient Portal Recruitment Patterns: Age Influences Response Speed, Timing Has Little Impact
Poster Number: 235
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Informatics Implementation, Real-World Evidence Generation, Patient Engagement and Preferences, Workflow
Programmatic Theme: Clinical Research Informatics
Efficient clinical research recruitment remains challenging, as traditional outreach is often slow and resource-intensive. Analysis of 85,601 Epic MyChart invitations at Emory Healthcare examined patient response patterns by age and timing. Invitation day had minimal impact, while age strongly influenced engagement: middle-aged and older adults responded fastest. Findings highlight how EHR-integrated informatics can support data-driven, targeted, and timely patient portal recruitment in real-world research settings.
Speaker(s):
Megan Schwinne, MPH
Emory University
Author(s):
Megan Schwinne, MPH - Emory University; Barney Chan, MS CIS - Emory University; Chad Robichaux, MPH - Emory University; Mugisha Niyibizi, MPH - Georgia CTSA; Neal Dickert, MD, Phd - Emory University; Gabriel Najarro, PA-C, MMSc - Emory University;
Megan Schwinne, MPH - Emory University
WAKE UP SAFE (Version 2): A National Adverse Event Case Reporting and Common Anesthesia Record Denominator Database (Anesthesia Informatics Initiatives)
Poster Number: 236
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Informatics Implementation, Information Visualization, Patient Safety, Surgery
Programmatic Theme: Clinical Informatics
Wake Up Safe (WUS) Version 2 is a national pediatric anesthesia quality and safety database redesigned by the Children’s Hospital of Philadelphia in collaboration with WUS leadership. Built on a secure cloud platform using a standard common data model, WUS collects adverse event reports (numerator data) and anesthesia records (denominator data). The system provides standardized reporting, dashboards, and a Cohort Explorer to support multi-institutional anesthesia quality improvement across member hospitals.
Speaker(s):
Fuchiang (Rich) Tsui, PhD, FAMIA, IEEE Senior Member
Children's Hospital of Philadelphia and University of Pennsylvania
Author(s):
Sachin Grover, Data Scientist - The Children's Hospital of Philadelphia; Sifei Han, PhD - Children's Hospital of Philadelphia; Eamonn Tweedy, PhD - Children's Hospital of Philadelphia; Allan Simpao, M.D.,M.B.I. - Children's Hospital of Philadelphia/ Perelman School of Medicine, Univ. of Pennsylvania; Rajeev Iyer, MD - Children's Hospital of Philadelphia; Jay Deshpande, MD - Wake Up Safe;
Fuchiang (Rich) Tsui, PhD, FAMIA, IEEE Senior Member - Children's Hospital of Philadelphia and University of Pennsylvania
CliNormLLM: Applying LLMs for Extraction and Normalization of Clinical Concepts
Poster Number: 237
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Information Extraction, Large Language Models (LLMs), Interoperability and Health Information Exchange, Natural Language Processing
Programmatic Theme: Clinical Informatics
Clinical narratives contain important information for downstream analytics, but extracting and normalizing clinical concepts remains challenging. This manuscript presents CliNormLLM, a modular framework for end-to-end clinical concept extraction and normalization. CliNormLLM uses structured extraction with bracket-marked spans and multiple schemas to investigate the generation of normalization signals, followed by concept linking through dictionary matching and LLM-guided semantic retrieval and re-ranking. The framework is evaluated on the BioCreative VIII Task 3 and 2019 n2c2 Track 3 benchmarks. Across both datasets, generating textual normalization targets, such as preferred terms or fully specified names, consistently outperforms direct concept ID generation, and dynamic few-shot selection improves performance. The best results are 0.820/0.747 F1 on BC8T3 and 0.583/0.500 F1 on 2019 n2c2 under normalization-only/strict end-to-end evaluation. Error analysis shows that extraction errors, especially missed concepts and modifiers, account for many failures, highlighting extraction as the main bottleneck for more complex inputs.
Speaker(s):
Ekin Soysal, MS
UTHealth Houston
Author(s):
Ekin Soysal, MS - UTHealth Houston; Kirk Roberts, PhD - University of Texas Health Science Center at Houston;
Ekin Soysal, MS - UTHealth Houston
A Pipeline for Extracting Structured EHR Phenotype Metadata at Scale
Poster Number: 238
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Information Extraction, Large Language Models (LLMs), Workflow
Programmatic Theme: Clinical Research Informatics
The collection of EHR phenotypes requires time and effort prohibitive to large-scale attempts. We propose an automated pipeline to collect structured phenotype metadata from printed articles at scale. Our multistep process uses a keyword search, a classification module, LLMs for structured data extraction, and concludes with manual review. Our workflow is sufficiently fast and scalable to collect large amounts of computable phenotypes.
Speaker(s):
Michael Murray, MS
VA Boston Healthcare System
Author(s):
Yuk-Lam Ho, MPH - VA Boston Healthcare System; Junghoon Chae, PhD - Oak Ridge National Laboratory; David Heise, M.S. Computer Engineering - Oak Ridge National Laboratory; Jacqueline Honerlaw, RN, MPH - VA Boston Healthcare System; Monika Maripuri, MBBS, MPH - VA Boston Healthcare System; Kelly Cho, PhD - VA Boston Healthcare/Harvard Medical School;
Michael Murray, MS - VA Boston Healthcare System
Extracting Medication Discontinuation from Clinical Notes Using an EHR-Adapted Large Language Model
Poster Number: 239
2026 Symposium 25x5 Presentation
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Information Extraction, Natural Language Processing, Large Language Models (LLMs)
Programmatic Theme: Clinical Informatics
Accurate identification of medication discontinuation is critical for defining treatment exposure in clinical research. We developed an EHR-adapted large language model trained through clinical continual pretraining, medical instruction tuning, and clinical task adaptation. Evaluated on newly annotated datasets from Mass General Brigham and MIMIC, our model achieved the best performance among tested LLMs. Results demonstrate that domain-adapted LLMs can improve scalable extraction of medication discontinuation information from clinical notes.
Speaker(s):
Jiageng Wu, PhD
Brigham and Women's Hospital and Harvard Medical School
Author(s):
Jiageng Wu, PhD - Brigham and Women's Hospital and Harvard Medical School; Bowen Gu, MS - Brigham and Women's Hospital; Richard Wyss, PhD - Brigham and Women’s Hospital and Harvard Medical School; Kueiyu Joshua Lin, MD, ScD - Brigham and Women's Hospital and Harvard medical school; Jie Yang, PhD, FACMI, FAMIA - Harvard Medical School;
Jiageng Wu, PhD - Brigham and Women's Hospital and Harvard Medical School
Automated Literature Screening Using a Scalable Rule-Based Framework: Evaluation in Heart Failure and Chronic Obstructive Pulmonary Disease
Poster Number: 240
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Information Retrieval, Natural Language Processing, Evaluation, Real-World Evidence Generation
Programmatic Theme: Clinical Research Informatics
We developed a scalable, rule-based automated literature screening pipeline to support evidence synthesis in high-volume clinical domains. Applied to heart failure and COPD, the framework uses deterministic keyword and fuzzy matching rules to prioritize relevant studies before manual review. Automated screening reduced records requiring manual review by up to 84% while maintaining high sensitivity (94% HF; 88% COPD). This approach demonstrates how transparent rule-based methods can improve efficiency and scalability in literature screening.
Speaker(s):
Omar Nafaa, MS
Mass General Brigham
Author(s):
Omar Nafaa, MS - Mass General Brigham; Ania Syrowatka, PhD - Brigham and Women's Hospital/Harvard Medical School; Carrie Wade, MLIS - Countway Library, Harvard Medical School; Christine Iannaccone, MPH - Brigham and Women's Hospital; Michelle Frits, N/A - Brigham and Women's Hospital; Lynn Volk, MHS - Mass General Brigham; David Bates, MD - Mass General Brigham; Harvard University;
Omar Nafaa, MS - Mass General Brigham
Augmenting Biomedical Literature Search with Large Language Models
Poster Number: 241
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Information Retrieval, Artificial Intelligence, Large Language Models (LLMs), Natural Language Processing, Evaluation
Programmatic Theme: Clinical Research Informatics
We investigate the role of large language models in biomedical literature search through systematic experiments with query refinement, hybrid retrieval, and re-ranking across multiple benchmarks. We find that LLMs can both help and hinder search. Informed by these findings, we develop a search system over ~40 million PubMed articles combining dense and lexical retrieval, accessible to LLM clients via the Model Context Protocol.
Speaker(s):
Maya Geer, B.S.
NIH
Author(s):
Maya Geer, B.S. - NIH; Qiao Jin, M.D. - National Institutes of Health; Zhiyong Lu, PhD - National Library of Medicine, NIH;
Maya Geer, B.S. - NIH
Revealing User Preference Profiles for an LLM-Generated Consumer Family Health History Report
Poster Number: 242
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Information Visualization, User-centered Design Methods, Patient Engagement and Preferences, Surveys and Needs Analysis, Natural Language Processing, Machine Learning
Programmatic Theme: Consumer Health Informatics
This study aims to reveal user preferences for LLM-generated family history reports along three key attributes: graphical format, educational guidance, and organization, based on a choice-based conjoint analysis. Clustered conjoint preferences and conjoint user reasoning comments from the conjoint task revealed distinct preferences for specific educational guidance and tabular graphical formats. These findings enable prioritization of family history report features and provide a framework for more user-centered designs of consumer-facing health materials.
Speaker(s):
Michelle Nguyen, MS
Johns Hopkins University School of Medicine
Author(s):
Michelle Nguyen, MS - Johns Hopkins University School of Medicine; Kamryn Chan, - - Johns Hopkins University; Casey Taylor, PhD - Johns Hopkins University;
Michelle Nguyen, MS - Johns Hopkins University School of Medicine
Seeing the Structure: A Network Analysis of Clinical AI Perception Studies
Poster Number: 243
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Information Visualization, Quantitative Methods, Artificial Intelligence, Information Extraction
Working Group: Visual Analytics Working Group
Programmatic Theme: Public Health Informatics
A network analysis of stakeholder perspectives on clinical AI reveals a clear core–periphery structure, with a dominant connected component and one isolated outlier. Across 18 studies, binary risk indicators, explainability indicators, and AI type are coded and compared using Jaccard similarity (≥0.35). The network identifies shared perception profiles, weak bridge ties, and distinct study patterns, strengthening research synthesis by making cross-study structure visible and highlighting areas for closer interpretation or further study.
Speaker(s):
Tauseef Ibne Mamun, Trainee
Vanderbilt University Medical Center
Author(s):
Laurie Novak, PhD, MHSA - Vanderbilt University Medical Center Dept of Biomedical Informatics; Megan Salwei, PhD - Vanderbilt University Medical Center;
Tauseef Ibne Mamun, Trainee - Vanderbilt University Medical Center
Smarter, Safer Dental Care: Integrating Medical Data into Dental Workflows
Poster Number: 244
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Interoperability and Health Information Exchange, Usability, User-centered Design Methods
Programmatic Theme: Clinical Research Informatics
This study evaluated dentists’ medical information needs and the usability of a FHIR®-based Patient Medical Summary (PMS®) application that retrieves data from an HIE. Semi-structured interviews with 50 dental clinicians identified medications, medical conditions, vital signs, and allergies as the most important information. Participants reported limited access to EHR/HIE data and usability challenges affecting information visibility and interpretation. Results suggested interoperable tools, clinician training, and AI-enabled tools may improve dentists’ access to medical information
Speaker(s):
Shuning Li, Ph.D.
Indiana University School of Dentistry
Author(s):
Shuning Li, Ph.D. - Indiana University School of Dentistry; Jason Mao, MS - IUPUI; Srinija Kasaraneni, Health Informatics - Indiana University Purdue University Indianapolis; Brian Dixon, MPA, PhD - Regenstrief Institute; Grace Gomez Felix Gomez, BDS, MPH, PhD - Indiana University School of Dentistry; Titus Schleyer, DMD, MBA, PhD - Regenstrief Institute; Eneida Mendonca, MD, PhD - Cincinnati Children's Hospital / University of Cincinnati; Laura Marie Romito, DDS, MS, MBA, FNAP - Indiana University School of Dentistry; Huiping Xu, Ph.D. - Indiana University School of Medicine; Burke Mamlin, MD - Regenstrief Institute; Thankam Thyvalikakath, DMD, PhD, FACMI - Regenstrief Institute, Indiana University School of Dentistry;
Shuning Li, Ph.D. - Indiana University School of Dentistry
Evidence-Gated Knowledge Graphs for AI-Assisted Synthesis of Environmental Exposure Evidence
Poster Number: 245
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Knowledge Representation and Information Modeling, Controlled Terminologies, Ontologies, and Vocabularies, Artificial Intelligence, Natural Language Processing, Information Extraction, Information Retrieval, Data Mining, Environmental Health and Climate Informatics
Programmatic Theme: Clinical Research Informatics
Synthesizing scientific evidence on microplastics exposure is challenging due to wide variability in measurement methods, detection limits, and reporting conventions across studies. We present a prototype evidence-constrained knowledge graph framework that systematically captures study provenance, analytical methods, and exposure observations from the literature. This system supports structured queries that preserve methodological context and identify evidence gaps, thus enabling AI-assisted synthesis of environmental exposure and health research.
Speaker(s):
Sudeepa Bhattacharyya, Professor/PhD in Bioinformatics
University of Missouri Kansas City
Author(s):
Sudeepa Bhattacharyya, Professor/PhD in Bioinformatics - University of Missouri Kansas City; Suman Maity, PhD Computer Science - Missouri S&T, Rolla, MO, USA; Maryam Salehi, PhD - University of Missouri; Melody Greer, PhD - University of Arkansas for Medical Sciences (UAMS);
Sudeepa Bhattacharyya, Professor/PhD in Bioinformatics - University of Missouri Kansas City
Imaging-Augmented Multimodal Biomedical Knowledge Graphs Improve Cognitive Outcome Modeling in Alzheimer’s Disease
Poster Number: 246
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Knowledge Representation and Information Modeling, Artificial Intelligence, Machine Learning
Programmatic Theme: Clinical Research Informatics
We developed an imaging-augmented multimodal biomedical knowledge graph to integrate neuroimaging biomarkers with genetic and clinical data for Alzheimer’s disease research. Using ADNI data, we constructed patient-centered graphs linking imaging, gene, and cohort-derived features and trained graph convolutional networks for cognitive outcome prediction. Knowledge-graph augmentation improved performance over tabular and patient-only graph baselines, demonstrating that explicitly modeling imaging biomarkers within knowledge graphs enhances cognitive outcome modeling in AD.
Speaker(s):
Haobo Zhang, Master
University of Michigan
Author(s):
Jiankun Wang, MS - University of Michigan; Zhe Huang, PhD - Weill Cornell Medicine, Cornell University; Qiannan Zhang, Ph.D. - Weill Cornell Medicine, Cornell University; Chang Su, PhD - Weill Cornell Medicine; Fei Wang, PhD - Weill Cornell Medicine; Jiayu Zhou, Ph.D. - University of Michigan;
Haobo Zhang, Master - University of Michigan
Large Language Models Underperform Baselines in Low-Data ED Return Prediction
Poster Number: 247
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Machine Learning, Evaluation
Programmatic Theme: Clinical Informatics
Clinical prediction models face limited data in real-world healthcare settings. This study evaluates large language models (LLMs), pretrained on large-scale corpora, as an approach to addressing limited training data in structured clinical prediction tasks. Using emergency department return visits with admission (RVA) as a case study, we compare an LLM with gradient boosting models across varying training dataset sizes using MIMIC-IV-ED data. Results show that gradient boosting models consistently outperform LLMs across all data regimes.
Speaker(s):
Shining Gu, Bachelor
Vanderbilt University
Author(s):
Shining Gu, Bachelor - Vanderbilt University; Katherine Brown, PhD - Vanderbilt University Medical Center; Bradley Malin, PhD - Vanderbilt University Medical Center;
Shining Gu, Bachelor - Vanderbilt University
A Knowledge Graph Approach to LLM Topic Boundary Control in Medical Education: A Case Study
Poster Number: 248
2026 Symposium LIEAF Presentation
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Knowledge Representation and Information Modeling, Artificial Intelligence, Natural Language Processing, Information Retrieval
Programmatic Theme: Academic Informatics / LIEAF
One challenge for educational chatbots powered by large language models (LLMs) is maintaining topical focus aligned with the content being reviewed. We present a knowledge graph-based approach for scoping conversational boundaries in medical education. Our method maps student queries against a domain-specific knowledge graph to classify whether questions fall within acceptable topical boundaries. With 78% accuracy in query classification, our knowledge graph structure provides a generalizable, verifiable method for controlling LLM behavior beyond medical education.
Speaker(s):
Firdaus Indradhirmaya, PhD Student
University of Pittsburgh
Author(s):
Firdaus Indradhirmaya, PhD Student - University of Pittsburgh; Andi Saptono, PhD - University of Pittsburgh School of Health and Rehabilitation Services; I Made Agus Setiawan, PhD - University of Pittsburgh; Maria Yuliana, Bachelor - University of Pittsburgh; Bambang Parmanto, PhD - University of Pittsburgh; Yong Choi, PhD - University of Pittsburgh;
Firdaus Indradhirmaya, PhD Student - University of Pittsburgh
Beyond De-identification: A Multi-Agent Generator–Critic Pipeline for Generating Realistic Korean Clinical Surrogate Data with Iterative Expert Validation
Poster Number: 249
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Natural Language Processing, Privacy and Security, Data Sharing, Information Extraction, Artificial Intelligence
Programmatic Theme: Clinical Research Informatics
De-identification replaces protected health information (PHI) with placeholder tokens, preserving legal compliance but sacrificing the linguistic naturalness required for downstream clinical NLP tasks. We argue that realistic surrogate data—in which PHI is replaced by contextually plausible, type-faithful values—is a prerequisite for high-quality Korean clinical NLP corpora. We present a multi-agent pipeline that generates realistic Korean surrogates through a Generator–Critic negotiation loop with iterative expert-guided validation. Applied to a Korean translation of 2,434 MIMIC-II nursing notes, our system achieves a 99.87% PHI replacement rate, 99.95% medical term retention, 98.27% type fidelity, and a Privacy–Utility harmonic mean of 99.18%. The negotiation loop exhibits a 10.28% retry rate and 2.61% fallback rate, with surrogate diversity reaching Shannon entropy H = 2.443 bits. We further characterize six Korean-specific linguistic challenges that distinguish Korean de-identification from English counterparts, with implications for multilingual clinical NLP. By enabling privacy-safe sharing of high-fidelity clinical text, this approach provides the corpus infrastructure necessary for data-driven learning health systems to continuously improve NLP-guided clinical workflows. The language-agnostic evaluation framework and Generator–Critic negotiation protocol are designed for extensibility, positioning this work as a replicable methodology for multilingual clinical NLP corpora beyond Korean.
Speaker(s):
Tong Min Kim, Ph.D.
The Catholic University of Korea College of Medicine
Author(s):
Sang-Ho Lee, M.D., Ph.D. - Department of Nephrology, Kyung Hee University Hospital at Gangdong, Seoul, Republic of Korea; Taehoon Ko, Ph.D. - The Catholic University of Korea College of Medicine;
Tong Min Kim, Ph.D. - The Catholic University of Korea College of Medicine
Evaluation of a Large Language Model for Identification and Characterization of Intimate Partner Violence in Emergency Department Clinical Notes
Poster Number: 250
2026 Symposium 25x5 Presentation
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Clinical Decision Support, Population Health, Documentation Burden, Healthcare Quality
Programmatic Theme: Clinical Informatics
Intimate partner violence (IPV) is common yet under-recognized in emergency departments (EDs), where identification depends on vague documentation and underused ICD-10 codes. We retrospectively analyzed 300 ED notes from a single urban health system (2023) retrieved using the keyword “domestic violence.” Human reviewers established ground truth for IPV history and IPV as the primary reason for the visit. GPT-4o classified these outcomes and generated IPV subtypes, ICD-10 codes, and resource recommendations. Agreement between GPT and human reviewers was substantial for IPV history and IPV as the primary reason for visit (both κ = 0.63), with sensitivity/specificity of 78.4%/93.1% for history and 93.4%/87.5% for primary reason. Errors most often reflected documentation ambiguity. Exploratory outputs demonstrated good agreement with reviewers in subtype classification and ICD-10 coding, as well as appropriate but non-specific resource recommendations. These findings suggest LLM-assisted review of ED documentation may support IPV identification, coding, and referral workflows.
Speaker(s):
Katerina Andreadis, MS
NYU Grossman School of Medicine
Author(s):
Aditya Jain, BA - NYU Grossman School of Medicine; Katerina Andreadis, MS - NYU Grossman School of Medicine; JaeEun Kwon, Master of Public Policy - NYU Langone Health; Hao Zhang - New York University School of Medicine/NYU Langone Health System; Devin Mann, MD - NYU Grossman School of Medicine;
Katerina Andreadis, MS - NYU Grossman School of Medicine
Agentic Medical Memory Update for Medical Question-Answering Agents
Poster Number: 251
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Artificial Intelligence, Clinical Decision Support, Workflow
Programmatic Theme: Clinical Informatics
Recent advancements in large language models have enabled the deployment of clinical agents capable of sophisticated diagnosis and treatment planning. Despite their potential, existing medical QA systems typically depend on stationary knowledge repositories, rendering them unable to adapt to the constant stream of new clinical evidence. In this work, we present the Agentic Medical Memory Update framework, which explicitly reconciles foundational medical consensus with shifting empirical data. Our findings indicate that a dynamic memory-update mechanism is essential for maintaining the clinical integrity of agent-generated responses in an evolving information landscape.
Speaker(s):
Shuai Wang, Ph.D.
Yale University
Author(s):
Shuai Wang, Ph.D. - Yale University; Yize Zhao, PhD - Yale University; Qingyu Chen, PhD - Yale University;
Shuai Wang, Ph.D. - Yale University
Scalable Extraction of Social Determinants of Health from Clinical Notes in a Sepsis Cohort Using Instruction-Tuned Language Models
Poster Number: 252
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Natural Language Processing, Information Extraction, Data Mining
Programmatic Theme: Clinical Research Informatics
We evaluated instruction-tuned decoder-only language models for multi-class extraction of social determinants of health (SDOH) from clinical notes in a sepsis cohort. In the first stage, prompt refinement improved performance, with GPT-oss achieving a macro-F1 of 0.79 versus 0.57 for a fine-tuned Flan-T5 baseline. In the second stage, we evaluated LLM ensemble agreement, which improved robustness and enabled high-confidence label assignment, supporting scalable enrichment of electronic health records with structured social context.
Speaker(s):
Diego Salazar, PhD
NIH
Author(s):
Pankaj Dipankar, PhD - National Institute of Health; Daniel Smolyak, PhD - National Institutes of Health; Quynh Nguyen, PhD - National Institute of Health;
Diego Salazar, PhD - NIH
Domain-Adapted Small Language Models for Reliable Clinical Triage
Poster Number: 253
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Natural Language Processing, Clinical Decision Support
Programmatic Theme: Clinical Informatics
Accurate and consistent Emergency Severity Index (ESI) assignment remains a persistent challenge in emergency departments, where highly variable free-text triage documentation contributes to mistriage and workflow inefficiencies. This study evaluates whether open-source small language models (SLMs) can serve as reliable, privacy-preserving decision-support tools for clinical triage. We systematically compared multiple SLMs across diverse prompting pipelines and found that clinical vignettes, concise summaries of triage narratives, yielded the most accurate predictions. The SLM, Qwen2.5-7B, demonstrated the strongest balance of accuracy, stability, and computational efficiency. Through large-scale domain adaptation using expert-curated and silver-standard pediatric triage data, fine-tuned Qwen2.5-7B models substantially reduced discordance and clinically significant errors, outperforming all baseline SLMs and advanced proprietary large language models (LLMs, e.g., GPT-4o). These findings highlight the feasibility of institution-specific SLMs for reliable, privacy-preserving ESI decision support and underscore the importance of targeted fine-tuning over more complex inference strategies.
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
Comparison of General-Purpose and Specialized Commercial Large Language Models for Clinical Question Answering with Respect to Accuracy and Completeness
Click to View Handout
Poster Number: 254
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Artificial Intelligence, Real-World Evidence Generation
Programmatic Theme: Clinical Informatics
We evaluated three widely available large language models (LLMs) on 50 clinically relevant questions spanning 10 common question types encountered by practicing clinicians. To simulate routine use by clinicians, questions were posed without prompt optimization. OpenEvidence, a specialized clinical AI system, demonstrated 98% accuracy and 84% completeness. General-purpose models ChatGPT 5.3 and Claude Sonnet 4.6 achieved comparable accuracy (98% each) but differed substantially in completeness (38% vs. 78%, p=0.002). ChatGPT generated the question set but did not demonstrate superior performance answering those questions. Questions were generated using an established clinical question taxonomy to minimize the possibility that models were trained on these questions. Both general-purpose and specialized LLMs demonstrated high accuracy for common clinical questions, though response completeness varied significantly. These findings suggest LLMs can reliably answer clinical questions phrased in novel ways, but completeness depends on model architecture and design.
Speaker(s):
Vishalakshi Lakshmanan, MD
UT Houston
Author(s):
Elmer Bernstam, MD - UTHealth Science Center Houston; Kevin Hwang, MD - University of Texas Medical School at Houston; Gabriel Aisenberg, MD - The University of Texas Health Science Center at Houston;
Vishalakshi Lakshmanan, MD - UT Houston
Performance-Adaptive Instruction Selection (PAIS) for Clinical Reasoning in Large Language Models
Poster Number: 255
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Natural Language Processing, Information Retrieval, Artificial Intelligence
Programmatic Theme: Clinical Informatics
Objective: Instruction tuning is widely used to adapt large language models (LLMs) for biomedical and clinical tasks, but most training pipelines rely on fixed or randomly sampled instruction templates. This study evaluates whether performance-adaptive instruction selection during fine-tuning improves clinical reasoning performance.
Materials and Methods: We fine-tuned three LLaMA-3-8B models using different instruction selection strategies: static instruction tuning with a fixed template, random instruction sampling from an instruction pool, and a performance-adaptive strategy that prioritizes instructions based on observed training performance using bandit-inspired heuristics. Models were evaluated on a clinical reasoning benchmark with additional evaluations on Clinical Diagnosis and Medical Question Answering tasks. Statistical significance was assessed using bootstrap resampling.
Results: Adaptive instruction selection improved conclusion accuracy compared with static tuning (+6.8%, $p=0.001$) and random selection (+3.8%, $p=0.044$), with consistent gains in reasoning quality metrics and improved performance on auxiliary tasks.
Conclusion: Performance-adaptive instruction selection during fine-tuning improves clinical reasoning performance in instruction-tuned LLMs.
Speaker(s):
Henry Mukalazi Serugunda, Msc
Makerere university
Author(s):
Henry Mukalazi Serugunda, Msc - Makerere university; Ouyang Jianquan, PhD - Xiangtan University;
Henry Mukalazi Serugunda, Msc - Makerere university
Large Language Model–Based Long COVID Identification and Symptom Extraction System (LCI-SES) for Clinical Notes Analysis Using Multi-Center Real-World Data
Poster Number: 256
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Real-World Evidence Generation, Data Mining, Population Health
Programmatic Theme: Clinical Research Informatics
Post-acute sequelae of COVID-19 (PASC) is often underdiagnosed. We developed a large language model (LLM) framework to harness data from clinical notes and applied it to identify patients without documented PASC across two health systems. By integrating structured EHR data with LLM-based free-text analysis, we increased PASC cases by 48% beyond ICD coding alone and characterized longitudinal symptom burden. This study highlights scalable LLM-based phenotyping for diseases that are under-reported or lack reliable diagnostic codes.
Speaker(s):
James Rogers, PhD
Merck & Co., Inc.
Author(s):
Michelle Wang, PharmD, PhD - UCSF; Allison Nguyen, M.S. - Merck & Co., Inc., Rahway, NJ, USA; James Roger, Ph.D. - Merck & Co., Inc., Rahway, NJ, USA; Carisa De Anda, PharmD - Merck & Co., Inc., Rahway, NJ, USA; Brynna Izquierdo, MPH - Merck & Co., Inc., Rahway, NJ, USA; Zeina Siam, Dr - Merck; Kai Zheng, PhD - University of California, Irvine; Tianchu Lyu, PhD - University of California, Irvine; Xiaoqiang Xiang, Ph.D. - University of California San Francisco; Alpesh Amin, MD - University of California Irvine; Andenet Emiru, MBA - University of California; YI-SHAN LAN, Master - University of California, San Francisco; Yiliang Zhou, PhD - University of California, Irvine; Liang Zhao, PhD, MBA - University of California San Francisco;
James Rogers, PhD - Merck & Co., Inc.
Beyond Diagnosis: Unveiling Medication and Procedure Context in Clinical Foundation Model
Poster Number: 257
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Artificial Intelligence, Machine Learning
Programmatic Theme: Clinical Informatics
Large language models (LLMs) can model longitudinal patient trajectories by treating structured clinical records as token sequences, yet the effect of non-diagnosis context and the explainability of the model remained insufficiently investigated. We evaluate and explain the prediction of a pretrained clinical foundation model on de-identified longitudinal administrative claims from over 10 million individuals in the Merative MarketScan Research Databases (2008-2024). The context-aware clinical sequences improve trajectory prediction and yield clinically coherent embedding structure across disease, medication, and procedure domains. We further apply SHapley Additive exPlanations (SHAP) to explore cross-domain influence structure, revealing that diseases retain strong aligned organization, medications contribute more distributed effect, and procedures provide broader contextual information to future disease prediction. This work demonstrates that incorporating richer clinical context beyond diagnoses improves patient trajectory prediction while supporting explainability analysis of the representations learned by LLMs.
Speaker(s):
Jun Ni, B.S.
Yale University
Author(s):
Jun Ni, B.S. - Yale University; Fan Ma, Phd - yale; Yuntian Liu, M.P.H - Yale University; Xiang Lan; Hua Xu, Ph.D - Yale University;
Jun Ni, B.S. - Yale University
Generative AI–Enabled Quality-of-Life Element Identification in Rare Diseases
Poster Number: 258
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Natural Language Processing, Information Extraction, Artificial Intelligence
Programmatic Theme: Clinical Research Informatics
This study presents a promising approach using large language models to systematically identify essential Quality of Life–related elements from rare disease (RD) biomedical literature. We used the HRQoL standard from the World Health Organization Quality of Life includes 6 assessment domains with 29 distinct facets. This study aims to support healthcare experts in establishing a foundation to more effectively manage patients with RDs.
Speaker(s):
Sungrim Moon, PhD
National Institutes of Health
Author(s):
Sungrim Moon, PhD - National Institutes of Health; Patrick Gahan, student - National Institutes of Health; Qian Zhu - National Institutes of Health;
Sungrim Moon, PhD - National Institutes of Health
An Agentic AI Framework for Dynamic Patient Profiling and Laboratory Test Result Interpretation
Poster Number: 259
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Delivering Health Information and Knowledge to the Public, Clinical Decision Support
Programmatic Theme: Clinical Research Informatics
Existing patient portals often provide static, non-personalized information. We developed a multi-agent AI framework using LangGraph to coordinate adaptive symptom collection and personalized laboratory result interpretation. By adjusting reference ranges for patient-specific factors like age and sex, the system generates context-aware insights grounded in medical knowledge. Preliminary results demonstrate the framework's ability to produce clinically valid, personalized interpretations, offering a dynamic approach to improving health literacy.
Speaker(s):
Balu Bhasuran, Ph.D
Florida State University
Author(s):
Balu Bhasuran, Ph.D - Florida State University; Zhe He, PhD, FIAHSI, FAMIA - Florida State University; Karim Hanna, MD - University of South Florida Health; Carl Yang, Ph.D. - Emory University; Dhruv Kale, Masters - Florida State University; Xiao Luo, PhD - Southern Methodist University; Samuel Jacob Chacko, MS - Florida State University;
Balu Bhasuran, Ph.D - Florida State University
A Large Language Model-based Framework for Automated Triage and Clinical Decision Support in Classical Hematology
Poster Number: 260
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Clinical Decision Support, Informatics Implementation, Data transformation/ETL, Workflow, Evaluation
Programmatic Theme: Clinical Informatics
Abnormal blood counts are among the most common reasons for referral to classical hematology, where specialists triage patients to be followed up with classical hematology, rerouted to malignant hematology, or declined pending further evaluation. Rising referral volumes and limited specialist availability outside academic centers delay urgent care, yet no automated frameworks exist to support this process. We present a hybrid system integrating large language models with rule-based workflows to automate triage and generate structured notes for clinical decision support. Using erythrocytosis as a model use case, we demonstrate the system workflow, evaluate performance for binary disease classification, and assess concordance between automated and clinician triage decisions in a retrospective cohort at Mayo Clinic. The system achieved 100% sensitivity and 100% specificity for binary classification across both development (n=29) and validation (n=59) cohorts.
Speaker(s):
Naif Ganadily, Ph.D. in Biomedical Informatics and Data Science
Arizona State University
Author(s):
Naif Ganadily, Ph.D. in Biomedical Informatics and Data Science - Arizona State University; Naif Ganadily, MS in Electrical Engineering - Arizona State University; Mohammed Khan, MBBS - Mayo Clinic; Ayub Umair, PhD - Mayo Clinic; Irbaz Riaz, PhD - Mayo Clinic; Li Liu, MD - Arizona State University; Julia Yennie, Doctor of Physical Therapy - Mayo Clinic; Tanios Bekaii-Saab, MD - Mayo Clinic; Surbhi Shah, MBBS - Mayo Clinic;
Naif Ganadily, Ph.D. in Biomedical Informatics and Data Science - Arizona State University
Evaluating Prompting Strategies for Surgical History Extraction for Pediatric Cancer Survivorship Care Plans
Poster Number: 261
2026 Symposium 25x5 Presentation
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Information Extraction, Artificial Intelligence, Surgery, Documentation Burden
Programmatic Theme: Clinical Informatics
Survivorship Care Plans (SCPs) require accurate surgical history documentation, but manual extraction is time-consuming and error prone. We evaluated four large language model (LLM) prompting strategies (zero-shot, definitions, few-shot, and few-shot with definitions) for automated extraction of twenty COG-aligned surgical procedures from 96 pediatric oncology patients' EHR notes. Few-shot prompting with clinical definitions achieved the highest global accuracy (96.61%), demonstrating that contextual grounding meaningfully improves LLM-based clinical entity extraction.
Speaker(s):
Harinishree Sathu, BS
Georgia Institute of Technology
Author(s):
Wayne Liang, MD MS FAMIA - Children's Healthcare of Atlanta & Emory University; Andrew Hornback, MS - Georgia Institute of Technology; Naveen Muthu, MD - Children's Healthcare of Atlanta; Yifei Wang, MS - Georgia Institute of Technology; Rebecca Williamson Lewis, MPH - Aflac Cancer & Blood Disorders Center, Children’s Healthcare of Atlant; Benjamin Lovelace, BS - Children's Healthcare of Atlanta; Paulette Djachechi, MS - CHOA; May Dongmei Wang, PhD - Georgia Institute of Technology; Karen Effinger, MD, MS - Aflac Cancer & Blood Disorders Center, Children’s Healthcare of Atlanta; Department of Pediatrics, Emory University School of Medicine;
Harinishree Sathu, BS - Georgia Institute of Technology
VERI-DPO: Evidence-Aware Alignment for Clinical Summarization via Claim Verification and Direct Preference Optimization
Poster Number: 262
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Natural Language Processing, Machine Learning, Deep Learning, Information Extraction
Programmatic Theme: Clinical Informatics
Brief Hospital Course (BHC) narratives must be clinically useful yet faithful to fragmented EHR evidence. LLM-based clinical summarizers still introduce unsupported statements, and alignment can encourage omissions (“say-less” degeneration). We introduce VERI-DPO, which uses claim verification to mine preferences and distill them into the summarizer with DPO. On MIMIC-III-Ext-VeriFact-BHC (100 ICU patients; patient-level splits), we train a retrieval-augmented verifier to label claim–evidence pairs as Supported, Not Supported, or Not Addressed via a single-token format. The verifier scores sentence-level claims from sampled BHC candidates and aggregates margins into a coverage-aware utility to mine length-controlled, contradiction-anchored preference pairs. On held-out patients, verifier-mined preferences separate candidates by contradiction density, and VERI-DPO reduces Not Supported claim rates from 10.7% to 1.9% (local verifier judge) and from 11.6% to 6.4% (GPT-4o judge), while improving validity from 76.7% to 82.5% and maintaining informative length.
Speaker(s):
Weixin Liu, PhD Student
Vanderbilt University
Author(s):
Weixin Liu, PhD Student - Vanderbilt University; Congning Ni, Ph.D. - Vanderbilt University Medical Center; Qingyuan Song, Master of Engineering - Vanderbilt University; Susannah Rose, PhD - Vanderbilt University Medical Center; Christopher Symons, PhD - Lirio; Murat Kantarcioglu, Ph.D. - Virginia Tech; Bradley Malin, PhD - Vanderbilt University Medical Center; Zhijun Yin, Ph.D. - Vanderbilt University Medical Center;
Weixin Liu, PhD Student - Vanderbilt University
Using Multiple Large Language Models As-a-Judge to Construct AI Categorization Framework in the All of Us
Poster Number: 263
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Natural Language Processing, Informatics Implementation
Programmatic Theme: Clinical Research Informatics
Identifying the scientific landscape of national research repositories such as All of Us and UK Biobank is important for tracking scientific impact and portfolio alignment. We developed and evaluated a multi-LLM framework to classify research workspaces and publications into program priority areas. Using prompt refinement and weighted aggregation across three LLMs, accuracy reached 80%–90% on gold-standard datasets, enabling scalable portfolio assessment and identification of gaps, uncategorized research, and emerging themes.
Speaker(s):
Lina Sulieman, PhD
Vanderbilt University Medical Center
Author(s):
Lina Sulieman, PhD - Vanderbilt University Medical Center; Justin Cook, MS - Vanderbilt University Medical Center; Cathy Shyr, PhD - Vanderbilt University Medical Center; Brandy Mapes, MSLB - Vanderbilt University Medical Center; Kevin Levy, MBA - Vanderbilt University Medical Center; Evan Ochsenfaber, MA - All of Us Research Program; Geoffrey Ginsburg, MD, PhD; Joshua Denny, MD, MS - National Institutes of Health; Melissa Basford, MBA - Vanderbilt University Medical Center; Paul Harris, PhD - Vanderbilt University;
Lina Sulieman, PhD - Vanderbilt University Medical Center
Large Language Model–Based Symptom Severity Extraction from Clinical Note Summaries and Validation Using Nursing Outcomes Classification Scores in Acute Myeloid Leukemia
Poster Number: 264
2026 Symposium LIEAF Presentation
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Controlled Terminologies, Ontologies, and Vocabularies, Information Extraction, Information Retrieval, Natural Language Processing, Knowledge Representation & Information Modeling
Programmatic Theme: Academic Informatics / LIEAF
This study develops and evaluates a large language model (LLM)-based pipeline for extracting symptom severity from clinical note summaries in acute myeloid leukemia. Using GPT‑4.1 summaries aligned with Nursing Outcomes Classification (NOC) scores, we assessed documentation fidelity and severity alignment across five symptom domains. Summaries demonstrated high expert‑rated accuracy, and LLM‑derived severity ratings showed strong proximity to corresponding NOC scores. Findings support the use of summarization‑first LLM pipelines as a scalable approach to NOC-aligned symptom‑severity monitoring in oncology.
Speaker(s):
Sena Chae, PhD, RN
University of Iowa
Author(s):
Pratik Maitra, MS - Iowa State University Department of Computer Science; Alaa Albashayreh, PhD, MSHI, RN - University of Iowa; Priscilla Adu Mateki Annan, MPA, RN - University of Iowa College of Nursing; Karen Dunn Lopez, PhD, MPH, RN, FAAN - University of Iowa College of Nursing;
Sena Chae, PhD, RN - University of Iowa
Automated Quality Auditing of Radiation Oncology Clinical Documentation Using Large Language Models
Poster Number: 265
2026 Symposium 25x5 Presentation
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Healthcare Quality, Natural Language Processing
Programmatic Theme: Clinical Informatics
We developed and validated a large language model (LLM)-based auditing pipeline to identify missing required elements in radiation oncology. The system achieved high recall for deficient documentation and processed the dataset approximately 11 times faster than projected manual review. This approach may support scalable, human-in-the-loop documentation quality assurance.
Speaker(s):
Bingyang Ye, PhD
Brigham and Women's Hospital
Author(s):
Bingyang Ye, PhD - Brigham and Women's Hospital; Tara Kosak, Med - Mass General Brigham; Vikram Goddla; Arpi Murray, BS - Brigham and Women's Hospital; Meghan Kearney, MS - Mass General Brigham; Neil Martin, MD - Brigham and Women's Hospital; Danielle Bitterman, MD - Harvard Medical School;
Bingyang Ye, PhD - Brigham and Women's Hospital
AI-assisted Longitudinal Adverse Event Documentation Workflow in Clinical Trials
Poster Number: 266
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Natural Language Processing, Information Extraction, Patient Safety, Workflow
Programmatic Theme: Clinical Research Informatics
Adverse event (AE) documentation in clinical trials relies on labor-intensive manual review and CTCAE-based reporting. We developed an AI-assisted framework that integrates large language models with ontology-aligned normalization and longitudinal patient-level tracking across clinical notes and laboratory data. Evaluated on oncology clinical trial datasets, the system achieved 86.3% end-to-end AE recall and high agreement with clinical adjudication, demonstrating the potential of LLM-assisted workflows for scalable safety monitoring.
Speaker(s):
Yao Ge, Ph.D.
National Institutes of Health (NIH)
Author(s):
Yao Ge, Ph.D. - National Institutes of Health (NIH); Charalampos Floudas, MD, DMSc, MS - NIH/NCI; Zhiyong Lu, PhD - National Library of Medicine, NIH;
Yao Ge, Ph.D. - National Institutes of Health (NIH)
Similar Uptake, Different Intensity: AI Use Among US and UK Clinicians
Poster Number: 267
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Artificial Intelligence, Informatics Implementation, Human-computer Interaction
Programmatic Theme: Clinical Research Informatics
We surveyed 295 US and UK clinicians to assess workplace use of generative AI. Overall, 50% reported at least weekly use, with higher weekly use in the United States than the United Kingdom (60% vs 42%, p<0.01). Documentation and literature summarization were the most common applications. Over one-third accessed AI via personal tools. Proof of accuracy and safety was the leading condition for increased use, emphasizing the need for validation, governance, and organizational support.
Speaker(s):
Emre Sezgin, PhD
Nationwide Children's Hospital / The Ohio State University College of Medicine
Author(s):
Alysha Taxter, MD - Nationwide Children's Hospital; Jennifer Lee, MD - Nationwide Children's Hospital;
Emre Sezgin, PhD - Nationwide Children's Hospital / The Ohio State University College of Medicine
Harnessing Generative AI for Non-Pharmacological Interventions in Alzheimer’s Disease and Related Dementias: A Systematic Review of Non-Clinical Applications
Poster Number: 268
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Artificial Intelligence, Usability, User-centered Design Methods, Delivering Health Information and Knowledge to the Public, Health Equity
Programmatic Theme: Consumer Health Informatics
We synthesized empirical evidence on generative AI (GAI)-enabled non-pharmacological interventions for non-clinician audiences affected by Alzheimer’s disease and related dementias (ADRD). Searches of PubMed, PsycINFO, CINAHL, and Web of Science (2018–2025) identified 14 studies evaluating chatbots and multimodal tools for caregiver education, psychosocial support, reminiscence, and cognitive training. Studies were mostly short-term and exploratory; personalization and cultural tailoring improved engagement, but gaps remain in longitudinal outcomes, privacy governance, and equitable real-world adoption.
Speaker(s):
Yiwen Li, MPH
George Mason University
Author(s):
Y. Alicia Hong, PhD - George Mason University;
Yiwen Li, MPH - George Mason University
An LLM-Assisted Hypothesis Framework for Drug-Outcome Evidence: A Proof-of-Concept Using GLP-1 RA
Poster Number: 269
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Natural Language Processing, Information Extraction
Programmatic Theme: Clinical Research Informatics
Constructing drug-health outcome reference sets from literature is labor-intensive but essential for pharmacovigilance and bias assessment in observational studies. We used GPT-4o-mini to extract drug-HOI relations from PubMed abstracts and aggregated them into structured hypothesis objects capturing polarity, certainty, and evidence snippets. From 4,868 abstracts, we generated 1,376 normalized drug-HOI hypotheses, with 43.5% supported by RxList and 65.6% appearing in FAERS.
Speaker(s):
Varada Khanna, MS
Yale University
Author(s):
Vipina K. Keloth, PhD - Yale University; Yong Chen, PhD - University of Pennsylvania; Hua Xu, Ph.D - Yale University;
Varada Khanna, MS - Yale University
Large Language Model-Based Analysis of Patient Goals in Electronic Health Records for Chronic Pain Care
Poster Number: 270
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Chronic Care Management, Artificial Intelligence, Information Extraction
Programmatic Theme: Clinical Informatics
Electronic health record (EHR) data may help measure patient-centered care, but documentation of patient goals in primary care is not well understood. We analyzed 6,093 free-text goals for adults with chronic non-cancer pain using prompt-based ChatGPT classification with human validation. While 36.4% were pain-related and focused on function over pain reduction, only 13% were patient-centered. Additionally, 42% were specific and measurable, and 35% were time-bound, highlighting gaps in guideline-concordant documentation.
Speaker(s):
MaKenzie Gee, MS
Regenstrief Institute
Author(s):
Randall Grout, MD - Indiana University; Jennifer Hillstrom, N/A - Eskenazi Health; Lindsey Sanner, MPH - Indiana University Richard M. Fairbanks School of Public Health; Christopher Harle, PhD - Indiana University; Olena Mazurenko, MD, PhD - Indiana University Fairbanks School of Public Health;
MaKenzie Gee, MS - Regenstrief Institute
Leveraging Foundational Models for Automated Assessment of Injection Laryngoplasty Skills
Poster Number: 271
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Evaluation, Surgery
Programmatic Theme: Clinical Research Informatics
The objective of the study was to evaluate whether a commercially available multimodal foundational model could assess simulated percutaneous injection laryngoplasty skill. Thirty videos (novice, intermediate, expert) were graded using zero-shot and few-shot prompting. Zero-shot prompting showed no correlation with training level, whereas few-shot prompting demonstrated significant discrimination (ρ=0.66, p=0.0002) and improved accuracy. Despite variability, repeated evaluations showed ordinal consistency, suggesting calibrated multimodal models may enable scalable procedural skill assessment.
Speaker(s):
Abdulla Elkhadrawy, BS
Weill Cornell Medical College
Author(s):
Anaïs Rameau, MD - Weill Cornell Medical College; Alex Cheng, BA - Weill Cornell Medical College; Adrienne Biskaduros, BS - Weill Cornell Medical College; Alexandra Li, BS - Weill Cornell Medical College; Sean Setzen, MD - Weill Cornell Medical College; Julianna Kostas, MD - Weill Cornell Medical College;
Abdulla Elkhadrawy, BS - Weill Cornell Medical College
Reliability of Large Language Models in Answering Premarital Screening Questions
Poster Number: 272
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Patient Safety, Public Health, Delivering Health Information and Knowledge to the Public
Programmatic Theme: Consumer Health Informatics
Large language models (LLMs) are increasingly used by the public for health information, yet their reliability in screening contexts remains uncertain. We conducted a blinded expert evaluation of ChatGPT-4o and DeepSeek-R1 using 66 premarital screening questions covering genetic and infectious disease topics. Both models demonstrated high overall accuracy and relevance, but clinically important reasoning errors, particularly in genetic risk assessment, were identified, highlighting the need for expert oversight before patient-facing use.
Speaker(s):
Samar Binkheder, PhD
King Saud University
Author(s):
Samar Binkheder, PhD - King Saud University; Munerah Hamed, PhD - College of Medicine, Umm Al-Qura University, Makkah, Saudi Arabia; Suha Sulimani, MBBS, MS - Healthy Marriage Program, Population Health deputyship, Ministry of Health, Riyadh, Saudi Arabia; Samar Ekram, PhD - College of Medicine, Umm Al-Qura University, Makkah, Saudi Arabia; Radi Alsafi, PhD - Faculty of Applied Medical Sciences, Umm Al-Qura University, Makkah, Saudi Arabia;
Samar Binkheder, PhD - King Saud University
Evaluating Demographic Equity in Clinical-LLM Diagnosis Extraction from MIMIC-IV Discharge Summaries
Poster Number: 273
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Natural Language Processing, Information Extraction, Health Equity, Artificial Intelligence, Evaluation, Racial Disparities, Fairness and elimination of bias
Programmatic Theme: Clinical Informatics
We evaluated demographic equity in LLM-based diagnosis extraction using 2,000 MIMIC-IV discharge summaries and Asclepius-Llama3-8B. Model outputs were classified as Exact, Relevant, or Incorrect against AHRQ Clinical Classifications Software ground truth across primary and top-5 diagnosis tiers. Black patients had lowest lenient accuracy (primary: 20.1%); Asian patients had the highest (29.6%), a 9.5% gap (p=0.043). This replicable, clinically grounded framework identifies race and age as priority dimensions for equity evaluation before clinical LLM deployment.
Speaker(s):
Shraddha Gupta, MS in Health Informatics
The Resource Group, Spend Management Solutions
Author(s):
Humayera Islam, PhD - The University of Chicago; Xing Song, PhD - University of Missouri;
Shraddha Gupta, MS in Health Informatics - The Resource Group, Spend Management Solutions
Leveraging Large Language Model to Identify Uncoded Disease in Electronic Health Record Data: A Case Study of Smoldering Multiple Myeloma
Poster Number: 274
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Natural Language Processing, Information Extraction, Information Retrieval
Programmatic Theme: Clinical Informatics
Identifying smoldering multiple myeloma (SMM), a premalignant condition for myeloma, is challenging and requires time-consuming manual chart review. We developed a large language model-based pipeline using Llama-3-8B to identify SMM diagnoses from unstructured Electronic Health Records. Our pipeline achieved a 92.1% F1-score, matched manual review diagnosis dates <30 days for 65.2% of patients, and significantly reduced processing time compared to manual chart review and rule/machine learning-based methods, enabling efficient, large-scale population-based studies for myeloma prevention.
Speaker(s):
Mei Wang, MS
Washington University in St. Louis
Author(s):
Yuan-Hung Kuan, BS - Washington Universitty in St. Louis; Martin Schoen, MD, MPH - Saint Louis University School of Medicine; Theodore Thomas, MD - Washington University School of Medicine; Jr-Shin Li, PhD - Washington University in St. Louis; Su-Hsin Chang, PhD - Washington University School of Medicine;
Mei Wang, MS - Washington University in St. Louis
CancerFAQ: A Multi-Agent RAG Architecture for Pediatric Oncology Question Answering
Poster Number: 275
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Artificial Intelligence, Informatics Implementation
Programmatic Theme: Clinical Informatics
Parents of children with cancer face significant distress following diagnosis. Current methods of communication include distribution of a large parent handbook, leading to possible information overload. We developed CancerFAQ, a multi-agent RAG chatbot grounded on a pediatric oncology parent handbook. In systematic testing, CancerFAQ correctly retrieved source pages for benchmark questions and produced the highest response quality score compared to baselines. These results support agentic RAG as a promising approach for institution-specific clinical communication support.
Speaker(s):
William Powell, MS
Washington University in St. Louis
Author(s):
William Powell, MS - Washington University in St. Louis; Ethan Hillis, MS - Institute for Informatics at Washington University School of Medicine in St. Louis; Philip Payne, PhD, FACMI, FAMIA - WashU Medicine and BJC Healthcare; Albert Lai, PhD, FACMI, FAMIA - Washington University; Bryan Sisk, MD, MSCI - WashU Medicine Physicians; Aditi Gupta - Washington University in St. Louis;
William Powell, MS - Washington University in St. Louis
LLM-Assisted Pre-Annotation for Clinical Named Entity Recognition: A Semi-Automated Pipeline with Dual Clinician Review
Poster Number: 276
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Natural Language Processing, Artificial Intelligence, Machine Learning, Information Extraction
Programmatic Theme: Clinical Informatics
Manual annotation of clinical notes for named entity recognition (NER) is resource-intensive and requires clinical expertise. This preliminary study used an LLM-assisted approach using de-identified MIMIC-IV notes targeting 25 entity types. A subset of 10 LLM-annotated notes were reviewed in Prodigy by two physicians, showing high interrater agreement (Cohen’s Kappa 0.691) and an estimated precision of 90.8%. Further research is underway to refine this process, to measure recall, and to create a gold annotation corpus.
Speaker(s):
Manhitha Mamillapalli, BS
University of Illinois
Author(s):
Manhitha Mamillapalli, BS - University of Illinois; Senthil Nachimuthu, MD, PhD, FAMIA - University of Utah School of Medicine;
Manhitha Mamillapalli, BS - University of Illinois
Evaluating Large Language model against Structured EHR Diagnoses: Agreement Patterns and Discrepancies
Poster Number: 277
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Natural Language Processing, Information Extraction, Artificial Intelligence, Clinical Decision Support, Machine Learning, Deep Learning, Data Mining
Programmatic Theme: Clinical Research Informatics
This study evaluates the GPT-OSS model’s ability to validate clinician-assigned ICD-10-CM codes against clinical notes. Comparing two prompting strategies, we found that granular, per-code evaluation achieved higher concordance than bulk prompting. Performance varied by ICD chapter: accuracy reached 100% for well-defined diagnoses (e.g., COVID-19) but dropped significantly for administrative (Z) or rare (Q) codes. We conclude that LLMs effectively validate narrative-supported diagnoses while highlighting auxiliary codes requiring human review.
Speaker(s):
MOHAN KASHYAP PARGI, Data Scientist/Masters in Machine Learning and Computing
CHILDRENS HOSPITAL OF PHILADELPHiA
Author(s):
Hanieh Razzaghi, PhD - Children's Hospital of Philadelphia; Levon Utidjian, MD, MBI - Children's Hospital of Philadelphia; Nhat Nguyen, PhD - CHILDRENS HOSPITAL OF PHILADELPHIA; Charles Bailey, MD, PhD - Children's Hospital of Philadelphia;
MOHAN KASHYAP PARGI, Data Scientist/Masters in Machine Learning and Computing - CHILDRENS HOSPITAL OF PHILADELPHiA
ICD-10 Codes Extraction Using Large Language Models With Synthetic Data Generation
Poster Number: 278
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Controlled Terminologies, Ontologies, and Vocabularies, Information Extraction
Programmatic Theme: Clinical Research Informatics
Automated ICD-10 coding from clinical notes is challenged by severe label imbalance and sparse training data for rare diagnoses. This study investigates synthetic clinical text generation using a biomedical large language model to augment training data for ICD prediction. Synthetic notes are generated using prompt constraints and ICD-derived semantic anchors to preserve clinical meaning. Experiments on the MIMIC-IV dataset suggest balanced mixtures of synthetic and real notes improve rare-code recall while maintaining overall model performance.
Speaker(s):
Quinn Chan, Computer science/ nursing
Northeastern University
Author(s):
Ritika Kumar, Master's - Northeastern University; Deahan Yu, PhD - Northeastern;
Quinn Chan, Computer science/ nursing - Northeastern University
Ontology-Guided Generation and Reliability Evaluation of Synthetic Lung Cancer Staging Notes
Poster Number: 279
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Patient-/Person-Generated Health Data, Artificial Intelligence
Programmatic Theme: Clinical Research Informatics
The generation of highly structured, realistic synthetic clinical data is critical for advancing cancer care while preserving patient privacy. We present an ontology-guided framework evaluating synthetic lung cancer metastasis notes using LLMs. Our pipeline integrates AJCC-aligned JSON schemas, structural validation, longitudinal checks, and SNOMED CT grounding. GPT-4o achieved up to 100% JSON validity, while ClinicalCamel-70B produced 0% valid JSON, underscoring major reliability barriers for open-source models in automated oncology data generation.
Speaker(s):
Laxmigayathri Challa, PhD
University of North Texas
Author(s):
Laxmigayathri Challa, PhD - University of North Texas;
Laxmigayathri Challa, PhD - University of North Texas
Machine Learning to Predict Undetected Cardiometabolic Conditions in US Adults using the National Health and Nutrition Examination Survey (2015-2018, 2021-2023)
Poster Number: 280
2026 Symposium Health Equity Presentation
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Machine Learning, Clinical Decision Support, Health Equity, Informatics Implementation, Chronic Care Management, Healthcare Quality
Working Group: Health and Healthcare Equity Working Group
Programmatic Theme: Public Health Informatics
We developed and temporally validated machine learning models to predict cardiometabolic conditions (CMCs) among US adults by using NHANES data (2015–2018, 2021–2023). Five algorithms were trained using 33 features. Among 4,790 adults, 17.6% had CMCs. CatBoost was the best model with sensitivity of 67.2% and negative predictive value (NPV) of 91.4%. Findings demonstrate that CatBoost can identify undetected CMCs using only routinely collected sociodemographic, behavioral, and anthropometric data, without requiring additional clinical testing.
Speaker(s):
Jingxin Yao, MS
University of Illinois Urbana-Champaign
Author(s):
Xuan Lin, MS - University of Illinois Urbana-Champaign; Minakshi Raj, PhD - University of Illinois at Urbana-Champaign; Bruno Pereira Nunes, PhD - University of Illinois Urbana-Champaign;
Jingxin Yao, MS - University of Illinois Urbana-Champaign
Integrating machine learning and linear mixed models to characterize gut microbial determinants of insulin resistance
Poster Number: 281
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Machine Learning, Quantitative Methods, Public Health, Causal Inference, Chronic Care Management, Causal Inference, Data transformation/ETL, Artificial Intelligence
Programmatic Theme: Translational Bioinformatics
We analyzed longitudinal 16S rRNA gut microbiome data with insulin resistance from the Integrative Human Microbiome Project (iHMP). Using a combination of LASSO feature selection, ensemble machine-learning, and longitudinal mixed modeling anchored on a physiologic insulin-resistance phenotype, we identified a microbiome signature of insulin resistance enriched for specific butyrate-producing and bile-acid-associated taxa. These findings reinforce the link between gut microbial ecology and host metabolic regulation and nominate specific taxa for functional follow-up and microbiome-targeted interventions.
Speaker(s):
Kesheng Wang, PhD
University of South Carolina
Author(s):
Kesheng Wang, PhD - University of South Carolina;
Kesheng Wang, PhD - University of South Carolina
Pharmacogenomics-Machine Learning Integration in Perioperative Care: An Unexamined Intersection
Poster Number: 282
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Machine Learning, Clinical Decision Support, Surgery, Interoperability and Health Information Exchange, Health Equity
Programmatic Theme: Translational Bioinformatics
Response to anesthetic, analgesic, and antiemetic medications varies widely across surgical patients, yet perioperative predictive tools remain largely population-based rather than biologically individualized. Machine learning (ML) and pharmacogenomics (PGx) have each advanced perioperative outcome prediction from distinct directions, but their integration remains limited. This structured literature review evaluated the extent of PGx-ML integration in perioperative research. Five targeted searches across Embase and PubMed retrieved 829 records; 28 met the inclusion criteria. Studies were categorized as perioperative ML without PGx (n=20), perioperative PGx without ML (n=5), and PGx-ML integration outside acute perioperative trajectories (n=3). No study integrated PGx with ML to model dynamic postoperative recovery. Despite sophisticated ML models and established genetic determinants of drug response, biologic variables remain absent from perioperative predictive frameworks. Integrating genomic profiles with longitudinal perioperative time-series defines a clear informatics frontier for precision perioperative decision support.
Speaker(s):
Renee Motew, MSN,CRNA
University of Florida
Author(s):
Renee Motew, MSN,CRNA - University of Florida; Lisiane Pruinelli, PhD, MS, RN, FAMIA, FAAN - University of Florida;
Renee Motew, MSN,CRNA - University of Florida
Topological Representations of Intracranial EEG Show Limited Cross-Center Generalization for Epilepsy Surgery Outcome Prediction
Poster Number: 283
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Machine Learning, Artificial Intelligence, Evaluation
Programmatic Theme: Clinical Research Informatics
Predicting seizure outcomes after epilepsy surgery remains challenging despite extensive presurgical evaluation with intracranial electroencephalography (iEEG). We evaluated whether dynamic topological features derived from iEEG using zigzag persistent homology (ZZPH) support prediction of 1-year seizure outcomes. Recordings from 102 patients across seven institutions were analyzed to construct time-resolved higher-order topological summaries of seizure dynamics. A Random Forest classifier was trained on a single-institution cohort and evaluated using internal cross-validation and external validation on independent datasets from other centers. Within the development cohort, models achieved moderate predictive performance (AUROC ≈ 0.62). However, performance declined substantially when the trained model was applied to external datasets (average AUROC ≈ 0.42). Similar variability was observed in leave-one-site-out analyses. These results indicate that predictive structure identified within individual iEEG datasets may not translate reliably across institutions, highlighting the importance of rigorous external validation when developing predictive models from complex neurophysiology data.
Speaker(s):
Katrina Prantzalos, MS
Case Western Reserve University
Author(s):
Katrina Prantzalos, MS - Case Western Reserve University; Pedram Golnari, MD - Case Western Reserve University; Dipak Upadhyaya, PhD - Case Western Reserve University; Saeideh Salehizadeh, MD - University Hospital Cleveland Medical Center; Akshaya Sivaji, MD - University Hospital Cleveland Medical Center; Geronimo Pacheco Aispuro, MD - University Hospital Cleveland Medical Center; Mohamed Ahmed Abdelsabour Hasan, MD - University Hospital Cleveland Medical Center; Sophia Rauda, N/A - Case Western Reserve University; Leonora Lipson, N/A - Case Western Reserve University; Suraj Thyagaraj, PhD - Case Western Reserve University; Guadalupe Fernandez-BacaVaca, MD - University Hospital Cleveland Medical Center; Jeffery Albert, PhD - Case Western Reserve University; Nick Gurski, PhD - Case Western Reserve University; Satya Sahoo, PhD, FAMIA - Case Western Reserve University;
Katrina Prantzalos, MS - Case Western Reserve University
Cross-Attention Enables Context-Aware Multimodal Skin Lesion Diagnosis
Poster Number: 285
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Machine Learning, Deep Learning, Data Mining
Programmatic Theme: Clinical Research Informatics
Clinical diagnosis of skin lesions integrates visual dermoscopic features with patient context such as age, skin type, and lesion characteristics. However, most artificial intelligence systems for dermoscopic analysis rely solely on image data and ignore structured clinical metadata. We developed a multimodal deep learning framework that combines dermoscopic images with patient metadata and evaluated whether cross-attention mechanisms better capture contextual interactions than conventional fusion strategies. Using 1,568 lesions from the PAD-UFES-20 dataset (69\% malignant) with associated metadata (age, sex, Fitzpatrick skin type, anatomical site, and lesion diameter), we compared four models: metadata-only logistic regression, image-only ResNet18, late fusion via feature concatenation, and cross-attention-based fusion. The image-only model achieved strong discrimination (AUC 0.9776), while late fusion slightly reduced performance (AUC 0.9717). The proposed cross-attention model achieved the best overall results (AUC 0.9818, AUPRC 0.9924) with improved calibration (ECE 0.0379). These findings suggest that attention-based multimodal learning enables more effective integration of patient context for automated skin lesion diagnosis.
Speaker(s):
Krishna Mridha, Phd Student
Case Western Reserve University
Author(s):
Krishna Mridha, Phd Student - Case Western Reserve University; Humayera Islam, PhD - The University of Chicago;
Krishna Mridha, Phd Student - Case Western Reserve University
Federated Systemic Atherosclerosis Prediction via Cross-disease Knowledge Distillation and Learnable Prototype-based Reasoning
Poster Number: 286
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Machine Learning, Deep Learning, Clinical Decision Support, Artificial Intelligence, Privacy and Security, Fairness and elimination of bias, Real-World Evidence Generation, Data Sharing
Programmatic Theme: Clinical Research Informatics
Despite sharing systemic atherosclerosis as a common etiology, integrated risk prediction for coronary artery disease (CAD) and peripheral arterial occlusive disease (PAOD) remains infeasible due to cross-departmental data silos and heterogeneous labels. In disjoint-label settings, standard federated averaging causes model collapse from incompatible gradients. To address this challenge, we propose a federated learning framework using cross-disease knowledge distillation and learnable prototypes, for cross-organ disease risk prediction without sharing raw data. From 5,181 patients across two clients, lightweight prototypes were extracted using 43 common clinical variables with confidence-based weighted server aggregation. In a validation cohort (n = 122), zero-shot cross-prediction area under the receiver operating characteristic curves (AUROC) reached 0.6523 (CAD-to-PAOD) and 0.6529 (PAOD-to-CAD), achieving 92.1% and 86.7% of direct prediction performance (0.7083 and 0.7529), with per-patient interpretability via gradient-based SHapley Additive exPlanations (SHAP). This framework demonstrates potential as a digital biomarker for integrated vascular risk management without additional examinations.
Speaker(s):
Jaemin Oh, Master student
Sejong University
Author(s):
Jaemin Oh, Master student - Sejong University; Seongwook Lee, M.S. Student - Sejong University; Minwoo Lee, M.D., Ph.D - Department of Neurology, Hallym University Sacred Heart Hospital, Anyang, South Korea; Hyunjun Moon, Ph.D - Department of Computer Engineering, Sejong University, Seoul, South Korea; Wonjong Noh, Ph.D - School of Software, Hallym University, Chuncheon, South Korea;
Jaemin Oh, Master student - Sejong University
Machine Learning-Based Risk Stratification for Left Ventricular Ejection Fraction Prediction: A Comparative Analysis in the Million Veteran Program
Poster Number: 288
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Machine Learning, Data Mining, Artificial Intelligence, Healthcare Quality, Clinical Decision Support, Natural Language Processing, Deep Learning
Programmatic Theme: Clinical Research Informatics
Left ventricular ejection fraction (LVEF) serves as both a diagnostic and prognostic indicator for heart failure and cardiovascular (CV) complications, yet optimal risk stratification using electronic health record data remains unclear. Fine-grained classification suffers from poor intermediate-category performance and reduced interpretability. We compared machine-learning approaches using CV and lifestyle features. Binary stratification achieved the best distinction (AUC: 0.747), with age-interaction features contributing 54.9% of predictive value, highlighting a need for age-stratified risk assessment.
Speaker(s):
Eungyoung Han, Biomedical Informatics/PhD
The Department of Veteran Affairs
Author(s):
Eungyoung Han, Biomedical Informatics/PhD - The Department of Veteran Affairs; Philip Tsao, PhD - Stanford University / VA Palo Alto Health Care System; Themistocles Assimes, MD, PhD - Stanford University / VA Palo Alto Health Care System; Shoa Clarke, MD, PhD - Stanford University; Austin Hilliard, PhD - VA Palo Alto Health Care System; Rodrigo Guarischi-Sousa, PhD - Stanford University; Jiayan Zhou, PhD - Stanford University;
Eungyoung Han, Biomedical Informatics/PhD - The Department of Veteran Affairs
Ensemble Modeling for Predicting Hypertension-Related Comorbidities Using EHR Data
Poster Number: 289
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Machine Learning, Clinical Decision Support, Population Health, Data Mining, Artificial Intelligence
Programmatic Theme: Clinical Research Informatics
This research utilized an EHR dataset to predict eight common hypertension comorbidities, including myocardial infarction, stroke, heart failure, and kidney failure. Five machine learning algorithms (Naive Bayes, Random Forest, XGBoost, CatBoost, AdaBoost) were evaluated. Ensemble models achieved the highest AUROC for heart failure (0.91) and heart attack (0.89), with the lowest for headache (0.82). These models demonstrate the potential of scalable ML approaches for predicting comorbidity risk in hypertensive patients.
Speaker(s):
Anjali Rajesh, Student
Rutgers University
Author(s):
Rahul Mittal, MD - Rutgers University; Srinivas Denduluri, PhD - Rutgers University; Michael Schelewski, B.S. - Rutgers University; Martin Pavlovski, PhD - Samsung Electronics America; Branimir Ljubic, MD, PhD - Rutgers University, NJ, USA;
Anjali Rajesh, Student - Rutgers University
Dynamic Prediction of AET Adherence to Inform Intervention Development in Breast Cancer Care
Poster Number: 290
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Machine Learning, Clinical Decision Support, Chronic Care Management, Population Health, Healthcare Quality, Real-World Evidence Generation, Geospatial (GIS) Data/Analysis
Programmatic Theme: Clinical Research Informatics
Adherence to adjuvant endocrine therapy (AET) is critical for improving breast cancer outcomes but remains suboptimal during long-term treatment. This study models adherence as a dynamic longitudinal process using linked electronic health record, pharmacy claims, and neighborhood socioeconomic data for 537 patients receiving tamoxifen or anastrozole. Recurrent neural network models were trained to predict future adherence risk. SHAP analysis identified prior adherence patterns as the strongest predictors, with demographic and clinical factors contributing smaller effects.
Speaker(s):
Yun Jiang, PhD, MS, RN, FAMIA
University of Michigan
Author(s):
Yun Jiang, PhD, MS, RN, FAMIA - University of Michigan; Xiayuan Huang, PhD. - Yale University;
Yun Jiang, PhD, MS, RN, FAMIA - University of Michigan
Learning-Based Prediction Models for Major Adverse Cardiovascular Events in Women Diagnosed with Breast Cancer: A Scoping Review
Poster Number: 291
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Machine Learning, Chronic Care Management, Quantitative Methods
Programmatic Theme: Clinical Research Informatics
Major adverse cardiovascular events are long-term and late effects experienced among women with breast cancer. We conducted a scoping review of learning-based models and identified substantial heterogeneity in algorithms, feature selection, and performance metrics across 10 eligible studies. Logistic regression and gradient boosting were predominant models, with discrimination metrics most reported. No studies incorporated genomic data or external validation. Additional data modalities and external validation are needed to support clinical translation.
Speaker(s):
Reindolf Oppong, BSN
University of Pittsburgh
Author(s):
Theresa Koleck, PhD, BSN, RN - University of Pittsburgh; Dennis Miezah, BSN, RN - University of Massachusetts, Boston;
Reindolf Oppong, BSN - University of Pittsburgh
Integrating Socioeconomic Factors to Improve Performance and Model Fairness in Machine Learning-based Diabetes Risk Prediction
Poster Number: 292
2026 Symposium Health Equity Presentation
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Machine Learning, Population Health, Health Equity
Programmatic Theme: Public Health Informatics
Type 2 diabetes prevention hinges on identifying risk early, yet many prediction tools overlook social context and equity. Using EHR data from OCHIN’s nationwide community health center network, we compared traditional and deep learning models for 5-year incident diabetes prediction and assessed fairness across racial groups. CatBoost performed best, and adding zip code–level community “vital signs” and SDoH Z codes improved both performance and fairness, highlighting the value of socioeconomically informed, equitable risk prediction.
Speaker(s):
Feifan Liu, PhD
University of Massachusetts Chan Medical School
Author(s):
ChandraHarsha Rachabathuni, MSc - Umass Chan Medical School; Feifan Liu, PhD - University of Massachusetts Chan Medical School; Ben Gerber, MD, MPH - University of Massachusetts Chan Medical School; Huanmei Wu, FAMIA, PhD - Temple University; Omar Martinez, JD, MPH, MS - University of Central Florida; Alana Biggers, MD, MPH, FACP - UIUC;
Feifan Liu, PhD - University of Massachusetts Chan Medical School
Multimodal Early Warning Prediction of Optic Pathway Glioma in NF1 Utilizing Large Language Models (LLMs)
Poster Number: 293
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Machine Learning, Large Language Models (LLMs), Information Extraction, Artificial Intelligence, Natural Language Processing
Programmatic Theme: Clinical Research Informatics
We developed multimodal machine learning models for 1-day, 1-month, and 6-month early prediction of optic pathway gliomas, tumors that can cause vision loss in children with NF1. Using structured (demographics, diagnoses) and unstructured (notes) data, we compared two fusion approaches: 1) concatenating LLM-extracted phenotypes with structured data for prediction, and 2) training BERT models to predict directly from multimodal texts. Both methods were successful, but using LLM-extracted features performed best, likely due to lower dimensionality.
Speaker(s):
Levi Kaster, BS
Washington University in St. Louis
Author(s):
Levi Kaster, BS - Washington University in St. Louis; Ethan Hillis, MS - Institute for Informatics at Washington University School of Medicine in St. Louis; Inez Oh, PhD - Institute for Informatics at Washington University in St. Louis, School of Medicine; Stephanie Morris, MD - Kennedy Krieger Institute; David Gutmann, MD, PhD - Washington University School of Medicine; Philip Payne, PhD, FACMI, FAMIA - WashU Medicine and BJC Healthcare; Aditi Gupta - Washington University in St. Louis;
Levi Kaster, BS - Washington University in St. Louis
How Reliably Can Wearable EEG Detect Seizures?
Poster Number: 294
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Machine Learning, Clinical Decision Support, Patient Safety
Programmatic Theme: Clinical Informatics
Wearable behind-the-ear EEG enables discreet ambulatory seizure monitoring, yet no published benchmark evaluates detection models with leave-one-subject-out cross-validation (LOSO-CV) on the SeizeIT2 dataset. We benchmarked six Machine Learning (ML) classifiers across 49 patients (367 seizures, 4,455 hours) using event-level metrics. Logistic Regression achieved the highest AUC (0.81); LightGBM offered the best clinical trade-off (74.5% sensitivity, 4.9 false alarms/hour), a rate suitable for ambulatory use.
Speaker(s):
Suman Kumar, MS
Cleveland State University
Author(s):
Debapriya Hazra, Ph.D. - Montclair State University; Melvin Speisman, MD - American College of AI and Medicine;
Suman Kumar, MS - Cleveland State University
Design of an Adaptive, Equity-Focused Digital System (PACD) for Improving Medication Adherence in Cardiovascular Disease: Evidence Synthesis and Proposed Framework
Poster Number: 295
2026 Symposium Health Equity Presentation
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Mobile Health, Clinical Decision Support, Health Equity, Artificial Intelligence
Programmatic Theme: Clinical Informatics
Non-adherence to cardiovascular medications kills more than 125,000 Americans annually and costs the healthcare system upward of $300 billion. A decade of mobile health investment has not significantly changed this, and the question is why. We argue the problem is a category error: existing tools are built to predict who will not adhere, when the clinical task is actually a sequential decision about what to do next, given unreliable feedback about what just happened. This paper reports Phase 0 of the Patient-Aligned Contextual Decision-making (PACD) framework. This review of the literature found three recurring structural failures: unreliable adherence signals, non-adaptive static interventions, and workflow disconnects. Each failure maps to a specific design response: an uncertainty-weighted reward function, a contextual bandit that updates continuously, and a co-design-derived feasibility mask enforced at runtime. We propose an adaptive, equity-focused Digital System (PACD) for improving medication adherence in cardiovascular disease.
Speaker(s):
Abhinav Bhat, Bachelor's
Honeywell
Author(s):
Yuri Quintana, PhD - Beth Israel Deaconess Medical Center;
Abhinav Bhat, Bachelor's - Honeywell
Computational phenotyping of long-distance runners from wearable device data and phenome-wide health associations in the All Of Us Research Program
Poster Number: 296
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Mobile Health, Population Health, Quantitative Methods, Infectious Diseases & Epidemiology
Programmatic Theme: Translational Bioinformatics
Using Fitbit data from 31,054 All of Us participants, we developed a multi-feature algorithm, incorporating step count, duration, and cadence, to identify long-distance runners. Among marathon-level runners, cumulative running days were associated with lower odds of obesity, type 2 diabetes, sleep disorders, and anxiety. Among ultra long-distance runners, uterine hypertrophy and oropharyngeal malignancy reached Bonferroni significance, though estimates were imprecise. These findings demonstrate the feasibility of deriving behavioral phenotypes from consumer wearables at scale.
Speaker(s):
Slavina Goleva, Ph.D.
National Institutes of Health
Author(s):
Evelynne Fulda, MSc - NIH/University of Oxford; Slavina Goleva, Ph.D. - National Institutes of Health; Bennett Waxse, MD, PhD - NIAID/NHGRI; Tracey Ferrara, PhD - National Human Genome Research Institute; Aiden Doherty, PhD - University of Oxford; Anya Topiwala, BM BCh MRCPsych DPhil FHEA - University of Oxford; Joshua Denny, MD, MS - National Institutes of Health;
Slavina Goleva, Ph.D. - National Institutes of Health
A Transparent Framework for Evaluating mHealth Dietary Apps Using Free-Text Meal Descriptions
Poster Number: 297
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Mobile Health, Evaluation, Patient-/Person-Generated Health Data
Programmatic Theme: Consumer Health Informatics
We present a transparent framework for evaluating mHealth dietary apps using free-text meal descriptions. The framework operationalizes unstructured text into structured meal balance scores, integrates randomized feature evaluation, and applies trajectory modeling to capture heterogeneous user responses. Applied to a 16-week randomized field experiment, the framework produced results consistent with dietitian assessments and demonstrated scalable, interpretable evaluation of user-generated dietary data. LLM-based labeling showed high concordance with rule-based classification as secondary validation.
Speaker(s):
Yi-Chin Lin (Kato-Lin), PhD
Hofstra University
Author(s):
Cheng Ma, MS - Carnegie Mellon University; Rema Padman, PhD - Carnegie Mellon University;
Yi-Chin Lin (Kato-Lin), PhD - Hofstra University
Real-Time NLP Routing for Patient Portal Messages Reduces Response Time, Resolution Time, and Staff Touches: Prospective Evaluation in Outpatient EHR Workflows
Poster Number: 298
2026 Symposium 25x5 Presentation
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Natural Language Processing, Workflow, Evaluation, Artificial Intelligence, Informatics Implementation, Documentation Burden, Administrative Systems
Working Group: Clinical Information Systems Working Group
Programmatic Theme: Clinical Informatics
We prospectively evaluated an EHR-integrated natural language processing (NLP) classifier that labels and routes patient portal messages into operational destinations. In 14 days across four outpatient sites, 469 routed conversations were compared with 402 unrouted controls. Routing reduced time to first staff interaction (median difference −1.0 hour), time to conversation completion (−22.5 hours), and total staff interactions (−2.0 touches). Model performance was high (overall accuracy 97.8%).
Speaker(s):
Blake Anderson, MD
Emory
Author(s):
Muhammad Zia ul Haq, MBBS, MPH - Switchboard, MD; Yuanda Zhu; Andrew Hornback, MS - Georgia Institute of Technology; Arash Harzand, MD;
Blake Anderson, MD - Emory
Depression Detection at the Point of Care: Automated Analysis of Linguistic Signals from Routine Primary Care Encounters
Poster Number: 299
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Natural Language Processing, Large Language Models (LLMs), Patient Engagement and Preferences, Artificial Intelligence, Clinical Decision Support
Programmatic Theme: Clinical Informatics
Depression is underdiagnosed in primary care, yet timely identification remains critical. Recorded clinical encounters, increasingly common with digital scribing technologies, present an opportunity to detect depression from naturalistic dialogue. We investigated automated depression detection from 1,108 audio-recorded primary care encounters in the Establishing Focus study, with depression defined by PHQ-9 (n=253 depressed, n=855 non-depressed). We compared three supervised approaches, Sentence-BERT + Logistic Regression (LR), LIWC+LR and ModernBERT, against a zero-shot GPT-OSS. GPT-OSS achieved the strongest performance (AUPRC=0.510, AUROC=0.774), with LIWC+LR competitive among supervised models (AUPRC=0.500, AUROC=0.742). Combined dyadic transcripts outperformed single-speaker configurations, with providers linguistically mirroring patients in depression encounters, an additive signal not captured by either speaker alone. Meaningful detection is achievable from the first 128 patient tokens (AUPRC=0.356, AUROC=0.675), supporting in-the-moment clinical decision support. These findings argue for passively collected clinical audio as a low-burden complement to existing screening workflows.
Speaker(s):
Feng Chen, MS
University of Washington
Author(s):
Feng Chen, MS - University of Washington; Manas Satish Bedmutha, BS; Janice Sabin - University of Washington; Andrea Hartzler, PhD FACMI - University of Washington; Nadir Weibel, PhD - UC San Diego; Trevor Cohen, MBChB, PhD - Biomedical Informatics and Medical Education, University of Washington;
Feng Chen, MS - University of Washington
Automated Extraction of Cancer Registry Data from Pathology Reports: Comparing LLM-Based and Ontology-Driven NLP Platforms
Poster Number: 300
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Natural Language Processing, Large Language Models (LLMs), Information Extraction, Data Standards
Programmatic Theme: Clinical Research Informatics
Cancer data standardization requires converting unstructured pathology reports into structured registry variables, a mostly manual and resource-intensive task. We evaluated two automated extraction platforms: Brim Analytics, an LLM-based system that guides and orchestrates abstraction, and DeepPhe, an ontology-driven system. Using 330 pancreatic adenocarcinoma and 34 breast cancer pathology reports from Johns Hopkins Hospital, we assessed both under deployment-realistic conditions. Brim Analytics achieved high accuracy across seven registry variables in pancreatic cancer (mean 96.7%), including T stage (96.4%) and histologic grade (97.0%), with a 3.0 p.p. decline on breast cancer (mean 93.7%). DeepPhe performed comparably for N stage (96.4% pancreatic, 94.1% breast) but had notable T stage deficits (83.6% pancreatic, 70.6% breast). Per-report processing times averaged 0.9 s (Brim, pancreatic), 4.6 s (Brim, breast), 1.1 s (DeepPhe, pancreatic), and 3.5 s (DeepPhe, breast). These results indicate that LLM-based extraction can achieve high accuracy across cancer types and support automated data workflows.
Speaker(s):
Thomas McPhaul, MS
Johns Hopkins University School of Medicine
Author(s):
Thomas McPhaul, MS - Johns Hopkins University School of Medicine; Kory Kreimeyer, MS - Johns Hopkins University School of Medicine; Alexander Baras, MD PhD - Johns Hopkins University School of Medicine; Taxiarchis Botsis, MSc, MPS, PhD - Johns Hopkins University School of Medicine;
Thomas McPhaul, MS - Johns Hopkins University School of Medicine
Multi-Class Identification of Justice Involvement in Clinical Notes Across Two Pediatric Institutions
Poster Number: 301
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Natural Language Processing, Information Extraction, Population Health, Large Language Models (LLMs), Artificial Intelligence, Deep Learning, Health Equity, Data Modernization
Programmatic Theme: Clinical Research Informatics
Youth exposure to family justice involvement is linked to adverse health outcomes but is not routinely identified in pediatric settings. We fine-tuned open-weight LLMs (Gemma-2, MedGemma) to classify 33 types of justice involvement in clinical notes across two pediatric institutions (~20,000 notes). Combined training achieved 0.85–0.90 AUPRC with minimal cross-site degradation, and self-training corrected annotation noise in rare classes. This tool can enable large-scale research on health impacts of justice involvement in youth.
Speaker(s):
Syed-Amad Hussain, BSE
The Ohio State University
Author(s):
Syed-Amad Hussain, BSE - The Ohio State University; Eric Fosler-Lussier, PhD - The Ohio State University; Samantha Boch, PhD, RN - Nationwide Children's Hospital;
Syed-Amad Hussain, BSE - The Ohio State University
Development of a Living Repository for Cataloging Publicly Available Clinical Narrative Datasets: A Community Effort by the AMIA NLP Working Group
Poster Number: 302
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Natural Language Processing, Data Sharing, Data Standards
Working Group: Natural Language Processing Working Group
Programmatic Theme: Clinical Research Informatics
Clinical NLP datasets are dispersed across various platforms and lack consistent documentation, which hampers their discoverability, accessibility, and reproducibility. To address this gap, we developed a FAIR-aligned metadata framework with 33 standardized variables across six categories to curate 60 publicly available corpora and datasets. Our analysis reveals variation in task coverage, data modalities, annotation practices, and access methods. The curated repository, hosted on NLP-WG GitHub, supports NLP/LLM benchmarking, education, and translational clinical NLP research.
Speaker(s):
Sunyang Fu, PhD, MHI
UTHealth Houston
Author(s):
Rongrong Wang, Master of Science in Health Informatics - UTHealth Houstom; Mohammad Arvan, PhD - University of Illinois at Chicago; Sarvesh Soni, PhD - National Library of Medicine (NLM); Chenyu Li, M.S. - University of Pittsburgh; Kevin Li, PhD Candidate - University of Washington; Sy Hwang, MS - University of Washington; Samah Fodeh-Jarad, PHD - Yale University; Sujani Kakumanu, MD - University of Wisconsin Hospital and Clinic and William S. Middleton Veterans Hospital; Feifan Liu, PhD - University of Massachusetts Chan Medical School; Swati Rajwal, MS - UMass Chan Medical School; Jakir Hossain Bhuiyan Masud, PhD - University of Alabama at Birmingham; Sonish Sivarajkumar, PhD - Eli Lilly; Anup Kumar Mishra, Ph.D., M.S. - Mayo Clinic; Xinsong Du, Ph.D. - Brigham and Women's Hospital/Harvard Medical School; Jiyeong Kim, PhD - Stanford University; Thomas Kinsley, MD - University of California, Los Angeles; Yanshan Wang, PhD - University of Pittsburgh; Yonghui Wu, PhD - University of Florida; Rui Zhang, PhD, FACMI, FAMIA, FIAHSI - University of Minnesota, Twin Cities; Sunyang Fu, PhD, MHI - UTHealth Houston;
Sunyang Fu, PhD, MHI - UTHealth Houston
MultiClinicalBERT: A Multilingual Transformer Pretrained on Real-World Clinical Notes
Poster Number: 303
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Natural Language Processing, Large Language Models (LLMs), Machine Learning
Programmatic Theme: Clinical Research Informatics
Multilingual clinical natural language processing remains limited by the lack of pretrained models trained on real-world clinical notes. We introduce MultiClinicalBERT, a multilingual transformer pretrained on a large corpus of clinical notes, biomedical literature, and general-domain text. The model is evaluated on eleven clinical NLP tasks across five languages and consistently outperforms mBERT and several language-specific baselines, demonstrating strong cross-lingual generalization and improved performance in both low-resource and high-resource clinical settings.
Speaker(s):
Enqi Liu, Master of Science
Harvard University
Author(s):
Enqi Liu, Master of Science - Harvard University; Yutong Yang, Master of Science Student - Harvard University; Jiageng Wu, Doctor of Philosophy - Brigham and Women’s Hospital; Richard Wyss, Doctor of Philosophy - Brigham and Women’s Hospital; Rishi J Desai, doctor of philosophy - Brigham and Women's Hospital; Kueiyu (Joshua) Lin, Doctor of philosophy - Brigham and Women's Hospital; Jie Yang, Doctor of philosophy - Brigham and Women's Hospital;
Enqi Liu, Master of Science - Harvard University
How Do We Talk To Doctors? An Evaluation of Clinician Interactions Through Lens of Narrative
Poster Number: 304
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Natural Language Processing, Patient-/Person-Generated Health Data, Patient Engagement and Preferences
Programmatic Theme: Consumer Health Informatics
In formal practice of medicine, the role of the clinician has never been so straightforward as simply identifying illness and prescribing treatment--patient-clinician interactions shape the process of practical medicine. Data on these interpersonal aspects of clinical care are therefore of great interest to medical informatics. This work examines the IlNeS corpus, a collection of autobiographical narratives from individuals with acute or chronic illness as told to distinct imagined audiences. The corpus functions as a lens surfacing how patients change the presentation of their stories of illness based on expected audience; in this way, the work evaluates illness narrative as designed for clinicians and identifies areas in which patient language may reveal friction in patient-clinician communication. While patient communication styles are highly individual, participants on average show honest and informative when speaking to a clinical professional. Exceptions surfaced by this work may hint at rifts in patient-provider communication necessitating further research.
Speaker(s):
Luke Breitfeller, PhD
Carnegie Mellon University
Author(s):
Luke Breitfeller, PhD - Carnegie Mellon University; Carolyn Rose, PhD - Carnegie Mellon University;
Luke Breitfeller, PhD - Carnegie Mellon University
Optimizing a Hybrid Algorithm for UMLS Concept Prediction
Poster Number: 305
2026 Symposium LIEAF Presentation
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Natural Language Processing, Deep Learning, Machine Learning
Programmatic Theme: Academic Informatics / LIEAF
We optimized a hybrid algorithm for UMLS concept prediction that combines deep learning (DL) and heuristic (HE) models to support editorial review of new terms in the UMLS 2025AB release (222,926 terms). Confidence scores were derived using empirical binning calibration to resolve DL–HE disagreements. Among three strategies, an HE step–specific calibration achieved the best performance, yielding 94.50 precision, 92.51 recall, and 93.49 F1, improving recall while maintaining high editorial accuracy and efficiency.
Speaker(s):
Chris Lu, Ph.D.
NIH/NLM/LHNCBC/ACIB/BCC
Author(s):
Chris J. Lu, Ph.D. - NIH/NLM/LHNCBC/ACIB/BCC; Randolph Miller, MD - NIH/NLM/LHNCBC/ACIB; Kin Wah Fung, MD, MS, MA - NIH/NLM/LHNCBC/;
Chris Lu, Ph.D. - NIH/NLM/LHNCBC/ACIB/BCC
MedGenie: Privacy-Preserving Clinical Text Annotation Platform with LLM Integration
Poster Number: 306
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Natural Language Processing, Large Language Models (LLMs), Artificial Intelligence
Programmatic Theme: Clinical Research Informatics
Clinical text annotation is a fundamental component of clinical natural language processing (NLP) research and has become increasingly critical in the era of generative artificial intelligence (GenAI), where large volumes of machine-generated clinical text require systematic evaluation for safety, accuracy, and clinical validity. Numerous annotation platforms exist, including open-source tools such as brat and CLAMP, as well as commercial solutions such as Label Studio and Generative AI Lab. However, existing open-source tools are often not trivial for non-NLP domain experts, particularly clinicians, and frequently lack integrated pre-annotation capabilities powered by modern language models. In contrast, commercial platforms commonly incorporate advanced large language models (LLMs) but raise substantial concerns regarding data governance, privacy, and regulatory compliance when clinical data are processed through cloud-based APIs due to protected health information (PHI). Currently, no open-source annotation platform simultaneously provides clinician-friendly usability and privacy-preserving integration with powerful local LLMs. To address this gap, we present MedGenie, a modular clinical text annotation platform that combines an intuitive user interface with a novel local LLM integration strategy enabling secure, privacy-preserving, and scalable LLM-assisted pre-annotation.
Speaker(s):
Yanshan Wang, PhD
University of Pittsburgh
Author(s):
Xiangyu Zeng, MS - University of Pittsburgh; Yanshan Wang, PhD - University of Pittsburgh;
Yanshan Wang, PhD - University of Pittsburgh
Note-to-Structured EHR Concordance: Pipeline and Drug Normalization
Poster Number: 307
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Natural Language Processing, Evaluation, Delivering Health Information and Knowledge to the Public, Clinical Decision Support, Artificial Intelligence
Programmatic Theme: Clinical Research Informatics
Incorporating clinical notes alongside structured EHR is important for biomedical machine learning, yet the degree of note-to-structured concordance remains poorly characterized. We developed a five-step, visit-level pipeline to compare note-derived entities with OMOP-structured data across 19,173 clinic visits at VUMC. Deterministic drug normalization (Path A) improved concordance (+1.56 pp containment, +2.35 pp overlap among 9,164 visits with ≥1 drug). The pipeline supports interpretable validation of note-derived entities and lexicon refinement.
Speaker(s):
Mingyang Jiang, Bachelor
Vanderbilt University
Author(s):
Congning Ni, Ph.D. - Vanderbilt University Medical Center; Zhijun Yin, Ph.D. - Vanderbilt University Medical Center;
Mingyang Jiang, Bachelor - Vanderbilt University
Design and Real-World Implementation of an AI-Driven Clinical Summarization Application
Poster Number: 308
2026 Symposium 25x5 Presentation
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Natural Language Processing, Large Language Models (LLMs), Documentation Burden, Information Extraction, Workflow, Human-computer Interaction, Informatics Implementation, Interoperability and Health Information Exchange
Programmatic Theme: Clinical Informatics
This study describes the design and real-world implementation of a SMART on FHIR-embedded, AI-driven clinical summarization tool that transforms unstructured EHR documentation into structured, source-traceable summaries. Evaluated by a sample of Canadian clinicians, the system demonstrated strong AI performance and high clinician-rated usability and accuracy. The tool reduced cognitive burden and improved efficiency, particularly in complex cases with high document volume.
Speaker(s):
CHRISTOPHER PETTENGELL, BMBCh
Pentavere Research Group Inc
Author(s):
Ruth Moulson, MPH - HEALWELL AI; Alexander Sachs, Masters of Mathematics in Computer Science, Artificial Intelligence - Healwell; Diana Shakhova, MaSc, System Design Engineering - Healwell; Anton Trublin, BEng, Computer Engineering - Healwell; Daniel D’Arolfi, Master of Applied Science, Electrical and Computer Engineering - Healwell; Elizaveta Garbuzova, MS - Healwell; Aaron Leibtag, B.Com. - Healwell; CHRISTOPHER PETTENGELL, BMBCh - Pentavere Research Group Inc; Steve Aviv, Bachelor of Business Science, Economics, Statistics, Actuarial Science - Healwell;
CHRISTOPHER PETTENGELL, BMBCh - Pentavere Research Group Inc
WiseNER: Meta-reasoning for Large Language Models in Biomedical Named Entity Recognition
Poster Number: 309
2026 Symposium LIEAF Presentation
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Natural Language Processing, Large Language Models (LLMs), Information Extraction
Programmatic Theme: Academic Informatics / LIEAF
Biomedical Named Entity Recognition (BiomedNER) is significant to understanding unstructured biomedical text. High-performing BiomedNER systems is constrained by the availability of high-quality labeled data. For out-of-domain entity types without sufficient labeled data, Large Language Models (LLMs) with zero-shot or one-shot prompting does not work well. Thus, in this work, we proposed WiseNER, a novel two-stage framework that addresses the zero-shot or one-shot generalization challenge in BiomedNER through Meta-Reasoning Distillation and Reinforcement Learning.
Speaker(s):
Wendong Ge, PhD
Merck
Author(s):
Hang Zhang, MS - University of Pittsburgh; Wendong Ge, PhD - Merck; Ruheng Wang, MS - UT Southwestern Medical Center, Dallas, TX, USA; Gregory Bryman, PhD - Merck; Antong Chen, PhD - Merck;
Wendong Ge, PhD - Merck
LLMs for Automated Subjective Cognitive Decline Detection in High-Risk Cardiovascular Populations
Poster Number: 310
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Natural Language Processing, Large Language Models (LLMs), Clinical Decision Support, Information Extraction, Evaluation, Real-World Evidence Generation, Artificial Intelligence, Informatics Implementation
Working Group: Natural Language Processing Working Group
Programmatic Theme: Clinical Informatics
Atrial fibrillation (AF) doubles dementia risk, yet subjective cognitive decline (SCD) mentions in clinical notes often go unrecognized. We developed a large language model (LLM)–based NLP pipeline to automatically extract SCD from the electronic health records of 2,547 AF patients. From 220 SCD-enriched notes segmented into 940 double-annotated chunks - labeled across nine cognitive categories with expert adjudication and high inter-annotator agreement (97.1%; Gwet's AC1 0.77–1.00)- we built an adjudicated benchmark. Using GPT-4o-mini, we tested three prompting approaches (zero-shot, few-shot, and few-shot with challenging cases) on 885 chunks. Few-shot prompting achieved the best performance (F1 0.918; Cohen's κ 0.879; perfect recall); adding challenging cases did not improve results. LLMs can reliably extract SCD against a rigorously adjudicated reference standard.
Speaker(s):
Afra Shamnath, Master of Public Health
NYU
Author(s):
Meghan Reading Turchioe, PhD, MPH, RN - Columbia University School of Nursing; Liqin Wang, PhD - Brigham and Women's Hospital; Marykay Pavol, PhD - Columbia University; David Slotwiner, MD - NYP; Angelo Biviano, MD - Columbia University;
Afra Shamnath, Master of Public Health - NYU
Leveraging Automated Radiology Report Analysis to Identify Incidental Pulmonary Findings and Follow-Up
Poster Number: 311
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Natural Language Processing, Large Language Models (LLMs), Workflow, Information Extraction, Chronic Care Management
Programmatic Theme: Clinical Research Informatics
Incidental pulmonary findings documented in radiology reports may warrant imaging follow-up to evaluate potential malignancy, yet identifying these findings and tracking follow-up at scale is challenging because the information is embedded in free text. We applied automated NLP pipelines to 39,473 radiology reports from 4,361 lung cancer patients to detect incidental findings and follow-up imaging. The system identified 550 patients with incidentalomas, demonstrating the feasibility of large-scale cohort construction from clinical radiology reports.
Speaker(s):
Zhaoyi Sun, Master of Science
University of Washington
Author(s):
Zhaoyi Sun, Master of Science - University of Washington; Farzad Ahmed, Ph.D. Student in Computer Science - George Mason University; Ozlem Uzuner, PhD - George Mason University; Martin Gunn, MBChB - University of Washington; Meliha Yetisgen, PhD - University of Washington;
Zhaoyi Sun, Master of Science - University of Washington
“Trying to Play Doctor Myself”: Clinical Note Use Among Healthy Adults
Poster Number: 312
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Patient Engagement and Preferences, Qualitative Methods, Evaluation, Delivering Health Information and Knowledge to the Public, Human-computer Interaction
Programmatic Theme: Consumer Health Informatics
Under the 21st Century Cures Act, U.S. healthcare organizations must provide patients with electronic access to clinical notes. Yet research often measures access using portal registration or log-ins, offering limited insight into how patients use patient-facing documentation. Semi-structured interviews with 25 adults without chronic conditions identified two patterns of note use: routine review and selective consultation. Note use intensified during diagnostic uncertainty, suggesting access metrics alone do not explain variation in how patients use clinical notes.
Speaker(s):
Roxanne Velazquez, MLIS
UW-Madison
Author(s):
Zachary Schultz, B.A. - University of Wisconsin-Madison; Catherine Arnott Smith, PhD - University of Wisconsin;
Roxanne Velazquez, MLIS - UW-Madison
Ambulatory patients’ trust in artificial intelligence tools and acceptability with ambient artificial intelligence scribes: A mixed-methods secondary analysis
Poster Number: 313
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Patient Engagement and Preferences, Artificial Intelligence, Informatics Implementation
Programmatic Theme: Clinical Informatics
Although ambient AI scribes have generally led to favorable outcomes, it remains unclear what the correlates are of patients’ acceptability of these tools. We conducted a secondary mixed-methods analysis of surveys and patient experience interviews (n=20) focusing on trust in AI. Trust in AI was highly variable in this sample. Higher acceptability was discussed in the context of low concerns with ethicality, supportive affective attitudes, and high perceived effectiveness. Intervention coherence had mixed impacts.
Speaker(s):
Oliver Nguyen, MSHI
University of Wisconsin at Madison
Author(s):
Majid Afshar, MD, MSCR - University of Wisconsin - Madison; Michael Jaeb, PhD, RN - University of Wisconsin at Madison; Mary Ryan Baumann, PhD - University of Wisconsin at Madison; Felice Resnik, PhD - University of Wisconsin at Madison; Anne Gravel Sullivan, PhD - University of Wisconsin at Madison; Graham Wills, PhD - UW Health; Jason Dambach, MD - UW Health; Leigh Ann Mrotek, PhD - University of Wisconsin at Madison; Mariah Quinn, MD, MPH - University of Wisconsin at Madison; Kirsten Abramson, MD - University of Wisconsin at Madison; Peter Kleinschmidt, MD - UW Health; Thomas Brazelton, MD, MPH - University of Wisconsin at Madison; Margaret Leaf, MS - UW Health; Heidi Twedt, MD - University of Wisconsin School of Medicine and Public Health; Brian Patterson, MD MPH - University of Wisconsin-Madison; Frank Liao, PhD - University of Wisconsin, Madison - UW Health; Stacy Rasmussen, BS - University of Wisconsin at Madison; Joel Gordon, MD - UW Health, University of Wisconsin School of Medicine and Public Health;
Oliver Nguyen, MSHI - University of Wisconsin at Madison
Caregiver Perspectives on Sharing Social Risk Information in Dementia Care for EHR Documentation
Poster Number: 314
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Patient Engagement and Preferences, Qualitative Methods, Workflow, Informatics Implementation
Programmatic Theme: Clinical Informatics
Dementia patients often face social challenges, yet social information in the electronic health record remains fragmented and inconsistently documented. As cognitive decline progresses, caregivers become the primary source of social information. This study examined caregiver perspectives on sharing social risk information with clinicians. All caregivers reported comfort sharing this information, but viewed screening as meaningful only when responses led to action. Participants preferred brief questionnaires completed privately through a portal, tablet, or paper before visits.
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; Nicole Hammer, B.A. - University of Florida - Department of Health Outcomes and Biomedical Informatics; Magda Knight, MS - University of Florida; Jennifer LeLaurin, PhD - University of Florida; Ramzi Salloum, PhD; Serena Jingchuan Guo, MD, PhD - Purdue University; Megan Gregory, Ph.D. - University of Florida;
Philipp Haessner, BS, MBA, MS - University of Florida
A Randomized Study of Mobile Phone Interface Complexity and Variations in In-Hospital Safety Reporting from Caregivers: Simpler is not always better
Poster Number: 315
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Patient Safety, Mobile Health, Patient Engagement and Preferences, Health Equity, Diversity, Equity, Inclusion, and Accessibility
Programmatic Theme: Consumer Health Informatics
Our objective was to compare a simplified mobile safety reporting tool to an existing version with detailed safety categories. We conducted a randomized controlled trial assessing the difference in safety observation reporting rates between the two user interfaces. The detailed version generated higher reporting rates than the simplified version (11.2 vs 7.0 reports/100 patient-days, p<0.001). This suggests that detailed interfaces providing safety scaffolding can meaningfully support family engagement in safety efforts.
Speaker(s):
Naomi Bardach, MD, MAS
UCSF
Author(s):
Danea Horn, PhD - UCSF; Anjana Sharma, MD - UCSF; Nancy Netherland, BA - California Children's Trust; Deborah Franzon, MD - UCSF; Helen Pu, MD - UCSF; Kiana Washington, BA - UCSF; Leslie Magana, BA - UCSF; Urmimala Sarkar, MD MPH - UCSF; Glenn Rosenbluth, MD - University of California San Francisco; John Boscardin, PhD - UCSF; Alicia Fernandez, MD - UCSF; Naomi Bardach, MD, MAS - UCSF;
Naomi Bardach, MD, MAS - UCSF
Understanding Clinicians’ Perceptions of the Usefulness of PGHD for Age-Friendly Health Care
Poster Number: 316
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Patient-/Person-Generated Health Data, User-centered Design Methods, Human-computer Interaction, Informatics Implementation, Surveys and Needs Analysis
Programmatic Theme: Clinical Informatics
Thoughtful and contextually appropriate implementation of patient generated health data (PGHD) in the electronic health record has the potential to facilitate improved Age-Friendly Health System care for geriatrics patients centered around the 4Ms (mobility, mentation, medication, and “what matters most”). This mixed-methods study evaluates clinicians’ views of the usefulness of 35 diverse PGHD elements at the point of care and can be used to prioritize elements for future implementation efforts.
Speaker(s):
Elise Garton, MSc
University of Utah
Author(s):
L Weaver, MLIS, PhD - University of Utah; Elise Garton, MSc - University of Utah; Christian Balbin, PhD - University of Utah; Rachel Dalrymple, MBA - University of Utah; Timothy W. Farrell, MD - University of Utah; Megan Puckett, MD - University of Utah; Kensaku Kawamoto, MD, PhD, MHS - University of Utah; Jorie Butler, PhD - University of Utah;
Elise Garton, MSc - University of Utah
Investigating empathy-control in LLM-based conversational health agents
Poster Number: 317
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Patient-/Person-Generated Health Data, Large Language Models (LLMs), Human-computer Interaction, Evaluation, Quantitative Methods, Curriculum Development
Programmatic Theme: Clinical Research Informatics
Generating synthetic clinical conversations requires modeling not just content, but affective behaviors such as empathy. We explore using large language models (LLMs) to simulate patient-provider conversations and dynamically control provider empathy. Using two clinical cases and three LLMs for patient and provider roles, we generated 216 augmented conversations across start, middle, and end segments. Empathy was evaluated by three LLMs using the EPITOME framework. Evaluators showed moderate–high agreement across the corpus (Krippendorff’s α = 0.912; pairwise Cohen’s κ ≈ 0.46-0.54), with consistent agreement across EPITOME dimensions of emotional reaction, interpretation, and exploration. Segment-level analysis showed that models reliably tracked controlled decreases in empathy, whereas attempts to increase empathy relative to baseline yielded smaller, non-significant shifts. These findings suggest that LLM-generated clinical conversations can support controlled manipulation and evaluation of empathy, highlighting their potential for adaptive training and scalable generation of synthetic clinical dialogues.
Speaker(s):
Manas Satish Bedmutha, BS
Author(s):
Manas Satish Bedmutha, BS; Kayla Hom, BS - UC San Diego; Feng Chen, MS - University of Washington; Victoria Chentsova, MS - UC San Diego Health; Nadir Weibel, PhD - UC San Diego;
Manas Satish Bedmutha, BS -
Decoding the Patient Voice: An Annotation Framework for Social Determinants of Health in Patient Portal Messaging
Click to View Handout
Poster Number: 318
2026 Symposium Health Equity Presentation
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Patient-/Person-Generated Health Data, Health Equity, Information Retrieval, Data Mining, Racial Disparities
Programmatic Theme: Consumer Health Informatics
This study analyzed 20,003 patient portal message sentences to identify social determinants of health (SDoH) using a structured annotation framework. After multi-round blinded review and consensus, 1,228 sentences were identified to contain SDoH data. Most common categories were healthcare access, neighborhood environment, and social context. Findings show that patient messages hold substantial SDoH information, supporting future development of automated natural language processing tools to extract this information and reduce manual review burden.
Speaker(s):
Mohamad Khair Abou Chaar, M.D.
Mayo Clinic
Author(s):
Nan Wang, Graduate Student - UTH; Sunyang Fu, PhD, MHI - UTHealth Houston; Hongfang Liu, PhD - University of Texas Health Science Center at Houston; Cornelius A Thiels, M.D., M.B.A. - Mayo Clinic; Ming Huang, PhD - UTHealth Houston;
Mohamad Khair Abou Chaar, M.D. - Mayo Clinic
A Computational Audit of the Global Health Data Governance Research Landscape: Bridging Policy and Implementation in the AI Era
Poster Number: 319
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Policy, Data Standards, Privacy and Security, Public Health
Programmatic Theme: Public Health Informatics
The rapid integration of AI into health systems has outpaced the development of unified global data governance. We present a large-scale computational audit of the research landscape using a scoping review and Latent Dirichlet Allocation (LDA) topic modeling. By analyzing 3,456 peer-reviewed publications, we identified twelve dominant thematic clusters that define the policy era, ranging from data privacy and ethics to AI implementation and Learning Health System (LHS) infrastructure. Our findings reveal a significant paradigm shift from theoretical privacy concerns toward the pragmatic challenges of systemic implementation. This work provides an empirical foundation for informatics leaders to harmonize global governance strategies with the adversarial and technical requirements of the AI-enabled healthcare landscape.
Speaker(s):
L. Raymond Guo, PhD.
West Virginia University
Author(s):
Shuimei Liu, S.J.D. - China University of Political Science and Law;
L. Raymond Guo, PhD. - West Virginia University
Do Social Vulnerability Risk Scores for a Geographic Region Influence Hospital Outcomes?
Poster Number: 320
2026 Symposium Health Equity Presentation
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Population Health, Health Equity, Healthcare Quality, Public Health, Diversity, Equity, Inclusion, and Accessibility
Programmatic Theme: Public Health Informatics
Social determinants of health play a critical role in shaping patient outcomes. Indices such as the Social Vulnerability Index (SVI) and the Vizient Vulnerability Index (VVI) provide ways to capture community-level vulnerability, yet their comparative utility in clinical populations remains insufficiently explored. We compare the associations of SVI and VVI with mortality and hospital readmission outcomes using electronic health record (EHR) data. We analyzed EHR data from 186,571 patients in a rural integrated health system. SVI and VVI were linked to patient records based on ZIP code and categorized into quartiles. Logistic regression models were used to examine associations with in-hospital, 30-day, and 90-day mortality, as well as 14-day and 30-day unplanned hospital readmissions, adjusting for demographic and clinical covariates. Both indices were associated with readmission outcomes, while mortality associations were weaker and non-linear. VVI showed stronger differentiation than SVI in the highest vulnerability quartile for readmissions.
Speaker(s):
Michelle Dempsey, Master of Science in Data Analytics
Geisinger
Author(s):
Biplab Bhattacharya, PhD - Geisinger Health System; David Vawdrey, PhD - Geisinger; Sonia Jahangiri, PhD Candidate - Rochester Institute of Technology;
Michelle Dempsey, Master of Science in Data Analytics - Geisinger
Assessment of a Routine Pediatric Screening Tool to Predict Autism Diagnosis
Poster Number: 321
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Population Health, Machine Learning, Real-World Evidence Generation, Quantitative Methods, Clinical Decision Support
Programmatic Theme: Clinical Research Informatics
We evaluated whether routinely collected Survey of Well-Being of Young Children (SWYC) developmental milestones scores can identify children likely to later receive an autism spectrum disorder diagnosis. Using 987 matched pairs from Duke University Health System, we found significant score differences beginning at 9 months. Predictive performance became meaningful by 12 months, with cumulative item-level models performing best and reaching an AUC of 0.847 at 18 months.
Speaker(s):
Gefei Li, MS
Duke University
Author(s):
Gefei Li, MS - Duke University; Elizabeth Erickson, MD - Duke University; Lauren Franz, MBChB, MPH - Duke University; Benjamin Goldstein, PhD - Duke University;
Gefei Li, MS - Duke University
Using Electronic Health Record Data to Measure Community Displacement
Poster Number: 322
2026 Symposium Health Equity Presentation
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Population Health, Health Equity, Geospatial (GIS) Data/Analysis, Public Health
Programmatic Theme: Public Health Informatics
The Indy Health “District,” a historically disinvested community in Indianapolis, Indiana, is the focus of an ongoing community development plan. Since neighborhood changes can involuntarily displace residents, we aimed to understand the utility of electronic health record (EHR) data for proactively signaling community displacement. We characterized the District population and measured displacement. EHR data showed higher prevalences of chronic conditions than population-level estimates. Future work will involve a longitudinal analysis of factors associated with displacement.
Speaker(s):
Leigh Anne Tang, PhD
Indiana University/Regenstrief Institute
Author(s):
Leigh Anne Tang, PhD - Indiana University/Regenstrief Institute; Tess Weathers, MPH - Indiana University; Curtis Williamson, MSW, LCSW - Indiana University; Valerie Yeager, DrPH, MPhil - Indiana University;
Leigh Anne Tang, PhD - Indiana University/Regenstrief Institute
Assessing Privacy Concerns in LLM: A Brief Questionnaire
Poster Number: 323
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Privacy and Security, Qualitative Methods, Large Language Models (LLMs), Surveys and Needs Analysis, User-centered Design Methods, Patient Engagement and Preferences, Evaluation
Programmatic Theme: Consumer Health Informatics
Privacy concerns are the primary barriers to adoption of large language model (LLM) applications. This study developed and validated a seven-item questionnaire to efficiently assess these concerns. Evaluation with 297 participants confirmed a single latent construct (privacy concern) with excellent internal consistency (Cronbach’s alpha = 0.914). The resulting instrument provides a reliable, streamlined tool for measuring user privacy perceptions in generative AI research.
Speaker(s):
Leming Zhou, PhD, FAMIA
University of Pittsburgh
Author(s):
Leming Zhou, PhD, FAMIA - University of Pittsburgh; Hongtao Wang, Bachelor of Arts - University of Pittsburgh; Yi Xu, Ph.D. - University of Pittsburgh; I Made Agus Setiawan, PhD - University of Pittsburgh; Haomin Hu, PhD - University of Pittsburgh; Andi Saptono, PhD - University of Pittsburgh School of Health and Rehabilitation Services; Yong Choi, PhD - University of Pittsburgh; Bambang Parmanto, PhD - University of Pittsburgh;
Leming Zhou, PhD, FAMIA - University of Pittsburgh
Med-Canary: Membership Inference Attacks on Fine-tuned Medical LLMs
Poster Number: 324
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Privacy and Security, Large Language Models (LLMs), Artificial Intelligence, Natural Language Processing, Deep Learning, Machine Learning, Quantitative Methods
Programmatic Theme: Clinical Research Informatics
Large language models (LLMs) fine-tuned on clinical data are increasingly used to generate clinical summaries, yet their memorization capabilities raise privacy concerns. We propose Med-Canary, membership inference attack (MIA), that leverages density-weighted medical terms to enhance detection . Experiments on four LLMs (Llama2, Llama3.1, Qwen2.5, Qwen3) demonstrate consistent improvements over baselines by 4.2%, achieving areas under receiver operating characteristic curve above 0.88. These results indicate that medical fine-tuned LLMs remain vulnerable to privacy leakage.
Speaker(s):
Jiayue Hou, M.S. Candidate
Shanghaitech University
Author(s):
Jiayue Hou, M.S. Candidate - Shanghaitech University; Wenjie Wang, PhD - ShanghaiTech University; Zhiyu Wan, PhD - ShanghaiTech University;
Jiayue Hou, M.S. Candidate - Shanghaitech University
Postpartum Depression Risk Prediction Using Electronic Health Records: Toward Internal Validation and Fairness-Aware Model Selection for Clinical Decision Support
Poster Number: 325
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Public Health, Machine Learning, Clinical Decision Support
Programmatic Theme: Clinical Research Informatics
This study developed machine learning models to predict clinically confirmed postpartum depression using electronic health record data from 13,895 deliveries (2017-2023). Primary evaluation used a stratified 80/20 split with discrimination and calibration metrics. Gradient Boosting achieved strong performance (AUC=0.89) and was identified as the leading candidate for clinical decision support. Stratified cross-validation and subgroup performance analyses are ongoing to strengthen robustness and fairness prior to deployment.
Speaker(s):
Mustafa Ozkaynak, PhD
University of Colorado-Denver | Anschutz Medical Campus
Author(s):
Virginia Lijewski, MPH - University of Colorado, Anschutz Medical Campus; Onur Tuysuzoglu, Master - Universtiy of Colorado Denver; Jessica Andersson, DNP - University of Colorado, Anschutz Medical Campus; Mustafa Ozkaynak, PhD - University of Colorado-Denver | Anschutz Medical Campus;
Mustafa Ozkaynak, PhD - University of Colorado-Denver | Anschutz Medical Campus
All-Payer Claims Data Quality for Glucagon-Like Peptide-1 Surveillance
Poster Number: 326
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Public Health, Healthcare Quality, Data Modernization
Programmatic Theme: Public Health Informatics
Use of GLP-1 receptor agonists is expected to grow as costs decline. Given Arkansas' high diabetes and obesity burden, surveillance is critical. Using the Arkansas APCD (2017–2024), we evaluated structural integrity, semantic concordance and timeliness of reporting. Most structural metrics met thresholds (≥95%); semantic concordance fell below the threshold (≥50%). Reporting lag was minimal (median 5 days). The Arkansas APCD is fit for GLP-1 surveillance but has modifiable limitations that require preprocessing by users.
Speaker(s):
Beatrice Boateng, PhD
University of Arkansas for Medical Sciences
Author(s):
Maria Fernandez, PhD - University of Arkansas for Medical Sciences; Güneş Koru, PhD, FAMIA - University of Arkansas for Medical Sciences, Northwest Regional Campus;
Beatrice Boateng, PhD - University of Arkansas for Medical Sciences
The MassCPR Federated Data Network for Pathogen Readiness: Governance, Network Status, Use Cases, and Future Directions
Poster Number: 327
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Public Health, Population Health, Infectious Diseases & Epidemiology, Geospatial (GIS) Data/Analysis, Information Visualization, Data Mining, Privacy and Security, Governance
Programmatic Theme: Public Health Informatics
The Massachusetts Consortium on Pathogen Readiness (MassCPR) launched a Federated Data Network connecting two dozen hospitals and public health datasets, with data on 8+ million patients. Based on the widely adopted i2b2 and SHRINE software platforms, the network supports both research and public health use cases related to infectious diseases. It serves as a model that can be extended to other states across the country.
Speaker(s):
Griffin Weber, MD, PhD
Harvard Medical School
Author(s):
Griffin Weber, MD, PhD - Harvard Medical School; Nick Benik, BS - Harvard Medical School; Nick Brown, MEng - Harvard Medical School; Marc Ciriello - Harvard Catalyst; Andrew Dion, . - Harvard Medical School; Allan Harris, . - Mass General Brigham; Andrew Hendrix, . - Harvard Medical School; Jeffrey Klann, PhD - Massachusetts General Hospital; Diane Keogh, . - i2b2 tranSMART Foundation; Michael Mendis, BS - Mass General Brigham; Shawn Murphy, MD, Ph.D. - Massachusetts General Hospital; Thomas Naughton, BA - Harvard Medical School; Marc-Danie Nazaire, BS in Computer Science - Harvard University; Ankit Panchamia, . - Harvard Medical School; Perminder Singh, . - Harvard Medical School; Philip Trevvett, M.A. - Harvard Medical School;
Griffin Weber, MD, PhD - Harvard Medical School
Uncovering Health Technology Design Opportunities for Supporting Hypertonic Chronic Pelvic Pain Syndrome Diagnosis and Management
Poster Number: 328
2026 Symposium Health Equity Presentation
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Qualitative Methods, Human-computer Interaction, Health Equity
Programmatic Theme: Consumer Health Informatics
This work delves into the experience of patients with hypertonic Chronic Pelvic Pain syndrome (CPPS), a condition affecting millions of women globally, yet an underinvested area of women’s healthcare. To address this gap, we conducted a thematic analysis of 122 publicly available patient-narrated stories of their experience with CPPS to uncover health technology design opportunities for future explorations. We present three themes suggesting proactive interventions to support treatment and management for both patients and providers.
Speaker(s):
Zainab Balogun, PhD(c), MA, MS
University of Maryland Baltimore County
Author(s):
Samon Nguyen, BS(c) - University of maryland Baltimore County; Tera Reynolds, PhD, MPH, MS - University of Maryland Baltimore County;
Zainab Balogun, PhD(c), MA, MS - University of Maryland Baltimore County
Photoplethysmography Pulse Morphologies Associated with Toe-Brachial Index
Poster Number: 329
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Quantitative Methods, Real-World Evidence Generation, Information Visualization, Chronic Care Management
Programmatic Theme: Clinical Research Informatics
The Toe-Brachial Index (TBI) is a diagnostic test for Peripheral Artery Disease (PAD), particularly in patients with arterial calcification. Photoplethysmography (PPG) is a noninvasive optical measurement with potential as a screening tool for PAD. We present the first large-scale analysis of the relationship between toe PPG pulse morphology and TBI (5,506 signals from N=2,456 patients). We identify multiple features correlated with TBI, supporting the potential for future PPG-based PAD screening.
Speaker(s):
Ava Fascetti, MS
University of California San Diego
Author(s):
Ava Fascetti, MS - University of California San Diego; Mustafa Naguib, MTM - University of California San Diego; Shamim Nemati, PhD - University of California San Diego; Pam Taub, MD - University of California San Diego; Karandeep Singh, MD, MMSc - University of California San Diego; Elsie Ross, MD, MSc - University of California San Diego; Edward Wang, PhD - University of California San Diego; Mattheus Ramsis, MD - University of California San Diego;
Ava Fascetti, MS - University of California San Diego
Estimating Mosaic Loss of Y Chromosome Using “All of Us” Data
Poster Number: 330
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Quantitative Methods, Workflow, Data Mining
Programmatic Theme: Translational Bioinformatics
The project was aimed at developing an analytical pipeline to estimate mosaic loss of Y chromosome using “All of Us” data. The resulting pipeline demonstrated high stability and yielded results comparable with previous reports. This novel analytical pipeline is available in open access and can be used for future analyses of All of Us data.
Speaker(s):
Joseph Finkelstein, MD, PhD
University of Arizona
Author(s):
Zhenyao Ye, MPH - University of Maryland, Baltimore; Wanting Cui, Masters - University of Arizona; Robin Black, BSc - University of Arizona; Dan Theodorescu, MD, PhD - University of Arizona;
Joseph Finkelstein, MD, PhD - University of Arizona
Identifying Implementation Gaps in Paxlovid Prescription for COVID-19 Outpatient Care
Poster Number: 331
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Racial Disparities, Infectious Diseases and Epidemiology, Population Health, Machine Learning
Programmatic Theme: Clinical Informatics
Paxlovid has shown to be highly effective in preventing severe COVID-19 or mortality when given in a timely manner. To identify implementation gaps, we analyzed prescription patterns in outpatient COVID care across demographic and clinical subgroups using multivariate logistic regression. Paxlovid seems appropriately guided by clinical factors such as age, contraindications and comorbidities, but substantial racial and ethnic disparities remain, indicating that equitable implementation of antiviral therapies for COVID-19 remains a critical challenge.
Speaker(s):
Da In Lee, MSE
Northwestern University Feinberg School of Medicine
Author(s):
Erika Juarez-Martinez, MD, PhD - Northwestern University Feinberg School of Medicine; Yuan Luo, PhD - Northwestern University;
Da In Lee, MSE - Northwestern University Feinberg School of Medicine
Transferable and Interpretable Early Disease Predictive Algorithms for Patients With AAV Using Large-Scale Clinical Real-World Data
Click to View Handout
Poster Number: 332
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Real-World Evidence Generation, Machine Learning, Artificial Intelligence, Knowledge Representation & Information Modeling, Data Mining, Clinical Decision Support, Quantitative Methods, Usability
Programmatic Theme: Clinical Informatics
Early diagnosis of AAV is challenging because of heterogeneous, multisystem manifestations. We conducted a retrospective claims-based study to develop and validate an interpretable machine learning model using longitudinal U.S. claims data to identify patients at risk for AAV before diagnosis. Models demonstrated good discrimination (AUC, 0.87–0.88). A parsimonious 11-feature model achieved comparable performance (AUC, 0.84). Claims-based modeling may support earlier risk stratification and diagnostic evaluation in routine clinical care.
Speaker(s):
Fang He, PhD
Amgen Inc.
Author(s):
Sam Oh, PhD - Amgen Inc.; John Page, MD, ScD - Amgen Inc.; Phuong Pham, PhD - Amgen Inc.;
Fang He, PhD - Amgen Inc.
Standardizing Survey, Vision Care, and Cognitive Assessment Data for AI-READI Research: A Generalizable Framework for Mapping Research Questionnaires to the OMOP Common Data Model
Poster Number: 333
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Real-World Evidence Generation, Surveys and Needs Analysis, Controlled Terminologies, Ontologies, Vocabularies, Governance, Fairness and elimination of bias, Artificial Intelligence, Patient-/Person-Generated Health Data, Data Standards
Programmatic Theme: Clinical Research Informatics
Multi-instrument surveys are widely used in large research cohorts to capture social determinants of health, behavioral factors, vision assessments, cognitive testing, and patient-reported history of diseases and medication use; however, heterogeneous questionnaire structures limit interoperability and secondary data reuse. We developed a reproducible framework for harmonizing survey instruments into the Observational Medical Outcomes Partnership (OMOP) Common Data Model (CDM), using the Artificial Intelligence Ready and Exploratory Atlas for Diabetes Insights (AI-READI) study as an exemplar. AI-READI, part of the NIH Bridge2AI initiative, collected participant-reported data from 47 validated and custom survey instruments across 3,574 participants. A governance-based semantic harmonization workflow mapped 2,004 survey data elements to OMOP concepts using paired question–response representations documented with the Simple Standard for Sharing Ontological Mappings (SSSOM). Unmappable elements were represented as AI-READI vocabulary extensions. The resulting framework enables scalable integration of patient-reported data into OMOP and supports interoperable analytics and AI-ready research.
Speaker(s):
Stephanie Hong, MS, FAMIA
Johns Hopkins University School of Medicine
Author(s):
Monique Bangudi, MS - Department of Medicine, Section of Biomedical Informatics and Data Science, Johns Hopkins University School of Medicine, Baltimore, MD; James Cavallon, BS - Department of Medicine, Section of Biomedical Informatics and Data Science, Johns Hopkins University School of Medicine, Baltimore, MD; Jessica Mitchell, MS - Department of Medicine, Section of Biomedical Informatics and Data Science, Johns Hopkins University School of Medicine, Baltimore, MD; Patricia Francis, MS - Department of Medicine, Section of Biomedical Informatics and Data Science, Johns Hopkins University School of Medicine, Baltimore, MD; Sigfried Gold, MFA MA - Johns Hopkins School of Medicine, Section of Biomedical Informatics and Data Science; Yvette Chen, MS - Department of Medicine, Section of Biomedical Informatics and Data Science, Johns Hopkins University School of Medicine, Baltimore, MD; Abigail Lucero, BS - Oregan Health & Science University; Dawn Matthies, Ph.D. - Department of Ophthalmology and Visual Sciences, Heersink School of Medicine, University of Alabama at Birmingham, Birmingham, AL,; Julie Mitchell, BS - Oregon Health & Science University, Portland, OR; Steven Chamberlin, ND, MS - Oregon Health & Science University, Portland, OR; Jorge Contreras, JD - S.J. Quinney College of Law and Department of Human Genetics, University of Utah, Salt Lake City, UT; Sally Baxter, MD, MSc - Division of Ophthalmology Informatics and Data Science, Viterbi Family Department of Ophthalmology and Shiley Eye Institute, University of California, San Diego, La Jolla, CA; Cindy Cai, MS, MD - Wilmer Eye Institute, Johns Hopkins University School of Medicine, Baltimore, MD; Michelle Haribar, PhD - Department of Ophthalmology and Visual Sciences, University of Illinois, College of Medicine; Christopher Chute, MD DrPH - Johns Hopkins University; Cecilia Lee, MD, MS - Washington University in St. Louis; Aaron Lee, MD MSCI - Washington University in St Louis;
Stephanie Hong, MS, FAMIA - Johns Hopkins University School of Medicine
Causal Model of Data Quality
Poster Number: 334
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Real-World Evidence Generation, Data transformation/ETL, Causal Inference, Large Language Models (LLMs), Quantitative Methods
Programmatic Theme: Clinical Research Informatics
Data Quality (DQ) issues of Real World Data are currently identified in terms of an ontological framework based on the type of DQ issue identified. We interrogated a set of DQ widgets to arrive at a hierarchical, causal model. Working to saturation, we identified a Core Model identifying causes and covariates from the Disease and its Biology down to Variables to be analyzed. A comparison of this Causal Model and the standard Ontological model was made with an LLM of 42 PCORnet®/PaTH-based articles. It identified 899 issues, with 35% issues identified by the Causal, but not the Ontological, framework, and 10%, the other way around. We offer the Causal DQ Model as a broader framework for DQ and one that can accommodate sharing across sites, as the basis of a DQ-causes Library, and one that can be quantified and therefore turned into decision support.
Speaker(s):
Hanieh Razzaghi, PhD
Children's Hospital of Philadelphia
Author(s):
Nicole Weiskopf, PhD - Oregon Health & Science University; Harold Lehmann, MD, PhD - Johns Hopkins University;
Hanieh Razzaghi, PhD - Children's Hospital of Philadelphia
Temporal Resolution of Longitudinal Clinical Data from EHR Determines Detectability of Pre-Dropout Warning: Evidence from a Glaucoma Cohort
Poster Number: 335
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Real-World Evidence Generation, Data Mining, Data Mining
Programmatic Theme: Clinical Research Informatics
EHR-based retention systems often rely on coarse bin summaries, masking temporal dynamics of patient disengagement. This study investigated how temporal resolution affects the inferential association between visual acuity and loss to follow-up in 9,756 glaucoma patients. Mixed-effects modeling across 3-month, 6-month, and 1-year windows revealed the 6-month resolution provided optimal fit. Results suggest that temporal precision is a critical parameter for extracting reliable clinical indicators from routine longitudinal EHR data.
Speaker(s):
Hai-Wei Liang, PhD
Department of Ophthalmology, University of Pittsburgh School of Medicine
Author(s):
Hai-Wei Liang, PhD - Department of Ophthalmology, University of Pittsburgh School of Medicine; Andrew Williams, MD - Department of Ophthalmology, University of Pittsburgh;
Hai-Wei Liang, PhD - Department of Ophthalmology, University of Pittsburgh School of Medicine
From Preparation to Recognition: Navigating the Accreditation Process for a DNP Nursing Informatics Specialty Program
Poster Number: 336
2026 Symposium LIEAF Presentation
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Teaching Innovation, Curriculum Development, Workforce Development
Programmatic Theme: Academic Informatics / LIEAF
This presentation describes the accreditation process of a DNP Nursing Informatics program aligned with AACN Essentials and CAHIIM/AMIA MHI competencies. Emphasizing rigorous, standardized education, the program mapped measurable informatics competencies across Miller’s Levels of clinical competence. The session highlights curricular alignment, performance-based assessment, and academic–practice partnerships as foundational strategies to ensure quality, accountability, and workforce-ready doctoral preparation in nursing informatics.
Speaker(s):
Robin Austin, PhD, DNP, DC, RN, NI-BC, FAMIA, FAAN
University of Minnesota, School of Nursing
Author(s):
Stuart Speedie, PhD - Institute for Health Informatics (IHI), University of Minnesota; Angela Kennedy, EdD - CAHIIM; Karen Monsen, PhD, RN, FAMIA, FNAP, FAAN - None; Connie Delaney, PhD - University of Minnesota, School of Nursing;
Robin Austin, PhD, DNP, DC, RN, NI-BC, FAMIA, FAAN - University of Minnesota, School of Nursing
Learning AI in Medicine: Permissions, Attitudes, and the Match
Poster Number: 337
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Teaching Innovation, Curriculum Development, Artificial Intelligence
Programmatic Theme: Clinical Informatics
Ambient AI scribes are rapidly entering clinical workflows, yet medical student access remains nearly absent. Three surveys (M2 students, n=97; residents, n=86; clerkship directors, n=94) reveal a 14× gradient in opposition to student access — from 3.1% of M2s to 42.6% of clerkship directors — while all groups converge on tiered access as preferred. Residents retrospectively underweighted AI scribe availability at Match time; M2s prospectively rate it far higher, suggesting a generational shift already underway.
Speaker(s):
Elizabeth Cramer, MD
University of Iowa, Carver College of Medicine
Author(s):
Emily Welder, MD - University of Iowa Health Care; Aaron Kunz, DO, MA, MME - University of Iowa Health Care; Yinghui Xu, MD - University of Iowa Health Care; Michael Maharry, MD - University of Iowa Health Care; Laura Seegmiller, MPH, MA - University of Iowa Health Care; Jennifer Van Tiem - University of Iowa; Kelly Skelly, MD - University of Iowa Health Care; Patrick Barlow, PhD - University of Iowa Health Care;
Elizabeth Cramer, MD - University of Iowa, Carver College of Medicine
Leveraging Health IT for Antibiotic Stewardship in Telemedicine: Lessons on Clinical Decision Support, Data Analytics, and Patient Communication from a Nationwide Quality Improvement Program
Poster Number: 338
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Telemedicine, Healthcare Quality, Infectious Diseases & Epidemiology, Clinical Decision Support, Patient Safety
Programmatic Theme: Clinical Informatics
Health information technology (IT) supports appropriate antibiotic prescribing, but its use in telemedicine is not well understood. As part of a program to improve antibiotic use, we examined telemedicine practices’ use of health IT for antibiotic stewardship. Using interviews, program implementation information, and surveys, we identified three health IT-enabled approaches: clinical decision support, data analytics, and patient communication tools. Findings highlight challenges/opportunities to integrate health IT into telemedicine to improve decision-making and antibiotic use.
Speaker(s):
Prashila Dullabh, MD
NORC at the University of Chicago
Author(s):
Jessica Fernandez, PhD - NORC at the University of Chicago; Mikaela Harmsen, MPH - NORC at the University of Chicago; Kiran Correa, MPP - NORC at the University of Chicago; Fiona Grady, MPP - NORC at the University of Chicago; Sara Keller, MD, MPH, MSPH - Johns Hopkins Medicine; Pranita Tamma, MD, MHS - Children's Hospital of Philadelphia; Susan Henderson, MD, MPH - AHRQ; Alison Laffan, PhD, MHS - NORC at the University of Chicago; Nova Liu, BS - Johns Hopkins Armstrong Institute for Patient Safety and Quality; Roy Ahn, ScD, ScM, MA - NORC at the University of Chicago; Carly Parry - NORC at the University of Chicago;
Prashila Dullabh, MD - NORC at the University of Chicago
Remote Patient Monitoring (RPM) and Healthcare Outcomes in Patients with COPD and CHF
Poster Number: 339
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Telemedicine, Data Mining, Chronic Care Management
Programmatic Theme: Clinical Informatics
Remote patient monitoring (RPM) could improve healthcare delivery and support long-term management of chronic diseases. This study examined the association between RPM participation and healthcare outcomes among patients with congestive heart failure (CHF) and chronic obstructive pulmonary disease (COPD). RPM-related services were identified in electronic health records between January 1, 2014, and December 31, 2018. The first qualifying RPM visit served as the index date, requiring at least two RPM visits within one year and no RPM use in the prior year. Adults aged 18 years or older with non-missing demographic information were included. Propensity score matching without replacement was performed at a 1:40 ratio based on gender, race, and 5-year birth-year groups, and five controls per case were randomly selected. The final analytic dataset included 1,260 patients. RPM users had higher outpatient and emergency department visit frequencies during follow-up, but lower inpatient admissions and shorter total hospital length of stay.
Speaker(s):
Wanting Cui, Masters
University of Arizona
Author(s):
Joseph Finkelstein, MD, PhD - University of Arizona;
Wanting Cui, Masters - University of Arizona
Bridging the Distance: A Scoping Review of Telehealth for Chronic Disease Management in Rural America.
Poster Number: 340
2026 Symposium Health Equity Presentation
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Telemedicine, Chronic Care Management, Health Equity, Population Health, Diversity, Equity, Inclusion, and Accessibility
Programmatic Theme: Public Health Informatics
This scoping review protocol addresses health disparities in rural populations by synthesizing evidence on telehealth interventions for chronic disease management. Despite telehealth's potential, its effectiveness and implementation in rural settings remain unclear. This review maps existing literature, identify gaps, and inform effective strategies. It systematically searches databases for studies from 2015, using two independent reviewers for data extraction and thematic analysis for synthesis.
Speaker(s):
Fatma AbdelSalam, Doctor of Health Administration
University of Texas at Tyler
Author(s):
Fatma AbdelSalam, Doctor of Business Administration - University of Texas at Tyler; Ifeolu David, PhD, MPH - University of Texas at Tyler;
Fatma AbdelSalam, Doctor of Health Administration - University of Texas at Tyler
Telehealth Appropriateness and Relative Advantage for ADRD: A Qualitative Study with Nursing Home Leaders
Poster Number: 341
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Telemedicine, Qualitative Methods, Chronic Care Management, Policy, Health Equity
Programmatic Theme: Clinical Informatics
We interviewed 31 nursing‑home leaders to identify when telehealth is appropriate for long‑stay residents with ADRD. Telehealth was favored for after‑hours triage, follow‑ups, medication reviews, behavioral observation, and selected specialty consults. However, in-person was preferred for new problems requiring physical exams. Successful telehealth visits for ADRD care hinged on reliable informants and staff facilitation. Current Medicare flexibilities through December 31, 2027 offer a window to standardize and evaluate these workflows for telehealth services.
Speaker(s):
Mauli Shah, MPH
Vanderbilt University Medical Center
Author(s):
Mauli Shah, MPH - Vanderbilt University Medical Center; Joyce Harris, MA - Vanderbilt University Medical Center; Sarah Zeng, B.A - Vanderbilt University; Bamlak Akilu, BA - Vanderbilt University; Paul Katz, MD - Florida State University; Mark Unruh, PhD - Weill Cornell Medical College; Hye-Young Jung, PhD - Weill Cornell Medical College; Jiani Yu, PhD - Weill Cornell Medical College; Jessica Ancker, MPH, PhD, FACMI - Vanderbilt University Medical Center;
Mauli Shah, MPH - Vanderbilt University Medical Center
Transition of Care AI Assisted Proof of Concept (TokAI) Leveraging the Clinical Health AI augmented Records Toolkit (CHART)
Poster Number: 342
2026 Symposium 25x5 Presentation
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Transitions of Care, Interoperability and Health Information Exchange, Artificial Intelligence, Information Extraction, Data transformation/ETL, Documentation Burden, Controlled Terminologies, Ontologies, Vocabularies, Data Standards
Working Group: Knowledge Representation and Semantics Working Group
Programmatic Theme: Clinical Informatics
Transitions of Care (ToC) often involve incomplete information exchange, driving readmissions and harm; 40% of Medicare acute-care patients transition to post-acute care. The PACIO Project developed TokAI, an AI-enabled ToC burden-reduction tool that automates extraction, harmonization, and FHIR bundling of post-acute data. TokAI uses AI agents to identify USCDI elements from structured and unstructured sources, apply PIQI-based quality checks, de-duplicate and resolve conflicts, and generate interoperable ToC bundles for clinician adjudication.
Speaker(s):
May Terry, M.Sc., BSEE, BSN
MITRE Corporation
Author(s):
James Patterson, M.Sc. - MITRE Corporation; Yolanda Liu, MD, M.Sc. - MITRE Corporation; Terrence O'Malley, MD, CMD - Harvard Medical School; Holly Miller, MD - Centauri Health Solutions; Lorraine Wickiser; Shelly Coyle, RN, MS, MBA - Centers of Medicaid and Medicare Services;
May Terry, M.Sc., BSEE, BSN - MITRE Corporation
Usability of an Integrated Patient Management System Connecting Remote Patient Support Systems
Poster Number: 343
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Usability, Clinical Decision Support, Telemedicine
Programmatic Theme: Clinical Informatics
Clinical decision support systems (CDSS) often function as standalone tools or lack sufficient integration with hospital workflows to provide standardized, patient-specific services. We developed and implemented an Integrated Patient Management System (IPMS) that connects clinicians’ prescribing and review processes with patient-side execution and data collection. IPMS integrates a clinician-facing web platform with the Remote Patient Support System (RPSS), a patient-facing mobile app. During the formative evaluation phase, we assessed usability, emphasizing workflow continuity from start to finish and data synchronization across modules. Forty researchers were recruited to perform scenario-based tasks and complete usability surveys while acting as both clinician and patient users. This formative usability assessment identifies usability issues and guides refinements during development, supporting better workflow alignment in future clinical settings and more stable deployment of comprehensive patient management functions.
Speaker(s):
Eun-Seon Yang, Researcher / Master's
Cerebrovascular Disease Research Center, Hallym University
Author(s):
Jinwon Kim, Bachelor of Science (BS) - Department of Artificial Intelligence Convergence, Hallym University; Jeonghwan Koh, Bachelor of Science (BS) - Department of Artificial Intelligence Convergence, Hallym University; Ung Park, Bachelor of Science (BS) - Department of Artificial Intelligence Convergence, Hallym University; Sunghan Lee, PhD - Department of Artificial Intelligence, Dankook University; In cheol Jeong, PhD - Hallym University;
Eun-Seon Yang, Researcher / Master's - Cerebrovascular Disease Research Center, Hallym University
The Usability-Security Gap in Nurse-Facing Clinical Decision Support Systems: Secondary Content Analysis of Privacy and Information Security Considerations
Poster Number: 344
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Usability, Clinical Decision Support, Privacy and Security
Working Group: Nursing Informatics Working Group
Programmatic Theme: Clinical Informatics
Clinical decision support (CDS) systems assist nurses’ decision-making by reducing cognitive workload and promoting adherence to evidence-based guidelines. Despite routine use of protected health information (PHI), privacy and information-security considerations are often underreported in usability evaluations. We conducted a secondary content analysis of 27 nurse-facing CDS usability studies published between 2010 and 2022, applying a deductive framework based on the HIPAA Security Rule and Privacy by Design principles. Six domains were assessed: PHI protection, technical safeguards, governance, risk assessment, workflow/architecture integration, and user training. Only eight studies (29.6%) referenced any privacy or security domain, most often as declarative statements (e.g., HIPAA/IRB references without implementation detail). Specific technical safeguards were described in four studies (14.8%). No study reported a formal privacy impact assessment or integrated security-relevant metrics into usability evaluation. These findings demonstrate a persistent usability–security gap and support sociotechnical approaches that evaluate usability and security concurrently to promote safe, effective clinical practice.
Speaker(s):
Hwayoung Cho, PhD, RN, FAMIA
University of Florida College of Nursing
Author(s):
Tamara Macieira, PhD, RN, FAMIA - University of Florida; Taeheon Lee, BS student - Ghent University Global Campus; Mei Liu, PhD - University of Florida; Megan Gregory, Ph.D. - University of Florida;
Hwayoung Cho, PhD, RN, FAMIA - University of Florida College of Nursing
Usability Evaluation of a Clinical Decision Support System for Cervical Cancer Screening, Triage, and Treatment in Women Living with HIV
Poster Number: 345
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: User-centered Design Methods, Usability, Global Health, Clinical Decision Support, Public Health
Programmatic Theme: Public Health Informatics
Despite expanded HPV testing and vaccination, cervical cancer remains the fourth most common cancer among women globally, with a disproportionate burden in low- and middle-income countries. Women living with HIV (WLH) face elevated risk due to persistent HPV infection and accelerated progression to precancerous lesions. Current clinical workflows in resource-limited settings often lack integrated decision support tools to guide evidence-based screening and treatment. To address this gap, we developed an NCI-funded clinical decision support (CDS) system to operationalize the screen-triage-and-treat (iSTAR) strategy for cervical cancer management among WLH at the point of care in the Dominican Republic. This study evaluated the system's usability through structured assessments with health informatics experts and end-users prior to clinical deployment.
Speaker(s):
Fabiana Dos Santos, PhD, MSN, RN
Purdue University School of Nursing
Author(s):
Fabiana Dos Santos, PhD, MSN, RN - Purdue University School of Nursing; Lauren Gerchow, PhD, RN - Columbia University School of Nursing; Mina Halpern, MPH - Clínica de Familia La Romana, La Romana, Dominican Republic; Yazmina Castro, MPH - Clínica de Familia La Romana, La Romana, Dominican Republic; Samantha Stonbraker, PhD, MPH, RN, FAAN, FAMIA - University of Colorado Anschutz Medical Campus; Pamela Baez, MD - Clínica de Familia La Romana, La Romana, Dominican Republic; Rebecca Schnall, BSN, MPH, PhD - Columbia University;
Fabiana Dos Santos, PhD, MSN, RN - Purdue University School of Nursing
Governing Risk Threshold Decisions in Clinical AI Deployment: A Case Study of Asthma Exacerbation Prediction
Poster Number: 346
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Workflow, Clinical Decision Support, Data transformation/ETL, Real-World Evidence Generation
Working Group: Climate, Health and Informatics Working Group
Programmatic Theme: Clinical Informatics
Risk prediction models require operational thresholds to translate continuous risk scores into clinical action. Although statistical methods identify ‘optimal’ cut-offs, threshold selection ultimately reflects value judgments regarding harm tolerance, resource allocation, and workflow feasibility. We describe a governance-informed approach to selecting a deployment threshold for an asthma exacerbation (AE) prediction model integrated into clinical workflows. Using prevalence-adjusted performance metrics and real-world provider capacity modeling, we evaluated multiple candidate thresholds and quantified downstream workload and missed-event trade-offs. We demonstrate that statistically optimal thresholds may produce operationally infeasible alert volumes or unacceptable miss rates. We propose a structured threshold governance framework integrating statistical performance, clinical utility, stakeholder input, and human oversight safeguards. This case illustrates how threshold decisions should be treated as organizational governance processes rather than purely technical optimizations.
Speaker(s):
Lu Zheng, Ph.D., M.S.
Mayo Clinic
Author(s):
Bhavani Singh Agnikula Kshatriya, Masters in Health Informatics - Mayo Clinic; Lauren Rost, PhD - Mayo Clinic; Joshua Ohde, PhD - Mayo Clinic; Momin Malik, PhD - Mayo Clinic; Kevin Peterson, PhD - Mayo Clinic; Tracey Brereton, Masters - Mayo Clinic; Brenna Loufek, MS - Mayo Clinic; Tara Pereira, MS - Mayo Clinic; Chenyu Gai, Master - Mayo Clinic; Chung Wi, MD - Mayo Clinic; Young Juhn, M.D., M.P.H. - Mayo Clinic; Shauna Overgaard, Ph.D. - Mayo Clinic; Joy Fladager Muth, APRN, C.N.P., M.S. - Mayo Clinic; Martha Hartz, MD - Mayo Clinic; Miguel Park, MD - Mayo Clinic;
Lu Zheng, Ph.D., M.S. - Mayo Clinic
Emergency Clinicians’ Workflow Adaptations to Workstation Changes
Poster Number: 347
2026 Symposium 25x5 Presentation
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Workflow, Human-computer Interaction, Usability
Programmatic Theme: Clinical Informatics
Technological transitions can disrupt emergency clinician (EC) workflows. We evaluated adjustments from Manual Login to Proximity Card Login workstations using a mixed-methods approach combining time-study observations (early, late time points) post-transition with surveys (early). The transition was followed by faster login times, improved usability ratings, and greater EC satisfaction despite increased workflow interruptions from auto-logouts. Clinicians and administrators should weigh efficiency and security gains against workflow burden (e.g., auto-logouts) when evaluating workstation technology transitions.
Speaker(s):
Arwen Declan, MD PhD
Prisma Health
Author(s):
Jennifer Rumsey, M3 - Prisma Health, USC School of Medicine Greenville; Sahil Sawant; Nicholas Perkins, DO - Prisma Health; Jeff Gerac, MD - Prisma Health; Frederick Lynch, MD - Prisma Health; Ronald Pirrallo, MD - Prisma Health; Sudeep Hegde, PhD - Clemson University;
Arwen Declan, MD PhD - Prisma Health
Parallel Linked Development of Future Mentors and Workforce in Biomedical Data Science and Informatics
Poster Number: 348
2026 Symposium LIEAF Presentation
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Workforce Development, Teaching Innovation, Curriculum Development
Programmatic Theme: Academic Informatics / LIEAF
This project links a mentoring‑development course with a new track in biomedical data science and informatics within institutional summer undergraduate research program to strengthen the research workforce pipeline. Graduate trainees learn mentoring skills while undergraduates complete full‑time research and data‑intensive coursework. Shared lectures, structured partnering, and faculty shadowing create an integrated environment where both groups advance technical, communication, and career‑readiness skills essential for biomedical informatics and data science.
Speaker(s):
Alexander Alekseyenko, PhD, FAMIA, FACMI
Medical University of South Carolina
Author(s):
Alexander Alekseyenko, PhD, FAMIA, FACMI - Medical University of South Carolina;
Alexander Alekseyenko, PhD, FAMIA, FACMI - Medical University of South Carolina
AI Morning Meet-Up (AIMM): A micro-intervention to build LLM implementation readiness in a health system
Poster Number: 349
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Workforce Development, Artificial Intelligence, Large Language Models (LLMs)
Programmatic Theme: Clinical Research Informatics
AI Morning Meet-Up (AIMM) is a learner-led micro-intervention designed to build health-system readiness for generative AI. Across 51 weekly sessions and five deep dives, participants explored emerging tools and policy implications. Survey results show increased AI literacy, confidence, and adoption intent, suggesting short, community-driven sessions can accelerate responsible AI implementation.
Speaker(s):
Katerina Andreadis, MS
NYU Grossman School of Medicine
Author(s):
JaeEun Kwon, Master of Public Policy - NYU Langone Health; Lynn Xu, MPH - NYU Grossman School of Medicine; Devin Mann, MD - NYU Grossman School of Medicine;
Katerina Andreadis, MS - NYU Grossman School of Medicine
Using Red and Green Teams to Quickly Evaluate and Deploy AI Documentation-Assisting Tools
Poster Number: 351
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support, Artificial Intelligence, Large Language Models (LLMs), Governance, Documentation Burden
Programmatic Theme: Clinical Informatics
We conducted a rapid, multidisciplinary Red and Green Team evaluation of an EHR‑embedded LLM documentation tool. In two hours, teams identified key vulnerabilities—including hallucinated clinical content, biased language, and circumvention of guardrails—as well as high‑value, low‑risk applications including shorthand expansion and patient‑friendly explanations. Findings directly informed an institutional governance framework defining risk, configuration needs, and oversight requirements for deploying clinical AI tools.
Speaker(s):
Anna Huffman, MD
UW Health/University of Wisconsin
Author(s):
Heidi Twedt, MD - University of Wisconsin School of Medicine and Public Health; David Schneider, MD - University of Wisconsin; Peter Kleinschmidt, MD - UW Health; Navin Pathak, MD - UW Health; Youngchan Park, MD, MPH - University of Wisconsin; SARA LAY, DO - University of Wisconsin;
Anna Huffman, MD - UW Health/University of Wisconsin
Diagnosing Biomedical Citation Faults: An Eight-Category Taxonomy and Multi-Design Evaluation
Click to View Presentation
Poster Number: 352
Presentation Time: 05:00 PM - 06:30 PM
Biomedical research depends on citations that accurately connect claims to evidence, yet citation errors can distort findings and propagate through literature. Existing evaluations often group distinct faults into broad labels or rely on synthetic examples. We developed Citation Diagnostic Engine (CiteDx), which combines language-model evidence extraction with deterministic rules to diagnose eight fault categories. Category-specific evaluations yielded 92.5% Wrong Reference precision in a held-out sample. In addition, Insufficient Support precision reached 90% among determinate findings.
Speaker(s):
Aston Liu, Student
UT Health
Aston Liu, Student - UT Health
Three Variables Carry the Signal: Cross-Model Permutation Ranking of Non-DXA Predictors of Femoral-Neck T-Score Category in NHANES
Click to View Presentation
Poster Number: 353
Presentation Time: 05:00 PM - 06:30 PM
Osteoporosis screening depends on bone-density imaging that is not universally accessible. Using four NHANES cycles (8,695 adults aged 50 and older), we classified femoral-neck T-score category from twelve non-DXA variables and ranked each by held-out performance lost when permuted. Gradient-boosted trees and logistic regression both ranked BMI, sex, and age first through third. The remaining nine variables changed accuracy by 1.53 points or less.
Speaker(s):
Andy Wu, High School Student
Dublin Jerome Highschool
Andy Wu, High School Student - Dublin Jerome Highschool
Yes, the other left: When Medical Vision-Language Models Confuse Left and Right: Evaluating Laterality Reasoning in Clinical Images
Click to View Presentation
Poster Number: 354
Presentation Time: 05:00 PM - 06:30 PM
Vision-language models (VLMs) increasingly interpret clinical photographs, but their ability to determine patient-centered left/right laterality remains poorly characterized. We benchmarked 13 general-purpose, medical, and physical-AI VLMs on 542 dermatology photographs using three prompts, horizontal mirroring, and four-stage reasoning analysis. No model exceeded the 50% constant-side baseline on left/right-determinable images, and predictions often failed to reverse after mirroring. Laterality should be evaluated as a distinct spatial-reasoning and abstention capability in clinical AI systems.
Speaker(s):
Mitchell Liu, Highschool
Brentwood
Mitchell Liu, Highschool - Brentwood
Mining Fascinomas from Online Health Communities with LLMs
Click to View Presentation
Poster Number: 355
Presentation Time: 05:00 PM - 06:30 PM
We developed a local large language model (LLM) pipeline to identify recurring themes and fascinomas in online health communities. Applied to endometriosis discussions on Reddit, it prioritized (i) syncope during pain, (ii) urinary symptoms, and (iii) pain during orgasm. In EHRs, 33.8% of patients with endometriosis and documented syncope had an episode within seven days of pain, heavy bleeding, anemia, or iron deficiency, suggesting a temporal association warranting further study.
Speaker(s):
Sitar Eswar, Student
Design Tech High School
Sitar Eswar, Student - Design Tech High School
LLM Moneyball: A Novel Methodology for Evaluating LLM Jury Performance for Scalable Artificial Intelligence Monitoring
Click to View Presentation
Click to View Handout
Poster Number: 356
Presentation Time: 05:00 PM - 06:30 PM
Large language model (LLM) juries may offer more scalable monitoring of deployed artificial intelligence (AI) than humans. Analogous to statistical frameworks for athlete and team performance, we developed a framework evaluating individual LLM jurors as Delphi panelists and overall jury effectiveness for monitoring AI output. We illustrate with a 4-agent jury evaluating AI-generated hospital discharge summaries, quantitatively identifying underperforming agents. This offers healthcare leaders pragmatic criteria to optimize Delphi-like juries for AI monitoring.
Speaker(s):
Rayna Kumar
Lynbrook High School
Rayna Kumar - Lynbrook High School
Poster Session 2
Description
Custom CSS
double-click to edit, do not edit in source
Date: Monday (11/09)
Time: 5:00 PM to 6:30 PM
Room: Trinity Exhibit Hall (Posters)
Time: 5:00 PM to 6:30 PM
Room: Trinity Exhibit Hall (Posters)