American Medical Informatics Association - Enhancing Student Learning and Engagement with AI Tutors in Health Informatics Higher Education

From Inbox to Revenue: Provider Adoption and Revenue Modeling of eVisit Billing at a Large Academic Health System

Poster Number: 100
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Administrative Systems, Documentation Burden, Evaluation, Workflow
Programmatic Theme: Clinical Informatics

Patient portal messaging increases clinician inbox workload, yet adoption of asynchronous online digital evaluation and management (“eVisit”) billing remains variable. We analyzed adult patient-initiated portal threads addressed to billing-eligible clinicians from October 1, 2024 to October 1, 2025 at a large academic health system. We quantified billed-thread frequency, provider adoption (≥1 billed eVisit), specialty-level patterns among high-burden specialties, and concentration of billing activity. Of 1,110,833 eligible threads to 2,860 providers, 9,315 (0.8%) were billed and only 351 providers (12.3%) billed ≥1 eVisit. Billing was highly concentrated: the top 10 providers accounted for 35.9% of billed eVisits. Using 2024 CMS RVU weights and conversion factor, billed eVisits corresponded to 5,508 RVUs ($183,349). A scenario-based extrapolation applying observed level distributions system-wide yielded $519,612, suggesting modest financial yield relative to message volume. These findings identify adoption gaps and high-burden specialties for workflow and educational interventions.

Speaker(s):
Yash Trivedi, MD
Vanderbilt University School of Medicine

Author(s):
Athira Sivadas, BA - Vanderbilt University School of Medicine; Nomongo Dorjsuren, BA - Vanderbilt University Medical School; Yash Trivedi, MD - Vanderbilt University School of Medicine; Ryan Buckley, MD - Vanderbilt University Medical Center; Bryan Steitz, PhD - Vanderbilt University Medical Center;

Prompt-Engineered Large Language Models for Clinical Concept Abstraction of Fall Risk Documentation

Poster Number: 101
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Artificial Intelligence, Healthcare Quality, Informatics Implementation
Programmatic Theme: Clinical Informatics

Falls are a major cause of morbidity among older adults and are targeted by Medicare quality measure C15. We evaluated a prompt-engineered large language model (LLM) for automated abstraction of fall risk and related clinician actions from unstructured clinical notes. Using a corpus of 400 annotated notes, the LLM achieved strong performance across tasks and outperformed supervised baselines. These findings support the potential of prompt-engineered LLMs for scalable abstraction of healthcare quality measures.

Speaker(s):
Max Hukill, AB
Kaiser Permanente School of Medicine

Author(s):
Kurtis Pruitt, BS - Kaiser Permanente School of Medicine; Max Hukill, AB - Kaiser Permanente School of Medicine; Manabu Torii, PhD - Kaiser Permanente; Diane Oliver, MD - Southern California Permanente Medical Group; Yang Huang, PhD - Kaiser Permanente Southern California; Daniel Zisook, MD - Kaiser Permanente-Southern California;

Fine-Tuning PubMedBERT for Hierarchical Condition Category Classification from Clinical Notes

Poster Number: 102
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Artificial Intelligence, Natural Language Processing, Healthcare Quality
Working Group: Clinical Research Informatics Working Group
Programmatic Theme: Clinical Informatics

Automating Hierarchical Condition Category (HCC) assignment directly from unstructured electronic health record (EHR) notes remains an important but understudied problem in clinical informatics. We present HCC-Coder, an end-to-end NLP system that maps narrative documentation to 115 Centers for Medicare & Medicaid Services(CMS) HCC codes in a multi-label setting. On the test dataset, HCC-Coder achieves a macro-F1 of 0.779 and a micro-F1 of 0.756, with a macro-sensitivity of 0.819 and macro-specificity of 0.998. By contrast, Generative Pre-trained Transformer (GPT)-4o achieves the highest macro-F1 of 0.735 and the highest micro-F1 of 0.708 under five-shot prompting. The fine-tuned model demonstrates consistent absolute improvements of 4%–5% in F1-scores over GPT-4o. To address severe label imbalance, we incorporate inverse-frequency weighting and per-label threshold calibration. These findings suggest that domain-adapted transformers provide more balanced and reliable performance than prompt-based large language models for hierarchical clinical coding and risk adjustment.

Speaker(s):
Xiangren Wang, MPH
University of Florida

Author(s):
Xiangren Wang, PhD - University of Florida; Noah Hammarlund, PhD - University of Florida; Mattia Prosperi, PhD, FAMIA - University of Florida; Yenan Zhu, PhD - University of Florida; Lee Revere, PhD - University of Florida;

Randomized Trial: Epic Generative AI Chart Summarization Tool to Reduce Ambulatory Provider Cognitive Task Load

Poster Number: 103
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Artificial Intelligence, Usability, Quantitative Methods
Programmatic Theme: Clinical Informatics

Epic Systems, a leading EHR vendor, is rapidly deploying generative AI (genAI) tools to enhance clinician workflows. This randomized controlled trial evaluates Epic’s genAI chart summarization tool’s impact on ambulatory clinician cognitive load and establishes a framework for future AI tool assessments. The study measures Physician Task Load (PTL) scores, chart time metrics, and professional fulfillment. Findings will guide responsible adoption of AI in healthcare, addressing usability, safety, and workflow integration.

Speaker(s):
Nina Zhu, MD
UCLA

Author(s):
Aaron Chin, MD - UCLA; Paul Lukac, MD, MBA, MS - UCLA Health; Thomas Kingsley, MD MPH MS - UCLA; Sitaram Vangala, MS - UCLA; John Mafi, MD, MPH - UCLA; Pallavi Mynampati, MBA - UCLA; Yan Phipps, MBA - UCLA; Artem Romanov, BA - UCLA; Maxwell Weng, BA - UCLA; Lauren Wisk, PhD - UCLA; Hawkin Woo - UCLA;

Acoustic Speech Deviations During Care Coordination Calls Reflect Cognitive Vulnerability in Heart Failure Patients

Poster Number: 104
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Artificial Intelligence, Natural Language Processing, Chronic Care Management, Clinical Decision Support
Programmatic Theme: Clinical Research Informatics

We analyzed 1,272 care coordination calls from 324 heart failure patients to examine whether speech acoustics reflect cognitive impairment. Compared with patients without cognitive impairment, patients with cognitive impairment showed modest baseline differences. However, among cognitively impaired patients, deviation in word count, utterance duration, jitter, and shimmer were significantly greater during discussions of risk factors. These findings suggest passive acoustic analysis of routine conversations may provide a scalable, non-invasive marker of cognitive and clinical vulnerability.

Speaker(s):
Sang Bin You, MSN, RN
University of Pennsylvania

Author(s):
Shuxuan Li, master - University of Pennsylvania; Yiheng Zhang, MA - University of Pennsylvania; Kathy Bowles, PhD - University of Pennsylvania; Margaret McDonald - Visiting Nurse Service of New York; Max Topaz, PhD, RN, MA, FAAN, FIAHSI, FACMI - Columbia University; Jiyoun Song, PhD - University of Pennsylvania School of Nursing;

How do Artificial Intelligence-enabled Decision Aids Impact Health Outcomes for Patient Centric Decision-Making? Evidence from Osteoarthritis Patients.

Poster Number: 105
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Artificial Intelligence, Clinical Decision Support, Patient Engagement and Preferences, Chronic Care Management, Surgery
Programmatic Theme: Clinical Research Informatics

How do AI-enabled decision aids (AI-DAs) affect long-term patient health outcomes through their impact on the quality of shared decision-making (SDM)? Does patients’ baseline depression status affect the relationship between AI-DAs and longer-term health outcomes? Leveraging randomized clinical trial (RCT) data, we find AI-DAs increase patients' long-term health outcomes by improving decision-quality. This effect is particularly pronounced for patients with depressive symptoms. Our research highlights how AI-DAs encourage compassionate, human-AI collaboration in high-stakes health decision-making.

Speaker(s):
Nidhish Nerur, PhD Student
University of Texas at Austin

Author(s):
Nidhish Nerur, PhD Student - University of Texas at Austin; Indranil Bardhan, PhD; Wen Wen, PhD - University of Texas at Austin; Kevin Bozic, M.D., MBA - University of Texas at Austin; Karl Koenig, M.D. - University of Texas at Austin; Prakash Jayakumar, MD PhD - The University of Texas at Austin;

Design and Requirements Analysis of a Conversational Agent for Mental Health: A Qualitative Study Based on Interviews with Patients and Therapists

Poster Number: 106
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Artificial Intelligence, Patient Engagement and Preferences, Qualitative Methods, User-centered Design Methods
Programmatic Theme: Clinical Research Informatics

To design and specify the functional and non-functional requirements for the development of a conversational agent focused on improving mental health. Our design is qualitative. The analysis enabled the identification of a structured set of functional requirements differentiated by role (Patient and Therapist), oriented toward basic emotional support, psychoeducation, monitoring, and the activation of crisis workflows, as well as non-functional requirements related to usability, privacy, informed consent, safety, reliability, and prevention of dependency.

Speaker(s):
David Villarreal-Zegarra, MPH
Department of Biomedical Informatics, University of Utah

Author(s):
David Villarreal-Zegarra, MPH - Universidad Científica del Sur; Irvin Dongo, PhD - Electrical and Electronics Engineering Department, Universidad Católica San Pablo; Milagros Isela Cahuana Cuentas, MSc - Psychology Department, Universidad Católica San Pablo; Yscenia Paredes-Gonzales, MSc - Digital Health Research Center;

A Mixed Methods Evaluation of Ambient Artificial Intelligence Scribes in Diabetes Primary Care: Physician and Patient Experiences

Poster Number: 107
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Artificial Intelligence, Informatics Implementation, Usability, Patient Engagement and Preferences, Human-computer Interaction
Programmatic Theme: Clinical Informatics

Ambient artificial intelligence (AI) scribes are promising to reduce documentation burden and improve care quality, but their impact on type 2 diabetes (T2D) care delivery is unknown. This study leverages a mixed-methods approach analyzing quantitative utilization data from the electronic health record and semi-structured qualitative interviews with physicians and patients. AI scribes were found to be a good fit for T2D care, aligning with physician and patient preferences. Opportunities for future improvements were identified.

Speaker(s):
Aaron Tierney, PhD
Kaiser Permanente Division of Research

Author(s):
Denise Payán, PhD, MPP - University of California, Irvine Joe C. Wen School of Population & Public Health; Mary Reed, DrPH - Kaiser Permanente Division of Research; Vincent Liu, MD, MSc - Kaiser Permanente;

Human–Model Confidence Alignment for Trustworthy Chest X-ray AI

Poster Number: 108
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Artificial Intelligence, Deep Learning, Machine Learning, Patient Safety, Evaluation
Programmatic Theme: Translational Bioinformatics

Deep learning models have achieved strong performance in medical imaging tasks, such as chest X-ray classification.
However, ensuring the trustworthy and reliability of these models remains critical.
In this paper, we report that when image resolution is progressively reduced, some models assign higher confidence to predictions on low-resolution images and lower confidence to predictions on high-resolution images.
This behavior contradicts our intuition and implies unreliable uncertainty estimation and potential random-guessing behavior.
To address this issue, we propose the human–model confidence Alignment Rate (AR), which measures the model's reliability when predicting images of varying resolutions.
We systematically demonstrate this phenomenon across state-of-the-art chest X-ray models using the CheXpert dataset for both multi-label classification and pixel-level segmentation tasks. Our preliminary solution improves overall accuracy on general benchmarks.
These findings highlight a key reliability risk in medical AI and motivate the development of targeted mitigation to enhance clinical trustworthiness.

Speaker(s):
Zheng Li, PhD student
New York Institute of Technology

Author(s):
Zheng Li, PhD student - New York Institute of Technology; Jerry Cheng, Ph.D. - New York Institute of Technology; Helen Gu, PhD - New York Institute of Technology;

A Unified Patient Journey Foundation Model for Cancer: Applications in Outcome Prediction and Data Imputation

Poster Number: 109
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Artificial Intelligence, Large Language Models (LLMs), Machine Learning, Real-World Evidence Generation
Programmatic Theme: Clinical Informatics

The architecture of large language models enables modeling of sequential clinical events across longitudinal patient
journeys. When pretrained on large-scale electronic health record (EHR) data, such models can learn joint event
distributions and generate plausible future patient trajectories. In this study, we present a unified patient journey
foundation model with use cases including predicting future cancer outcomes and imputing information, such as
missing units of measurement (UoM) in EHR data. The model demonstrated strong performance comparable to
statistical machine learning baselines for cancer prediction without task-specific training, and achieved 98% accuracy in UoM imputation, with a weighted F1-score of 0.90. Its versatility enables the applications in forecasting, trajectory simulation and data imputation without additional training.

Speaker(s):
Wilson Lau, PhD
Truveta

Author(s):
Wilson Lau, PhD - Truveta; Ehsan Alipour, MD PhD - Truveta; Youngwon Kim, Ph.D. - Truveta; Sihang Zeng, BS - University of Washington; Anand Oka, PhD - Truveta; Jay Nanduri, MBA - Truveta;

AgentCP: Agentic LLM for Accurate and Calibrated Computational Phenotyping

Poster Number: 111
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Artificial Intelligence, Large Language Models (LLMs), Data Mining, Information Extraction
Programmatic Theme: Clinical Informatics

Computational Phenotyping is hindered by the complexity and verbosity of unstructured data, as well as expensive and time-consuming manual reviews. We introduce AgentCP, a multi-agent system integrating large language models and machine learning tools that simulate clinical chart review workflows to identify disease and subphenotypes. Regarding identifying heart failure, dementias, and their various subtypes, the AgentCP achieves near-human accuracy and robust calibration, outperforming traditional models, while assisting experts and reducing labor costs.

Speaker(s):
Chengxi Zang, PHD
Weill Cornell Medicine

Author(s):
Zhqi Lyu, PhD - WCMC; Haoyang Li, PhD - Weill Cornell Medicine; Fei Wang, PhD - Weill Cornell Medicine; Chengxi Zang, PHD - Weill Cornell Medicine;

Evaluating Generative AI Performance in Personalized Health Plan Generation: Insights from Expert Evaluations

Poster Number: 112
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Artificial Intelligence, Evaluation, Qualitative Methods
Programmatic Theme: Consumer Health Informatics

This study evaluates the performance of generative artificial intelligence in creating personalized one-week exercise and dietary plans for older adults with multimorbidity. Expert evaluations found that the generated plans were well structured and generally safe; however, they exhibited limitations in feasibility, comprehensiveness, personalization, and clinical validity. These results suggest that generative AI should serve as a supplement to, rather than a replacement for, professional clinical judgment.

Speaker(s):
Yuanying Pang, Master
Florida State University

Author(s):
Yuanying Pang, Master - Florida State University; Lynn Panton, PhD - Florida State University; Rayven Nairn, MS - Florida State University; Ravinder Nagpal, PhD - Florida State University; Zhe He, PhD, FIAHSI, FAMIA - Florida State University;

Large Language Model–Based Identification of Post-Stroke Cognitive Impairment from Longitudinal Clinical Notes in Electronic Health Records

Poster Number: 113
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Artificial Intelligence, Chronic Care Management, Large Language Models (LLMs), Workflow, Informatics Implementation, Clinical Decision Support
Programmatic Theme: Clinical Research Informatics

Background:
Post-stroke cognitive impairment (PSCI) is common but frequently underdiagnosed in routine clinical practice. Reliance on structured diagnostic codes may underestimate PSCI because cognitive symptoms are often documented in narrative clinical notes rather than as coded diagnoses.
Objective:
To evaluate whether a large language model (LLM)–based approach improves identification of PSCI from clinical notes compared with ICD code–based detection.
Methods:
We conducted a retrospective study of patients with ischemic stroke within a large academic health system. PSCI detection was performed using ICD codes and an LLM-based agent applied to longitudinal clinical notes. Chart review was conducted for random 100 patients to establish reference-standard PSCI status.
Results:
Among 3,048 stroke patients, ICD codes identified cognitive impairment in 1,006 patients (33.0%), whereas the LLM identified 2,114 patients (69.4%). Compared with chart review, the LLM approach showed higher sensitivity (98.3% vs 52.5%) but lower specificity (43.9% vs 73.2%) (p = 0.040).
Conclusions:
LLM-based analysis of clinical notes substantially improves detection of PSCI and may serve as a scalable EHR-based screening approach to identify stroke survivors who warrant further cognitive evaluation.

Speaker(s):
Yuhua Wu, BSN
School of nursing, Emory University

Author(s):
Ali Tfaily, MS - Emory University; Fadi Nahab, MD - Emory University, school of medicine; Jessica Saurman, PhD - Emory University, School of medicine; Ghada Mohamed, MD - Emory University, School of Medicine; Runze Yan, PhD - Emory University, School of Nursing; Xiao Hu, PhD - Emory University, School of Nursing;

AI triage system for head and neck tumors

Poster Number: 114
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Artificial Intelligence, Clinical Decision Support, Documentation Burden, Large Language Models (LLMs)
Programmatic Theme: Clinical Informatics

We developed an AI pipeline to triage external head and neck surgery referrals by integrating Epic requests, external records, OCR, document identification, and LLM-based guideline review to determine surgical eligibility with rationale generation. In 301 retrospective cases, the AI classified 72% as likely surgical, 12% as non-surgical, and 16% for review. All non-surgical classifications had no subsequent surgery, supporting improved triage efficiency and surgical yield within a human-in-the-loop model.

Speaker(s):
Valentina Carducci, MS
Mayo Clinic

Author(s):
Santiago Romero-Brufau, MD, PhD - Mayo Clinic; Doug Snyder, M.D., M.S. - Mayo Clinic; Jill Gruenwald, Au.D. - Mayo Clinic; Daniel Price, M.D. - Mayo Clinic; Valentina Carducci, MS - Mayo Clinic;

Future Design Considerations for AI-Enabled Healthcare Technologies to Address Systemic Barriers in Healthcare Systems: Disabled and Chronically Ill Community Perspectives

Poster Number: 115
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Artificial Intelligence, Disability, Accessibility, and Human Function, User-centered Design Methods
Programmatic Theme: Consumer Health Informatics

This user-centered design study explored how future AI-enabled technologies could better support disabled and chronically ill communities navigating health systems. Qualitative interviews were conducted with N= 25 disabled and/or chronically ill participants and data was analyzed thematically informed by the principles of disability justice. Design considerations for future AI-enabled technologies, included designing to address systemic health system barriers and centering disabled community expertise. Results suggest that community-led design is essential to rebuild trust in healthcare systems, and meaningfully support the health and care engagement of disabled and chronically ill communities.

Speaker(s):
Bradley Iott, MPH, PhD
University of Michigan

Author(s):
Morgan Gray, MSLIS - University of Michigan School of Information; Bradley Iott, MPH, PhD - University of Michigan; Megan Threats, PhD, MSLIS - University of Michigan - Ann Arbor;

ACARE: Agentic AI Reasoning Engine for Cancer Care with Medical Knowledge Grounding

Poster Number: 116
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Artificial Intelligence, Natural Language Processing, Knowledge Representation & Information Modeling, Machine Learning, Clinical Decision Support, Deep Learning
Programmatic Theme: Clinical Research Informatics

Recent advances in large language models (LLMs) show promise for healthcare but remain limited by weak grounding, poor transparency, and unreliable reasoning in complex clinical contexts. We present ACARE, an Agentic Cancer AI Reasoning Engine that supports oncology decision-making through structured agent collaboration, knowledge retrieval, and reasoning. By integrating curated oncology knowledge with verification and reasoning modules, ACARE improves factual consistency and reduces hallucinations. Experiments across cancer-related reasoning tasks demonstrate improved accuracy, robustness, and interpretability.

Speaker(s):
Xiang Li, PhD
Massachusetts General Hospital and Harvard Medical School

Author(s):
Sophia Yunjia Liu, High School - Shanghai American School; Yi Pan, PhD - The University of Georgia; Fang Zeng, PhD - Massachusetts General Hospital; Xiang Li, PhD - Massachusetts General Hospital and Harvard Medical School;

ACES: Anchored Coordination for Reliable Annotation and Effective Semantics in Clinical Narratives

Poster Number: 117
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Artificial Intelligence, Clinical Decision Support, Natural Language Processing
Programmatic Theme: Clinical Research Informatics

Clinical narrative annotation is often inconsistent due to ambiguous span boundaries and label interpretations. We introduce ACES, a framework that anchors annotation with large language model (LLM) suggestions and derives empirical label semantics using predicate-based pattern discovery. In a study of 25 motivational interviewing transcripts, ACES improved span-level agreement from 0.52 to 0.76 F1. Human evaluation across 102 predicates confirmed that discovered patterns accurately represent recurring semantic structures in patient language.

Speaker(s):
Hadeel Elyazori, PhD
George Mason University

Author(s):
Kevin Lybarger, PhD - George Mason University; Rusul Abdulrazzaq, Undergraduate Student (Biology) - George Mason University; Ande Leonard Mitchell, Undergraduate - George Mason University; Blessing Nguyen, Undergraduate - George Mason University; Sasanka Sreedevi-Naresh, Undergraduate - George Mason University; Stephanie Elizabeth Krafsig, Undergraduate - George Mason University; Thomas H Rossi, Undergraduate - George Mason University; Jay Shah, MD - National Institutes of Health; Lynn H. Gerber, MD - Inova Fairfax Hospital, Inova Health System; Secili DeStefano, PT, DPT - Optimal Motion Physical Therapy; Siddhartha Sikdar, PhD - George Mason University; Samuel Acuña, PhD - George Mason University;

Essential Artificial Intelligence Competencies for Nursing Education: A Scoping Review

Poster Number: 118
2026 Symposium LIEAF Presentation

Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Artificial Intelligence, Curriculum Development, Large Language Models (LLMs), Teaching Innovation
Programmatic Theme: Academic Informatics / LIEAF

The increasing integration of artificial intelligence (AI) into nursing practices has created a need to update nursing curricula to prepare students at all levels necessary to provide safe and effective care in this AI era. The purpose of this scoping review was to examine the current state of AI competency in nursing education and map the findings to Bloom’s taxonomy of knowledge, skills, and attitudes (KSA) framework. Three databases were used in the search, and 36 articles were included. Abstracted competencies were mapped to KSA domains respectively. Research on this topic is characterized by a strong international presence and suggests that the integration of AI into nursing education requires a comprehensive, multi-dimensional approach across all KSA core domains. To prepare a future-ready workforce, nursing curricula must balance technical skills such as prompt engineering and AI use with the preservation of core nursing values such as ethical integrity and clinical judgment.

Speaker(s):
Grace Gao, PhD, DNP
St Catherine University

Author(s):
Grace Gao, PhD, DNP - St Catherine University; Sarah White, MSN, RN, CNE - St Catherine University; Tobias Baumann, BA - St Catherine University; Alvin Jeffery, PhD, RN - Vanderbilt University Medical Center; Christie Martin, PhD, MPH, RN-BC, LHIT-HP - University of Minnesota School of Nursing; Jenna Marquard, PhD - University of Minnesota; Catherine Graeve, PhD, MPH, CNE, AHN-BC, RN, PHN - St Catherine University;

Small Language Models in Medicine: A Scoping Review of Performance, Efficiency, and Clinical Deployment

Poster Number: 119
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Artificial Intelligence, Large Language Models (LLMs), Diversity, Equity, Inclusion, and Accessibility, Health Equity
Programmatic Theme: Clinical Informatics

Small language models (SLMs) offer efficient alternatives to large language models for healthcare applications, but their clinical readiness remains unclear. We conducted a scoping review of 110 studies evaluating SLMs in medical tasks. While SLMs often achieved competitive benchmark performance, prospective clinical validation, standardized efficiency reporting, ethical considerations (privacy, bias, safety) and deployment in low-resource environments were rare. We propose a roadmap to support safe and scalable clinical integration.

Speaker(s):
Matthew Wong, Medicine
Cambridge University

Author(s):
Matthew Wong, Medicine - Cambridge University; Rui Yang, Master - Duke-NUS Medical School; Huitao Li, Msc - Duke-Nus Medical School; Nan Liu, PhD - National University of Singapore; Yulan He, BASc, MEng, PhD - Kings College London; Lorainne Tudor Car, MD, PhD - Kings College London;

Evaluating Large Language Models’ Confidence on Predicting Glaucoma Progression to Surgery using Free-Text Clinical Notes

Poster Number: 120
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Artificial Intelligence, Large Language Models (LLMs), Evaluation
Programmatic Theme: Clinical Research Informatics

This study’s goal was to benchmark the ability of open-source LLMs in predicting patients’ progression to glaucoma surgery using clinical ophthalmology notes. Utilizing Free-text Clinical Notes, Models were prompted to answer yes/no and to estimate the probability of progression to surgery(extrinsic probability).The probability associated with the generated yes/no token was also extracted (intrinsic probability).Results show LLMs were able to perform modestly on this prognostic task.

Speaker(s):
Jawwad Javeed, N/A
Stanford University School of Medicine

Author(s):
Sophia Wang, MD, MS - Stanford University; Jawwad Javeed, N/A - Stanford School of Medicine;

Comparison and analysis of human experts and LLMs in evaluating and criticizing clinical research hypotheses: a case study

Poster Number: 121
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Artificial Intelligence, Qualitative Methods, Evaluation
Programmatic Theme: Clinical Research Informatics

Hypothesis dictates hypothesis-driven studies. Hypothesis generation and refinement are time-consuming and
demanding expertise and experience. The paper explores whether large language models (LLMs, ChatGPT-5,
Gemini, Claude Opus 4.6) can rate and refine hypotheses and compares their ratings with those of human domain
experts (HDEs) on 10 hypotheses, randomly selected from a human-subject study. Significance, validity, and
feasibility were the three primary dimensions of hypothesis quality. LLMs rated hypotheses show a positive bias
compared to HDEs ratings overall. LLMs did not show consistent results as HDEs ratings, except for Claude Opus
in the validity rating. The three LLMs are highly divergent in their validity, significance, and feasibility ratings.
LLMs demonstrated some helpful traits in refinement, such as making the language read fluently and smoothly.
However, LLMs missed the invalid hypotheses entirely (40%) and generated unsupported statements with confidence,
assertive tones, suggesting that LLMs require users' caution and supervision for high-level use.

Speaker(s):
Xia Jing, MD, PhD
Clemson University

Author(s):
Xia Jing, MD, PhD - Clemson University; Chang Liu, PhD - Ohio University; Yuchun Zhou, PhD - Ohio University; James Wang, PhD - Clemson University;

Seed-Guided Human-in-the-Loop AI Curation of Perioperative Organ Injury Literature: Development and Evaluation

Poster Number: 122
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Artificial Intelligence, Controlled Terminologies, Ontologies, and Vocabularies, Knowledge Representation & Information Modeling, Large Language Models (LLMs), Surgery
Working Group: Clinical Research Informatics Working Group
Programmatic Theme: Clinical Informatics

To describe and evaluate a human-in-the-loop curation workflow for perioperative organ injury (POI) literature, using acute respiratory distress syndrome (ARDS) as the initial use case. The current work focuses on the curation stage that transforms expert interest into structured, machine-readable evidence objects; ontology and knowledge-graph construction are downstream steps and are not the focus of this manuscript.
Methods: We began with 14 expert-curated landmark ARDS trials that exemplified the desired study elements (clinical question, population context, intervention, comparator, outcomes, strengths, limitations, and context qualifiers). These seed summaries informed a reusable curation schema and guided AI-assisted extraction from full-text PDFs and abstracts into structured JSON format. An expanded ARDS literature corpus was retrieved, subjected to AI-based screening and ranking, and prioritized via topic modeling to identify high-value papers for subsequent curation rounds. Human experts intervened at critical checkpoints to validate extractions, resolve contextual ambiguities, and approve candidates for inclusion.
Results: From an initial PubMed-derived ARDS corpus of 15,192 records, the seed-guided prioritization workflow incorporating AI screening, semantic ranking, and topic-based selection yielded a curated set of 103 papers (six per domain across 17 POI-relevant domains), enriched for randomized controlled trials and multicenter designs. Human-in-the-loop corrections ensured accuracy, as demonstrated by iPROVE-OLV (PMID 38065200), where AI output was revised from an erroneous abdominal-surgery reference to the correct thoracic one-lung ventilation context, retaining the precise beneficial effect estimate (RR 0.39, 95% CI 0.28-0.56).
Discussion: This seed-guided approach bridges manual expert curation and scalable AI extraction, preserving domain intent while efficiently generating high-quality structured evidence. It mitigates common AI limitations through targeted human oversight and supports future POI knowledge base expansion.
Conclusion: The validated workflow operationalizes expert priorities, identifies and curates high-value studies, and produces reliable evidence objects as a foundational substrate for subsequent ontology and knowledge-graph development in perioperative organ injury.

Speaker(s):
jinlian wang, PhD
UTHealth

Author(s):
Yafen Liang, MD - McGovern Medical School at UTHealth Houston; Holger Eltzschig, MD. PhD - McGovern Medical School at UTHealth Houston;

N3CFM: Development of Clinical Foundation Model Using N3C

Poster Number: 123
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Artificial Intelligence, Large Language Models (LLMs), Clinical Decision Support
Programmatic Theme: Clinical Research Informatics

Foundation models enable transferable representation learning from large electronic health record (EHR) datasets. We present N3CFM, a clinical foundation model developed within the NIH National COVID Cohort Collaborative (N3C) enclave using multi-institutional EHR data. Transformer-based pretraining on up to 22 million patients produced robust patient representations, achieving AUROC up to 99.1% for long COVID prediction and outperforming baseline machine learning, recurrent neural network, and pretrained embedding approaches.

Speaker(s):
Humayera Islam, PhD
The University of Chicago

Author(s):
Humayera Islam, PhD - The University of Chicago; Pablo Napan Molina, MS - UT Health; Rafael Tinajero Ayala Gonzalez Arce, MS - Rice University; Michael Ghebranious, B.S. - Texas A&M University; Sepideh Sepahi, Master - University of Texas Health Science Center at Houston; Degui Zhi, Ph.D. - The University of Texas Health Science Center at Houston (UTHealth) McWilliams School of Biomedical Informatics; Laila Rasmy, PhD, MSc, MBA, RPh. - UTHealth MSBMI;

A Heterogenous Graph Neural Network Framework for Multi-Cohort Integration and MMSE Prediction in Alzheimer’s Disease

Poster Number: 124
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Artificial Intelligence, Data Mining, Knowledge Representation & Information Modeling
Programmatic Theme: Clinical Research Informatics

Alzheimer’s disease (AD) is a complex neurodegenerative disorder characterized by progressive cognitive decline that ultimately leads to significant loss of independence and quality of life. The Mini-Mental State Examination (MMSE) is a widely used cognitive screening measure in Alzheimer’s disease research and clinical assessment. Accurate prediction of MMSE from multimodal patient data may support disease characterization, cohort stratification, and downstream modeling of cognitive decline. However, predicting cognitive scores from observational cohort data remains challenging because relevant signals are fragmented across heterogeneous modalities such as demographics, genetics, imaging biomarkers, and clinical symptoms. While deep phenotyping research cohorts such as the Alzheimer’s Disease Neuroimaging Initiative (ADNI) provide high-fidelity multimodal measurements, including neuroimaging biomarkers, genetic data, and standardized cognitive assessments such as MMSE, these cohorts are often limited in size and contain relatively sparse information about real-world comorbidities and clinical conditions. Conversely, large-scale electronic health record (EHR)–based cohorts such as the All of Us Research Program capture extensive information about diagnoses, comorbidities, medications, and diverse patient populations, but they generally lack detailed neuroimaging measurements and standardized cognitive scores. Consequently, each cohort provides only a partial view of the disease process, motivating the development of methods that can integrate complementary information across cohorts. Here we propose a heterogeneous graph neural network framework for multi-cohort integration and MMSE prediction in Alzheimer’s disease. Our approach leverages biomedical knowledge graphs (BKG) in two main ways. First, we use the BKG as a bridge to connect disparate cohorts by linking patients through shared BKG entities. Second, the integration of structured biomedical knowledge provides a critical biological prior that transcends individual patient-level features. By capturing curated relationships among genes, symptoms, and pathways, the KG enriches patient representations with domain knowledge derived from large-scale biomedical literature.

Speaker(s):
zhe huang, PhD
Weill Cornell

Author(s):
Zuoyu Yan, Ph.D. - Weill Cornell Medicine; Chang Su, PhD - Weill Cornell Medicine; Fei Wang, PhD - Weill Cornell Medicine;

Enhancing Laboratory Data Harmonization with a Multistep NLP Approach Leveraging Test Result Information

Poster Number: 125
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Artificial Intelligence, Informatics Implementation, Natural Language Processing, Interoperability and Health Information Exchange
Programmatic Theme: Clinical Informatics

Laboratory clinical data provides insights on patient’s current state and evolution, making it critically valuable for AI model training and clinical research. However, harmonization into terminologies such as LOINC remains a non-trivial task. This work develops a methodology that leverages on both laboratory test description and results metadata (units of measure and type of result) to standardize tests into LOINC. The approach implemented combines different steps: entity annotation into UMLS Concept Unique Identifiers (ScispaCy), specimen and component UMLS synonyms, and property pipeline based on example LOINC units. Results indicated a 12% increase in accuracy. When considering laboratory results information, Top-5 accuracy increases to 54%. Scalability of the methodology was assessed using local catalogues from more than 128 different healthcare organizations, showing 30% for the Top-5 accuracy in a random 500 test subset. The combination of different NLP approaches, to handle the laboratory test elements, provide a step towards data interoperability.

Speaker(s):
Matvey Palchuk, MD, MS, FAMIA
TriNetX, LLC.

Author(s):
Aída Muñoz Monjas, MSc - Universidad Politécnica de Madrid; David Pérez del Rey, PhD - Universidad Politécnica de Madrid; Matvey Palchuk, MD, MS, FAMIA - TriNetX, LLC.; John Doole, Pharm. D., MFA - TriNetX, LLC;

AI-Driven Case Simulation as Learning Infrastructure to Identify Educational Gaps in EMS Training

Poster Number: 126
2026 Symposium LIEAF Presentation

Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Artificial Intelligence, Clinical Decision Support, Qualitative Methods, Evaluation, Human-computer Interaction, Real-World Evidence Generation
Programmatic Theme: Academic Informatics / LIEAF

EMS training programs have limited tools to systematically evaluate how learners make clinical decisions across diverse scenarios outside formal training environments. We present an AI-driven case simulation platform adapted to function as learning infrastructure by capturing structured learner interaction signals during EMS scenarios. Analysis of these signals reconstructs decision pathways and maps behaviors to competency domains, which allows for identification of recurring performance gaps and informing targeted, data-driven improvements in EMS curricula.

Speaker(s):
Annika Coleman, BS
Vanderbilt University

Author(s):
Logan Pasquariello, Student - Vanderbilt University; Annika Coleman, BS - Vanderbilt University; Carly Eckert, MD, PhD, MPH - Avante; Dana Knueven, MPH, PM-CC, I/C - Vanderbilt University Medical Center; Kevin Sexton, MD - Vanderbilt University Medical Center (VUMC);

LLM4Rules: A Framework for Scalable LLM-Guided Symbolic Reinforcement Learning

Poster Number: 127
Presentation Time: 05:00 PM - 06:30 PM

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

Rule-based clinical natural language processing (NLP) systems remain widely used for extracting clinical concepts from electronic health record (EHR) narratives, but maintaining rule resources requires extensive manual error analysis and rule refinement. This study investigates whether large language models (LLMs) can assist in identifying extraction errors and generating candidate rules to improve rule-based clinical NLP systems. Using error reports derived from a multi-site evaluation of a previously validated rule-based model for cognitive and neuropsychiatric-related clinical concepts, we developed a human-in-the-loop framework, LLM4Rule. The framework first uses LLMs to classify extraction errors and generate explanatory reasoning, which can then be incorporated into prompts for rule generation. Three LLMs (GPT-5.2, GPT-4o, GPT-4o-mini) were evaluated under four prompting conditions. LLM-generated rule sets improved performance compared with the baseline NLP-CAM system, increasing F1-score from 0.37 to 0.58. These findings suggest that LLMs can support scalable rule refinement for rule-based clinical NLP systems.

Speaker(s):
Nan Wang, Graduate Student
UTH

Author(s):
Rongrong Wang, Master of Science in Health Informatics - UTHealth Houstom; Jaerong Ahn, PhD - UT Health of Houston; Min Ji Kwak, MD, DrPH - McGovern Medical School; Sunyang Fu, PhD, MHI - UTHealth Houston;

When Temperature Zero Isn't Enough: Factual Inconsistency of an Agentic Retrieval-Augmented Generation (RAG) System in Patient Chart Review

Poster Number: 128
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Artificial Intelligence, Large Language Models (LLMs), Evaluation
Programmatic Theme: Clinical Research Informatics

We evaluated the session to session factual consistency of an LLM powered agentic RAG system for melanoma chart review. Five canonical clinical questions were run in 10 independent sessions per patient for 10 patients. Despite tested at temperature-zero, the system frequently produced multiple distinct facts for the same patient–question pair. Log analysis revealed this non determinism stems from autonomous query conflation and lexical abstraction. These findings highlight critical reproducibility and safety limits, making human-in-the-loop validation essential.

Speaker(s):
Heekyong Park, PhD
Mass General Brigham

Author(s):
Nich Wattanasin, MS - Mass General Brigham; Martin Rees, BS - Mass General Brigham; Yichuan Grace Hsieh, Ph.D. - Mass General Brigham; Marykate E Murphy, RN - Massachusetts General Hospital; Janet Boyle-Kelly, RN - Massachusetts General Hospital; Gregory Belsky, BS - Mass General Brigham; Tran Le, PhD - Mass General Brigham; Wendy Bossons, MS - Software Engineer II; Bhaswati Ghosh, MS - Mass General Brigham; Allan J. Harris, BS - Mass General Brigham; Thomas McShane, MS - Mass General Brigham; Christopher D. Herrick, MBA - Mass General Brigham; Shawn Murphy, MD, Ph.D. - Massachusetts General Hospital;

Enhancing Student Learning and Engagement with AI Tutors in Health Informatics Higher Education

Poster Number: 129
2026 Symposium LIEAF Presentation

Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Artificial Intelligence, Curriculum Development, Evaluation, Teaching Innovation
Programmatic Theme: Academic Informatics / LIEAF

Artificial Intelligent (AI) tutors are increasingly used to support student learning. This study evaluates the impact of AI-assisted learning activities in a graduate-level Health Informatics course, Data Mining in Healthcare. An AI-assisted course section (n = 19) was compared with similar course sections without AI tutors (n = 12) to evaluate student performance. The AI-assisted cohort improved from a pre-AI baseline mean of 91.3% to a post-AI mean of 97.82%. Compared with control groups (mean = 90.5%), the AI cohort achieved higher scores and lower variability. Statistical testing indicated a significant improvement (t(18) = 9.73, p < .000001) and a significant difference between the AI and control groups (t ≈ 2.97, p ≈ .012) with a large effect size (Cohen’s d ≈ 1.18). These findings suggest that integrating AI tutors into the health informatics curriculum may enhance learning outcomes and promote more consistent student performance.

Speaker(s):
Abdul Hafeez, PhD
George Mason University

Author(s):
Abdul Hafeez, PhD - George Mason University; Sanja Avramovic, PhD - George Mason University;

From Clinical Notes to Patient Understanding: An Evidence-Grounded Agentic System for Explaining Medical Concepts in EHR Narratives

Poster Number: 130
Presentation Time: 05:00 PM - 06:30 PM

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

This study illustrates a memory-augmented agentic system that identifies medically complex concepts in clinical narratives and generates patient-oriented explanations grounded in biomedical evidence. Unlike prior work focused on simplifying entire notes, our framework performs concept-level reasoning, retrieves supporting knowledge from literature and clinical guidelines, and validates explanations for consistency. Experiments using the EHRCon benchmark demonstrate accurate concept extraction and transparent, evidence-supported explanations for patient-facing interpretation of EHR information.

Speaker(s):
Yuqi Wu, PhD, MPH, MPA
University of North Florida

Author(s):
Yuqi Wu, PhD, MPH, MPA - University of North Florida; Hanadi Hamadi, PhD - University of North Florida;

What Should the AI Era Doctor Know? A Scoping Review of Proposed Artificial Intelligence Competencies for Medical Education

Poster Number: 131
2026 Symposium LIEAF Presentation

Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Artificial Intelligence, Curriculum Development, Knowledge Representation & Information Modeling
Programmatic Theme: Academic Informatics / LIEAF

Artificial intelligence is reshaping clinical practice, yet AI training in undergraduate medical education is fragmented. We conducted a scoping review (PubMed, Embase, Web of Science, ERIC; inception–July 28, 2025) to synthesize proposed AI competencies for medical students. Fifty-four studies from 22 countries yielded 564 statements, consolidated into a 7-domain framework spanning 37 competencies and 170 learning objectives. Most sources were recent and propositional, highlighting the need for consensus.

Speaker(s):
Victor Hunt, MSc
The Warren Alpert Medical School of Brown University

Author(s):
Victor Hunt, MSc - The Warren Alpert Medical School of Brown University; Laurine Sprehe, PhD Candiate - Massachusetts General Hospital; Weston de Lomba, MD Candidate - The Warren Alpert Medical School of Brown University; Rodrigo Gameiro, MD, MPH, MMSc - Department of Biomedical Informatics, Harvard Medical School; Naira Woite, MD, MPH - Department of Internal Medicine, Yale School of Medicine; Carrie Wade, MLIS - Countway Library, Harvard Medical School; Jonathan Chen, MD, PhD - Stanford University Hospital; Steven Rougas, MD, MSc - The Warren Alpert Medical School of Brown University; Hamish Fraser, MBChB, MRCP, MSc - Brown University; Amy Sullivan, EdD - Beth Israel Deaconess Medical Center Shapiro Institute Center for Education; Felipe Fregni, MEd, MPH, MMSc, MD, PhD - Harvard T.H. Chan School of Public Health;

Concordance Between Large Language Models and Transplant Surgeons in Deceased-Donor Kidney Evaluation

Poster Number: 132
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Artificial Intelligence, Large Language Models (LLMs), Surgery, Clinical Decision Support, Evaluation
Programmatic Theme: Clinical Research Informatics

Deceased-donor kidney evaluation requires the rapid consideration of complex donor and organ data, yet organ discard rates remain high and criteria for acceptance vary. We evaluated the concordance between accept/decline recommendations by large language models (LLMs) and transplant surgeons’ using historical decisions on 761 kidney offers made to Rhode Island Hospital between 2023 and 2025. Agreement between LLMs and surgeons was limited across trials (Cohen's κ ≤ 0.039). These findings highlight important differences between LLM and clinician decision-making and inform future development of AI decision-support tools in transplantation.

Speaker(s):
Victor Hunt, MSc
The Warren Alpert Medical School of Brown University

Author(s):
Victor Hunt, MSc - The Warren Alpert Medical School of Brown University; Laurine Sprehe, PhD Candidate - Massachusetts General Hospital; Nikolas Montaquila, MD Candidate - Warren Alpert Medical School of Brown University; Adena Obsand, MD - Brown University Health; Hamish Fraser, MBChB, MRCP, MSc - Brown University; Paul Morrissey, MD - Brown University Health; Dicken Ko, MD - Brown University Health;

Question-Aware Multimodal Fusion for Detecting Depression in Older Adults with Mild Cognitive Impairment from Remote Interviews

Poster Number: 133
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Artificial Intelligence, Clinical Decision Support, Deep Learning, Machine Learning, Telemedicine
Programmatic Theme: Clinical Informatics

Depression affects 32% of older adults with mild cognitive impairment (MCI) and accelerates cognitive decline. Existing multimodal approaches for depression detection in MCI often ignore the interviewer-question context and fine-grained temporal multimodal interactions. We propose a question-aware multimodal fusion approach that models temporal interactions among facial, acoustic, language, and non-contact heart rate signals from remote MCI interviews. The model achieved an F1 score of 0.75, outperforming a baseline without question context or multimodal interaction modeling.

Speaker(s):
Merna Bibars, PhD.
Georgia Institute of Technology

Author(s):
Merna Bibars, PhD. - Georgia Institute of Technology; Bolaji Omofojoye, MS. - Emory University; Allan Levey, MD, PhD - Emory University; Rachel Hershenberg, PhD - Emory University; Gari Clifford, DPhil - Emory University - BMI; Hyeokhyen Kwon, Ph.D. - Emory University;

From Narrative Guideline to Machine-Readable Eligibility Criteria: ChatGPT-Assisted Extraction of Germline Testing Recommendations from an ASCO-SSO Breast Cancer Guideline

Poster Number: 134
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Artificial Intelligence, Clinical Decision Support, Large Language Models (LLMs), Information Extraction, Documentation Burden
Programmatic Theme: Clinical Research Informatics

This study describes a prompt-engineered workflow using ChatGPT 5.2 to identify and extract germline testing eligibility recommendations from the ASCO-SSO breast cancer guideline into a structured JSON schema suitable for clinical decision support. Extracted recommendation sections were compared to a manually annotated gold standard, and performance was evaluated at the recommendation-section level using precision and recall to assess extraction accuracy.

Speaker(s):
Leo Alvarez, Biomedical Informatics
University of Utah

Author(s):
Jiantao Bian, PhD - Biomedical Informatics Department, University of Utah; Anne Madeo, Master of Sciences in Human Genetics - Biomedical Informatics Department, University of Utah; Richard Bradshaw, MS - University of Utah Health Sciences; Wendy Kohlmann, Master's Degree in Genetic Counseling - University of Utah; Kensaku Kawamoto, MD, PhD, MHS - University of Utah; Caitlin Allen, PhD, MPH - MUSC; Guilherme Del Fiol, MD, PhD - University of Utah; Ravi Sharaf, MD, MS, BS - Weill Cornell Medicine;

Multimodal AI to Predict Persistent Fever After Antibiotics in Hospitalized Cancer Patients

Poster Number: 135
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Artificial Intelligence, Clinical Decision Support, Large Language Models (LLMs), Machine Learning
Programmatic Theme: Clinical Research Informatics

Persistent fever after antibiotic initiation is common in hospitalized cancer patients and often prompts antimicrobial escalation. We developed and validated an early-fusion multimodal AI model to predict persistent fever at the 48–72-hour reassessment window using structured clinical data, time-series forecasts, note-derived phenotypes, and CT features. The model showed strong internal performance (AUC 0.83, 95% CI 0.81–0.86), maintained external performance (MIMIC-IV AUC 0.79, 95% CI 0.74–0.85), and outperformed clinicians in vignette-based evaluation.

Speaker(s):
Gernot Pucher, MSc
University Medicine Essen

Author(s):
Gernot Pucher, MSc - University Medicine Essen; Kevin Kopp, Data Science - University Hospital Essen; Aman Deep, PhD - University Medicine Essen; Anna-Maria Stark, MD Student - University Medicine Essen; Martin Schuler, MD - University Medicine Essen; Felix Nensa, MD - University Medicine Essen; Christian H. Reinhardt, MD - University Medicine Essen; Jens Kleesiek, MD PhD - University Medicine Essen; Christopher M. Sauer, MD PhD - University Medicine Essen;

Anatomy AI Education: A Retrieval-Augmented Generation System for Interactive Anatomy Education

Poster Number: 136
2026 Symposium LIEAF Presentation

Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Artificial Intelligence, Curriculum Development, Large Language Models (LLMs)
Programmatic Theme: Academic Informatics / LIEAF

Anatomy AI is an intelligent conversational assistant using Retrieval-Augmented Generation (RAG) to support human anatomy education at the University of Pittsburgh. Integrating LangChain, GPT-4, and Pinecone, the system delivers accurate, course-grounded responses across question answering, video timestamp localization, and quiz generation via a unified prompt architecture that eliminates intent classification.

Speaker(s):
Yanshan Wang, PhD
University of Pittsburgh

Author(s):
Mariano De Leon, MS - University of Pittsburgh; Reivian Berrios Barillas, DPT - University of Pittsburgh; Yanshan Wang, PhD - University of Pittsburgh;

Using LLMs to Analyze Medical Malpractice Cases at Vanderbilt University Medical Center

Poster Number: 137
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Artificial Intelligence, Clinical Decision Support, Large Language Models (LLMs), Healthcare Quality, Patient Safety
Programmatic Theme: Clinical Research Informatics

Medical malpractice represents a significant challenge for healthcare systems. This study leveraged ChatGPT (model: GPT-4.1-mini) to analyze over 1,200 malpractice case files at Vanderbilt University Medical Center, categorizing complaints and rating the effectiveness of different strategies in addressing the issue. Results indicate that almost 70% of cases could have been prevented through EHR-based interventions, with Clinical Decision Support Systems alone reducing medical malpractice incidents by 65%.

Speaker(s):
Eric Chen, High School Student
Montgomery Bell Academy

Author(s):
Adam Wright, PhD - Vanderbilt University Medical Center; Eric Chen, High School Student - Montgomery Bell Academy;

Evaluation of Ambient and Artificial Intelligence on Documentation Burden of Emergency Department Providers

Poster Number: 138
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Artificial Intelligence, Documentation Burden, Human-computer Interaction
Programmatic Theme: Clinical Informatics

This study evaluated the impact of generative AI–supported ambient documentation on emergency department residents. Twenty residents completed surveys before and after implementation. Ambient documentation led to a 60% reduction in perceived documentation burden, 2.5 fewer after shift documentation hours weekly, and a 43% decrease in burnout scores. EHR data showed a 20% faster time to note completion. AI-assisted documentation improved efficiency and reduced resident workload.

Speaker(s):
Khoa Tang, DO
Geisinger

Author(s):
Kyle Marshall, MD, MBI, FACEP, FAAEM, FAMIA - Geisinger; Nicholas Maffetone, DO - Geisinger;

Do Vision Language Models Enhance Medical Diagnostic Process?: A Systematic Review

Poster Number: 139
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Artificial Intelligence, Clinical Decision Support, Machine Learning
Programmatic Theme: Clinical Informatics

Novel vision-language models (VLMs) emulate the diagnostic reasoning of a physician by integrating textual patient history with visual imaging data. This systematic review evaluates diagnostic VLMs across diverse medical domains, focusing on data sources, model architectures, performance, and limitations. Results show that multimodal VLMs outperform models relying solely on text or image data. Although direct comparisons between VLMs and human performance remain conflicting, physicians assisted by VLMs achieved higher accuracy than unassisted physicians.

Speaker(s):
Lattawat Eauchai, MD
Mayo Clinic

Author(s):
Laura Otalora Gonzalez, M.D. - Mayo Clinic; Yifan Shi, - - Mayo Clinic; Michele McGinnis, M.S. - Mayo Clinic; Alexandor Yovchev, - - Mayo Clinic; Brian Pickering, MD, FFARCSI - Mayo Clinic; Vitaly Herasevich, MD, PhD, FCCM, FAMIA - Mayo Clinic;

Design and Evaluation of a Model-Based Automated Feedback Agent Across Academic Domains

Poster Number: 140
2026 Symposium LIEAF Presentation

Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Artificial Intelligence, Teaching Innovation, Evaluation
Programmatic Theme: Academic Informatics / LIEAF

Providing consistent, high-quality feedback remains a challenge across academic disciplines. We developed and evaluated a model-based automated feedback agent. Across 50 synthetic writing samples evaluated by expert faculty, the agent generated useful, structured feedback in most domains, while highlighting areas requiring domain-specific refinement.

Speaker(s):
Jeeyae Choi, PhD
University of North Carolina Wilmington

Author(s):
Donald Crawford, DNP - University of North Carolina Wilmington; Sang Teck Oh, PhD - University of North Carolina Wilmington; Xuemei Chen, PhD - University of North Carolina Wilmington; Barbara Carlson, PhD - University of North Carolina Wilmington;

USING AMBIENT LISTENING TO IMPROVE USE OF EHR EFFCIENCY AND DOCUMENTATION

Poster Number: 141
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Artificial Intelligence, Documentation Burden, Usability
Programmatic Theme: Clinical Informatics

Physicians face increasing burnout due to the documentation demands of electronic health records (EHR). Generative AI, particularly Ambient AI, offers innovative solutions to enhance patient care and streamline operations. This poster presents findings from a six-month pilot program using Ambient AI technology to reduce documentation burden. Key metrics showed improvements in time spent on notes and system usage, pajamas time reduction , increase in number of patient seen though some trends remained flat. Expanding AI-powered tools may alleviate physician burnout and improve efficiency in healthcare settings and increase physician efficiency

Speaker(s):
Olasunkanmi Adeyinka, MD
ut health

Author(s):
Lindy Anderson-Papke, MHA - UTHealth Houston; julie bortolotti, md - uthealth;

Youth Use and Trust in Commercial Generative Artificial Intelligence Tools for Health

Poster Number: 142
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Artificial Intelligence, Large Language Models (LLMs), Surveys and Needs Analysis, Delivering Health Information and Knowledge to the Public, Patient Engagement and Preferences, Health Equity
Programmatic Theme: Consumer Health Informatics

Commercial generative AI is increasingly accessible, yet little is known about youth engagement with these tools for health. We surveyed U.S. youth aged 14–24 to characterize use and trust. Nearly half reported using AI for health, driven by convenience, though most expressed limited trust in AI-generated advice. Hispanic, non-Hispanic Black, non-Hispanic Asian, and rural youth were significantly more likely to do so, suggesting underserved populations may turn to these tools to fill gaps in care access.

Speaker(s):
Karan Desai, BS
University of Michigan Medical School

Author(s):
Karan Desai, BS - University of Michigan Medical School; Natalie Guzman, BA - Univeristy of Michigan; Tony Chang, MS - University of Michigan Medical School; Lydia Perry, BS - University of Michigan Medical School; Abigail Kappelman, MA - University of Michigan Medical School; Marika Waselewski, MPH - University of Michigan Department of Family Medicine; Tammy Chang, MD, MPH, MS - University of Michigan Department of Family Medicine; Andrew Wong, MD, MS - University of Michigan;

Undergraduate Student Usage of Generative AI Chatbots for Health Advice

Poster Number: 143
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Artificial Intelligence, Delivering Health Information and Knowledge to the Public, Surveys and Needs Analysis, Qualitative Methods, Quantitative Methods
Programmatic Theme: Consumer Health Informatics

This mixed-methods study examines how and why undergraduates use generative AI chatbots for health advice. Using national Healthy Minds Study data and qualitative survey data from Vanderbilt students, we found that GenAI use was associated with perceived need and barriers to traditional care, while convenience and accessibility drove adoption. Findings highlight how student health information-seeking behaviors are shifting and suggest implications for campus health resources and informatics design.

Speaker(s):
Sarah Zeng, B.A
Vanderbilt University

Author(s):
Sarah Zeng, B.A - Vanderbilt University; Jonathan Metzl, MD, PhD - Vanderbilt University; Jessica Ancker, MPH, PhD, FACMI - Vanderbilt University Medical Center;

Cost effectiveness of AI-based Interventions in Mental Health: A Systematic Review

Poster Number: 144
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Artificial Intelligence, Policy, Real-World Evidence Generation, Usability, Healthcare Quality, Informatics Implementation, Global Health, Diversity, Equity, Inclusion, and Accessibility
Programmatic Theme: Public Health Informatics

This systematic review included 13 studies evaluating cost-effectiveness of AI-based mental health interventions. Most reported ICERs below willingness-to-pay thresholds using QALYs as primary outcomes, but relied on static Markov models and short time horizons, potentially overestimating benefits by not capturing AI's adaptive learning capacity. Non-medical and indirect costs were frequently underreported. A comprehensive cost component framework was synthesized to guide policymakers and healthcare institutions in resource allocation when implementing AI-based mental health interventions.

Speaker(s):
Lidan Tian, MSC
the hong kong polytechnic university

Author(s):
Lidan Tian, MSc - School of Nursing, The Hong Kong Polytechnic University; Xinyu Feng, MSN - School of Nursing, The Hong Kong Polytechnic University; Jinxiao Lian, PhD - School of Optometry, The Hong Kong Polytechnic University, Hong Kong; Grace Wing Ka Ho, RN, PhD - School of Nursing, The Hong Kong Polytechnic University; Vivian Hui, RN, PhD - School of Nursing, The Hong Kong Polytechnic University & School of Nursing, University of Pittsburgh;

Developing a Dual-Agentic AI Framework for Automated Evidence Synthesis in ADRD Research

Poster Number: 145
Presentation Time: 05:00 PM - 06:30 PM

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

The rapid growth of Alzheimer's Disease and Related Dementias (ADRD) literature has outpaced traditional manual review methods. This study presents a dual-agent AI framework decoupling data retrieval from domain-specific classification using LangChain, LangGraph, and OpenAI function-calling. The system achieved a 47.7x speed improvement over human reviewers in screening ADRD literature. While precision and recall remain areas for improvement, findings support integrating a third Decision Agent to enhance autonomous validation and approach human-level accuracy.

Speaker(s):
Duo Wei, PhD, FAMIA
Stockton University

Author(s):
Tasnim Raisa, BS - Stockton University; Riya Goyal, BS - New Jersey Institute of Technology;

Evaluating the Clinical Reliability of AI Prediction Models in Hemophilia: A Systematic Methodological Audit

Poster Number: 146
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Artificial Intelligence, Clinical Decision Support, Machine Learning, Qualitative Methods, Patient Safety
Programmatic Theme: Clinical Research Informatics

Artificial intelligence–based predictive models are increasingly explored for risk prediction and treatment optimization in hemophilia care. We conducted an algorithmic audit of published AI prediction models (2015–2025) using structured evaluation frameworks (CHARMS) and risk-of-bias assessment (PROBAST). Across 11 studies describing 20 predictive models, traditional machine learning accounted for 55%, deep learning 27%, and graph neural networks 18%. Only 18% reported external validation and none assessed calibration, highlighting methodological gaps affecting clinical reliability.

Speaker(s):
Pravallika Manchu, Masters
Grand Valley State University

Author(s):
Pravallika Manchu, Masters - Grand Valley State University; Swarna Bharathi Kathi, Masters of Science in Health and Bioinformatics, Bachelors in Dental Surgery - Grand Valley State University; Suhila Sawesi, PhD - GVSU;

Agentic AI for Smart Analytical Visualization of Clinical Studies Using Graph Data Models on Research Data Commons

Poster Number: 147
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Artificial Intelligence, Large Language Models (LLMs), Information Visualization, Information Retrieval, Knowledge Representation & Information Modeling
Programmatic Theme: Clinical Research Informatics

Large language models can support data analysis but often misinterpret the structural logic of clinical studies. We introduce an agentic AI framework that organizes visualization generation into coordinated planning, code generation, and execution stages. Interpretation and reasoning agents transform user queries and study metadata into analytical plans, which are translated into graph queries and executed through a LangGraph workflow. Experiments on ARDaC and IPO datasets demonstrate protocol-aware visualization generation with transparent and auditable analytical workflows.

Speaker(s):
Hao WANG, PhD Student
Purdue University

Author(s):
Hao Wang, PhD Student - Purdue University; Haining Wang, PhD - Indiana University; Yingjie Chen, PhD - Purdue University; Wanzhu Tu, PhD - Indiana University; Jing Su, PhD - Indiana University School of Medicine;

Designing a Pragmatic Randomized Controlled Trial for Ambient Listening in Nursing Through a Learning Health System

Poster Number: 148
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Artificial Intelligence, Evaluation, Documentation Burden, Governance, Informatics Implementation, Real-World Evidence Generation
Working Group: Nursing Informatics Working Group
Programmatic Theme: Clinical Informatics

A health system and university Learning Health System (LHS) team partnered to plan and design a pragmatic EHR-embedded stepped wedge, cluster randomized controlled trial for a hospital-wide ambient listening deployment in inpatient nursing. Integrated LHS team members contributed to development of rollout strategies, governance structures, and evaluation metrics.

Speaker(s):
Ann Wieben, PhD, RN NI-BC FAMIA
University of Wisconsin-Madison School of Nursing

Author(s):
Jann Pfaff, PhD RN - UW Hospitals and Clinics; Mary Ryan Bauman, PhD - University of Wisconsin Madison; Felice Resnik, PhD - University of Wisconsin Madison; Sarah Brzozowski, PhD RN - UW Hospitals and Clinics; Chelsey Langer, MSN - UW Hospitals and Clinics; Anne Gravel Sullivan, PhD - University of Wisconsin Madison; Caitlin Voegele, MBA - University of Wisconsin Madison; Leigh Ann Mrotek, PhD - University of Wisconsin Madison; Majid Afshar, MD, MSCR - University of Wisconsin - Madison; Elizabeth Burnside, MD - University of Wisconsin-Madison; Jenn Hankwitz, MSN - UW Hospitals and Clinics; Stacy Rasmussen, BS - UW Hospitals and Clinics; Becky Kohler, MPH RN - UW Hospitals and Clinics;

Artificial Intelligence in Medication Management in Critical Care: Where Do We Stand?

Poster Number: 149
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Artificial Intelligence, Critical Care, Clinical Decision Support, Transitions of Care
Programmatic Theme: Clinical Informatics

Artificial intelligence (AI) methods are increasingly explored to support medication dosing and management in intensive care units (ICUs), where physiological variability complicates treatment. We conducted a systematic scoping review (2011–2025) across PubMed, Embase, and Cochrane, identifying 47 studies applying AI to ICU medication management. Most studies were retrospective and focused on antibiotics. Despite promising model performance, few studies reported external validation or clinical deployment, highlighting gaps in translating AI models into real-world ICU practice.

Speaker(s):
Dhruvin Patel, Computer Science
Loyola University Chicago

Author(s):
Dhruvin Patel, Computer Science - Loyola University Chicago; Nazanin Azarvash, MD - Stritch School of Medicine, Loyola University Chicago; Danica Quickfall3, MD - Department of General Internal Medicine, Mayo Clinic, Arizona; Horace Rhodes Hambrick, MD, MS - Northwestern University Feinberg School of Medicine; kianoush B Kashani, MD, MS - Department of Nephrology and Hypertension, Mayo Clinic, Minnesota; Samie Tootooni, PhD - Loyola University Chicago;

LangGraph-Orchestrated Agentic AI Scribes for Home Health Documentation: System Architecture and Fact-Level Evaluation

Poster Number: 150
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Artificial Intelligence, Documentation Burden, Evaluation, Large Language Models (LLMs), Information Extraction
Programmatic Theme: Clinical Informatics

Patient-clinician conversations during home health care (HHC) visits are long and information-dense. Single-pass LLM scribes can blur evidence vs inference, causing errors. We developed a multi-step LangGraph-orchestrated SOAP-note generation agentic pipeline and evaluated on 15 HHC recordings using 428 atomic facts (gold-dataset). The pipeline was run with two model configurations (GPT-5 and GPT-5.1). Both the configurations exhibited a precision-recall tradeoff: former achieved higher precision (~94%), later achieved higher recall (~97%) with hallucination rate (~11%).

Speaker(s):
Pallavi Gupta, PhD
Columbia University

Author(s):
Pallavi Gupta, PhD - Columbia University; Zhihong Zhang, PhD - Columbia University; William Ho, BS, MS - Columbia University; Yu-Wen Chen, BS, MS - Columbia University; Sasha Vergez, B.S. in Biology - VNS Health; Margaret McDonald - Visiting Nurse Service of New York; Zoran Kostic, PhD - Columbia University; Julia Hirschberg, PhD - Columbia University; Max Topaz, PhD, RN, MA, FAAN, FIAHSI, FACMI - Columbia University;

Adjusting Queries Can Optimize the Cost-Utility Tradeoff in LLMs

Poster Number: 151
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Artificial Intelligence, Large Language Models (LLMs), Evaluation
Programmatic Theme: Clinical Informatics

Commercial large language models (LLMs) charge based on input/output length and LLM type. Since healthcare organizations have limited budgets, it is critical to optimize cost and utility of LLMs for long-term sustainability. We implement multi-objective optimization to determine if varying how much information provided to an LLM can improve performance at lower costs. Our results suggest mixed prompting strategies can yield equal or higher performance at a lower cost than a single prompting strategy.

Speaker(s):
Katherine Brown, PhD
Vanderbilt University Medical Center

Author(s):
Katherine Brown, PhD - Vanderbilt University Medical Center; Jinsong Liu, PhD - Weill Cornell Medicine College; Nüma Molière, BS - Weill Cornell Medicine; Sarah Ayton, PhD, MPH - Weill Cornell Medicine; Yevgeniy Vorobeychik, PhD - Washington University; Murat Kantarcioglu, Ph.D. - Virginia Tech; Jessica Ancker, MPH, PhD, FACMI - Vanderbilt University Medical Center; Yiye Zhang, PhD - Weill Cornell Medicine; Bradley Malin, PhD - Vanderbilt University Medical Center;

PRIMES: A novel approach for automated patient eligibility screening for clinical programs and beyond

Poster Number: 152
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Artificial Intelligence, Large Language Models (LLMs), Interoperability and Health Information Exchange, Clinical Decision Support, Transitions of Care, Information Retrieval
Programmatic Theme: Clinical Informatics

Patient eligibility screening for clinical programs such as Hospital at Home relies on manual chart review or brittle rule-based systems, which miss unstructured clinical data. PRIMES reframes eligibility as a retrieval-evaluation problem, using embedding-based retrieval to surface semantically relevant patient resources and LLM-based evaluation against natural-language criteria. Preliminary results on synthetic FHIR data suggest PRIMES matches rule-based accuracy on structured criteria while extending coverage to clinical notes and multimodal sources.

Speaker(s):
Tim Schwirtlich, MS
Northwestern University

Author(s):
Tim Schwirtlich, MS - Northwestern University; Abel Kho, MD, FACMI - Northwestern University;

Multi-Region Brain-Based Epigenetic Aging Clock

Poster Number: 153
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Artificial Intelligence, Machine Learning, Deep Learning
Programmatic Theme: Translational Bioinformatics

While pan-tissue epigenetic clocks effectively estimate biological aging, they often lack tissue-specific precision. To address this, we developed a brain-specific aging clock using 23 DNA methylation datasets (about 3,250 samples). Our support vector regression model outperformed existing clocks, accurately predicting chronological age (r=0.97, median absolute error = 3.56 years) across multiple brain regions. Furthermore, its age gap estimates exhibited nominal associations with MMSE, CERAD scores, and Braak stage in ROSMAP and Mount Sinai Brain Bank cohorts.

Speaker(s):
Travyse Edwards, PhD
University of Pennsylvania

Author(s):
Li Shen, Ph.D., FAIMBE, FACMI, FAMIA - University of Pennsylvania; Qi Long, Ph.D. - University of Pennsylvania;

Clinical Question to Reproducible Model: An AI Copilot for Clinician-Driven Predictive Modeling in Healthcare

Poster Number: 154
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Artificial Intelligence, Clinical Decision Support, Healthcare Quality
Programmatic Theme: Clinical Research Informatics

We propose a local-first AI copilot that translates clinician-generated research questions into reproducible predictive modeling workflows using healthcare data. The system is designed to bridge clinical intent and technical implementation through structured study planning, data profiling, methodological safeguards, local code generation and execution, evaluation, and clinician-readable reporting. By preserving data privacy and supporting transparent, methodologically guided analysis, the proposed framework aims to improve collaboration, reproducibility, and clinical relevance in healthcare machine learning research.

Speaker(s):
Yuqi Wu, PhD, MPH, MPA
University of North Florida

Author(s):
Yuqi Wu, PhD, MPH, MPA - University of North Florida; Hanadi Hamadi, PhD - University of North Florida;

Dialysis Risk Prediction and Treatment Effect Estimation for AKI patients using Longitudinal Electronic Health Records

Poster Number: 155
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Causal Inference, Machine Learning, Deep Learning, Artificial Intelligence, Real-World Evidence Generation, Data Mining, Quantitative Methods, Population Health
Programmatic Theme: Clinical Research Informatics

Progression to dialysis or end-stage renal disease is a rare but clinically important outcome. Clinicians need evidence on how medication exposures influence downstream risk. We constructed a fixed-window EHR cohort (90-day observation, 730-day prediction; N=81401; dialysis/ESRD prevalence: 1.1%) and modeled sequences of diagnoses, procedures, and medications with kidney laboratory trends (creatinine, BUN, eGFR). A transformer-based causal multi-head model was trained to estimate drug- and ingredient-level average treatment effects (ATEs) using counterfactual exposure removal and insertion under a full medication history setup. On test set, predictive performance reached an AUC of 0.694 and PR-AUC of 0.094. At the selected decision threshold (0.883), the model achieved an F1 score of 0.201 with a Brier score of 0.018. Post-hoc causal analyses of lab changes (eGFR, creatinine, BUN) using IPTW, AIPW, naive, and covariate-adjusted OLS methods assessed clinical directionality. Results showed partial protective-direction support for ACE/ARB exposures and worsening-direction signals for loop diuretics.

Speaker(s):
Kalyani Pande, Master of Science
Stony Brook University

Author(s):
Kalyani Pande, Master of Science - Stony Brook University; Evan Yang, BS - Stony Brook University; Bryan Zhu, High School - Stony Brook University; Sandeep K. Mallipattu, MD - Stony Brook University; Alisa Yurovsky, PhD - Stony Brook University; Tengfei Ma, PhD - Stony Brook University;

Revisiting the High-Benefit Patient: Causal Generic Machine Learning Inference for Intensive Blood Pressure Control

Poster Number: 156
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Causal Inference, Machine Learning, Artificial Intelligence, Chronic Care Management
Programmatic Theme: Clinical Research Informatics

We applied the model-free GenericML framework using robust estimators (the Best Linear Predictor, Group Average Treatment Effects, and baseline profile classification) to reassess treatment effect heterogeneity (HTE) of intensive blood pressure (BP) using pooled data from two large-scale randomized control trials on BP controls. Among 10,712 participants with up to three years of follow-up, we estimated the differential risks of the primary cardiovascular outcome. Intensive BP control produced a modest average risk reduction (BLP β₁ = −0.0136), but evidence for HTE was not statistically significant (BLP β₂ p = 0.443; GATES top–bottom contrast p = 0.471). Although CLAN grouped participants with metabolically adverse profiles into the highest predicted-benefit stratum, their treatment response did not differ significantly from that of the lowest-benefit group. Overall, while machine learning can identify clusters of high-risk baseline phenotypes, group-level inference revealed no meaningful HTE, underscoring the need for caution when interpreting individualized treatment predictions.

Speaker(s):
Amir Habibdoust Lafmajani, PhD
University of Missouri

Author(s):
Amir Habibdoust Lafmajani, PhD - University of Missouri; Xing Song, PhD - University of Missouri;

Developing a breast cancer risk prediction model for clinical use

Poster Number: 157
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Causal Inference, Clinical Decision Support, Human-computer Interaction
Programmatic Theme: Clinical Research Informatics

Current breast cancer risk models suffer from limited predictive power and poor clinical integration. We developed a Random Survival Forest model using EHR data from 66,583 patients to predict risk. By identifying 17 key features from 759 clinical variables, the model achieved performance comparable to deep learning alternatives. Designed for clinical utility, our approach uses passively collected data and interpretable outputs, facilitating seamless integration into provider workflows for improved early detection.

Speaker(s):
Matthew Murrow, PhD
Vanderbilt University Medical Center

Author(s):
Matthew Murrow, PhD - Vanderbilt University Medical Center; Zhijun Yin, Ph.D. - Vanderbilt University Medical Center;

Complex Care Management Governance: Health System & Health Plan

Poster Number: 158
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Chronic Care Management, Population Health, Informatics Implementation, Administrative Systems
Working Group: Clinical Information Systems Working Group
Programmatic Theme: Clinical Informatics

Mass General Brigham recently launched several complex care management programs co-designed and jointly staffed by the provider system and health plan. These programs align plan and provider care management efforts to improve quality and patient experience, leveraging shared technology. To support coordinated reporting, outreach, and clinical management, two governance bodies were established: a Joint Governance Committee providing executive and regulatory oversight, and a Joint Epic Technical Subcommittee responsible for technical implementation of approved decisions.

Speaker(s):
Todd King, PA-C
Mass General Brigham

Author(s):
Elaine Goodman, MD, MBA - Mass General Hospital/Brigham & Women's Hospital/ Harvard Med School;

The Impact of Social Determinants of Health on Telehealth Utilization and Patient Experiences in Hypertension and Diabetes Care

Poster Number: 159
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Chronic Care Management, Health Equity, Telemedicine
Programmatic Theme: Consumer Health Informatics

This study aims to evaluate the utilization patterns, effectiveness, and patient satisfaction of telehealth services among individuals with hypertension and/or diabetes, and to investigate the influence of social determinants of health (SDOH) on telehealth access and utilization in this population. We conducted a cross-sectional analysis using data from the 2022 Health Information National Trends Survey (HINTS 6) by the National Cancer Institute. The study sample included 3,009 respondents with self-reported diabetes, hypertension, or both conditions. Telehealth usage was assessed through 14 survey questions, and participant characteristics were analyzed using sociodemographic, baseline health, and SDOH data. Of the 6,252 HINTS 6 survey respondents, 3,009 met the inclusion criteria. Significant sociodemographic differences were observed across the diabetes and/or hypertension groups. Common reasons for telehealth use included provider recommendation, convenience, and infection avoidance. Social determinants of health, such as food insecurity and transportation issues, were more prevalent among individuals with both conditions, though no significant differences in telehealth experiences were noted across groups.

Speaker(s):
Jiancheng Ye, PhD
Weill Cornell Medicine

Author(s):
Haoxin Chen, MS - The University of Hong Kong;

Clinical Validation of a Data Integration and Visualization Platform for Wearable-derived Patient-Generated Health Data in Type 2 Diabetes

Poster Number: 160
2026 Symposium LIEAF Presentation

Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Chronic Care Management, Data Standards, Informatics Implementation, Patient-/Person-Generated Health Data
Programmatic Theme: Academic Informatics / LIEAF

Wearable sensors measuring sleep, heart rate variability, and physical activity provide vital indicators for glycemic control in type 2 diabetes. However, data overload and fragmentation hinder clinical application. To address this, Validation and Inspection Tool for Armband-based Lifelog Data (VITAL) was evaluated using Oura ring data from affected patients. VITAL integrates fragmented sleep and heart rate variability metrics into standardized visualizations. Overlaying data enabled intuitive identification of physiological correlations, facilitating practical use at the point-of-care.

Speaker(s):
Bomin Jeon, PhD
Seoul National University College of Nursing

Author(s):
Eunyoung Im, PhD - Seoul National University College of Nursing; Jinsun Jung, PhD - Seoul National University College of Nursing; Hyeoneui Kim, PhD - Seoul National University;

Cross-Disease Analysis of Clinical Trial Eligibility Traits in Comparative Effectiveness Research Using LLM-Based Extraction

Poster Number: 161
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Chronic Care Management, Natural Language Processing, Clinical Decision Support
Programmatic Theme: Clinical Research Informatics

We applied LLM-based extraction to 595 comparative effectiveness research trials from PCORI and ClinicalTrials.gov, identifying 21,653 eligibility trait mentions consolidated into 509 canonical traits across 10 disease categories. Of these, 322 were classified as Predictable and Necessary for EHR-based inference. Cross-disease analysis revealed 151 traits shared across five or more therapeutic areas, supporting a tiered prescreening architecture combining generalizable and disease-specific modules.

Speaker(s):
Abdul Muqeeth, MS Bioinformatics; PhD Student, Health Informatics
Indiana University

Author(s):
Abdul Muqeeth, MS Bioinformatics; PhD Student, Health Informatics - Indiana University; Yan Zhuang, PhD, FAMIA - Indiana University;

Ontology-Grounded Imaging Phenotype Knowledge Graphs for Explainable Oncology Reasoning from Abdominal CT

Poster Number: 162
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Clinical Decision Support, Knowledge Representation and Information Modeling, Controlled Terminologies, Ontologies, Vocabularies, Artificial Intelligence, Machine Learning, Information Retrieval, Information Extraction
Programmatic Theme: Clinical Informatics

Clinical decision-making in oncology requires reasoning over relationships among anatomical structures, tumor characteristics, and disease progression. Although deep learning models can accurately detect and segment tumors from medical images, most imaging AI systems present results as voxel-level predictions or latent embeddings, which are difficult to interpret clinically. Consequently, translating imaging-derived evidence into structured clinical reasoning remains a major challenge for medical AI systems.
We propose an ontology-grounded Imaging Phenotype Knowledge Graph (IPKG) framework that transforms segmentation-derived tumor measurements from abdominal CT scans into structured clinical representations aligned with biomedical ontologies. Using a pancreatic CT dataset, the framework extracts lesion-centric imaging phenotypes including tumor burden, lesion multiplicity, organ containment, and vascular proximity. These phenotypes are mapped to standardized clinical concepts defined in ontologies such as SNOMED CT, RadLex, and the NCI Thesaurus and organized into a knowledge graph that supports interpretable reasoning and query answering over clinically meaningful predicates.
To assess clinical relevance, we analyze oncology narrative reports from the CORAL dataset and examine semantic correspondence between imaging-derived phenotypes and terminology used in radiology and oncology documentation. Because the imaging and clinical datasets originate from different cohorts, this analysis is performed at the concept level rather than at the patient level.
The proposed framework demonstrates how imaging-derived phenotypes can be integrated with ontology-aligned clinical knowledge to support explainable reasoning over oncology findings and facilitate knowledge-driven clinical decision support.

Speaker(s):
Yugyung Lee, PhD
University of Missouri - Kansas City

Author(s):
Udiptaman Das, MS - University of Missouri–Kansas City; Krishnasai Atmakuri, MS - University of Missouri–Kansas City; Saeed Alqarni, PhD - Saudi Electronic University; Duy Ho, PhD - California State University Fullerton; Yugyung Lee, PhD - University of Missouri - Kansas City; Chi Lee, PhD - University of Missouri-Kansas City; John Park, MD - NKC Health; John Kang, MD, PhD - University of Washington; Chul Ha, MD - UT Health San Antonio MD Anderson Cancer Center;

Balancing Evidence-Based Ambition and EHR Constraints: Lessons from Implementing Multidomain CDS in Skilled Nursing Facilities

Poster Number: 164
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Clinical Decision Support, Informatics Implementation, User-centered Design Methods
Programmatic Theme: Clinical Informatics

Implementing Clinical decision support (CDS) in commercial electronic health record (EHR) requires balancing evidence-based design with technical constraints. We describe tradeoffs encountered while deploying a multidomain CDS system in skilled nursing facilities. Planned automation features, including dynamic eligibility detection and cross-domain alerts, required simplification due to vendor limitations. Iterative refinement prioritized workflow integration over maximal automation, offering transferable lessons for scalable CDS implementation in constrained EHR environments.

Speaker(s):
Veysel Baris, PhD, RN
Brigham and Women's Hospital

Author(s):
Min Jeoung Kang, PhD - Brigham and Women's Hospital/ Harvard Medical School; Alice Kim, MS - Brigham and Women's Hospital; Rosa Recio-Garcia, RN - Brigham and Women's Hospital; Kumiko Schnock, RN, PH.D - Brigham and Women's Hospital/ Harvard Medical School; Pamela Garabedian, MS - Mass General Brigham Inc.; Nancy Latham, PhD PT - Brigham and Women's Hosptial; Jason Falvey, PhD - University of Maryland; Elizabeth Dennis, PhD - University of Maryland; Jay Magaziner, PhD - University of Maryland; Rodrigo Valderrábano, MD - Brigham and Women's Hospital; Ling Tang, Ms - University of Maryland; Chance Backert, RN - PointClickCare Inc., Boston; Brittin Wagner, Ms - PointClickCare Inc.,; Richard White, BSc, MBA - PointClickCare Inc.,; Denise Orwig, PhD - University of Maryland; PATRICIA C DYKES, PhD, MA, RN - Emory University;

Design and Formative Usability Evaluation of an Interoperable Clinical Decision Support Tool for Cannabidiol-Drug Interactions

Poster Number: 165
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Clinical Decision Support, Patient Safety, Human-computer Interaction, Interoperability and Health Information Exchange
Programmatic Theme: Clinical Research Informatics

Cannabis-drug interactions pose a growing clinical challenge as medicinal cannabis use expands across the United States. Despite known pharmacokinetic risks via cytochrome P450 enzyme inhibition by cannabidiol (CBD), clinicians lack targeted digital tools to support real-time interaction management. We developed an evidence-based clinical decision support (CDS) web application for CBD-drug interactions, implemented as a SMART-on-FHIR application integrated with a local OMOP-on-FHIR instance containing synthetic data. The tool addresses three high-priority interactions: 1) CBD-warfarin, 2) CBD-clopidogrel, and 3) direct oral anticoagulants (DOACs). In a formative usability study with six pharmacists using a within-subjects design, participants completed clinical scenarios significantly faster with the CDS tool versus standard resources (4 min 54 sec vs. 8 min 22 sec; p = 0.031). The average system usability scale (SUS) score was 89 (SD = 9.0). These preliminary findings demonstrate the feasibility of building interoperable, context-specific CDS applications to support CBD-drug interaction management in clinical workflows.

Speaker(s):
Richard Boyce, PhD
University of Pittsburgh

Author(s):
Sandra Kane-Gill, PharmD - University of Pittsburgh; McKenna Anderson, PharmD - University of Pittsburgh; Kojo Abanyie, PharmD/Post-doctoral Scholar - University of PIttsburgh;

Detecting and Governing Unsafe Vendor-Supplied EHR Defaults: A Learning Health System Case of Medication Instruction Errors

Poster Number: 166
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Clinical Decision Support, Patient Safety, Usability, Governance, Informatics Implementation, Real-World Evidence Generation
Programmatic Theme: Clinical Informatics

EHR vendors increasingly embed standards‑based defaults intended to improve usability, yet the safety implications of these choices may only emerge after deployment across health systems. We describe a Learning Health System case in which a vendor‑supplied medication instruction default produced recurrent dosing‑interval hazards. Using frontline safety signals, interdisciplinary review, and enterprise governance, we identified failure modes, evaluated mitigation strategies, and modified deployed functionality—demonstrating how health systems can detect and govern unsafe EHR defaults in real‑world practice.

Speaker(s):
John McGreevey, MD
University of Pennsylvania

Author(s):
Adrienne Terico, PharmD - University of Pennsylvania Health System; Nishaminy Kasbekar, BSPharm, PharmD, FASHP, CPEL - Penn Medicine; Joel Betesh, MD - University of Pennsylvania Health System;

A Machine Learning–Driven Clinical Chatbot for Infection Control in a Congenital Heart Disease ICU

Poster Number: 167
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Clinical Decision Support, Critical Care, Machine Learning
Programmatic Theme: Clinical Research Informatics

Hospital-acquired infections (HAIs) remain a leading cause of morbidity in pediatric Cardiac Intensive Care Units (CICUs). We present Bot4KidsHeart, a multi-phase, data-driven informatics framework designed to reduce HAIs in infants under one year. Structured observations (>1,400) and stakeholder interviews informed a behavioral classification system categorizing actions by type, risk, and severity. Analysis revealed >50% procedural non-compliance in critical tasks such as dressing changes, oral care, and extubation, with open-room layouts and family non-compliance (hand hygiene lapses, phone or food use) contributing to risk. Bot4KidsHeart integrates machine learning and natural language processing to deliver real-time, context-aware infection-prevention guidance, combining hospital guidelines with clinician input, achieving 99.2% response accuracy. Simulation-optimization using RealOpt projected improvements in workflow adherence and HAI reduction. In a pilot study with 100 families, 65% retained and applied training. This work demonstrates a scalable, AI-driven approach that empowers families, supports clinical teams, and advances infection prevention in high-risk pediatric care.

Speaker(s):
Eva Lee, PhD
The Data and Analytics Innovation Institute

Author(s):
Guanlin Chen, PhD - Georgia Institute of Technology; Peijue Zhang, MS - Georgia Institute of Technology; Meiyi Guo, MS - Georgia Institute of Technology; Michael Wright, MBA - Grady Health System;

Predicting Cognitive Impairment in Older Adults with Dementia from Longitudinal Electronic Health Record Notes Using Large Language Models

Poster Number: 168
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Clinical Decision Support, Large Language Models (LLMs), Artificial Intelligence
Programmatic Theme: Clinical Informatics

Cognitive impairment is a key determinant of outcomes in older adults with dementia, yet routine cognitive assessments are resource-intensive and not always available. We developed a large language model framework to identify cognitive impairment from one year of longitudinal EHR notes. Using cohorts linked to OASIS and MDS assessments, our approach achieved macro-F1 scores of 68.76 and 69.30. These findings demonstrate the potential of LLMs to enable scalable cognitive impairment identification from routine clinical documentation.

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; Richard Wyss, PhD - Division of Pharmacoepidemiology and Pharmacoeconomics, Department of Medicine, Brigham and Women’s Hospital, Harvard Medical School, Boston, MA, USA; Jie Yang, PhD, FACMI, FAMIA - Harvard Medical School; Kueiyu Joshua Lin, MD, ScD - Division of Pharmacoepidemiology and Pharmacoeconomics, Department of Medicine, Brigham and Women’s Hospital, Harvard Medical School, Boston, MA, USA;

Clustering Clinical Note Embeddings to Identify Patterns Related to Clinical Decision Making: A Feasibility Study

Poster Number: 169
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Clinical Decision Support, Chronic Care Management, Machine Learning, User-centered Design Methods, Patient Safety, Qualitative Methods, Quantitative Methods, Population Health
Programmatic Theme: Clinical Informatics

Electronic health records contain vast amounts of data used in clinical decision making for complex chronic diseases. With the goal of integrating unstructured data into algorithms that facilitate decision making, this exempt feasibility study used text embeddings to identify patterns in unstructured clinical narratives related to guideline-directed medical therapy for patients with heart failure. Patterns emerged among embeddings, suggesting this as a viable first step toward scalable characterization and validation of clinical narratives.

Speaker(s):
Carly Daley, PhD
Parkview Health

Author(s):
Amelia Roebuck, PhD - Parkview Mirro Center for Research and Innovation; Michael Mirro, MD - Parkview Mirro Center for Research and Innovation; Shion Guha, PhD - Parkview Mirro Center for Research and Innovation;

Machine Learning–Based Clinical Decision Support Systems in Low- and Middle-Income Countries: A Scoping Review of Clinical Evaluation Studies

Poster Number: 170
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Clinical Decision Support, Machine Learning, Global Health, Real-World Evidence Generation
Working Group: Global Health Informatics Working Group
Programmatic Theme: Clinical Research Informatics

Machine learning–based clinical decision support systems (ML-CDSS) are often proposed to improve care in low- and middle-income countries (LMICs), but real-world evidence is limited. We conducted a scoping review (PubMed, Embase, Web of Science; 2013–2024) of prospectively evaluated, non-imaging ML-CDSS deployed in routine LMIC clinical settings. Of 2,113 records, only four studies met criteria (two randomized trials; two prospective cohorts), highlighting a major evaluation gap.

Speaker(s):
Macton Mgonzo

Author(s):
Macton Mgonzo; Victor Hunt, MSc - The Warren Alpert Medical School of Brown University; Keyana Zahiri, BS - The Warren Alpert Medical School; Allan Kimaina, MSc - Moi University, Eldoret, Kenya; Allison DeLong, PhD - Brown University; Arman Oganisian, PhD - Brown University; Ian Bacher - Brown University; Rami Kantor, MD - Warren Alpert Medical School, Brown University; Nicole Kim, BA - Brown Center for Biomedical Informatics; Ann Mwangi, PhD - Moi University, Eldoret, Kenya; Joseph Hogan, PhD - Department of Biostatistics, School of Public Health, Brown University; Hamish Fraser, MBChB, MRCP, MSc - Brown University;

Predicting Pediatric Critical Events Using Kolmogorov Arnold Networks

Poster Number: 171
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Clinical Decision Support, Critical Care, Machine Learning
Programmatic Theme: Clinical Informatics

Most machine learning-based clinical decision support tools developed to recognize deteriorating children lack interpretability. In this study, we evaluate the feasibility of Kolmogorov-Arnold Networks (KANs), a modeling framework that outputs symbolic equations explicitly describing underlying model behavior, for predicting pediatric clinical deterioration. Our trained KAN model demonstrated improved discrimination over elastic net and XGBoost methods, underscoring that interpretability is achievable without sacrificing predictive performance. Future work entails balancing KAN formulaic simplification with model complexity.

Speaker(s):
Sierra Strutz, PhD Student in Biomedical Data Science
University of Wisconsin - Madison

Author(s):
Sierra Strutz, PhD Student in Biomedical Data Science - University of Wisconsin - Madison; Kyle Carey, MPH - University of Chicago; Priti Jani, MD, MPH - University of Chicago; Almas Syed, MD - Loyola University; Neil Munjal, MD - University of Wisconsin; Majid Afshar, MD, MSCR - University of Wisconsin - Madison; Matthew Churpek, MD, MPH, PhD - University of Wisconsin-Madison; Anoop Mayampurath, PhD - University of Wisconsin - Madison;

Data Suitability of Early Automated Clinical Decision Support Tools in Emergency Care

Poster Number: 172
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Clinical Decision Support, Informatics Implementation, Data Standards, Workflow, Natural Language Processing
Programmatic Theme: Clinical Informatics

Clinical decision support (CDS) tools can improve emergency department decision-making but remain inconsistently used due to workflow and implementation challenges. We evaluated 17 guideline-directed CDS tools for cardiopulmonary conditions by categorizing their required data elements by source and structure. Most variables were documented in structured electronic health record fields, indicating strong potential for automation. These findings highlight opportunities and inform strategies for scalable integration of automated CDS within emergency care workflows.

Speaker(s):
Swara Chokshi, University Student
Virginia Commonwealth University

Author(s):
Swara Chokshi, University Student - Virginia Commonwealth University; Alan Storrow, MD - Vanderbilt University Medical Center; Jesse Wrenn, MD, PhD - VUMC;

Interpretable Phenotyping of Chronic Pain Tailored to Rehabilitation

Poster Number: 173
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Clinical Decision Support, Chronic Care Management, Artificial Intelligence, Machine Learning, Large Language Models (LLMs)
Programmatic Theme: Clinical Research Informatics

Chronic pain presents heterogeneous manifestations, making identification in electronic health records (EHRs) challenging. We validated and improved computable phenotype algorithms using inpatient and outpatient EHR data from a healthcare system in north central Florida. Using chart derived ground truth, we evaluated existing phenotypes, code combinations defined by experts, logistic regression, and a large language model–proposed algorithm. Existing methods showed high sensitivity but low specificity, while expert rules and a weighted logistic score improved balance.

Speaker(s):
Tiancheng Zhou, M.S
University of Florida

Author(s):
Mark Bishop, PT, PhD - University of Florida; Tiancheng Zhou, M.S - University of Florida; Jason Beneciuk, DPT, PhD - University of Florida; Mattia Prosperi, PhD, FAMIA - University of Florida;

Implementation Feasibility of the CONCERN Early Warning System in Korean Inpatient EHRs: A Gap Analysis Using the HPM-ExpertSignals Framework

Poster Number: 174
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Clinical Decision Support, Informatics Implementation, Data Standards, Interoperability and Health Information Exchange, Patient Safety
Programmatic Theme: Clinical Informatics

International deployment of EHR-based early warning systems requires more than feature mapping. Using 163,769 inpatient encounters from an OMOP Common Data Model (CDM) database, we evaluated the structural alignment and signal portability of the CONCERN nurse-driven early warning system in a Korean hospital. Of five core documentation features, three demonstrated full matches and two partial matches in the Korean EHR. Documentation-derived signals were influenced by local workflows and sociotechnical context, indicating that effective CDS transfer requires alignment across structural, behavioral, and contextual domains.

Speaker(s):
Youngjin Lee, PhD
Ajou University

Author(s):
Youngjin Lee, PhD - Ajou University; Rachel Lee, PhD, RN - Columbia University; Sujin Gan, Integrated PhD course student; Min Jeoung Kang, PhD - Brigham and Women's Hospital/ Harvard Medical School; Sarah Rossetti, RN, PhD, FAAN, FACMI, FAMIA, FIAHSI - Columbia University Irving Medical Center; Kenrick Cato, PhD, RN, CPHIMS, FAAN, FACMI - University of Pennsylvania/ Children's Hospital of Philadelphia; PATRICIA C DYKES, PhD, MA, RN - Emory University;

An Imaging Informatics Pipeline for Quantifying Surgical Wounds in Phonomicrosurgery

Poster Number: 175
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Clinical Decision Support, Quantitative Methods, Informatics Implementation
Programmatic Theme: Clinical Informatics

We developed a multimodal pipeline that transforms intraoperative images from phonomicrosurgery into quantitative predictors of surgical outcomes, demonstrating how clinician-led informatics workflows can extract actionable metrics from existing EMR. Associations with post-voice-rest stroboscopic outcomes were evaluated using multiple linear regression. Bivariate correlations were non-significant; however, multivariable analysis revealed a suppression pattern: incision size negatively predicted mucosal wave recovery (β = −1.82, p = 0.025) while lesion area showed positive association (β = 13.27, p = 0.037).

Speaker(s):
Ruiqing "Stephanie" Fan, MA
University of Houston

Author(s):
Ruiqing "Stephanie" Fan, MA - University of Houston; Yin Yiu, MD - Department of Otolaryngology, Houston Methodist Hospital, Houston Methodist, Weill Cornell Medical College; Andrew Tritter, MD - Department of Otorhinolaryngology-Head and Neck Surgery, UTHealth Houston, McGovern Medical School; Ashwini Joshi, PHD, CCC-SLP - Department of Communication Sciences and Disorders, University of Houston;

Task-Dependent Acoustic Feature Profiles for Vocal Resonance Classification: A Machine Learning Comparison of Sustained Phonation and Connected Speech

Poster Number: 176
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Clinical Decision Support, Machine Learning, Artificial Intelligence
Programmatic Theme: Clinical Research Informatics

Vocal resonance is assessed subjectively in clinical voice evaluation, limiting reliability. We trained two independent random forest classifiers on 64 female voice samples to classify forward- versus back-focused resonance separately for sustained phonation and connected speech using 88 eGeMAPSv02 features. The CS model (accuracy: 75.1%) outperformed the SP model (66.5%). SHAP analysis revealed non-overlapping features across tasks, demonstrating perceived resonance is task-dependent. These findings inform task-aware design of voice assessment tools.

Speaker(s):
Ruiqing "Stephanie" Fan, MA
University of Houston

Author(s):
Ruiqing "Stephanie" Fan, MA - University of Houston; Shaheen Awan, PhD, CCC-SLP - University of Central Florida; Zachary Abrams, PhD - Institute for Informatics at Washington University School of Medicine in St. Louis; Ashwini Joshi, PhD, CCC-SLP - University of Houston;

Augmenting the Primary Care Encounter: An AI-Enabled Clinical Decision Support System Prototype Using Structured Terminologies

Poster Number: 177
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Clinical Decision Support, Artificial Intelligence, Controlled Terminologies, Ontologies, Vocabularies, Data Standards, Informatics Implementation, Chronic Care Management, Natural Language Processing, User-centered Design Methods
Programmatic Theme: Clinical Informatics

Primary care remains largely excluded from the benefits of artificial intelligence due to the absence of structured, labeled clinical data. We present a prototype clinical decision support system that augments the primary care encounter through a symptom-first architecture, mapping conditions to quantifiable diagnostic criteria using standardized terminologies including SNOMED CT and ICD-10. The system captures granular, time-sensitive data at the point of care, establishing the foundation for AI-driven precision diagnosis and improved clinical outcomes.

Speaker(s):
Manasa Devi Chinta, PharmD, MHIT, PhD Student
University of South Carolina

Author(s):
Manasa Devi Chinta, PharmD, MHIT, PhD Student - University of South Carolina; Stephen Lloyd, MD, PhD - University of South Carolina; Elizabeth Regan, PhD - University of South Carolina;

Agentic Delphi for Phenotyping Immune–Related Acute Kidney Injury in Cancer Patients with Immunotherapies

Poster Number: 178
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Clinical Decision Support, Large Language Models (LLMs), Artificial Intelligence, Patient Safety, Real-World Evidence Generation
Programmatic Theme: Clinical Research Informatics

Immune-related acute kidney injury (irAKI) in patients receiving immune checkpoint inhibitors is underrecognized and lacks diagnostic guidelines. We present the first agentic Delphi protocol using eight LLM-based specialist agents and a clinician-validated 14-question survey grounded in EHR evidence via retrieval-augmented generation. Agents first independently assess, then debate, and finally reassess irAKI probability. Across 509 post-ICI AKI cases, the system identified 147 irAKI cases (4.5%, 147/3,259 of ICI-treated patients; 28.9%, 147/509 of post-ICI AKI cases) and produced complete auditable reasoning traces with measured consensus refinement across rounds.

Speaker(s):
Haining Wang, PhD
Indiana University

Author(s):
Haining Wang, PhD - Indiana University; Hao Wang, PhD Student - Purdue University; Yao Chen, MS - Indiana University School of Medicine; Chenxi Xiong, MS - Indiana University; Shihui Jiang, MBA - Indiana University School of Medicine; Qianqian Song, Ph.D. - University of Florida; Xing He, Ph.D. - Indiana University; Lin Wang, PhD - Purdue University; Michael Eadon, MD - Indiana University School of Medicine; Jiang Bian, PhD - Indiana University/Regenstrief Institute; Jing Su, PhD - Indiana University School of Medicine;

Evaluation of Health Level Seven International’s Fast Healthcare Interoperability Resources (FHIR) to Represent Output of a Clinical Decision Support Knowledge Base Encoded in Arden Syntax

Poster Number: 179
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Clinical Decision Support, Data Standards, Data Sharing
Programmatic Theme: Clinical Informatics

Introduction: Arden Syntax medical logic module (MLM) outputs remain unstructured despite the use of FHIR as the standard for encoding data elements. Methods: WRITE statements were extracted from a convenience sample of 340 MLMs and inspected. Results: All WRITE statements were representable by the FHIR R5 GuidanceResponse resource except 2 MLMs using the more specific RiskAssessment and ImmunizationRecommendation resources. Conclusion: FHIR R5 is adequate for representing Arden Syntax output in standard format, enhancing MLM sharing.

Speaker(s):
Robert Jenders, MD, MS, FACP, FACMI, FHL7, FAMIA
Charles Drew University/UCLA

Author(s):
Robert Jenders, MD, MS, FACP, FACMI, FHL7, FAMIA - Charles Drew University/UCLA;

SafeRecall: A BI-RADS-Guided Uncertainty Framework for Safe Mammography Recall Reduction

Poster Number: 180
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Clinical Decision Support, Artificial Intelligence, Deep Learning
Programmatic Theme: Translational Bioinformatics

Over 10% of mammography patients are recalled unnecessarily, yet AI recall-reduction tools ignore predictive uncertainty—risking missed cancers in grey-zone cases. SafeRecall combines a missing-view-robust longitudinal architecture, BI-RADS-aware composite loss, and an uncertainty-guided three-zone decision framework via Monte Carlo Dropout. On EMBED (50k+ exams, 42% African American), SafeRecall achieves AUC 0.86, safely eliminates 29.5% of unnecessary recalls at 95% sensitivity, with consistent cross-demographic performance.

Speaker(s):
Yiran Song, doctor
University of Minnesota

Author(s):
Yiran Song, doctor - University of Minnesota; Rui Yin, PhD - University of Florida; Nian Wang, Ph.D. - utsouthwestern; Mingquan Lin, PhD - University of Minnesota;

Enhancement of ICD-11 Foundation synonyms from Mondo Disease Ontology

Poster Number: 181
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Controlled Terminologies, Ontologies, and Vocabularies, Data Standards, Interoperability and Health Information Exchange, Natural Language Processing, Knowledge Representation & Information Modeling
Programmatic Theme: Translational Bioinformatics

Mondo Disease Ontology classes have been mapped to ICD-11 Foundation entities. This correlation of semantically similar concepts provides a resource that can use automated rules to find synonyms that can be added to the ICD-11 Foundation to enhance coverage. Mondo is an actively curated source referencing many other OBO Foundry resources. We found that over 4,000 mappings offered over 12,000 potential synonyms. Application of various automated heuristic rules produced a list of over 5,000 synonyms recommended for addition to the ICD-11 Foundation.

Speaker(s):
Charles Morrey, PhD
Utah Valley University

Author(s):
Charles Morrey, PhD - Utah Valley University; Xiaohan Zhang, MD,MS; Christopher Chute, MD DrPH - Johns Hopkins University;

Development of a Multi-Stage Pipeline for Extracting Drug Response Associations from the GWAS Catalog

Poster Number: 182
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Controlled Terminologies, Ontologies, and Vocabularies, Information Extraction, Clinical Decision Support, Data Mining, Natural Language Processing, Knowledge Representation & Information Modeling, Data transformation/ETL, Quantitative Methods
Programmatic Theme: Translational Bioinformatics

The GWAS Catalog contains 891,200 variant-trait associations, but drug response
associations are not systematically classified. We developed a four-stage pipeline combining
UMLS Semantic Types with text patterns and validated drug names using an integrated
ChEMBL-SIDER lexicon (47,552 drugs). Applied to 891,200 associations, the pipeline
extracted 9,521 drug response associations across 730 studies, achieving a precision of
98.3% and a recall of 78.4%.

Speaker(s):
Sukyoung Park, MSc
Asan Medical Center, University of Ulsan College of Medicine

Author(s):
Sukyoung Park, MSc - Asan Medical Center, University of Ulsan College of Medicine; Sujung Jang, MSc - Asan Medical Center, University of Ulsan College of Medicine; Grace Juyun Kim, PharmD, PhD - Asan Medical Center; Kye Hwa Lee, Research Associate Professor - Asan Medical Center, University of Ulsan College ofMedicine;

Towards a Conceptual Model of Complementary and Integrative Health

Poster Number: 183
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Controlled Terminologies, Ontologies, and Vocabularies, Knowledge Representation and Information Modeling, Interoperability and Health Information Exchange, Data Sharing
Programmatic Theme: Translational Bioinformatics

Complementary and integrative health (CIH) approaches are increasingly used in healthcare, yet semantic resources supporting the integration and analysis of CIH evidence data remain limited. This work presents an ongoing effort to develop a conceptual model for the CIH domain. Guided by competency questions and expert input, the model structures relevant concepts to support the extraction, integration, analysis and assessment of CIH evidence data.

Speaker(s):
César Bernabé, PhD
University of Illinois Urbana-Champaign

Author(s):
Marcelo Fiszman, MD, Ph.D. FACMI, DipIBLM - Semedy, Inc; Robin Austin, PhD, DNP, DC, RN, NI-BC, FAMIA, FAAN - University of Minnesota, School of Nursing; Cui Tao, PhD - Mayo Clinic; Rui Zhang, PhD, FACMI, FAMIA, FIAHSI - University of Minnesota, Twin Cities; Halil Kilicoglu, PhD - University of Illinois Urbana-Champaign;

Optimal blood glucose targets for critically ill patients with sepsis in the intensive care unit

Poster Number: 184
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Critical Care, Infectious Diseases and Epidemiology, Patient-/Person-Generated Health Data
Working Group: Student Working Group
Programmatic Theme: Clinical Research Informatics

Background: Hyperglycemia and hypoglycemia are key risk factors for morbidity and mortality in critically ill septic patients. Despite ongoing research, glucose management guidelines for critically ill patients remain inconsistent, especially for diabetic patients. This study identifies optimal glucose targets to reduce mortality in critically ill patients with sepsis.
Research Question: What is the optimal blood glucose range associated with reduced in-hospital mortality for critically ill patients with sepsis, and how does this vary by diabetic status?
Study Design and Methods: This cohort study analyzed 22,374 adult intensive care unit (ICU) patients with sepsis from the MIMIC-IV database. Non-linear logistic regression models assessed the relationship between 72-hour median blood glucose levels and in-hospital mortality, adjusting for age, gender, and Sequential Organ Failure Assessment (SOFA) score. Subgroup analyses explored variations based on diabetic status and other clinical factors.
Results: The study found a U-shaped relationship between blood glucose levels and mortality, with the lowest risk at 6.3 mmol/L overall and 6.8 mmol/L for diabetic patients. A 5–8 mmol/L glucose range during the first 72 hours was associated with a mortality risk below 10%, representing up to a 5% reduction in mortality compared to the guideline targets of 7.8–10.0 mmol/L. No significant differences were found between patients with or without skin and soft tissue infection.
Interpretation: The findings suggest that a tighter glycemic control range of 5–8 mmol/L could improve survival in ICU patients with sepsis, challenging current guidelines. Further randomized controlled trials are necessary to validate and optimize glycemic control strategies for critically ill septic patients.

Speaker(s):
GOUSIA HABIB, Postdoctoral Research Fellow
University of Helsinki

Author(s):
Kevin Soon Teo, MBBS Division of Neurology, Department of Medicine, - National University Hospital, Singapore; Willem van de Bonds, Research Scientist - Institute for Human Development and Potential, Agency for Science, Technology and Research, Singapore; Mengling Feng, PhD - National University of Singapore; Kay Choong See, Adjunct Associate Professor Medicine - Division of Respiratory and Critical Care Medicine, Department of Medicine, National University Hospital, Singapore;

Outcomes for the Doctorate in Health Informatics (DHI): An Evidenced-Based Practice, Applied Doctorate

Poster Number: 185
2026 Symposium LIEAF Presentation

Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Curriculum Development, Teaching Innovation, Workforce Development, Informatics Implementation, Workflow, Evaluation
Programmatic Theme: Academic Informatics / LIEAF

The Doctorate in Health Informatics (DHI) is a hybrid, advanced-practice doctorate that prepares experienced professionals to lead evidence-based, practice-focused healthcare transformation. We report program outcomes through February 2026, including retention, progression, and graduation metrics, to inform best practices in academic informatics training and workforce development. Retention remained high (80.6%–96.2%) and overall progression was 85%, with increasing completions as cohorts matured.

Speaker(s):
Angela Ross, DNP, RN, MPH, PMP, DASM, PHCNS-BC
University of Texas Health McWilliams School of Biomedical Informatics

Author(s):
Chelsea Overstreet, MA - McWilliams School of Biomedical Informatics at UTHealth Houston; Susan Fenton, PhD - UTHealth Houston McWilliams School of Biomedical Informatics; Juliana Brixey, PhD, MPH, RN - UTHealth Houston; Erika Cavazos-Juarez, DHI, MBBS, MS, PMP - University of Texas (UTH) Health Science Center at Houston McWilliams School of Biomedical Informatics (SBMI); Angie Hayes, DHI, MS - UTHealth MSBMI; V. Gail Turner, DNP, MBA, NI-BC, PMP - McWilliams School of Biomedical Informatics at UTHealth Houston;

Protective Effects of Increased EHR-Mediated Information Sharing Within- and Between-Healthcare Teams on Patient Outcomes

Poster Number: 186
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Data Mining, Healthcare Quality, Quantitative Methods, Workflow
Programmatic Theme: Clinical Research Informatics

Electronic Health Records (EHRs) provide a common repository for patient information, and while many challenges persist, EHRs allow asynchronous coordination among healthcare professionals (HCPs). Such coordination is inherently challenging, as large multidisciplinary teams, or teams-of-teams, of HCPs jointly manage complex patient care. These teams-of-teams naturally fit under the Multiteam System (MTS) framework, where evidence suggests that both stronger within- and between-team communication improves overall team performance. Yet little is known how about these patterns of information sharing mediated through the EHR correlate with patient outcomes. We extracted over 53.9 million EHR access logs for 482 patients to construct novel within- and between-team information sharing measures which were then used to predict future emergency department (ED) visits and hospitalizations admitted through the ED. These network measures showed both statistical and practical significance, suggesting that this modifiable dimension of teamwork- information sharing through the EHR- could reduce acute care utilization.

Speaker(s):
Daniel Sewell, PhD, MS
University of Iowa

Author(s):
Shin-Ping Tu, MD, MPH - University of California, Davis; Daniel Sewell, PhD, MS - University of Iowa; Brittany Garcia, PhD - University of California, Davis; Quinn Stoddard-O'Neill, MS - University of Iowa; Michael Hogarth, MD, FACMI, FACP - University of California, San Diego; Marissa Shuffler, PhD, MA - Clemson University; Aaron Boussina, PhD, MA, MS - University of California, San Diego; Helen Chew, MD - University of California, Davis; David Cooke, MD, FACS - University of California, Davis; Patrick Romano, MD, MPH - University of California, Davis; Joann Elmore, MD, MPH - University of California, Los Angeles; Alan Dow, MD, MSHA - Virginia Commonwealth University; Xi Zhu, PhD, MS - University of California, Los Angeles;

MedEdu Co-Pilot: An LLM-Driven Diagnostic Reasoning System for Medical Education

Poster Number: 187
2026 Symposium LIEAF Presentation

Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Data Mining, Natural Language Processing, Information Retrieval
Programmatic Theme: Academic Informatics / LIEAF

We present MedEdu Co-Pilot, an LLM-driven system designed to support the training of clinicians in diagnostic reasoning. Using over 20,000 structured clinical case reports curated from PubMed Central Open Access, the system guides learners through stepwise, case-based reasoning with retrieval-augmented knowledge support. By simulating authentic clinical diagnostic workflows, MedEdu Co-Pilot enables scalable, interactive training to strengthen clinicians’ diagnostic reasoning skills.

Speaker(s):
Rui Yang, Master
Duke-NUS Medical School

Author(s):
Rui Yang, Master - Duke-NUS Medical School; Huitao Li, Msc - Duke-Nus Medical School; Siqi Ge, BS - Duke-NUS Medical School; Nan Liu, PhD - National University of Singapore;

Operational Data for Process Improvement in Outbreak Response: A Measles Case Study of Public Health Data Infrastructure Gaps

Poster Number: 188
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Data Modernization, Infectious Diseases and Epidemiology, Public Health, Data Sharing, Informatics Implementation, Workflow
Programmatic Theme: Public Health Informatics

Measles outbreaks burden health departments, yet public health lacks standardized infrastructure to capture operational experience for efficiency gains. A cross-jurisdictional data collection exercise across seven jurisdictions with 2024–2025 measles outbreaks documented availability of epidemiological, workforce, and cost metrics. Findings reveal a fragmented data ecosystem focused on surveillance rather than operational data for process improvement, underscoring the need to incorporate core operational data into a minimum standardized dataset for outbreak response.

Speaker(s):
Elizabeth Campbell, MS, MSPH, PhD
Johns Hopkins Bloomberg School of Public Health

Author(s):
Haley Farrie, MPH - Johns Hopkins Center for Outbreak Response Innovation; Sarah Gillani, MPH - Center for Outbreak Response Innovation at Johns Hopkins Center for Health Security; Sutyajeet Soneja, PhD - Johns Hopkins Center for Outbreak Response Innovation; Daniel Jernigan, MD - Johns Hopkins Center for Outbreak Response Innovation; Caitlin Rivers, PhD - Johns Hopkins Center for Outbreak Response Innovation;

Building Cross-Organizational Learning Health System Data Infrastructure Using a Bulk FHIR-Based Extraction Architecture

Poster Number: 189
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Data Sharing, Interoperability and Health Information Exchange, Informatics Implementation
Programmatic Theme: Clinical Research Informatics

We evaluated a scalable, Bulk FHIR-based reference architecture (RADIANT) enabling automated extraction of consented patient‑level data across multiple pediatric health systems. Deployed at CHOP and UCSF, the system aggregated over 70 million FHIR resources and demonstrated feasible near‑real‑time refresh workflows despite vendor‑specific limitations and multi‑site governance variation. Findings show that standardized Bulk FHIR extraction can reduce manual abstraction burdens and provide a foundation for cross‑organizational learning‑health‑system activities.

Speaker(s):
Jeritt Thayer, PhD
The Children's Hospital of Philadelphia

Author(s):
Jeritt Thayer, PhD - The Children's Hospital of Philadelphia; Allison Heath, PhD - Children's Hospital of Philadelphia; Charles Haynes, BS - Children's Hospital of Philadelphia; Natasha Singh, MS - Children's Hospital of Philadelphia; Meen Chul Kim, Doctor of Philosophy - The Children's Hospital of Philadelphia; Hannah Calkins, MSIS - Children's Hospital of Philadelphia; JD Bank, BS - Seattle Children's Hospital; Louis DeNonno, BS - Seattle Children's Hospital; Jory Purvis; Adam Resnick, PhD - Children's Hospital of Philadelphia;

Coopetition Works: A New Playbook for FAIR Biomedical Data Infrastructure

Poster Number: 190
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Data Sharing, Data Standards, Policy, Workforce Development
Programmatic Theme: Translational Bioinformatics

The NIH‑supported Generalist Repository Ecosystem Initiative (GREI) addresses fragmented biomedical data infrastructure through a structured coopetition model that aligns seven independent repositories on shared met
adata standards, persistent identifiers, FAIR Signposting, and harmonized open metrics. GREI’s governance framework, built on a common code of conduct, agile task groups, and strategic partnerships, has accelerated interoperable, FAIR‑aligned practices while preserving repository autonomy. This talk highlights key outcomes and lessons for coordinating heterogeneous organizations to strengthen FAIR data sharing.

Speaker(s):
Kristi Holmes, PhD
Northwestern University

Author(s):
Rebecca Li, MD - Vivli; Matthew Carson, PhD - Northwestern University;

Data Harmonization of Multiple Datasets in Stroke Recurrence Kinetics

Poster Number: 191
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Data Sharing, Data Standards, Data transformation/ETL
Programmatic Theme: Clinical Research Informatics

Clinical trial datasets (POINT, SOCRATES, THALES) were harmonized and standardized, enabling kinetic modeling to distinguish transient “vulnerable” versus “stabilized” patient states after stroke. The harmonized dataset thus makes it possible to identify and analyze these two distinct patient states and their associated stroke recurrence rate and allow for future research on these relationships.

Speaker(s):
Simmer Beniwal, MPH
University of Chicago

Author(s):
Julie Johnson, PhD, MPH, RN - University of Chicago; Danielle Landron, MSN - University of Chicago; James Brorson, MD - University of Chicago; Stacie Landron, MS Computer Science - University of Chicago; Mihai Giurcanu, PhD - University of Chicago; Shyam Prabhakaran, MD, MS - University of Chicago; James Siegler, MD - University of Chicago;

Design and Development of a Layered Clinical Research Data Commons Infrastructure to Enable Cohort Exploration and Discovery within a Neuroscience Learning Health System

Poster Number: 192
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Data Sharing, Data Modernization, Data transformation/ETL, Knowledge Representation & Information Modeling, Governance, Information Visualization, Information Retrieval
Programmatic Theme: Clinical Research Informatics

We developed a layered clinical research data architecture to integrate fragmented neuroscience datasets and enable secure cohort and specimen discovery within a Learning Health System. The infrastructure preserves immutable raw data, harmonizes multi-study clinical and cognitive data from approximately 4,000 participants, and provides a governed web-based exploration interface for investigator for cohort requests, and visualizations. This approach supports scalable cohort identification, reproducibility, and improved data stewardship while maintaining project-specific governance.

Speaker(s):
Nelly-Estefanie Garduno-Rapp, MD, MSHI
UTSW

Author(s):
Nelly-Estefanie Garduno-Rapp, MD, MSHI - UTSW; Mounika Thakkallapally, M.S. Data Science - University of Texas Southwestern Medical Center; Fangjiang Wu, MS - UTSW; Qinbo Zhou, PhD - UTSW; Jenny Weon, MD, PhD - UT Southwestern Medical Center; Donghan Yang, PhD - UTSW; Yang Xie, PhD - UTSW; Justin Rousseau, MD, MMSc - University of Texas Southwestern Medical Center;

Harmonization and Standardization of Mental Health Data: A Hybrid Framework for the IMPACT-MH Initiative

Poster Number: 193
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Data Sharing, Data Standards, Data Mining
Programmatic Theme: Clinical Research Informatics

Mental health research requires data with significant diversity within diagnostic groups for better clinical-decision making. The Individually Measured Phenotypes to Advance Computational Translation in Mental Health (IMPACT-MH) initiative was formed to collect, harmonize, standardize, and submit multimodal mental health data including data derived from patient EHR, EMA, Audio/Visual, and Neurocognitive assessments on behalf of 13 research partners to the National Institute for Mental Health (NIMH) Data Archive (NDA).

Speaker(s):
Kalpana Raja, PhD, MRSB, CSci
School of Medicine, Yale University

Author(s):
Kalpana Raja, PhD, MRSB, CSci - School of Medicine, Yale University; Polly Painter, MSR - Yale Univeristy; Na Hong, PhD - Yale University; Josha Kenney, MS - Yale Unviersity; Christian Horgan, BS - Yale University; Ahmed Abdelhady, BS - Yale University; Fang Li, PhD - Mayo Clinic; Sarah Lichenstein, PhD - Yale University; Hamada Hamid Altalib, DO, MPH, FAES - Yale University; Avanti Bhandarkar, PhD - Yale University; Ellen Zhen, BS, BA - Yale School of Medicine; Yong Chen, PhD - University of Pennsylvania; Cui Tao, PhD - Mayo Clinic; Hua Xu, Ph.D - Yale University;

Configuring Business-focused Platforms and Research-focused Data Capture Management of a Community-based Clinical Trial

Poster Number: 194
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Data Standards, Data Sharing, Data transformation/ETL, Data Modernization, Documentation Burden
Programmatic Theme: Clinical Research Informatics

The mHealth-4-Mhealth intervention is a community-driven clinical trial with materials that require Integration of information from multiple sources to ensure its success. Clinical Trial Management Systems do not incorporate all data necessary to ensure operational compliance and efficiency, particularly for studies conducted in community or non-clinical settings. This work summarizes our development of an integrated clinical trials operational data framework that could be adapted to other community-based trials or settings.

Speaker(s):
Ricky Flores, Master of Science
Children's Nebraska

Author(s):
Ricky Flores, Master of Science - Children's Nebraska; Jonathan Figliomeni, Master of Science - University of Nebraska Medical Center; Michelle Warren, PhD - UNK; Sachi Verma, PhD, MPH, MTech - University of Nebraska Medical Center; Jana Broadherst, MD, PhD - UNMC; Betty Oberley, BS - UNMC; Samantha Beisel, AS - UNMC; Ellen Kerns, PhD, MPH - UNMC; Russell McCulloh, MD - UNMC;

Consensus-based development of Common Data Elements to Enhance the Interoperability of Myositis Research Data

Poster Number: 195
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Data Standards, Knowledge Representation and Information Modeling, Interoperability and Health Information Exchange
Programmatic Theme: Clinical Research Informatics

Common data elements (CDEs) are defined as standardized question-answer pairs that are designed for reuse across clinical studies to improve the findability, accessibility, interoperability, and reusability of clinical data. In line with Congressional priorities to develop CDEs for autoimmune disease research, we have developed CDEs for 24 myositis instruments, with an emphasis on CDE modularity and machine-readability through the use of reusable CDE bundles and standardization of contextual information.

Speaker(s):
Matthew Diller, PhD
National Institutes of Health

Author(s):
Matthew Diller, PhD - National Institutes of Health; Mark Bodkin, BS - DLH; Varsha Surampudi, PhD - DLH; Juan Rodriguez, BS - DLH; Payam Noroozi Farhadi, MD - National Institute of Environmental Health Sciences; Didem Saygin, MD - Rush University Medical Center; Christopher Mecoli, MD, MHS - Johns Hopkins University School of Medicine; Audrey Kessel, BS - DLH; Meridith Thanner, PhD - DLH; Steven Ramsey, BS - DLH; Adam Schiffenbauer, MD - National Institute of Environmental Health Sciences; Lisa Rider, MD - National Institute of Environmental Health Sciences; Richard Scheuermann, PhD - Division of Intramural Research (DIR), National Library of Medicine (NLM), NIH, DHHS;

Accelerating CDE and FHIR Portfolio Analysis Leveraging Generative AI

Poster Number: 196
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Data Standards, Interoperability and Health Information Exchange, Artificial Intelligence, Large Language Models (LLMs)
Programmatic Theme: Clinical Research Informatics

This study presents a portfolio analysis leveraging generative AI to examine NIH research artifacts and systematically categorize disease areas and the primary reasons for using Common Data Elements (CDEs) and HL7 Fast Healthcare Interoperability Resources (FHIR), requiring minimal manual curation. The analysis uncovered domain-specific adoption trends and highlighted the complementary but distinct roles that CDEs and FHIR play across clinical and biomedical research.

Speaker(s):
Sungrim Moon, PhD
National Institutes of Health

Author(s):
Sungrim Moon, PhD - National Institutes of Health; Wenling Chang, PhD - NIH; Snipta Mallick, BS - The National Institutes of Health; Hsinyi Tsang, PhD - National Institutes of Health; Joseph Croghan, MS - National Institutes of Health; Belinda Seto, Ph.D. - Office of Data Science Strategy, NIH;

Building Analytics-Ready Tables from FHIR Bundles with Spark SQL

Poster Number: 197
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Data Transformation/ETL, Interoperability and Health Information Exchange, Data Modernization
Programmatic Theme: Public Health Informatics

Public health reporting workflows based on Fast Healthcare Interoperability Resources (FHIR) suffer from the highly nested and extensible structure of FHIR bundles, which makes it difficult to perform exploratory data analysis and compile tabular reports. This study describes our experience using Spark SQL to reduce this burden by implementing a generic FHIR ingestion pipeline to transform heterogeneous FHIR bundles arriving as JSON text files into SQL tables as a data warehouse in Databricks.

Speaker(s):
Joseph Korpela, PhD
Washington State Department of Health

Author(s):
Joseph Korpela, PhD - Washington State Department of Health; Samantha Siebman, MPH - Washington State Department of Health; Ander Pierce, MA - Washington State Department of Health; Puneet Parashar, MBA - Washington State Department of Health;

Mapping Sensor Data to Clinical Notes using a Transformer Architecture

Poster Number: 198
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Data Transformation/ETL, Deep Learning, Informatics Implementation, Machine Learning
Programmatic Theme: Clinical Research Informatics

Ambient in-home sensors generate longitudinal signals that may reflect changes in health status, but translating these observations into clinically meaningful language remains challenging. We present Sense2Note, a transformer-based approach that maps daily ambient sensor patterns to structured clinical terms derived from EHR encounter documentation. Using PyTorch and a leave-one-patient-out design, the model was trained on 47 preprocessed sensor features and evaluated across five patients, with zero-padding and feature reordering used to harmonize differences in sensor availability. In group-level holdout evaluation, the model achieved an AUROC of 0.67, AUPRC of 0.26, and Brier score of 0.08; across held-out patients, mean AUROC was 0.70 ± 0.10 and mean AUPRC was 0.34 ± 0.10. These findings suggest that ambient sensor data contain predictive signal for clinical term inference, supporting probabilistic prioritization of clinical status rather than stand-alone classification in this early proof-of-concept study.

Speaker(s):
Noah Marchal, PhD
University of Missouri

Author(s):
Xing Song, PhD - University of Missouri; Mihail Popescu, PhD - University of Missouri;

Improving Deep Learning Aneurysm Detection in Cerebral Angiography: Relative Impact of Vessel-Enhancement Filtering and Dataset Diversity

Poster Number: 199
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Deep Learning, Machine Learning, Information Extraction
Programmatic Theme: Clinical Research Informatics

Deep learning models have shown promise for automated detection of vascular abnormalities in fluoroscopic angiography; however, their reliability under heterogeneous clinical imaging conditions remains suboptimal. This study evaluates the effects of vessel-enhancement preprocessing and dataset variability on the effectiveness of convolutional neural networks (CNNs) for aneurysm detection. A ResNet-18 architecture was trained and evaluated across two cerebral angiography datasets: a controlled open-source dataset and a clinical dataset derived from interventional radiology procedures. Four preprocessing conditions were compared: raw images and three Hessian-based vessel enhancement filters (Frangi, Sato, and Meijering). Within-dataset experiments demonstrated improved discrimination with vessel-enhancement preprocessing under stable imaging conditions. However, models trained in a single imaging environment showed substantial performance degradation when evaluated across environments. Training on combined heterogeneous datasets improved cross-environment generalization. These findings suggest that while vessel-enhancement preprocessing improves vascular signal detection, dataset diversity plays a greater role in achieving robust, clinically relevant deep learning performance.

Speaker(s):
Naveen Gulati, Masters
Georgia Institute of Technology

Author(s):
Naveen Gulati, Masters - Georgia Institute of Technology;

Foundation Models to Unlock Real-World Evidence from Nationwide Medical Claims

Poster Number: 200
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Deep Learning, Artificial Intelligence, Knowledge Representation & Information Modeling
Programmatic Theme: Clinical Research Informatics

ReClaim is a claims-native clinical foundation model trained on nationwide U.S. MarketScan data from 2008-2024. Using decoder-only Transformers, large-scale pretraining, and task-specific post-training, it models longitudinal patient trajectories spanning diagnoses, procedures, medications, and costs. ReClaim improves disease prediction, healthcare expenditure forecasting, and real-world evidence analyses, showing that harmonized administrative claims can support foundation modeling at national scale.

Speaker(s):
Fan Ma, Phd
yale

Author(s):
Fan Ma, Phd - yale; Yuntian Liu, M.P.H - Yale University; Xiang Lan, Ph.d - Yale; Jun Ni, B.S. - Yale University; Weipeng Zhou, PhD - Yale University; Mauro Giuffrè, MD PhD - Yale; Laila Rasmy, PhD, MSc, MBA, RPh. - UTHealth MSBMI; Lingfei Qian, PHD - Yale University; Xueqing Peng, PhD - Yale University; Yujia Zhou, M.S. - Yale University; Ruey-Ling Weng, MS. - Yale University; Huan He, Ph.D. - Yale University; Andrew Loza, MD PhD - Yale University School of Medicine | VA Connecticut Healthcare System, US Department of Veterans Affairs; Qingyu Chen, PhD - Yale University; Hua Xu, Ph.D - Yale University;

Multi-modal Deep Learning Model for Predicting Early HNSCC Cancer Recurrence

Poster Number: 201
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Deep Learning, Artificial Intelligence, Informatics Implementation
Programmatic Theme: Translational Bioinformatics

Recent advancements in machine learning and deep learning have demonstrated success in oncology; there are models being developed with different goals, like prediction and classification. We developed a DL model with the aim of predicting early and late head and neck squamous carcinoma recurrence; the model utilizes clinical, mutation, gene expression, and whole slide image data as inputs. We found that the multi-modality model outperformed traditional ML models.

Speaker(s):
Michael Adolphus, Bachelors Degree Candidate
Brown Center for Clinical Cancer Informatics and Data Science

Author(s):
Michael Adolphus, Bachelors Degree Candidate - Brown Center for Clinical Cancer Informatics and Data Science; Jessica Patricoski-Chavez, MS - Brown University Center for Computational Molecular Biology; Ece Uzun, PhD - Brown University Health/Brown University;

High Accuracy, Limited Data: A Decade of AI Research in Malaria Detection

Poster Number: 202
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Deep Learning, Artificial Intelligence, Machine Learning, Governance, Public Health
Programmatic Theme: Public Health Informatics

Machine learning and deep learning approaches have been widely studied for automated malaria detection from microscopy images. We conducted a systematic review of AI-based malaria detection studies published over the past decade. Among 173 included studies, convolutional neural networks dominated and reported a mean diagnostic accuracy of 95.9%. However, only 12 unique datasets were identified across all studies, indicating substantial dataset reuse. These findings highlight the need for more diverse datasets and stronger real-world validation to support reliable AI deployment in malaria-endemic settings.

Speaker(s):
Ghofrane Bousbih, Master's in Science in Health Informatics and Bioinformatics
Grand valley state university

Author(s):
Suhila Sawesi, PhD - GVSU;

Understanding Information Needs in Chronic Kidney Disease Care: Perspectives of Patients, Carers, and Healthcare Professionals

Poster Number: 203
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Delivering Health Information and Knowledge to the Public, Chronic Care Management, Qualitative Methods, Data Sharing, Transitions of Care
Working Group: Primary Care Informatics Working Group
Programmatic Theme: Clinical Informatics

Patients with chronic kidney disease increasingly access health information through digital technologies, including patient portals, online resources, and generative AI tools. This qualitative study explores how patients, carers and health professionals navigate and verify health information across fragmented healthcare systems. Interviews with 55 participants revealed multi-source information seeking, AI-assisted information exploration and carers acting as informal coordinators of patient information. Findings indicate implications for the design of patient-centred health information systems.

Speaker(s):
Jun Wang, Phd student
University of Sheffield

Author(s):
Denis Newman-Griffis, PhD - University of Sheffield;

A Modular Agent-Based Chatbot for Natural Language Clinical Query Answering in OpenMRS

Poster Number: 204
Presentation Time: 05:00 PM - 06:30 PM

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

Clinical decision support in pediatrics lacks real-time EHR integration at point of care. We developed a natural language chatbot querying live OpenMRS data through a five-stage validation pipeline integrating EHR records, FDA and pediatric knowledge bases, with local LLM inference for offline deployment. A validation agent prevents hallucinated responses by enforcing evidence existence. Clinicians receive evidence-based recommendations and patients receive plain language explanations for medication contraindications, allergies, immunizations, and dosing.

Speaker(s):
Meghana Hudgi, Master’s in Health Informatics
Indiana University

Author(s):
Animisha Chitikeneni, Master student - Indiana University Indianapolis; Hao Liu, PhD - Montclair State University; Yan Zhuang, PhD, FAMIA - Indiana University;

Evaluating a Prostate Cancer Health Information Platform for Rural Patients and Families: A Community-based Usability Study

Poster Number: 205
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Delivering Health Information and Knowledge to the Public, Usability, Health Equity
Programmatic Theme: Consumer Health Informatics

Prostate cancer patients in rural areas have higher mortality rates than those in urban areas. This study evaluates the usability of the Interactive Prostate Cancer Information, Communication, and Support Program (iPICS), an evidence-based, user-centered consumer health information platform, by partnering with three rural public libraries in North Carolina. Usability testing has been conducted with rural patients and caregivers to assess user experience and the feasibility of delivering reliable health information in underserved rural regions.

Speaker(s):
Fei Yu, PhD
UNC at Chapel Hill

Author(s):
Fei Yu, PhD - UNC at Chapel Hill;

LimbRehabCaRES: Implementing a Learning Health System for Limb Loss Rehabilitation

Poster Number: 206
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Disability, Accessibility, and Human Function, Healthcare Quality, Patient-/Person-Generated Health Data, Informatics Implementation, Workforce Development, Patient Engagement and Preferences, Governance, Transitions of Care
Programmatic Theme: Clinical Informatics

Access to coordinated limb loss care in the United States remains limited, contributing to delayed rehabilitation and increased healthcare costs. A Learning Health System, LimbRehabCaRES, was designed using the Limb Loss Rehabilitation Continuum (LLRC) framework and implemented at Penn State to integrate coordinated care pathways, EHR data, and analytics dashboards. The center supports patient journey optimization and data governance while enabling digital health innovation, education initiatives, and community engagement to improve limb loss rehabilitation care delivery.

Speaker(s):
Prateek Grover, MD PhD MHA
Penn State

Author(s):
Shadi Hijjawi, MD - Penn State Health; Anilchandra Attaluri, PhD - Penn State University; Syed Reza, PhD - Penn State University; Truong Tran, PhD - Penn State University; Thiru Annaswamy, MD MA - Penn State Health;

Seen, Safe, and Supported? Understanding the Need for Affirming Online Spaces for Black LGBTQ+ Youth

Poster Number: 207
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Diversity, Equity, Inclusion, and Accessibility (DEIA), Health Equity, Population Health, Public Health, Surveys and Needs Analysis
Programmatic Theme: Public Health Informatics

Black LGBTQ+ youth experience elevated mental health risks driven by stigma and discrimination. This study analyzed survey data from 164 Black LGBTQ+ youth (ages 15–24) to explore support within online communities. Results showed online spaces are critical for identity exploration, affirmation, and peer support, particularly for transgender youth. Participants frequently provided relational support, though material aid was rare. Despite benefits, youth reported missing in-person connection, highlighting the need for complementary online and in-person support spaces.

Speaker(s):
Wisdom Dayok, BS
Yale University

Author(s):
Wisdom Dayok, BS - Yale University; Kayla Hightower, PhD Psychology - Dr. Terika McCall; Mary Peng, MPH, MS - Department of International Health, Johns Hopkins School of Public Health;

Contributors to Communication Burden in the Electronic Health Record

Poster Number: 208
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Documentation Burden, Usability, User-centered Design Methods, Information Extraction
Programmatic Theme: Clinical Informatics

Secure messaging in electronic health records is a ubiquitous communication tool in hospitals that has had the unintended consequence of increasing the communication burden on providers. Clinicians have some control to set preferences over which secure messages create interruptive notifications. We evaluated the influence of secure message volume and related push notifications to assess their relative impact on time in communication activities. Push notifications have a stronger influence on time in communication activities than messages sent and received.

Speaker(s):
John Will, MPA
NYU Langone Health

Author(s):
William Small, MD, MBA - NYU Langone Health; Jonathan Austrian, MD - NYU Langone Health; Paul Testa, MD, JD, MPH - NYU Langone Health; Jonah Feldman, MD, FACP - NYU Langone Health;

Longitudinal Changes in Ambulatory Clinicians’ Perceived Workload After Implementation of Ambient Artificial Intelligence Scribes

Poster Number: 209
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Documentation Burden, Artificial Intelligence, Informatics Implementation
Programmatic Theme: Clinical Informatics

As it is unclear how perceived workload changes over time after ambient artificial intelligence (AI) scribe implementation nor what the correlates of changes in perceived workload are, we developed linear mixed-effects models using longitudinal survey data (n=67 ambulatory clinicians). The ambient AI scribe was not significantly associated with perceived workload immediately and over time after implementation. Additional implementation strategies are needed to meaningfully reduce perceived workload.

Speaker(s):
Oliver Nguyen, MSHI
University of Wisconsin at Madison

Author(s):
Majid Afshar, MD, MSCR - University of Wisconsin - Madison; Mary Ryan Baumann, PhD - University of Wisconsin at Madison; Michael Jaeb, PhD, RN - 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 II, 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 - UW Health; Joel Gordon, MD - UW Health, University of Wisconsin School of Medicine and Public Health;

Environmental Impact of Digital Healthcare Technologies

Poster Number: 210
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Environmental Health and Climate Informatics, Telemedicine, Evaluation
Programmatic Theme: Consumer Health Informatics

Climate change is a global challenge, and digital healthcare technologies (DHTs) may help reduce healthcare-related carbon footprints. This scoping review synthesised evidence on their environmental impact. Seventy studies were included, mainly examining telehealth and travel-related emission reductions. Most reported decreased greenhouse gas emissions. Some noted increased electricity use and lifecycle impacts. Methodological variability and limited life cycle assessments indicate the need for more rigorous, comprehensive evaluations.

Speaker(s):
Sarath Rathnayake, RN, BScN, MScN, PhD
University of Bradford, United Kingdom

Author(s):
Sarath Rathnayake, RN, BScN, MScN, PhD - University of Bradford, United Kingdom; Chinasa Odo, BSc, MSc, PhD - University of Bradford, United Kingdom; Hadiza Ismaila, Bsc, MSc, PhD - University of Bradford, United Kingdom; Natasha Alvarado, BA, MA, PhD - University of Bradford, United Kingdom; Veronica Parisi, BA, MSc - University of Bradford, UK; Joshua Pink, MMath, MSc, PhD - University of Bradford, UK; Ruth Agbakoba, BSc< MSc, PhD - University of Bradford, UK; Rebecca Randell, PhD - University of Bradford;

Toward the "Social E": An AI-Assisted Framework for Integrating Social, Environmental, and Occupational Risk Factors into the Electronic Health Record

Poster Number: 211
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Environmental Health and Climate Informatics, Artificial Intelligence, Clinical Decision Support, Infectious Diseases & Epidemiology, Teaching Innovation, Usability
Working Group: Climate, Health and Informatics Working Group
Programmatic Theme: Clinical Informatics

Despite growing EHR adoption of structured social risk data, environmental determinants—particularly those interacting with social vulnerability—remain poorly integrated despite increasing clinical relevance. We describe creation of a clinical training curriculum that demonstrates the importance of synthesizing social and environmental data elements for preventive intervention and propose an AI-assisted approach to social history documentation that systematically captures actionable social, environmental, and occupational risk factors, advancing a comprehensive "Social E" framework within the EHR.

Speaker(s):
Peter Rabinowitz, MD MPH
University of Washington

Author(s):
Peter Rabinowitz, MD MPH - University of Washington; Manijeh Berenji, MD MPH - UC Irvine School of Medicine/Joe C Wen School of Population and Public Health (at UC Irvine); VA Long Beach Healthcare System;

Evaluation of an EHR-derived pregnancy episode identification algorithm against a clinical obstetric registry

Poster Number: 212
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Evaluation, Real-World Evidence Generation, Population Health, Data Mining, Public Health
Programmatic Theme: Clinical Research Informatics

We evaluated an EHR-derived pregnancy episode identification algorithm by comparing algorithm-derived episodes and complications with a gold-standard clinical obstetric registry. Pregnancy episodes were matched using temporal overlap, and agreement in gestational age and pregnancy complications was assessed. Results showed high agreement across gestational age estimates and major pregnancy complications. These findings show a promising potential of the use of EHR-derived pregnancy timelines for large-scale maternal health research and evaluation of time-sensitive exposures during pregnancy.

Speaker(s):
Minqi Xiong, MS
Johns Hopkins University School of Medicine

Author(s):
Robert Barrett, BS - Johns Hopkins University; Theresa Boyer, PhD - Johns Hopkins Bloomberg School of Public Health; Benjamin Martin, PhD - Johns Hopkins School of Medicine; Anum Minhas, MD, MHS - Johns Hopkins School of Medicine;

Generating Realistic Missing Data using HGB-based Framework

Poster Number: 213
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Evaluation, Data Mining, Machine Learning
Programmatic Theme: Clinical Research Informatics

Real-world data in biomedicine often has missing values, requiring robust imputation strategies. Comparing these methods is difficult without ground truth values. While logistic regression has been used to simulate missing at random mechanisms, it may not capture longitudinal data complexities. We propose a framework using Histogram-based Gradient Boosting (HGB) for artificial missing data generation. HGB handles non-linear relationships and models complex missingness while preserving longitudinal dynamics. This provides a benchmark for evaluating imputation methods.

Speaker(s):
Yein Jeon, M.S.
University of Washington

Author(s):
Yein Jeon, M.S. - University of Washington; Jordan Gauthier, MD, MSc - Fred Hutch Cancer Center; Qian Wu, PhD - Fred Hutch Cancer Center;

Ping, Don’t Ring: A Randomized Trial of Text‑Based Post‑Discharge Follow‑Up

Poster Number: 214
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Evaluation, Mobile Health, Delivering Health Information and Knowledge to the Public
Programmatic Theme: Consumer Health Informatics

Text messaging offers a resource-efficient alternative to post-discharge phone calls. We conducted a rapid-cycle quality-improvement study comparing phone calls versus texts for elective surgery patients. ED visits and readmissions did not differ significantly between groups, though power was limited. To assess the likelihood of missed harm, bootstrap simulations were conducted. Text patients had worse outcomes in only 12% of simulations for ED visits and 31% for admissions, suggesting texts are a safe, resource-efficient alternative.

Speaker(s):
Yin Jian, MPH
NYU Langone Health

Author(s):
Nathan Klapheke, BS - NYU Langone Health;

Enabling Visual Cohort Comparison in Notebook Environments

Poster Number: 215
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Fairness and Elimination of Bias, Information Visualization, Health Equity
Programmatic Theme: Clinical Research Informatics

We present a prototype of a Python package and interactive widget that healthcare researchers can use to analyze EHR-derived cohorts. This builds on previous work to bring interactive visualizations and dynamic reweighting to the Jupyter Notebook environment, a platform that is widely used for data analytics. We aim to iteratively develop this tool with the goal of empowering researchers to visualize and mitigate selection bias, and thus enhance the generalizability and validity of their research.

Speaker(s):
Viola Goodacre, MSc, MPS
University of North Carolina at Chapel Hill

Author(s):
Viola Goodacre, MSc, MPS - University of North Carolina at Chapel Hill; Hong Yi, Ph.D. - University of North Carolina at Chapel Hill; Shuhan Lu, MS - University of North Carolina at Chapel Hill; Estella Calcaterra, BS - University of North Carolina at Chapel Hill; Natthawut Adulyanukosol - University of North Carolina at Chapel Hill; David Borland, PhD - RENCI, The University of North Carolina at Chapel Hill; David Gotz, PhD - University of North Carolina;

Fairness Evaluation of Large Language Models in Biomedical Question Answering: Dual Perspectives of Medical Researchers and Patients

Poster Number: 216
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Fairness and Elimination of Bias, Large Language Models (LLMs), Artificial Intelligence
Programmatic Theme: Clinical Research Informatics

Large Language Models (LLMs) have demonstrated impressive performance in biomedical question answering (QA) tasks; however, concerns remain regarding their fairness across demographic groups. This study systematically evaluates group fairness of LLMs in biomedical QA from dual perspectives of medical researchers and patients, using the PubMedQA, Chinese NMLE, and USMLE datasets. Under researcher identity framing, models produced consistent but less accurate responses, with nationality introducing measurable bias, providing contextual information improved both accuracy and fairness. In patient-group evaluation, models performed better on Chinese datasets. A reasoning mode significantly improved response consistency for USMLE data. All models satisfied fairness criteria across gender and age dimensions. These findings suggest that identity and patient attributes can influence model performance, but such biases can be mitigated through deliberate mechanism design, providing a foundation for developing fairer biomedical LLMs.

Speaker(s):
Yiran Shu, BS
ShanghaiTech University

Author(s):
Yiran Shu, BS - ShanghaiTech University; Yihan Wang, BS - ShanghaiTech University; Junyao Tang, BS - ShanghaiTech University; Zhouyuan Fan, BS - ShanghaiTech University; Sihan Xie, Bachelor - Shanghaitech University; Hongzhu Jiang, bachelor - ShanghaiTech University; Zhiyu Wan, PhD - ShanghaiTech University;

Regional Disparities in Digital Health Adoption Across Indian States: An Analysis of ABHA IDs, eSanjeevani Teleconsultations, and ABDM Infrastructure

Poster Number: 217
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Global Health, Public Health, Delivering Health Information and Knowledge to the Public, Geospatial (GIS) Data/Analysis, Data transformation/ETL, Diversity, Equity, Inclusion, and Accessibility, Evaluation, Population Health
Working Group: Public Health Informatics Working Group
Programmatic Theme: Public Health Informatics

India is moving toward a more digital healthcare system through the Ayushman Bharat Digital Mission (ABDM),
but progress is not equal across all states. This study looks at how different regions are using digital health tools
by analyzing three main indicators: the number of ABHA health IDs created, eSanjeevani teleconsultations, and
ABDM-verified healthcare facilities. We used official data sources and adjusted the numbers based on population to make fair comparisons. The results show that southern states are leading in digital health use, while eastern, central, and northeastern states are falling behind. We also found that being a richer state does not always mean better digital health adoption other factors like good planning, infrastructure, and public awareness matter more. Based on these findings, we suggest region based solutions to help all areas benefit from digital health and move closer to a connected, nationwide healthcare system.

Speaker(s):
Mukesh Patel, Research Software Developer
Temple University

Author(s):
Simran Balhara, MSHI - Temple University; Aditi Vaid, MSHI - Temple University; Tsedey Tsegaye, MSHI - Fox Chase Cancer Center; Tulay G Soylu, PhD - Barnett College of Public Health: Health Services Administration, Temple Univeristy;

Exploring the Impact of Temperature Anomalies on Global Hypertension Trends Using Population and Environmental Data: A Regional and Sex-Based Informatics Approach

Poster Number: 218
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Global Health, Geospatial (GIS) Data/Analysis, Policy
Programmatic Theme: Public Health Informatics

This study explores the impact of climate change, particularly temperature anomalies, on global hypertension trends, highlighting regional and sex disparities. We utilized large-scale population and environmental data to analyze the impact of climate trends, particularly increasing temperature anomalies, on hypertension prevalence across various regions. While global hypertension prevalence hasn't worsened, regions with better healthcare, like Europe and North America, show improvements. In contrast, Africa faces higher hypertension rates due to limited healthcare. Women generally have better hypertension control than men. Our findings suggest the need for targeted interventions and stronger healthcare infrastructure, particularly in lower-income regions. Sex differences also indicate that women tend to have more favorable hypertension outcomes than men, suggesting a need for targeted interventions to address these disparities.

Speaker(s):
Jiancheng Ye, PhD
Weill Cornell Medicine

Author(s):
Haoxin Chen, MS - The University of Hong Kong;

Gap Analysis of Reporting Frameworks: Assessing Alignment with FDA Transparency Requirements for AI/ML-Enabled Software

Poster Number: 219
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Governance, Artificial Intelligence, Policy, Machine Learning, Data Standards, Fairness and elimination of bias, Patient Safety, Healthcare Quality
Programmatic Theme: Clinical Research Informatics

FDA authorizations of AI/ML-enabled medical devices have grown exponentially since 2016, yet transparency deficits persist. The FDA's 2024 transparency guidance, developed with Health Canada and the UK's MHRA, establishes 22 disclosure principles, but no assessment exists of whether AI/ML frameworks align with these recommendations. We deconstructed these principles into 99 binary-assessable questions and evaluated six frameworks against them. Preliminary findings indicate coverage gaps of over 30%, concentrated in user-facing transparency and lifecycle management.

Speaker(s):
Yash Raka, Bachelors
University of Pittsburgh, School of Medicine

Author(s):
Yash Raka, Bachelors - University of Pittsburgh, School of Medicine; Shyam Visweswaran, MD PhD - University of Pittsburgh;

How Can 211 Connect Clinics to Community? Discussing Lessons Learned from a Longterm Statewide Partnership between Michigan 211 and Specialty Care Providers to Address Unmet Non-Medical Needs in Michigan

Poster Number: 220
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Health Equity, Population Health, User-centered Design Methods
Programmatic Theme: Clinical Informatics

As healthcare delivery increasingly relies on electronic referrals to address patients’ non-medical needs, evidence remains scarce regarding the real-world effectiveness, feasibility, and patient acceptability of these platforms. While commercial solutions (e.g., Unite Us) garner attention, they raise concerns around cost, data ownership, and directory reliability. In contrast, our team has piloted integration of Michigan 211—a statewide, nonprofit social resource referral system—using low-touch electronic modalities across multispecialty clinical environments. This panel presents findings from qualitative and quantitative evaluations of CQI pilot sites and participatory design processes, highlighting strategies for optimizing provider workflows and engaging patients in referral linkage design. Panelists will discuss the unique capacities of 211 for social informatics interventions, lessons learned in multispecialty implementation, and approaches for maximizing patient engagement. Attendees will gain actionable insights for designing scalable, equitable, and patient-centered referral tools that bridge healthcare and social service systems.

Speaker(s):
Emma Lentini, LLMSW
Michigan 211

Author(s):
Bradley Iott, MPH, PhD - University of Michigan;

Operationalizing Post-Market Equity Surveillance for AI-Enabled Medical Devices: A Manufacturer-Centered Informatics Architecture

Poster Number: 221
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Health Equity, Artificial Intelligence, Governance, Policy, Real-World Evidence Generation, Fairness and elimination of bias, Evaluation, Data Modernization
Programmatic Theme: Public Health Informatics

AI-enabled medical devices are evaluated pre-market but lack structured post-market equity monitoring. We propose a manufacturer-centered Equity Surveillance Architecture (ESA) that requires periodic subgroup performance reporting, structured generation of fairness metadata, cross-site drift detection, and integration of regulatory feedback. By embedding equity surveillance into lifecycle obligations, ESA reframes fairness as a post-market accountability function rather than an institutional quality improvement task.

Speaker(s):
Sahar Zahid, MD
Arizona State University

Author(s):
Sahar Zahid, MD - Arizona State University;

Digital Health Equity: A Concept Analysis

Poster Number: 222
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Health Equity, Informatics Implementation, Policy, Artificial Intelligence, Telemedicine
Working Group: Health and Healthcare Equity Working Group
Programmatic Theme: Consumer Health Informatics

Digital health tools are integral to care delivery, yet definitional inconsistency in digital health equity impedes research, policy, and practice. Using the Walker and Avant eight-step method, this analysis identifies four defining attributes: equitable reach, usable design, supported engagement, and systemic equity monitoring. It also outlines the key antecedents, consequences, and measurable referents of the concept. An operational definition is proposed to advance conceptual clarity and guide implementation.

Speaker(s):
Abdul-Ghaniyu Alidu, PhD
University of Michigan

Author(s):
Yun Jiang, PhD, MS, RN, FAMIA - University of Michigan;

Standardizing electronic health record portal-based financial hardship and health-related social needs screening into a structured dashboard

Poster Number: 223
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Health Equity, Patient-/Person-Generated Health Data, Real-World Evidence Generation, Surveys and Needs Analysis
Programmatic Theme: Clinical Informatics

In this project, we developed a standardized electronic health record (EHR) portal–based screening process for financial hardship and health-related social needs. Screening results are integrated into a structured dashboard that enables clinical and research teams to systematically identify unmet needs, support efficient and timely referrals, and improve precision social care delivery that may ultimately improve patient outcomes.

Speaker(s):
Melissa Beauchemin, PhD, RN, MS, CPNP, FAAN
Columbia University

Author(s):
Melissa Beauchemin, PhD, RN, MS, CPNP, FAAN - Columbia University; David DeStephano, MS - Columbia University; Natalie Benda, PhD - Columbia University School of Nursing; Nickolas Dreher, MD - Columbia University; Claire Sathe, MD, JD - Columbia University; Dawn Hershman, MD, MS - Columbia University;

Leveraging Large Language Models to Identify Empathic Communication in Patient Portal Messages

Poster Number: 224
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Healthcare Quality, Natural Language Processing, Machine Learning, Large Language Models (LLMs), Artificial Intelligence
Programmatic Theme: Clinical Informatics

Large language models (LLMs) may enable scalable measurement of empathic communication in patient–clinician portal messaging. Using annotated patient and clinician messages, we evaluated two open-source LLMs under multiple prompting strategies and compared performance with supervised models. LLMs achieved strong performance identifying patient empathic opportunities and clinician empathy, often matching or exceeding supervised approaches, suggesting LLMs can support large-scale assessment of empathic communication in digital healthcare settings.

Speaker(s):
Vishal Shetty, PhD
Geisinger

Author(s):
Brendan O'Connor, PhD - University of Massachusetts Amherst; Christina Gregor, BA - Geisinger; Lorraine Tusing, BA - Geisinger; Airín Martínez, PhD - University of Massachusetts Amherst; Eric Wright, PharmD, MPH - Geisinger; David Chin, PhD;

GENE-QC: A Browser-Based Multi-Omics Data Quality Management Platform Adapting the Clinical Data Quality Framework

Poster Number: 225
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Healthcare Quality, Workflow, Administrative Systems, Evaluation
Working Group: Clinical Information Systems Working Group
Programmatic Theme: Translational Bioinformatics

We developed GENE-QC, a browser-based platform that operationalizes the clinical data quality management life cycle framework—encompassing the planning, construction, operation, and utilization stages—for file-based multi-omics and clinical metadata validation. The platform implements 34 quality metrics across three core dimensions (Completeness, Plausibility, Conformance) in a File–Column–Value hierarchy, and incorporates MOCHI, an adversarial NMF-Transformer imputation model, enabling end-to-end quality assessment and remediation

Speaker(s):
Doyeon An, M.P.H.
Gachon University

Author(s):
Minsik Lim, B.A. - Gachon University, Republic of Korea; Suehyun Lee; Suehyun Lee, Ph D - Gachon University;

EmoBot: Explainable Three-Tier Risk Stratification for Cantonese-Speaking Youth Seeking Mental Health Support

Poster Number: 226
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Human-computer Interaction, Large Language Models (LLMs), Workflow, Clinical Decision Support
Programmatic Theme: Consumer Health Informatics

EmoBot is an explainable three-tier risk stratification chatbot for Cantonese-speaking youth, integrating case-based retrieval and constrained LLM reasoning to support nurse-in-the-loop triage. Evaluated on 382 expert-consensus statements, EmoBot achieved 95.8% accuracy and 95.2% macro F1, without tier confusions. By providing tier predictions, highlighted risk cues, and exemplar-based evidence, EmoBot enables scalable and explainable workflows for low-resource language settings, addressing barriers to youth mental health access while maintaining safety through transparent human review.

Speaker(s):
Vivian Hui, RN, PhD
The Hong Kong Polytechnic University

Author(s):
Hin Chi Kwok, Bachelor - The Hong Kong Polytechnic University; Xinyu Feng, Master - The Hong Kong Polytechnic University; Perlie Chung, BSc - The Hong Kong Polytechnic Univeristy; Deborah Lee, MSc - The Hong Kong Polytechnic University; Joey Lam, RN (Mental Health) - Hong Kong Polytechnic University; Gregor Stiglic, PhD - University of Maribor; Shaowei GUAN, BSc - The Hong Kong Polytechnic University;

Enabling Adaptive Patient Check-Ins: Generative AI Conversational Agents for Collecting Patient-Reported Outcomes

Poster Number: 227
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Human-computer Interaction, Large Language Models (LLMs), Patient-/Person-Generated Health Data, Artificial Intelligence
Programmatic Theme: Consumer Health Informatics

Incorporating generative AI within conversational agents can support more dynamic and flexible collection of patient-reported outcomes (PROs), improving user experience. We describe the development of a conversational agent designed to collect PROs for incretin mimetic medication management. Through an iterative prompt engineering process, we developed a two-phase prompt architecture to manage PRO collection within the agent. In preliminary testing, the agent demonstrated high accuracy in appropriately eliciting and capturing PROs.

Speaker(s):
Angela Mastrianni, PhD
NYU Grossman School of Medicine

Author(s):
Angela Mastrianni, PhD - NYU Grossman School of Medicine; Natalie Henning; Katerina Andreadis, MS - NYU Grossman School of Medicine; Javier Gonzalez, BS - NYU Langone Health; Danissa Rodriguez Caraballo, PhD Computer science - NYU Grossman School of Medicine; Tiffany Rose Martinez, MPH - NYU Grossman School of Medicine; Devin Mann, MD - NYU Grossman School of Medicine; Elizabeth Stevens, PhD, MPH - NYU Grossman School of Medicine;

How the Digital World Shapes Sobriety: The Role of Digital Tools in Recovery for Alcohol Use Disorder

Poster Number: 228
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Human-computer Interaction, Patient Engagement and Preferences, Qualitative Methods
Programmatic Theme: Consumer Health Informatics

Alcohol use disorder (AUD) is a debilitating illness that has affected approximately 29 million people in the United States with numerous health consequences and impact on individuals’ social lives. To understand how digital tools factor into sobriety, we surveyed and interviewed 12 adults with AUD. Using inductive analysis, we identified four major themes which reveal how digital resources, online communities, and social media play a role in maintaining sobriety for those living with AUD.

Speaker(s):
Elizabeth Byrd, Information Science
UNC-Chapel Hill

Author(s):
Elizabeth Byrd, Information Science - UNC-Chapel Hill; Abigail Egnatz, BS - UNC Chapel Hill; William Payne, PhD - UNC Chapel Hill;

From Patient Perspectives to an Ethics-First Architecture for AI-Powered Smart Glasses for Health Self-Management

Poster Number: 229
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Human-computer Interaction, Artificial Intelligence, Privacy and Security, Patient-/Person-Generated Health Data, Interoperability and Health Information Exchange, Patient Engagement and Preferences
Programmatic Theme: Consumer Health Informatics

AI-powered smart glasses can capture continuous visual context for health self-management, but patient perspectives on their ethical implications remain underrepresented. Through a literature review of academic, policy, and legal sources, we synthesize patient concerns into four challenge domains: surveillance and privacy, data ownership, autonomy, and trust & bias. The research aims to translate identified challenges into an actionable ethical architecture for open-source smart-glasses platforms and applications.

Speaker(s):
Tim Schwirtlich, MS
Northwestern University

Author(s):
Tim Schwirtlich, MS - Northwestern University; Abel Kho, MD, FACMI - Northwestern University;

Possible ICD-10 miscoding of coccidioidomycosis as Rift Valley Fever in the U.S. Veterans Health Administration (VHA) detected by EHR-based surveillance

Poster Number: 230
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Infectious Diseases and Epidemiology, Information Extraction, Information Visualization, Patient-/Person-Generated Health Data, Population Health, Quantitative Methods
Programmatic Theme: Public Health Informatics

Leveraging the VA SHIELD Vector-borne, Zoonotic, and Environmentally-transmitted Diseases (VBZED) surveillance dashboard, we evaluated possible miscoding of coccidioidomycosis as Rift Valley fever. We analyzed diagnostic frequency within the VA and compared it to CDC reports. Findings suggest miscoding occurred in the outpatient setting following introduction of ICD-10 in 2015. This has implications for using ICD-10 codes in disease surveillance.

Speaker(s):
Julienne Reynolds, M.Sc
Durham VA Medical Center / Institute for Medical Research

Author(s):
Jack Anderson, BS - Durham VA Medical Center; Ariana Paredes-Vincent, MPH - Veterans Health Administration / Aptive HTG; Pankaj Agarwal, MSc - Durham VA Medical Center / Institute for Medical Research; Lauren Epstein, MD MSc - Atlanta VA Healthcare System; Jimmy Efird, PhD - Jamaica Plains VA Medical Center; Michael Gelman, MD, PhD - Veterans Administration Bronx; Robert Bonomo, MD - Louis Stoke Cleveland VA Medical Center; Maria Rodriguez-Barradas, MD - Michael E DeBakey VA Medical Center; Christopher Woods, MD MPH - Durham VA Medical Center; Sheldon Brown, MD - Hudson Valley VA Healthcare System;

Identifying Clinical Features Associated with Severe Outcomes and Subphenotypes in Multisystem Inflammatory Syndrome in Children

Poster Number: 231
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Infectious Diseases and Epidemiology, Patient-/Person-Generated Health Data, Machine Learning
Programmatic Theme: Clinical Research Informatics

Multisystem inflammatory syndrome in children (MIS-C) is a rare hyperinflammatory condition following SARS-CoV-2 infection. We assembled a multicenter cohort of 147 adjudicated MIS-C cases and evaluated laboratory and clinical features against nine outcomes. Neutrophilia was independently associated with vasoactive agent use, external oxygenation, and ICU admission. Spectral clustering identified three subphenotypes with divergent outcomes, revealing that older age alone does not determine severity. Phenotype-based risk stratification may outperform single markers for MIS-C.

Speaker(s):
Vijeeth Guggilla, BA
Northwestern University

Author(s):
Vijeeth Guggilla, BA - Northwestern University; Michael Semanik, MD - University of Wisconsin - Madison; Dominic Co, MD/PhD - University of Wisconsin-Madison; Brian Nolan, MD - Northwestern University Feinberg School of Medicine; Lacey Gleason, MSPH - Northwestern University Feinberg School of Medicine; Alona Furmanchuk, PhD - Center for Health Information Partnerships, Northwestern University, , Feinberg School of Medicine; Theresa Walunas, PhD - Northwestern University; Judith Smith, MD/PhD - Unviersity of Wisconsin-Madison;

Firefly: An EHR-Integrated Platform for Scalable Competency-Based Assessment and Learning Analytics in Surgical Training

Poster Number: 232
2026 Symposium LIEAF Presentation

Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Informatics Implementation, Artificial Intelligence, Information Extraction, Human-computer Interaction, Surgery, Teaching Innovation, Workflow, User-centered Design Methods
Programmatic Theme: Academic Informatics / LIEAF

Competency-based medical education requires scalable systems capable of generating reliable competency signals from clinical workflows. We describe Firefly Lab, an EHR-interfaced platform that uses HL7 and SMART on FHIR interfaces to capture procedural assessment data during routine clinical care. By combining human-centered workflow design with learning analytics, including learning curve modeling and utility analysis, the platform transforms clinical activity into longitudinal competency datasets and enables learning health systems approaches to graduate medical education.

Speaker(s):
Ruchi Thanawala, MD, MS
Oregon Health and Science University

Author(s):
Ruchi Thanawala, MD, MS - Oregon Health and Science University; Phillip Jenkins, MD - Oregon Health & Science University; Jonathan Jesneck, PhD - Firefly Foundation;

Evaluating a Digital Decision Aid for Atrial Fibrillation Rhythm Control: A Hybrid Implementation-Effectiveness Trial

Poster Number: 233
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Informatics Implementation, Patient Engagement and Preferences, Clinical Decision Support
Programmatic Theme: Consumer Health Informatics

Digital decision aids hold promise for improving shared decision-making, but effectiveness depends on more than good design. In a hybrid implementation-effectiveness trial of a web-based atrial fibrillation decision aid, we found high acceptability but heterogeneous patient outcomes, moderated by patient readiness, health literacy, delivery timing, and clinical context. We present lessons learned for embedding decision support into learning health system workflows to reach the right patient, with the right tool, at the right time.

Speaker(s):
Meghan Reading Turchioe, PhD, MPH, RN
Columbia University School of Nursing

Author(s):
Afra Shamnath, Master of Public Health - NYU; Yihong Zhao, PhD - Columbia University; David Slotwiner, MD - NYP Queens/ Weill Cornell Medicine; Angelo Biviano, MD - Columbia University Irving Medical Center;

From Grant to User Story: The Iterative Thematic User Story Framework (ITUSF) for Agile Ideation in a Diabetes Research Context

Poster Number: 234
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Informatics Implementation, User-centered Design Methods, Mobile Health, Patient Engagement and Preferences, Population Health, Artificial Intelligence
Programmatic Theme: Consumer Health Informatics

Consumer health informatics often adopts agile development methods designed for commercial settings, which can conflict with the fixed scope and timelines of grant-funded research. We developed the Iterative Thematic User Story Framework (ITUSF), a three-phase model (Exploration, Prioritization, Refinement) that integrates thematic coding and structured consensus to guide user story ideation for a type 2 diabetes digital health trial while maintaining alignment with grant constraints.

Speaker(s):
Lynn Xu, MPH
NYU Grossman School of Medicine

Author(s):
Lynn Xu, MPH - NYU Grossman School of Medicine; Amelia Shunk, MMCi - NYU Grossman School of Medicine; Katerina Andreadis, MS - NYU Grossman School of Medicine; JaeEun Kwon, Master of Public Policy - NYU Langone Health; Danissa Rodriguez Caraballo, PhD Computer science - NYU Grossman School of Medicine; Javier Gonzalez, BS - NYU Langone Health; Soumik Mandal, PhD - New York University Grossman School of Medicine; Priscilla D’Antico, PhD - NYU Langone Health; Gina-Maria Arena, MA - NYU Langone Health; Ronaldo Patino Flores, BA - NYU Langone Health; Jocelyn Cruz, MPH - NYU Langone Health; Antoinette Schoenthaler, EdD - NYU Langone Health; Devin Mann, MD - NYU Grossman School of Medicine;

Longitudinal EHR-Based Prediction of Cardiovascular Disease Risk in Autism Spectrum Disorder Using Machine Learning Models

Poster Number: 235
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Informatics Implementation, Machine Learning, Population Health, Healthcare Quality
Programmatic Theme: Clinical Research Informatics

Autistic individuals have 3.46 times higher odds of cardiovascular disease (95% CI: 3.38-3.54). Three ML models were developed using up to ten years of longitudinal EHR data from 42,624 autistic patients below the age of 30. Longitudinal LSTM performed best with AUROC 0.720. Late autism diagnosis, sleep disorders, and use of anti-anxiety and anticonvulsant medications were the strongest predictors across all models, highlighting the need for early cardiovascular screening in these higher-risk groups.

Speaker(s):
Alina Rohulia, BS
University of Missouri-Columbia

Author(s):
Alina Rohulia, BS - University of Missouri-Columbia; Adnan Qureshi, MD - University of Missouri-Columbia; Benjamin Black, MD - University of Missouri - Columbia; Chi-Ren Shyu, PhD, FACMI, FAMIA - University of Missouri-Columbia;

Contextualizing Peripartum Mental Health to Model Dyadic Ties in Population Health Informatics

Poster Number: 236
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Informatics Implementation, Population Health, Knowledge Representation & Information Modeling
Programmatic Theme: Public Health Informatics

This study employs an informatics approach to evaluate peripartum depression by establishing the parental dyad as the primary data unit. Utilizing the NIMHD Framework, we mapped 11 conceptual features across biological, interpersonal, and structural domains, including insurance and housing as critical drivers for connecting the dots in underserved couples. This multi-level modeling bridges clinical indicators with disparate social domains to analyze how these determinants shape health trajectories in fragile families.

Speaker(s):
Iswaria Gnanadass, Phd
University Of Texas Health Houston

Author(s):
Sarah Jaleel, BS - University of Texas Houston Health Science Center; Sahiti Myneni, PhD - University of Texas Health Science Center- Houston;

Using Social Determinants of Health to Guide Digital Health Interventions in a Learning Health System

Poster Number: 237
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Informatics Implementation, Telemedicine, Transitions of Care
Programmatic Theme: Clinical Informatics

Heart failure is a leading cause of hospitalization, and identifying patients who benefit most from digital health interventions remains challenging. Using data from the randomized Mighty-Heart trial, we examined whether neighborhood-level social determinants of health improve prediction of 30-day readmission. Social deprivation measures were among the strongest predictors and interacted with the mobile integrated health intervention. Incorporating social risk into informatics systems may help target digital health resources at the point of care.

Speaker(s):
Sarah Danziger, MD
Columbia/New York Presbyterian

Author(s):
Ruth Masterson Creber, PhD, MSc, RN - Columbia University; Meghan Reading Turchioe, PhD, MPH, RN - Columbia University School of Nursing; Melani Ellison, MPH - Columbia University School of Nursing; Stacey Dai, MPH - Columbia University School of Nursing; Brock Daniels, MD, MPH - Weill Cornell Medicine; Yujing Fu, BS - Columbia University; John Won, BS - Columbia University; Rita Xiong, BS - Columbia University;

Temporal Network Analysis of Activities of Daily Living Preceding Incident Mild Cognitive Impairment

Poster Number: 238
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Information Extraction, Information Visualization, Information Retrieval
Programmatic Theme: Clinical Informatics

This study characterizes the temporal evolution of activities of daily living (ADL) co-impairment networks in the five years preceding incident mild cognitive impairment (MCI) using data from the Mayo Clinic Study of Aging. Network analysis revealed distinct functional trajectories: individuals progressing to dementia showed denser and more strongly connected ADL networks, whereas unstable MCI exhibited sparse connectivity. Increasing ADL network connectivity and clustering preceded dementia progression, suggesting network-based functional measures may provide early indicators of cognitive decline.

Speaker(s):
Muskan Garg, Postdoctoral Research Associate
Mayo Clinic

Author(s):
Muskan Garg, Postdoctoral Research Associate - Mayo Clinic; Xingyi Liu, Ph.D. - Mayo Clinic; Eunji Jeon, Ph.D. - Mayo clinic; Maria Vassilaki, MD, PhD - Mayo Clinic; Ronald C Petersen, MD, PhD - Mayo Clinic; Jennifer St Sauver, PhD, MPH - Mayo Clinic; Sunghwan Sohn, PhD - Mayo Clinic;

Multimodal Clinical NLP Integration for Alcohol Use Disorder Phenotyping in HIV Care

Poster Number: 239
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Information Extraction, Clinical Decision Support, Natural Language Processing, Artificial Intelligence, Health Equity
Programmatic Theme: Clinical Informatics

Alcohol use disorder (AUD) among people with HIV is underdiagnosed in clinical settings, limiting treatment opportunities. We are developing an interpretable NLP-based phenotyping approach that integrates structured EHR data with active learning-assisted, expert annotated clinical text using the INCEpTION platform. By creating intermediary NLP-based variables to enhance explainability and applying XGBoost within a multimodal data aggregation architecture, we aim to improve AUD case detection. We will present preliminary results of our predictive approach.

Speaker(s):
William Bradford, MD, MSPH
University of Alabama Birmingham

Author(s):
William Bradford, MD, MSPH - University of Alabama Birmingham; Micah Cochran, MSCS - University of Alabama Birmingham; Hale Thompson, PhD, MPH - University of Alabama Birmingham; Jordi Garcia-Diaz, MD - University of Alabama Birmingham; Davis Bradford, MD - University of Alabama Birmingham; John Osborne, PhD - University of Alabama at Birmingham;

Aligning hospital EHRs with Age Friendly Care: A qualitative study

Poster Number: 240
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Information Retrieval, Usability, Health Equity, Qualitative Methods
Programmatic Theme: Consumer Health Informatics

We interviewed N=20 clinicians to understand their workflows in the EHR when providing care to older adults to improve 4Ms EHR navigation (i.e., mobility, mentation, medication, what matters). Interview transcripts were analyzed thematically and resulted in five themes including difficulty finding information about the 4Ms, poor 4Ms data quality, sparse EHR data sharing, variable intra-facility practices, and homegrown 4M EHR tools. These findings can inform the development of a 4Ms EHR tool.

Speaker(s):
Daniel Gilmore, PhD
The Ohio State University

Author(s):
Daniel Gilmore, PhD - The Ohio State University; Lauren Southerland, MD - The Ohio State University Wexner Medical Center; Sadie Chen, MLIS - The Ohio State University Wexner Medical Center; Sean Huang, MD - Vanderbilt University; Fernanda Bellolio, MD - Mayo Clinic; Juliessa Pavon, MD, MHS - Duke University Health System; Elizabeth Goldberg, MD - University of Colorado School of Medicine; Daniel Walker, PhD, MPH - Ohio State University;

CareLens: A Multimodal Clinical Assistant Leveraging Hybrid RAG and Agentic Reasoning

Poster Number: 241
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Information Retrieval, Large Language Models (LLMs), Documentation Burden, Qualitative Methods, Natural Language Processing, Patient-/Person-Generated Health Data, Artificial Intelligence, Data transformation/ETL
Programmatic Theme: Clinical Informatics

Dissolution of electronic health records (EHR) into structured data, unstructured clinical notes, and images slow down the decision-making in critical clinical settings. Although large language models (LLMs) have indicated favorable results in clinical decision-making, the application of these models is hindered by the "numerical hallucinations" issue that occur when the LLMs interact with structured electronic health records (EHR). In this paper, we introduce and evaluate the CareLens clinical decision support system, that incorporates an intent recognition component that routes user queries through one of the two dedicated execution pathways: deterministic Text-to-SQL and agentic RAG, which enable the retrieval and synthesis of structured MIMIC-IV data and unstructured clinical notes, respectively. The proposed system offers a scaled-up, real-time safety net of clinical administration and triage between the precision of data and attention to clinical understanding.

Speaker(s):
Gowtham Vuppaladhadiam, Masters
University of North Texas

Author(s):
Gowtham Vuppaladhadiam, Masters - University of North Texas; Baby Jahnavi Kovelamudi, Masters - University of North Texas; Samuel Jerome Sibbi Rayan, Masters - University of North Texas; Ana D cleveland, Ph.D - University of North Texas; Tozammel Hossain, Ph.D - University of North Texas;

A Strategy for Transparent Risk Assessment of School Violence

Poster Number: 242
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Information Visualization, Usability, Clinical Decision Support
Programmatic Theme: Clinical Informatics

Current school violence prevention strategies rely on school personnel, clinical evaluation, and subjective, manual risk assessments. Our team created the Automated Risk Assessment (ARIA) pipeline that uses natural language processing of standardized interviews to predict risk of prospective aggression. in this work we created automated reports to explain ARIA’s underlying decision-making process, display individualized risk probabilities, contributing factors, and phrase-level evidence aligned with established risk themes.

Speaker(s):
Lara Kanbar, PhD
CCHMC

Author(s):
Siddique Korakottil, BSc - CCHMC; Alexander Osborn, MS - CCHMC; Andrew Cifuentes, BSc - CCHMC; Jennifer Combs, BSc - CCHMC; Drew Barzman, MD - University of Cincinnati; Judith Dexheimer, PhD - Cincinnati Children's Hospital Medical Center;

From Eligibility Criteria to Patient Cohorts: An Agent-based Visual Analytics System for EHR Cohort Identification

Poster Number: 243
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Information Visualization, Artificial Intelligence, Information Retrieval
Programmatic Theme: Translational Bioinformatics

Identifying patient cohorts from electronic health records (EHRs) is essential for clinical research but often requires translating complex eligibility criteria into executable database queries. We present an agent-based visual analytics system that enables researchers to construct cohort criteria as an interpretable graph and automatically translate them into SQL queries over OMOP-formatted EHR data. This approach combines interactive visualization and AI agents to improve the transparency and efficiency of EHR cohort identification.

Speaker(s):
Vincent Zhang, MS
Yale University

Author(s):
Huan He, Ph.D. - Yale University; Vincent Zhang, MS - Yale University; Lingfei Qian, PHD - Yale University; Vipina K. Keloth, PhD - Yale University; Ruey-Ling Weng, MS. - Yale University; Yujia Zhou, M.S. - Yale University; Na Hong, PhD - Yale University; Hua Xu, Ph.D - Yale University;

Assessing the Relationship between HIE and Patient Outcomes at National Scale

Poster Number: 244
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Interoperability and Health Information Exchange, Healthcare Quality, Transitions of Care, Evaluation
Programmatic Theme: Clinical Informatics

There is a need for robust evaluations of HIE impact. We use national claims data to identify individuals that visited two unaffiliated provider organizations and compare outcomes based on whether the two organizations were connected to the same HIE network(s). HIE is associated with slightly lower probability of being admitted to the hospital after an ED visit but not other types of utilization. Efforts are needed to address factors impeding benefit realization from HIE.

Speaker(s):
Julia Adler-Milstein, PhD, FACMI
UCSF School of Medicine

Author(s):
Ariel Linden, DrPH - UCSF; Jordan Everson, PhD - Georgetown University School of Medicine;

Feasibility of Identifying Oncology Trial Common Data Elements by Clustering Foundation Model Embeddings

Poster Number: 245
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Knowledge Representation and Information Modeling, Large Language Models (LLMs), Data Mining
Programmatic Theme: Clinical Research Informatics

Insufficient patient enrollment impedes oncology clinical trials’ progress, and variance in eligibility criteria representation limits our insights about trial recruitment. Identifying trials’ common data elements (CDEs) augments eligibility criteria’s utility, and foundation models demonstrate robustness in capturing semantic information. In this study, we cluster foundation model embeddings of breast cancer trials’ recruitment criteria to identify and evaluate CDEs. We show that foundation model embeddings robustly represent oncology trial criteria at various levels of clinical specificity.

Speaker(s):
Adit Anand, M.A.
Columbia University

Author(s):
Adit Anand, M.A. - Columbia University; Karthik Natarajan, PhD - Columbia University Dept of Biomedical Informatics;

Enhancing Knowledge Retrieval and AI‑Driven Insights in E‑Health Startups: An Ontological Framework

Poster Number: 246
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Knowledge Representation and Information Modeling, Telemedicine, Governance
Programmatic Theme: Consumer Health Informatics

E‑health startups have emerged as influential actors in the healthcare industry, transforming how care is
delivered, monitored, and accessed. The term “e‑health startup” refers to health‑tech ventures that
leverage technologies such as telehealth and mobile health applications. We have developed a
prototype ontology (OntoStar) tailored to support the e-health startup domain. It contains 189 concepts
and covers relevant topic areas including Organizations, Products, Stakeholders, Clinical Settings, Users,
Technology, AI, Sustainability, Social Impact, and Ethical Issues.

Speaker(s):
Navya Martin Kollapally, PhD in Computer Science
Kean University

Author(s):
Navya Martin Kollapally, PhD in Computer Science - Kean University; Ana Cristina Siqueira, PhD - William Paterson University; Zhi Wei, PhD - New Jersey Institute of Technology; James Geller, PhD - NJIT;

Enhancing Extraction in Multi-Turn Clinical Cases: PhaseInjectRAG and the Retrieval Bottleneck

Poster Number: 247
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Large Language Models (LLMs), Information Retrieval, Information Extraction, Evaluation, Healthcare Quality
Programmatic Theme: Clinical Informatics

PhaseInjectRAG (PI-RAG) is a multi-agent RAG architecture designed to automate the extraction of guideline-adherent criteria in unfolding clinical cases. Evaluated against the AMEGA benchmark, PI-RAG achieved an 88.1% relative F1 improvement over a single-pass baseline. A diagnostic audit revealed that retrieval strategy (Query Gaps), rather than corpus coverage, is the primary bottleneck. PI-RAG provides a foundational architecture for retrospective clinical auditing and benchmarking of emergent healthcare AI models.

Speaker(s):
Kudakwashe Mundove, MD
Lehigh University

Author(s):
Kudakwashe Mundove, MD - Lehigh University; Ana-Iulia Alexandrescu-Anselm, MS - Lehigh University; Eric Obeysekare, PhD - Lehigh University;

A Parameter-Efficient Transfer Learning Approach through Multitask Prompt Distillation and Decomposition for Clinical NLP

Poster Number: 248
Presentation Time: 05:00 PM - 06:30 PM

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

Existing prompt-based fine-tuning methods typically learn task-specific prompts independently, imposing significant computing and storage overhead at scale when deploying multiple clinical natural language processing (NLP) systems. We present a multitask prompt distillation and decomposition framework that learns a single shared meta-prompt from 21 diverse clinical source tasks and adapts it to unseen target tasks with fewer than 0.05% trainable parameters. Evaluated across five clinical NLP task types (named entity recognition, relation extraction, question answering, natural language inference, and summarization) on 10 held-out target datasets using three backbone models (LLaMA 3.1 8B, Meditron3 8B, gpt-oss 20B), our framework consistently outperforms LoRA by 1.5-1.7% despite using orders of magnitude fewer parameters, and exceeds single-task prompt tuning by 6.1–6.6%. The gpt-oss 20B model achieves the highest overall performance, particularly on clinical reasoning tasks. The strong zero- and few-shot performance demonstrates better transferability of the shared prompt representation.

Speaker(s):
Cheng Peng, PhD
University of Florida

Author(s):
Cheng Peng, PhD - University of Florida; Mengxian Lyu, Master - University of Florida; Ziyi Chen, Master of Science - Department of Health Outcomes and Biomedical Informatics, College of Medicine, University of Florida; Yonghui Wu, PhD - University of Florida;

Lightweight Open-Source LLM for Automated Screening in Scoping Review

Poster Number: 249
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Large Language Models (LLMs), Natural Language Processing, Health Equity, Information Extraction
Programmatic Theme: Public Health Informatics

Introduction: Literature screening for evidence synthesis is labor-intensive. Large language models (LLMs) may accelerate this process.
Methods: A lightweight open-source LLM (gpt-oss-20b) screened 20,182 PubMed entries through a three-stage sequential pipeline for a health equity-focused scoping review.
Results: Screening completed in 30 hours, yielding 144 candidates with estimated costs between $266–$661, compared to approximately 224 hours and $9,419 for manual screening.
Conclusion: LLMs show promise as a tool to reduce screening time/costs.

Speaker(s):
Adam Wilcox, PhD
Washington University in St. Louis

Author(s):
David Dávila-García, MA - Washington University School of Medicine in Saint Louis, Institute for Informatics, Data Science & Biostatistics; Amanda Gilbert, PhD - School of Public Health at Washington University in St. Louis; Adam Wilcox, PhD - Washington University in St. Louis;

Benchmarking Accuracy and Reasoning Alignment in Proprietary Clinical LLM Systems: A Standardized Evaluation Study

Poster Number: 250
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Large Language Models (LLMs), Artificial Intelligence, Evaluation, Quantitative Methods
Programmatic Theme: Clinical Informatics

We benchmark three widely adopted proprietary clinical LLM systems, ExpertAI, OpenEvidence, and DoxGPT, using MedThink-Bench, an expert-annotated medical reasoning dataset. The systems achieve comparable accuracy, on par with most open-source models, reflecting the benchmark's challenging multi-step reasoning tasks. However, they show differences in reasoning alignment. ROUGE-L indicates more verbose responses from DoxGPT, whereas BERTScore shows stronger alignment for OpenEvidence and ExpertAI. These findings suggest proprietary systems with comparable accuracy can exhibit differences in reasoning alignment.

Speaker(s):
Dongping Zhang, PhD
Endeavor Health

Author(s):
Dongping Zhang, PhD - Endeavor Health; Anthony Solomonides, PhD, MSc(Math), MSc(AI), FAMIA, FACMI - Research Institute, Endeavor Health; Sonia Sultan, MPH, MBA, MHA - Endeavor Health; Rema Padman, PhD - Carnegie Mellon University; Nirav Shah, MD MPH - NorthShore University HealthSystem;

How Humans Detect Hallucinations and Omissions: Reasoning-Informed Design for Safer Mental Health Support Chatbots

Poster Number: 251
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Large Language Models (LLMs), Qualitative Methods, Human-computer Interaction
Programmatic Theme: Consumer Health Informatics

General-purpose chatbots are increasingly used to support mental health concerns such as anxiety and depression. Yet large language models produce hallucinations and omissions that shape perceived reliability and potential harm in emotionally sensitive contexts. We conducted 53 reasoning cue interviews with diverse stakeholders who annotated Llama 3 responses to depression- and anxiety-related prompts. Their reasoning showed harm severity assessment involves more than detecting factual errors. We introduce three conceptual framings that explain how annotators detect errors and assign harm severity: (1) cue-based user reasoning, where annotators rely on subtle linguistic cues for error identification; (2) hesitancy as epistemic labor, reflecting the cognitive effort of evaluating ambiguous or incomplete responses; and (3) expectation-driven judgment, where harm severity depends on alignment with expected chatbot roles such as peer, coach, or professional. We further translate these insights into design implications for future chatbots, better aligning them with users’ expectations of care and support.

Speaker(s):
Hyeyoung Ryu, PhD
Vanderbilt University Medical Center

Author(s):
Janelle Faiman, BS - Vanderbilt University Medical Center; Susannah Rose, PhD - Vanderbilt University Medical Center; Shelagh Mulvaney, PhD, FAMIA - Vanderbilt University; Laurie Novak, PhD, MHSA - Vanderbilt University Medical Center Dept of Biomedical Informatics;

Augmenting clinical feature space with multimodal data and leveraging LLMs for RVD Detection Enhancement

Poster Number: 252
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Large Language Models (LLMs), Machine Learning, Public Health, Clinical Decision Support
Programmatic Theme: Clinical Research Informatics

Right ventricular dysfunction (RVD) is common and strongly associated with mortality but remains under-coded and hard to detect from transthoracic echocardiography because there is no specific ICD-10 code and NLP tools are limited. We used ChatGPT-4o to generate scalable RVD labels from 331,793 MIMIC-IV discharge summaries (2008–2019), achieving 96–98% agreement with human reviewers and yielding a cohort of 32,517 TTE admissions with 4,417 (13.6%) RVD cases. Using structured pre-echo EHR features, we trained logistic regression, random forest, XGBoost, and LightGBM models and conducted race-stratified and calibration analyses. Discrimination ranged from modest to good (AUC 0.59–0.76), with tuned XGBoost providing the best balance (AUC 0.75, recall 0.59, F1 0.36); calibration further improved probability accuracy for XGBoost and LightGBM (AUC ≈0.76, Brier ≈0.10). Overall, LLM-derived labels combined with structured EHR data enabled scalable, well-calibrated prediction of RVD and offer a generalizable approach for other under-coded cardiovascular phenotypes.

Speaker(s):
MANAL ALHUSSEIN, PhD Student
George Mason University

Author(s):
Janusz Wojtusiak, PhD - George Mason University; ARWA ALZEER, phD Student - George Mason University;

T2D-Bench: Evidence-Gated Evaluation of LLM Outputs for Type 2 Diabetes Using a Multi-Layer Clinical–Lifestyle Knowledge Graph

Poster Number: 253
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Large Language Models (LLMs), Knowledge Representation and Information Modeling, Clinical Decision Support, Artificial Intelligence, Controlled Terminologies, Ontologies, Vocabularies, Evaluation, Patient Safety, Chronic Care Management
Programmatic Theme: Clinical Research Informatics

Large language models (LLMs) can produce clinically fluent recommendations for type 2 diabetes while failing to satisfy guideline constraints or explicitly justify lifestyle-related glycemic claims. We present T2D-Bench, a reproducible benchmark and evidence-gated evaluation framework for testing whether LLM outputs satisfy explicit, graph-checkable evidence requirements. T2D-Bench is built on a multi-layer clinical–lifestyle knowledge graph that combines a biomedical spine (UMLS, DrugBank, SIDER), computable ADA Standards of Care rules, and lifestyle knowledge connected through a mechanistic bridge to glycemic laboratory effects. Across 100 structured vignettes spanning diagnosis, medication safety, and adversarial lifestyle conflicts, baseline outputs failed benchmark-defined evidence-path checks in 35% of cases for GPT-4o-mini and 33% for GPT-4o. The evidence gate detects unsupported omissions and uses constrained revision to bring outputs into verifier-level compliance with benchmark-defined evidence requirements. These results show that computable evidence constraints can make unsupported clinical omissions explicit, measurable, and correctable in diabetes-focused LLM outputs.

Speaker(s):
Saba A.Farahani, PhD candidate in Electrical Engineering and Computer Science
University of California Irvine

Author(s):
Saba A.Farahani, PhD candidate in Electrical Engineering and Computer Science - University of California Irvine; Hung Cao, PhD - University of California, Irvine; Ramesh Jain, PhD - University of California, Irvine; Amir Rahmani, PhD - University of California, Irvine;

An LLM-based Conversational Agent for Problem Solving in Dementia Caregiving

Poster Number: 254
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Large Language Models (LLMs), Chronic Care Management, Telemedicine
Programmatic Theme: Clinical Research Informatics

Family caregivers of individuals with dementia face daily challenges from behavioral and psychological symptoms of dementia (BPSD), leading to significant burden and stress. CUIDA, an evidence-based virtual intervention for Latino caregivers managing BPSD, integrates a large language model (LLM)-based conversational agent (CA) that provide real-time problem-solving guidance. Specifically, the CA guides caregivers through ABC problem solving, which involves identifying a target behavior (B), gathering contextual information including activators (A) and consequences (C), and then brainstorming, selecting, and experimenting with strategies to modify the target behavior. We implement and evaluate various approaches of CA development, including zero-shot and few-shot prompting, single-agent and multi-agent structures, and dialogue control. Results show that all configurations achieved high completion rates. Few-shot prompting and multi-agent structure supported deeper exploration of behavior-related and problem-solving content, whereas state control supported more regulated transitions. These findings highlight an important trade-off in CA conversations between open-ended exploration and structured progression.

Speaker(s):
Ziqing Ji, Biomedical and Health Informatics
University of Washington

Author(s):
Ziqing Ji, Biomedical and Health Informatics - University of Washington; Weichao Yuwen, PhD, RN - University of Washington Tacoma; Daniil Filienko, BS in Computer Science and Systems - University of Washington Tacoma; Miriana Duran, M.D., MPH - University of Washington; Susan McCurry, PhD - University of Washington; Robert Penfold, PhD - University of Washington, Kaiser Permanente Washington Health Research Institute; Trevor Cohen, MBChB, PhD - Biomedical Informatics and Medical Education, University of Washington; Magaly Ramirez, PhD, MS, MS - University of Washington;

Does Country Matter? Evaluating Geographic Sensitivity of Large Language Models in Zero-Shot Chronic Kidney Disease Screening

Poster Number: 255
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Large Language Models (LLMs), Artificial Intelligence, Fairness and elimination of bias, Clinical Decision Support, Evaluation, Population Health, Global Health
Programmatic Theme: Clinical Informatics

Large language models (LLMs) show strong zero-shot reasoning capabilities, but their predictions can be sensitive to contextual information embedded in prompts. This study investigates whether geographic context influences LLM-based chronic kidney disease (CKD) screening. Using a community-based CKD dataset, we systematically vary the country attribute in prompts while keeping patient features unchanged. Results show that prediction stability differs across geographic contexts and models, indicating that country tokens can influence clinical inference despite identical clinical inputs.

Speaker(s):
Sirajam Munira, PhD
Rensselaer Polytechnic Institute

Author(s):
Ashad Kabir, PhD - Charles Sturt University;

Pragmatic Evaluation and Implementation of Large Language Models in Hospital Workflows

Poster Number: 256
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Large Language Models (LLMs), Evaluation, Quantitative Methods
Programmatic Theme: Clinical Informatics

A growing number of health systems are deploying LLM-based tools to assist with clinical and administrative tasks. However, there is a dearth of real-world studies illustrating their use and application to hospital decision-making. We address this gap by providing the evaluation results of three LLM-based tools considered for adoption at a large academic medical center. We demonstrate how realistic evaluations have guided operational decision-making and enabled effective optimization of LLMs for the healthcare setting.

Speaker(s):
Ashley Oliver, MPH
Children's Hospital of Philadelphia

Author(s):
Dhineshvikram Krishnamurthy, Master of Science - Childrens Hospital of Philadelphia; James Urick, MS - Children's Hospital of Philadelphia; Hojjat Salmasian, MD, MPH, PhD, FACMI, FAMIA - Children's Hospital of Philadelphia; Abdul Tariq, PhD - Children's Hospital of Philadelphia;

An Interpretable Mixture of Multimodal Agent Model to Predict Outcomes in Patients with Substance Misuse

Poster Number: 257
Presentation Time: 05:00 PM - 06:30 PM

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

Patients with substance misuse are at risk for negative outcomes after being discharged from the hospital. Identifying which patients are at increased risk can facilitate early intervention. In this work we incorporate patient's prior pharmacy claims, EMS incidents, and EHR data to predict which patients are at increased risk. In addition we add explainability to our model, which provides highlighted text indicating why a patient is at increased risk.

Speaker(s):
Tim Gruenloh, MS
University of Wisconsin - Madison

Author(s):
Tim Gruenloh, MS - University of Wisconsin - Madison; Jennie Martin, MS - University of Wisconsin - Madison; Askar Safipour Afshar, MS - University of Wisconsin - Madison; Preeti Gupta, MD, MPH - Scripps Research, University of California San Diego; John Caskey - University of Wisconsin-Madison; Madeline Oguss, MS - University of Wisconsin - Madison; Elizabeth Salisbury-Afshar, MD, MPH - University of Wisconsin - Madison; Ryan Westergaard, MD, PhD, MPH - Univsersity of Wisconsin - Madison; Matthew Churpek, MD, MPH, PhD - University of Wisconsin-Madison; Majid Afshar, MD, MSCR - University of Wisconsin - Madison; Anoop Mayampurath, PhD - University of Wisconsin - Madison;

Uncovering Latent Themes of Neonatal Pain Assessment and Treatment in Routine Clinical NICU Nursing Notes Using Large Language Model-Based Semantic Topic Modeling

Poster Number: 258
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Large Language Models (LLMs), Information Extraction, Critical Care
Programmatic Theme: Clinical Research Informatics

In premature infants, pain assessment is challenging due to limited behavioral expression, and nurses' clinical observations and reasoning are rarely captured in structured fields. Large language model-based semantic topic modeling enables new approaches to analyzing unstructured clinical documentation, not routinely accessible. This study applied unsupervised semantic topic modeling to nursing documentation to identify latent themes describing patterns in pain assessment, treatment decisions, and nonpharmacologic comfort practices embedded within routine electronic health record narratives.

Speaker(s):
Julie Vignato, PhD, RN, RNC-LRN, CNE
University of Iowa College of Nursing

Author(s):
Alaa Albashayreh, PhD, MSHI, RN - University of Iowa; You Wang, MS in Data Science - University of Iowa; Jae Bae, PhD, MS, RN - University of Iowa College of Nursing; Kristi Haughey, MS, RN, RNC_NIC - University of Iowa Stead Family Children's Hospital; Amanda Karstens, MS, RN, RNC-NIC - University of Iowa Stead Family Children's Hospital; Emily Spellman, DNP, MSN, RNC-NIC - University of Iowa Health Care, Stead Family Children's Hospital; Anna Krupp, PhD, MSHP, RN - University of Iowa, College of Nursing; Kimberly Powell, PhD, RN, FAMIA, FAAN - University of Missouri - Columbia; Lindsey Knake, MD, MS - The University of Iowa Carver College of Medicine; Andrew Boyd, MD - University of Illinois Chicago; Catherine Craven, PhD, MA, MLS, FAMIA - Dell Medical School, UT Austin; Karen Dunn Lopez, PhD, MPH, RN, FAAN - University of Iowa College of Nursing;

Computable Phenotype for Maternal Discharge Readiness from NICU Unstructured Electronic Health Record Data

Poster Number: 259
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Large Language Models (LLMs), Clinical Decision Support, Critical Care
Programmatic Theme: Clinical Research Informatics

Maternal infant data are often fragmented across separate, unlinked electronic health records. This obscures longitudinal assessment within the maternal-infant dyad, which is especially problematic during prolonged NICU hospitalizations. Maternal concerns are frequently embedded within infant nursing documentation, limiting scalable evaluation of maternal discharge readiness. We applied large language models leveraging named entity recognition and semantic topic modeling to routine unstructured nursing narratives to derive computable maternal phenotypes operationalizing maternal discharge readiness for clinical decision support.

Speaker(s):
Julie Vignato, PhD, RN, RNC-LRN, CNE
University of Iowa College of Nursing

Author(s):
Alaa Albashayreh, PhD, MSHI, RN - University of Iowa; You Wang, MS in Data Science - University of Iowa; Jae Bae, PhD, MS, RN - University of Iowa College of Nursing; Kristy Haughey, MS, RN, RNC-NIC - University of Iowa Stead Family Children's Hospital; Amanda Karstens, MS, RN, RNC-NIC - University of Iowa Stead Family Children's Hospital; Emily Spellman, DNP, MSN, RNC-NIC - University of Iowa Health Care, Stead Family Children's Hospital; Anna Krupp, PhD, MSHP, RN - University of Iowa, College of Nursing; Kimberly Powell, PhD, RN, FAMIA, FAAN - University of Missouri - Columbia; Lindsey Knake, MD, MS - The University of Iowa Carver College of Medicine; Andrew Boyd, MD - University of Illinois Chicago; Catherine Craven, PhD, MA, MLS, FAMIA - Dell Medical School, UT Austin; Karen Dunn Lopez, PhD, MPH, RN, FAAN - University of Iowa College of Nursing;

Measuring Empathy in Consumer Health AI Chatbots: A Case Study in Acute Spinal Cord Injury

Poster Number: 260
2026 Symposium LIEAF Presentation

Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Large Language Models (LLMs), Patient Engagement and Preferences, Delivering Health Information and Knowledge to the Public
Programmatic Theme: Academic Informatics / LIEAF

Measuring empathy in Consumer Health AI chatbots lacks standardized evaluation methods, with most studies relying on custom rubrics. We applied five-item Consultation and Relational Empathy (CARE) measure to assess empathy in chatbot responses to 15 acute SCI questions. Two chatbots (Luma and GPT-4o) were evaluated by three raters. CARE items were scorable in 99.3% of responses, and empathy scores differed across chatbots, demonstrating the feasibility of using a validated clinical instrument for standardized chatbot empathy evaluation.

Speaker(s):
Bayu Aryoyudanta, Master
University of Pittsburgh

Author(s):
Maria Yuliana, Bachelor - University of Pittsburgh; I Made Agus Setiawan, PhD - University of Pittsburgh; Andi Saptono, PhD - University of Pittsburgh School of Health and Rehabilitation Services; Brad E. Dicianno, MD - University of Pittsburgh; Yong Kyung Choi, PhD, MPH - University of Pittsburgh; Bambang Parmanto, PhD - University of Pittsburgh;

Leveraging LLMs and Knowledge Graphs to Identify Multi-System Aging Biomarkers and Build an EHR-Based Aging Clock

Poster Number: 261
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Large Language Models (LLMs), Controlled Terminologies, Ontologies, and Vocabularies, Machine Learning
Programmatic Theme: Clinical Informatics

The growing incidence of multimorbidity in aging populations presents a major healthcare challenge. While many aging biomarkers have been identified experimentally, translating these findings into reliable clinical risk models using real-world Electronic Health Record (EHR) data remains difficult. We developed an end-to-end framework to address this gap by integrating literature-derived biomarkers with large-scale clinical data. Using large language models and hybrid retrieval, 358 aging biomarkers from the literature were mapped to standardized LOINC codes, yielding 101 validated features through expert review. These mappings were applied to a harmonized institutional data lake covering 1.36 million patient records. Biological baselines were established using rolling-window medians derived from the NHANES 2021–2023 cohort. After removing over 12 million anomalous records through strict data quality checks, a LightGBM regressor estimated biological age with a mean absolute error of 11.32 years.

Speaker(s):
Md Kamruz Zaman Rana, MSHI
University of Missouri - Columbia

Author(s):
Yaswitha Jampani - University of Missouri - Columbia; Xing Song, PhD - University of Missouri; Abu Mosa, PhD, MS, FAMIA - University of Alabama at Birmingham;

At Scale and Uncorrected: Fabricated References Across 2.5 Million Biomedical Papers

Poster Number: 262
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Large Language Models (LLMs), Artificial Intelligence, Information Retrieval, Real-World Evidence Generation
Programmatic Theme: Clinical Research Informatics

We developed an automated pipeline to detect fabricated references (citations to nonexistent studies) across 2,471,758 PubMed Central papers. The pipeline identified 4,046 fabricated references in 2,810 papers, with rates rising 14-fold since 2023. Affected papers accumulated 11,241 forward citations spanning 457 systematic reviews and 12 clinical guidelines, yet 98.4% remain uncorrected. Metadata features cannot reliably distinguish fabricated from legitimate references (AUC=0.64). Automated pre-publication verification is both feasible and urgently needed.

Speaker(s):
Maxim TOPAZ, PhD, RN, MA, FAAN, FIAHSI, FACMI
Columbia University

Author(s):
Nir Roguin, MD - Ben-Gurion University of the Negev, Beer-Sheva, Israel; PALLAVI GUPTA, PhD - Columbia University; Zhihong Zhang, PhD - Columbia University; Laura-Maria Peltonen, PhD - University of Eastern Finland;

Evaluating Human-Led and LLM–Assisted Hybrid Deductive–Inductive Workflows for Qualitative Analysis

Poster Number: 263
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Large Language Models (LLMs), Natural Language Processing, Qualitative Methods, Artificial Intelligence, Patient Engagement and Preferences, Information Extraction, Evaluation, Patient-/Person-Generated Health Data
Programmatic Theme: Clinical Research Informatics

This study evaluated LLM-assisted qualitative analysis using 122 transcripts from a heart failure study. A structured eleven-prompt workflow implemented hybrid deductive–inductive coding and was compared with human-led thematic analysis. LLM analysis generated complementary cross-case themes but revealed issues with quotation fidelity and sensitivity to batching strategies. LLMs may support large-scale qualitative analysis but require structured prompting and human verification to ensure analytic rigor.

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; Meghan Reading Turchioe, PhD, MPH, RN - Columbia University School of Nursing; Melani Ellison, MPH - Columbia University School of Nursing; Stacey Dai, MPH - Columbia University School of Nursing; Soojeong Han, PhD, AGNP, RN - Columbia University School of Nursing; Mary Beth Happ, PhD, RN - The Ohio State University; David Russell, PhD - Appalachian State University; Ruth Masterson Creber, PhD, MSc, RN - Columbia University;

Keeping Students on Track: A Scope Management Framework for AI-Assisted Clinical Exam Review

Poster Number: 264
2026 Symposium LIEAF Presentation

Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Large Language Models (LLMs), Teaching Innovation, Evaluation
Programmatic Theme: Academic Informatics / LIEAF

We developed a scope management framework for AI-assisted clinical exam review.
The framework includes a behavioral taxonomy of out-of-scope student queries,
scope-aware response strategies, and an LLM-based implementation. Evaluation
against expert (n=42) and trained non-expert labels (n=182) showed strong
accuracy on clear-boundary queries (93.6%), while relatedness-judgment queries
(27.3%) remain a challenge.

Speaker(s):
I Made Agus Setiawan, PhD
University of Pittsburgh

Author(s):
Maria Yuliana, Bachelor - University of Pittsburgh; Firdaus Indradhirmaya, PhD Student - University of Pittsburgh; Bayu Aryoyudanta, Master - University of Pittsburgh; Dipu Patel, DMSc - University of Pittsburgh; David C. Beck, EdD - University of Pittsburgh; Andi Saptono, PhD - University of Pittsburgh School of Health and Rehabilitation Services; Bambang Parmanto, PhD - University of Pittsburgh;

Evaluating Large Language Models for De-identification of Clinical Text with Synthetic Protected Health Information

Poster Number: 265
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Large Language Models (LLMs), Privacy and Security, Artificial Intelligence
Programmatic Theme: Clinical Research Informatics

Electronic health records (EHRs) are essential for AI-driven healthcare research but contain sensitive Protected Health Information (PHI). Traditional de-identification methods struggle with unstructured clinical narratives, and public datasets lack ground-truth PHI for evaluation. We propose a synthetic PHI injection framework to construct a labeled dataset and evaluates the de-identification performance of two large language models (LLMs) (DeepSeek-V3 and Gemini-3-Pro) under aggressive and conservative prompt strategies. Our multi-dimensional evaluation measures privacy recall, utility similarity, and risk scores across both unstructured and structured datasets. Results show that LLMs achieve strong de-identification in unstructured narratives, with Gemini achieving near-perfect privacy recall (>98%) across all strategies. However, they face significant challenges in structured data, where DeepSeek exhibited a cell leakage rate of approximately 20%. Overall, Gemini demonstrates more consistent performance, while DeepSeek exhibits higher variance. We establish a benchmark for LLM-based clinical text anonymization and provides a methodological foundation for privacy-preserving EHR data sharing.

Speaker(s):
Yuzhe Lin, BEng
ShanghaiTech University

Author(s):
Yuzhe Lin, BEng - ShanghaiTech University; Siyu Li, BEng - ShanghaiTech University; Dalin Li, BEng - ShanghaiTech University; Jianhao Zhu, BEng - ShanghaiTech University; Zhiyu Wan, PhD - ShanghaiTech University;

RDEmbed: A Sentence-Embedding Framework for Rare Disease Phenotyping and Phenotypic Similarity Analysis

Poster Number: 266
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Large Language Models (LLMs), Natural Language Processing, Deep Learning, Clinical Decision Support
Programmatic Theme: Clinical Research Informatics

RDEmbed is a transformer-based framework that applies sentence embeddings from clinical narratives to rare disease phenotyping and phenotypic similarity analysis. Evaluated on simulated and real-world datasets, RDEmbed outperformed ontology-based approaches in identifying similar patients and disease-specific clustering. These results demonstrate that language-based embeddings provide a scalable, context-aware method that complements traditional ontology-driven analyses and supports improved rare disease diagnosis and patient matching.

Speaker(s):
Xinghua Wang, Master
The University of Texas Health Science Center at Houston

Author(s):
Xinghua Wang, Master - The University of Texas Health Science Center at Houston; jinlian wang, PhD - UTHealth; Hongfang Liu, PhD - University of Texas Health Science Center at Houston; Arif Harmanci, PhD - The University of Texas Health Science Center at Houston;

Accelerating Age-Friendly Care Through AI-Enabled Care Coordination: A Learning Health System Approach to Discharge Summarization

Poster Number: 267
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Large Language Models (LLMs), Clinical Decision Support, Transitions of Care
Programmatic Theme: Clinical Informatics

Continuity of care for older adults transitioning from acute-care to post-acute settings remains a major challenge. Communication gaps across multidisciplinary teams result in fragmented care coordination. Age-Friendly Health Systems framework emphasizes four evidence-based components, known as 4Ms. Current discharge processes lack systematic integration of these components. 4M4YOU, an AI-enabled discharge summary prototype, supports learning health system principles by bridging the gap between complex EMR data and actionable care coordination information for post-acute care providers.

Speaker(s):
Rajashree Dahal, MS
University of Illinois Chicago

Author(s):
Kimberly Powell, PhD, RN, FAMIA, FAAN - University of Missouri - Columbia; Mihail Popescu, PhD - University of Missouri; Pardis Hossein Pour, PhD - University of Illinois Chicago; Natalie Parde, Ph.D. - University of Illinois Chicago; Jessica Hansen, PT/DPT - University of Missouri; Keaton Mullins, MS - University of Illinois Chicago;

Evaluating Large Language Models Performance in UMLS-Derived Ontology Classification

Poster Number: 268
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Large Language Models (LLMs), Controlled Terminologies, Ontologies, and Vocabularies, Evaluation, Workflow
Programmatic Theme: Clinical Informatics

We evaluated GPT-4 performance in classifying concepts within a UMLS-derived focal neurologic deficit ontology. Using a reproducible hierarchical extraction workflow, 5,102 concepts were extracted from root concept Neurological Deficit (102957003) which were then labeled by a neurologist to establish gold standard FND ontology. GPT-4, deployed via the Azure Open AI API and prompted with neurologist-defined guidelines and few-shot examples, achieved 91.4% accuracy (k=0.79), demonstrating substantial agreement in ontology classification.

Speaker(s):
Mounika Thakkallapally, Data Scientist II
University of Texas Southwestern Medical Center

Author(s):
Mounika Thakkallapally, Data Scientist II - University of Texas Southwestern Medical Center; Nelly-Estefanie Garduno-Rapp, MD, MSHI - UTSW; Wenqi Shi, PhD - UT Southwestern Medical Center; Justin Rousseau, MD, MMSc - University of Texas Southwestern Medical Center;

Leveraging Open-Source Large Language Models for Automated Tumor Matching at Central Cancer Registries

Poster Number: 269
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Large Language Models (LLMs), Natural Language Processing, Artificial Intelligence, Data Modernization, Workflow, Public Health
Programmatic Theme: Public Health Informatics

Tumor-level record linkage in central cancer registries is largely manual and requires reviewing structured tumor attributes and unstructured pathology reports. We evaluated open-source large language models (LLMs) for tumor matching using data from the Missouri Cancer Registry (2021 to 2024). After supervised fine-tuning, generative models achieved macro F1 scores up to 0.922, outperforming encoder-based models. These results suggest LLMs can support scalable tumor-level linkage and reduce manual registry workload.

Speaker(s):
Mohammad Beheshti, MSHI
University of Missouri

Author(s):
Mohammad Beheshti, MSHI - University of Missouri; Lovedeep Gondara, PhD - University of British Columbia; Jeffrey Steffens, ODS-C - University of Missouri; Mihail Popescu, PhD - University of Missouri; Prasad Calyam, PhD - University of Missouri; Iris Zachary, PhD - University of Missouri, Department of Public Health;

Domain-Specific Multi-Agent Collaboration for Domestic Violence Triage and Actionable Safety Support: A Benchmark and System Evaluation

Poster Number: 270
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Large Language Models (LLMs), Delivering Health Information and Knowledge to the Public, Patient Safety, Artificial Intelligence, User-centered Design Methods
Programmatic Theme: Clinical Informatics

Generic AI models often lack the safety protocols and specialized knowledge required for high-stakes domain intervention, such as domestic violence (DV). This study introduces a "Dispatch-Analyze-Synthesize-Critique" multi-agent framework that simulates a multidisciplinary team of nurse, legal, psychological, and social work experts with expert-curated prompts. By outperforming single-LLM baselines in safety compliance and actionability, this system demonstrates how specialized multi-agent architectures can provide reliable, trauma-informed support in low-resource clinical settings.

Speaker(s):
Shaowei GUAN, BSc
The Hong Kong Polytechnic University

Author(s):
Xinyu Feng, Master - The Hong Kong Polytechnic University; Lidan Tian, MSC - the hong kong polytechnic university; Tipparat Udmuangpia, PhD - Boromarajonani College of Nursing; Tina Bloom, PhD - Notre Dame of Maryland University; Vivian Hui, RN, PhD - The Hong Kong Polytechnic University;

Interactive Clinical Trial Matching with AI Agents

Poster Number: 271
Presentation Time: 05:00 PM - 06:30 PM

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

Patient information available at screening is often incomplete, and referral coordinators frequently ask follow-up questions to assess trial eligibility. We develop an agent-based framework to help clinicians automate this process: a clinician agent identifies missing eligibility-related patient information and asks targeted follow-up questions to obtain it, while a trial matching agent uses the updated patient profile to rank candidate trials. On the TREC 2022 benchmark, this approach substantially improved trial ranking performance.

Speaker(s):
Yin Fang, Ph.D.
National Institutes of Health

Author(s):
Yin Fang, Ph.D. - National Institutes of Health; Qiao Jin, M.D. - National Institutes of Health; Lauren He, B.A. - National Institutes of Health; SHUBO TIAN, PhD - NLM; Zhiyong Lu, PhD - National Library of Medicine, NIH;

Med-V1: Small Language Models for Zero-shot and Scalable Biomedical Evidence Attribution

Poster Number: 272
Presentation Time: 05:00 PM - 06:30 PM

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

Med-V1 is a three-billion-parameter language model for biomedical evidence attribution, designed to perform scalable verification of whether claims are supported by source articles. Trained on 1.5 million synthetic instances with standalone claims, retrieved articles, agreement labels, and rationales, Med-V1 strongly outperforms its backbone across five benchmarks and matches GPT-5 accuracy. Case studies on LLM-generated statements and clinical guidelines demonstrate utility for hallucination detection and citation auditing at scale.

Speaker(s):
Qiao Jin, M.D.
National Institutes of Health

Author(s):
Qiao Jin, M.D. - National Institutes of Health; Yin Fang, Ph.D. - National Institutes of Health; Lauren He, B.A. - National Institutes of Health; Yifan Yang, Ph.D. - Genentech; Guangzhi Xiong, BA - University of Virginia; Zhizheng Wang, Ph.D - National Institutes of Health; Nicholas Wan, Bachelor of Engineering - University of Michigan Medical School; Joey Chan, M.S. - University of Illinois Urbana-Champaign; Donald Comeau, PhD - National Institutes of Health; Robert Leaman - NCBI/NLM/NIH; Charalampos Floudas, MD, DMSc, MS - NIH/NCI; Aidong Zhang, PhD - University at Buffalo; Michael Chiang, MD - National Institutes of Health, National Eye Institute; Yifan Peng, PhD - Weill Cornell Medicine; Dept of Population Health Sciences; Div of Health Informatics; Zhiyong Lu, PhD - National Library of Medicine, NIH;

Large Language Models for Quality Assurance of Breast Cancer Biomarkers in Pathology Reports

Poster Number: 273
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Large Language Models (LLMs), Information Extraction, Public Health, Data Modernization, Artificial Intelligence, Natural Language Processing, Clinical Decision Support, Workforce Development
Programmatic Theme: Public Health Informatics

Manual abstraction of breast cancer biomarkers is time-consuming and limited by workforce shortages. We evaluated large language models (LLMs) on pathology reports, comparing extracted biomarkers to registry-validated data. Mistral Small-24B achieved the highest accuracy across ER, PR, HER2, and Ki-67. Discrepancies were flagged for targeted review. Our framework combines LLM extraction with consistency auditing to improve efficiency while supporting quality assurance.

Speaker(s):
Maryam Seifaddini, PhD Student in Health Informatics
University of Missouri

Author(s):
Maryam Seifaddini, PhD Student - University of Missouri; Mohammad Beheshti, MSHI - University of Missouri; Magda Esebua, MD, Board-Certified in Anatomic & Clinical Pathology and Cytopathology - University of Missouri; Iris Zachary, PhD - University of Missouri, Department of Public Health;

Identifying Laterality of Nephrolithiasis Surgical Procedures Using Large Language Models

Poster Number: 274
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Large Language Models (LLMs), Natural Language Processing, Informatics Implementation
Programmatic Theme: Clinical Informatics

Electronic health records (EHRs) contain valuable information for identifying disease phenotypes and treatment patterns. However, key clinical attributes of nephrolithiasis, such as stone laterality, is often incompletely captured in structured fields. Laterality of nephrolithiasis is important for understanding burden of disease as well the location of surgical procedures, which are only partially encoded through Current Procedural Terminology (CPT) codes and only if the modifier is added. Natural language processing (NLP) approaches can recover this information from clinical narratives, but traditional rule-based systems have limitations in capturing complex linguistic patterns. The purpose of this study was to assess the capabilities of large language models (LLMs) for extracting surgical laterality of nephrolithiasis procedures from clinical notes.

Speaker(s):
Jasleen Gandhi, Graduate Student
Vanderbilt University

Author(s):
Jasleen Gandhi, Graduate Student - Vanderbilt University; Cosmin Bejan, PhD - Vanderbilt University Medical Center; Layth Qassem, PharmD - VUMC; Paul Heider, PhD - Medical University of South Carolina; Nhat Nguyen, DPhil - The Children's Hospital of Philadelphia; Ryan Hsi, MD - Vanderbilt University Medical Center; Gregory Tasian, MD, MSc, MSCE - Children's Hospital of Philadelphia;

Public Use of LLMs for Health Information: A Rapid Review

Poster Number: 275
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Large Language Models (LLMs), Artificial Intelligence, Human-computer Interaction
Programmatic Theme: Consumer Health Informatics

People increasingly turn to LLMs for healthcare information daily, yet the research focuses on AI model performance, limiting our understanding of how the public uses generative AI tools when seeking healthcare information and its impact on decision-making. This rapid review reveals that most people use LLMs for symptom interpretation and self-diagnosis, often without the prompting skills needed to elicit safe and effective responses that inform healthcare decisions.

Speaker(s):
Madison Horton, PhD
Columbia University School of Nursing

Author(s):
Afra Shamnath, Master of Public Health - NYU; Maria Clara Dragut, MS - Columbia University; Shiveen Kumar, BS - New York Institute of Technology Old Westbury; Uday Suresh, MS - Vanderbilt University Department of Biomedical Informatics; Natalie Benda, PhD - Columbia University School of Nursing; Meghan Reading Turchioe, PhD, MPH, RN - Columbia University School of Nursing;

Literature-Based Extraction of Environmental Exposure Metadata Elements

Poster Number: 276
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Large Language Models (LLMs), Public Health, Information Extraction
Programmatic Theme: Public Health Informatics

The literature on environmental exposure studies has grown significantly in the past two decades. However, the lack of a standardized exposure data repository makes these studies difficult to find, access, interoperate, and reuse (FAIR). To address this issue, we used CEDAR tools to create an ontology-linked template that describes a wide variety of environmental exposure studies. We have used large language models to automatically extract relevant metadata from environmental publications to complete the template fields.

Speaker(s):
Kei-Hoi Cheung, PhD
Yale University

Author(s):
Zixian Yu, BS - Yale University; Weipeng Zhou, PhD - Yale University; Martin O'Connor, MS - Stanford University; Nicole Deziel, PhD - Yale University; Kei-Hoi Cheung, PhD - Yale University;

Evaluation of OpenEvidence, DoxGPT, UpToDate Expert AI, and Ada Health Using Emergency Department Patient Data

Poster Number: 277
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Large Language Models (LLMs), Artificial Intelligence, Evaluation, Patient Safety, Clinical Decision Support
Programmatic Theme: Clinical Informatics

The role of large language models (LLMs) and symptom checkers (SCs) in clinical decision support has been rapidly expanding, yet without thorough evaluation of reliability. Using emergency department patient data, we compared the diagnostic accuracy and triage safety of OpenEvidence, DoxGPT, UpToDate Expert AI, and Ada. While LLM diagnostic performance was comparable to Ada and physician panels, models missed high acuity diagnoses and variably produced unsafe triage recommendations, raising concerns about use in clinical settings.

Speaker(s):
Hamish Fraser, MBChB, MRCP, MSc
Brown Center for Biomedical Informatics, Brown, University

Author(s):
Nikolas Montequila, ScB - The Warren Alpert Medical School; Victor Hunt, MSc - The Warren Alpert Medical School of Brown University; Anita Zahiri, N/A - Brown Center for Biomedical Informatics, Brown, University; Nadine Naja, ScD - The Warren Alpert Medical School, Brown, University; Jessica Roa, ScB - The Warren Alpert Medical School, Brown University; Ross Hilliard, MD - Maine Medical Center, Portland, Maine; Hamish Fraser, MBChB, MRCP, MSc - Brown University;

Protocol Authoring and Biomarker Graph Validation

Poster Number: 278
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Large Language Models (LLMs), Knowledge Representation and Information Modeling, Evaluation, Natural Language Processing, Workflow
Programmatic Theme: Clinical Research Informatics

We observed protocol–registry and protocol–recruitment drift gaps when comparing IRB-approved protocols with corresponding ClinicalTrials.gov records. These gaps are acute for biomarkers because biomarker eligibility and endpoints are often embedded in narrative text and inconsistently reused across downstream tasks. We propose a schema-constrained authoring pipeline that standardizes core protocol sections using a Multiple Constraints for Precision (MCP) checklist and a planner–generator–critic architecture design, and populates a biomarker knowledge graph for drift checks and trial discovery.

Speaker(s):
Ramya Sri Baluguri, Postdoctoral Scholar
University of California, Davis

Author(s):
Ramya Sri Baluguri, Postdoctoral Scholar - University of California, Davis; Nicholas Anderson, PhD - University of California, Davis;

A Hybrid Language Framework for Ontology-Based Clinical Concept Extraction

Poster Number: 279
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Large Language Models (LLMs), Clinical Decision Support, Information Retrieval, Artificial Intelligence
Programmatic Theme: Clinical Research Informatics

This study presents a hybrid framework for extracting standardized clinical concepts from narrative clinical notes. The approach integrates linguistic preprocessing, semantic retrieval, large language models (LLMs), and ontology-based normalization within a multi-stage pipeline. Evaluation with physician-reviewed annotations indicates that the framework can improve the accuracy and reliability of clinical concept extraction and support better standardization of information in electronic health records.

Speaker(s):
Behnaz Eslami, PhD Student
Loyola University Chicago

Author(s):
Samie Tootooni, PhD - Loyola University Chicago; Dmitriy Dligach, Ph.D. - Loyola University Chicago; Nazanin Azarvash, BS - Loyola University Medical Center; Paula De La Pena, PhD - Loyola University Medical Center; Benjamin Strickland, DO - Northern Illinois and Indiana;

Mapping Free-Text Chief Complaints to a Structured Clinical Taxonomy Using LLM-Based Soft Classification

Poster Number: 280
Presentation Time: 05:00 PM - 06:30 PM

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

This study presents CCMapper-LLM, a large language model–based framework for mapping free-text emergency department chief complaints to a structured clinical taxonomy. The approach performs taxonomy-constrained multi-label classification to align narrative triage descriptions with standardized clinical categories. Evaluation with clinician-reviewed annotations demonstrates strong agreement and highlights the potential of LLM-based methods to support more consistent triage documentation and improved usability of emergency department clinical data.

Speaker(s):
Behnaz Eslami, PhD Student
Loyola University Chicago

Author(s):
Samie Tootooni, PhD - Loyola University Chicago; Dmitriy Dligach, Ph.D. - Loyola University Chicago; Kathleen Bobay, PhD, RN, FAAN - Loyola University Chicago; Mark E Cichon, DO - Loyola University Medical Center;

A Systematic Evaluation of Mixture-of-Experts for ICU Outcome Prediction and Patient Phenotyping

Poster Number: 281
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Machine Learning, Deep Learning, Critical Care
Programmatic Theme: Clinical Research Informatics

Clinical outcome prediction models often adopt a one-size-fits-all approach and fail to capture the heterogeneity of intensive care unit (ICU) patient subgroups. Mixture-of-Experts (MoE) models can jointly predict outcomes and discover subgroups, yet application to structured ICU data remains limited. In this work, we applied MoE to predict in-hospital mortality in the MIMIC-III dataset (N=38,512), evaluating 462 configurations through a two-stage model selection procedure. Our goal was to examine whether MoE can achieve competitive outcome prediction while extracting subgroups. The results show that predictive performance remained stable across configurations, whereas clustering quality varied substantially. A model with k=4 experts produced subgroups with distinct presentation profiles. Geometric separation and distributional divergence captured complementary aspects of subgroup quality, motivating joint evaluation. These findings indicate that a MoE pipeline achieves competitive mortality prediction while producing subgroups, and highlight the necessity of joint clustering evaluation for phenotyping in structured clinical data.

Speaker(s):
Yingchuan Sun, Master’s Student in Computer Science
Emory University

Author(s):
Yingchuan Sun, Master’s Student in Computer Science - Emory University; Shengpu Tang, PhD - Emory Unversity; Hyunjung Gloria Kwak, PhD - Emory University;

Comparative Evaluation of Machine Learning and Deep Learning Models for Predicting Telehealth Use among U.S. Adults

Poster Number: 282
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Machine Learning, Deep Learning, Informatics Implementation, Artificial Intelligence, Telemedicine, Mobile Health, Public Health, Chronic Care Management
Programmatic Theme: Public Health Informatics

A total of 7,095 U.S. adults from HINTS 7 was used to compare machine learning (ML) approaches in prediction of telehealth use. ML identified 31 informative features; random forest achieved the highest F1 score (0.820), followed by deep neural network and gradient boosting (0.810 each), while Tabular Prior-data Fitted Network showed the highest AUC (0.797). Multivariable regression confirmed 19 independent factors. Telehealth use was strongly associated with digital engagement, chronic diseases, healthcare, and providers facilitation.

Speaker(s):
Kesheng Wang, PhD
University of South Carolina

Author(s):
Kesheng Wang, PhD - University of South Carolina;

A Framework for Evaluating Predictive and Causal Roles of Genomic and Clinical Features Across Patient Subgroups: Evidence from axSpA

Poster Number: 283
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Machine Learning, Causal Inference, Real-World Evidence Generation, Health Equity, Quantitative Methods, Population Health
Programmatic Theme: Translational Bioinformatics

Genomic testing for disease lacks systematic methods to assess sex-differential clinical utility. Our proposed three-step framework quantifies genomic value beyond EHR data, identifies sex-differential clinical predictors, and tests causal effects independent of genetic risk. Using NIH All of Us data for axSpA prediction, genomic markers improved AUROC ~8× more in females but AUPRC ~2× more in males. Height strongly predicted axSpA in males, and double/debiased machine learning confirmed a differential signal after genomic adjustment.

Speaker(s):
Gayathri Donepudi, PhD Student
University of Washington

Author(s):
Gayathri Donepudi, PhD Student - University of Washington; Jennifer Hadlock, MD - Institute for Systems Biology;

Automated Pain Detection: Impact of Psychosocial Factors on Physiology-based Models

Poster Number: 284
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Machine Learning, Chronic Care Management, Deep Learning, Clinical Decision Support
Programmatic Theme: Clinical Research Informatics

Automated assessment of chronic pain remains challenging due to variability in how individuals perceive and report pain. Physiological responses reflect autonomic activity associated with pain but often fail to generalize across individuals. Psychosocial factors such as coping style and pain-related beliefs also influence pain perception, yet they are rarely incorporated into computational models. This study proposes a deep neural network framework that integrates physiological features with participant-level psychosocial variables for pain level classification. Data were collected from 40 adults (30 with chronic pain and 10 healthy controls) in an IRB-approved study and evaluated using a leave-one-participant-out approach. Models trained using physiological features alone showed poor performance, achieving a median accuracy of 49.2% and median macro-F1 score of 0.468. Incorporating psychosocial variables improved performance, increasing median accuracy to 87.8% and median macro F1-score to 0.766. These findings highlight the value of biopsychosocial modeling for more reliable automated pain assessment.

Speaker(s):
Tonushree Dutta, M.S.
University of Minnesota Twin Cities

Author(s):
Tariq Azmy, MS - University of Minnesota; Quan Guan, MS - University of Minnesota; Beth Groenke, DDS, MS - University of Minnesota; John Sartori, PhD - University of Minnesota;

Early Prediction of Alzheimer's Disease and Related Dementias Using Machine Learning and Claims Data

Poster Number: 285
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Machine Learning, Patient-/Person-Generated Health Data, Real-World Evidence Generation
Programmatic Theme: Public Health Informatics

We conducted a retrospective study using IBM® MarketScan® claims (2015–2019) to predict incident Alzheimer's Disease and Related Dementias. Machine learning models were developed using propensity score-matched cohorts across 12- to 30-month predictive windows and 6- or 12-month observation windows. XGBoost demonstrated the best discriminative power, particularly with a 12-month observation window (AUC 0.96–0.99), maintaining robust performance nearly three years before clinical diagnosis. SHAP analysis identified outpatient service frequency, ophthalmology procedures, diagnostic imaging, and acute care intensity as the primary predictors of ADRD risk. These findings indicate that longitudinal healthcare utilization patterns provide a scalable signature of cognitive decline long before clinical diagnosis. Integrating such models into healthcare utilization could significantly enhance early risk identification. Future work will focus on external validation in extended populations and integration with EHR-based clinical features to enhance model generalizability and clinical utility.

Speaker(s):
Xingyue Huo, Master
University of Arizona

Author(s):
Xingyue Huo, Master - University of Arizona; Joseph Finkelstein, MD, PhD - University of Arizona;

Enhancing Mortality Prediction in the Intensive Care Unit by Combining Structured EHR Data and LLM-Extracted Clinical Features

Poster Number: 286
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Machine Learning, Artificial Intelligence, Large Language Models (LLMs), Data Mining, Natural Language Processing
Programmatic Theme: Clinical Research Informatics

Accurate mortality prediction in the Intensive Care Unit (ICU) may benefit from leveraging both structured and unstructured EHR data. We propose a multi-modal framework combining structured clinical covariates with clinically interpretable features extracted from clinical notes via zero-shot Large-Language Model (LLM) prompting. Evaluated on MIMIC-IV data using Logistic Regression and XGBoost, the combined approach yielded AUROC improvements of 0.02–0.049 over structured-only baselines, with the LLM-extracted malignancy feature emerging as the strongest predictor.

Speaker(s):
Ni Gao, BS
YALE UNIVERSITY

Author(s):
Ni Gao, BS - YALE UNIVERSITY; Yan Wang, PhD - Yale University; Tsung-Ting Kuo, PhD, FAMIA - Yale University;

Integrating Topic Modeling and Cosine Similarity for Accurate and Interpretable Prediction of Sepsis Mortality

Poster Number: 287
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Machine Learning, Natural Language Processing, Information Visualization
Programmatic Theme: Clinical Research Informatics

Predicting patient outcome in sepsis is a critical challenge in clinical settings. We developed a machine learning framework that combines structured electronic health record data with topic modeling and cosine similarity from word embeddings to improve outcome prediction in sepsis patients. Using a publicly available cohort of 200 sepsis patients, models enhanced with NLP-extracted features achieved significantly higher accuracy than those using structured data alone. Feature importance analyses identified topics related to sepsis shock and worsening conditions, along with prototype similarities, as key predictors of negative outcomes. Temporal analysis of topic proportions revealed dynamic semantic changes in clinical narratives during hospitalization. External validation with the MIMIC-III dataset confirmed the robustness and generalizability of our approach, with consistent improvements in predictive performance. Our findings demonstrate that integrating structured and unstructured data via interpretable NLP techniques can substantially enhance sepsis risk prediction and support clinical decision-making.

Speaker(s):
Ilaria Lonoce, PhD
University of Chicago

Author(s):
Ilaria Lonoce, PhD - University of Chicago; Yoav Gilad, PhD - University of Chicago; Julie Johnson - University of Chicago; Tomasz Oliwa, PhD - University of Chicago;

Machine Learning-Based Prediction of Falls in Geriatric Patients Using Real World Electronic Health Record Data

Poster Number: 288
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Machine Learning, Artificial Intelligence, Quantitative Methods, Population Health, Healthcare Quality
Programmatic Theme: Clinical Research Informatics

Falls are the leading cause of injury-related mortality in U.S. adults aged 65 and older. Using multi-domain EHR data from 29,862 UTMB geriatric patients, we trained five machine learning models to predict fall occurrence. LightGBM achieved the highest discrimination (AUROC 0.825), substantially outperforming logistic regression (AUROC 0.681) and conventional screening benchmarks. SHAP analysis identified healthcare utilization, polypharmacy, and age as dominant predictors. These findings support automated, scalable fall risk stratification in academic health system EHRs.

Speaker(s):
Elizaveta Naydanova, M.D./Ph.D. Student
University of Texas Medical Branch

Author(s):
Erin Hommel, MD MS - University of Texas Medical Branch at Galveston; Duane Morrow, Ph.D. - University of Texas Medical Branch; Melissa Morrow, Ph.D. - University of Texas Medical Branch; Kamil Khanipov, PhD - University of Texas Medical Branch; George Golovko, PhD - University of Texas Medical Branch;

Using EHR data to identify toxic epidermal necrolysis risk factors in a large health system

Poster Number: 289
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Machine Learning, Large Language Models (LLMs), Patient Safety
Programmatic Theme: Clinical Research Informatics

Adverse drug reactions (ADRs) are underreported in coded clinical systems. We aim to use a combination of structured and unstructured EHR data to classify cases of harmful and costly ADRs found in the VA Healthcare System. Our approach incorporates NLP, zero-shot entailment classification, and ICD diagnosis codes to identify cases for genome-wide association studies (GWAS), where we aim to detect the genetic determinants of drug-related safety events.

Speaker(s):
Brian Ferolito, MSc
Department of Veterans Affairs

Author(s):
Brian Ferolito, MSc - Department of Veterans Affairs; Daniel Golden, MSc - Department of Veterans Affairs; Kai Gravel-Pucillo, MSc - Department of Veterans Affairs; J Michael Gaziano, MD, PhD - Department of Veterans Affairs; Kelly Cho, PhD - VA Boston Healthcare/Harvard Medical School; Alexandre Pereira, MD, PhD - Department of Veterans Affairs;

A Multi-Protein Biomarker Panel for Non-Invasive MASLD Progression Detection

Poster Number: 290
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Machine Learning, Population Health, Clinical Decision Support, Causal Inference, Public Health, Healthcare Quality, Quantitative Methods, Evaluation
Programmatic Theme: Translational Bioinformatics

MASLD affects 30% of people worldwide and can progress to cirrhosis, yet most patients are undiagnosed. Current diagnosis requires invasive liver biopsy. This study developed a blood plasma biomarker panel (ALB, APOB, SERPINA1, AHSG, A2M) capable of classifying normal, MASLD, and cirrhosis cases. A two-stage hierarchical voting ensemble classifier achieved 95.8% disease detection sensitivity/specificity and AUROC values of 0.987, 0.799, and 0.827, respectively. LiNGAM causal analysis confirmed disease severity as the primary driver of biomarker expression.

Speaker(s):
Tejasvi Vivek, High School
The Quarry Lane School

Author(s):
Tejasvi Vivek, High School - The Quarry Lane School;

Identifying American Indian and Alaska Native Patients at Risk for Unsuppressed Viral Load using Electronic Health Record Data

Poster Number: 291
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Machine Learning, Health Equity, Quantitative Methods
Programmatic Theme: Clinical Informatics

Achieving viral suppression is important for people living with HIV (PLWH). Barriers to viral suppression can be especially prominent for American Indian and Alaska Native (AIAN) people and early identification of unsuppressed viral load risk can support timely intervention. We present preliminary research to identify AIAN PLWH at risk of unsuppressed viral load using electronic health records data, achieving modest performance. Further research can support healthcare delivery for a marginalized and underserved population.

Speaker(s):
Haoyun Hong, BA
University of Washington, Seattle

Author(s):
Haoyun Hong, BA - University of Washington, Seattle; Taryn Tye, BS - University of Washington, Seattle; Amanda Hayes, BA - University of Washington, Seattle; Danner Peter, MPH - University of Washington; Talon Slater, N/A - Indigenous Social Drivers of Health Community Advisory Board; Derrick Belgrade, MPA - Indigenous Social Drivers of Health Community Advisory Board; Kara Harvill, MD - Indigenous Social Drivers of Health Community Advisory Board; Jason Deen, MD - University of Washington, Seattle; Heidi Crane, MD - University of Washington, Seattle; Vanessa Simonds, ScD - Montana State University; Amandalynne Paullada; Andrea Hartzler, PhD FACMI - University of Washington; Oliver Bear Don't Walk, PhD - University of Washington;

Predicting Complete Diabetes Care Disengagement from Social Determinants of Health: A Machine Learning Study

Poster Number: 292
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Machine Learning, Population Health, Diversity, Equity, Inclusion, and Accessibility
Programmatic Theme: Public Health Informatics

A subset of adults with diabetes remains entirely disconnected from healthcare, receiving neither HbA1c monitoring nor physician visits. Using data from 60,440 adults in the 2017 Behavioral Risk Factor Surveillance System, we developed interpretable machine learning models using social determinants of health to predict complete care disengagement. Random Forest achieved the highest discrimination (AUC 0.74). Key predictors included the absence of a personal physician and socioeconomic barriers, enabling population-level identification of patients invisible to clinical registries.

Speaker(s):
Md Mohaimenul Islam, PhD
University at Buffalo

Author(s):
Tahmina Nasrin Poly, PhD - Taipei Medical University; Arinze Okere, PharmD - University at Buffalo;

Explainable AI Insights Into Clinical Factors and Care Pathways Driving Longer Time to Pulmonary Arterial Hypertension Diagnosis

Poster Number: 293
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Machine Learning, Real-World Evidence Generation, Clinical Decision Support, Quantitative Methods
Programmatic Theme: Clinical Informatics

Explainable AI methods were applied to large-scale administrative claims data to characterize drivers of prolonged time to pulmonary arterial hypertension (PAH) diagnosis among adults with chronic unexplained dyspnea. Using random forest, boosting, and LASSO models with interpretability techniques, we evaluated demographic, clinical, and care-seeking patterns associated with time to right heart catheterization–confirmed PAH. Results reveal modifiable system- and pathway-level contributors to diagnostic delay, informing potential decision support strategies for earlier PAH recognition.

Speaker(s):
Arielle Marks-Anglin, PhD
Merck & Co, Inc.

Author(s):
Yiru Wang, PhD - Merck; Anna Watzker, MHS - Merck & Co., Inc., Rahway, NJ, USA; Boshu Ru, Ph.D. - Merck;

Cross-Institutional Reproducibility of Machine Learning Models for Clinical Prediction in Pediatric Neurofibromatosis Type 1

Poster Number: 294
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Machine Learning, Informatics Implementation, Evaluation, Natural Language Processing
Programmatic Theme: Clinical Research Informatics

Clinical prediction models built from electronic health records often fail to reproduce across institutions due to differences in documentation and patient populations. Using neurofibromatosis type 1 (NF1) as a test case, we transferred a machine learning pipeline combining structured EHR data and NLP-derived clinical note features from one institution to another. Multimodal models maintained or improved performance across sites, demonstrating the potential for reproducible and portable EHR-based prediction pipelines.

Speaker(s):
Levi Kaster, BS
Washington University in St. Louis

Author(s):
Levi Kaster, BS - Washington University in St. Louis; Saki Amagai, Student - Northwestern University; Inez Oh, PhD - Institute for Informatics at Washington University in St. Louis, School of Medicine; Stephanie Morris, MD - Kennedy Krieger Institute; Robert Listernick, MD - Northwestern University Feinberg School of Medicine; Carlos Prada, MD - Feinberg School of Medicine at Northwestern University; Yuan Luo, PhD - Northwestern University; David Gutmann, MD, PHD - Washington University School of Medicine in St. Louis; Philip Payne, PhD, FACMI, FAMIA - WashU Medicine and BJC Healthcare; Aditi Gupta - Washington University in St. Louis;

Development of a Mobile Health Application for Latent Tuberculosis Self-Management: A Community Engaged Approach

Poster Number: 295
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Mobile Health, Population Health, Public Health
Working Group: Nursing Informatics Working Group
Programmatic Theme: Public Health Informatics

Self-management of latent TB infection (LTBI) is essential for tuberculosis prevention. Using a Community-Based Participatory Research approach, we developed a culturally tailored mHealth application for Haitian immigrants in Indiana. LTBI self-management knowledge was extracted from clinical guidelines and discussions with community partners. The application includes educational resources, three algorithms, 50 user-interface screens, and features to collect images for future TB detection algorithm development. Heuristic evaluation by experts and usability evaluation by patients will be measured.

Speaker(s):
Soojung Jo, PhD RN
Purdue University

Author(s):
Alfu Laily, PhD - Purdue University; Young L Kim, PhD - Purdue University; Jennifer L. Brown, PhD - Purdue University; Kathy Sullender, MS - Daviess County Health Department; Merle Holsopple, MD - Daviess County Health Department; Jennifer Stefancik, MS - Purdue Extension Daviess County; Pouagnel Elizaire, BS, CCHW - Daviess County Health Department;

Lessons Learned from an Academic-Industry Partnership to Develop an eHealth CIRCLE Platform

Poster Number: 296
2026 Symposium LIEAF Presentation

Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Mobile Health, User-centered Design Methods, Delivering Health Information and Knowledge to the Public
Programmatic Theme: Academic Informatics / LIEAF

This study presents a framework for academic-industry partnerships in developing CIRCLE, an eHealth intervention platform for dementia caregivers. Using an Agile methodology over five months, an interdisciplinary team of researchers, UX designers, and developers collaborated to translate evidence into a functional platform. This partnership provides insights and a strong pathway for future development of eHealth health interventions.

Speaker(s):
Bianca Shieu, PhD
UT San Antonio

Author(s):
Payton Strozier, BA - UT Health San Antonio; Shamim Ashrafi, Masters - InNeed Intelligent Cloud; Lixin Song, PhD - UTHSA;

Development and Pilot Testing of an mHealth Intervention on Health Behavior Self-Efficacy and Overall Wellness in Family Dementia Caregivers

Poster Number: 297
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Mobile Health, Diversity, Equity, Inclusion, and Accessibility, Informatics Implementation, Chronic Care Management
Programmatic Theme: Consumer Health Informatics

Family caregivers of dementia patients face high stress and deteriorating health. We developed and pilot-tested a 3-
week, app-based mobile health (mHealth) intervention integrating lifestyle medicine—physical activity, emotional
management, and social connection—for dementia caregivers in South Korea (N=58). Using a quasi-experimental
design, we found the mHealth intervention significantly improved health behavior self-efficacy (p=.024) and holistic
wellness (p<.001) compared to a control group, demonstrating the efficacy of digital frameworks for caregiver
support.

Speaker(s):
HyeJin Park, PhD RN
FSU College of Nursing

Author(s):
Myungeun Suh, MSN - Dong-A University of Health; Ruda Lee, BSN - Shepherd Rehabilitation Center; Eunjoo Lee, PhD - Kyungpook National University; Boyoung Kim, PhD - Chonnam National University;

HierCamemICD: Hierarchical Deep Learning for French ICD-10 Coding with Retrieval-Augmented Evaluation and Clinical Trust Analysis

Poster Number: 298
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Natural Language Processing, Deep Learning, Controlled Terminologies, Ontologies, Vocabularies, Information Retrieval, Artificial Intelligence, Clinical Decision Support, Evaluation, Information Extraction
Programmatic Theme: Clinical Informatics

We present HierCamemICD, a hierarchical framework for automated French ICD-10 coding combining frozen CamemBERT features with WHO ClaML semantic enrichment. Evaluated on 241,763 death certificate lines across 9,541 codes, the framework achieves 82.8% code accuracy under oracle enrichment (verification workflow) and 35.2% with the best automated baseline (hybrid TF-IDF+CamemBERT). Retrieval experiments comparing BM25, CLS cosine, and mean-pooling retrieval reveal that BM25 (10.3% Recall@1) outperforms both dense methods (~5%), diagnosing the clinical-to-formal terminology gap as the deployment bottleneck rather than vocabulary coverage (99.9%). Frequency-stratified analysis shows enrichment transforms mid-frequency code accuracy from 3.9–11.8% to 64.8–88.2%. Selective prediction at 50% coverage achieves 98.6% accuracy, supporting human-in-the-loop deployment. Text-level robustness analysis identifies case sensitivity (93.4% flip rate) as the primary vulnerability, motivating input normalization for production systems.

Speaker(s):
Grace Esther DONG, Master Of Science
Aivancity

Author(s):
Grace Esther DONG, Master Of Science - Aivancity;

Evaluating Human–LLM Agreement in Clinical Name Entity Recognition

Poster Number: 299
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Natural Language Processing, Artificial Intelligence, Clinical Decision Support
Programmatic Theme: Clinical Informatics

We evaluated agreement between humans and LLMs in clinical NER using 66 double-annotated, de-identified clinical notes from three academic health systems. We compared scispaCy and three LLMs using exact string and embedding-based semantic matching. Exact matching underestimated performance, while semantic evaluation revealed higher agreement and more meaningful rankings. MedGemma achieved the highest F1 (0.55). Results highlight how evaluation definitions shape conclusions and inform the use of LLM-based NER in downstream clinical informatics applications.

Speaker(s):
Mihail Popescu, PhD
University of Missouri

Author(s):
Yibo Chen, B. S. - University of Missouri-columbia; Mihail Popescu, PhD - University of Missouri;

PLACID: Privacy-preserving Large language models for Acronym Clinical Inference and Disambiguation

Poster Number: 300
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Natural Language Processing, Privacy and Security, Large Language Models (LLMs), Patient Safety
Programmatic Theme: Clinical Research Informatics

Large Language Models (LLMs) offer transformative solutions across many domains, but healthcare integration is
hindered by strict data privacy constraints. Clinical narratives are dense with ambiguous acronyms, misinterpretation
these abbreviations can precipitate severe outcomes like life-threatening medication errors. While cloud-dependent
LLMs excel at Acronym Disambiguation, transmitting Protected Health Information to external servers violates pri-
vacy frameworks. To bridge this gap, this study pioneers the evaluation of small-parameter models deployed entirely
on-device to ensure privacy preservation. We introduce a privacy-preserving cascaded pipeline leveraging general-
purpose local models to detect clinical acronyms, routing them to domain-specific biomedical models for context-
relevant expansions. Results reveal that while general instruction-following models achieve high detection accuracy
(∼0.988), their expansion capabilities plummet (∼0.655). Our cascaded approach utilizes domain-specific medical
models to increase expansion accuracy to (∼0.81). This novel work demonstrates that privacy-preserving, on-device
(2B-10B) models deliver high-fidelity clinical acronym disambiguation support.

Speaker(s):
Manjushree Aithal, Ph.D.
University of Colorado Anschutz, Co

Author(s):
Alexander Kotz, Ph.D. Student - University of Colorado Anschutz, Co; James Mitchell, Ph.D. - University of Colorado Anschutz, CO;

Domain-Adaptive Transformer Fine-Tuning for Theory-Grounded Detection of Caregiver Stress in Online Dementia Support Communities

Poster Number: 301
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Natural Language Processing, Machine Learning, Patient-/Person-Generated Health Data, Population Health, Information Extraction
Programmatic Theme: Consumer Health Informatics

Family caregivers of people living with dementia experience chronic stress, and existing clinical assessments rely on structured survey instruments that are difficult to apply at scale. Online support communities contain rich narratives that may provide caregiver stress signals. Natural language processing (NLP) research often relies on generic linguistic features and rarely operationalizes psychological theoretical constructs. This study introduces a manually annotated dataset of dementia caregiving narratives grounded in established frameworks and evaluates whether domain-adaptive transformer fine-tuning improves automated detection of caregiver stress severity. We compare frozen sentence embeddings with classical classifiers and fine-tuned transformer models. Fine-tuned MPNet achieved the best held-out test performance for three-level stress classification (Accuracy=0.714, Macro F1=0.688). Across five-fold cross-validation, frozen MPNet embeddings with a linear SVM produced lower but more stable performance (Macro F1=0.639). Findings demonstrate the potential of theory-grounded NLP approaches for scalable digital phenotyping of caregiver stress in online support communities.

Speaker(s):
Jessica Hernandez Chilatra, PhD Student
University of Texas Health Science Center at Houston

Author(s):
Jessica Hernandez Chilatra, PhD Student - University of Texas Health Science Center at Houston; Xiaoqian Jiang, PhD - University of Texas Health Science Center at Houston;

OMOPCompass: A Model Context Protocol Approach for EHR Queries in an OMOP CDM Database Augmented with SDOH Data—A Comparative Study with OHDSI Atlas

Poster Number: 302
2026 Symposium LIEAF Presentation

Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Natural Language Processing, Large Language Models (LLMs), Population Health, Information Retrieval
Programmatic Theme: Academic Informatics / LIEAF

OHDSI Atlas provides a robust analytics platform for cohort building and characterization. However, it requires significant domain knowledge, has a steep learning curve and can be time-consuming to be used for simple exploratory queries. In addition, scope of Atlas does not extend beyond the OMOP CDM without ingesting the external data into the CDM and mapping and loading the new concept sets into the concept table. OMOPCompass introduces an MCP-based workflow with a simple, intuitive UI to allow for natural language exploration of clinical data and extending the exploration into non-clinical domains through an augmented OMOP database. OMOPCompass has integrated SDOH data as freestanding tables linked to the EHR through patient location. OMOPCompass has been architected to incorporate a variety of datasets like billing and claims, omics, mortality and social vulnerability indices, positioning it as a versatile and user-friendly tool for clinicians and public health professionals.

Speaker(s):
Nita Deshpande, PhD
Emory University

Author(s):
Venkatesh Javvaji, MS - Emory University; Dileep Gunda, MS - Emory University; Chad Robichaux, MPH - Emory University; Sheida Habibi, Masters - Emory University; Jon Duke, MD - Georgia Tech Research Institute; Tony Pan, PhD - Emory University;

Evaluating Large Language Model Translation Fidelity for Medical Documents Across High- and Low-Resource Languages

Poster Number: 303
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Natural Language Processing, Health Equity, Large Language Models (LLMs), Artificial Intelligence, Evaluation
Programmatic Theme: Clinical Informatics

We evaluated four frontier large language models (GPT-5.1, Claude Opus 4.5, Gemini 3 Pro, Kimi K2) translating 22 CDC and American Cancer Society patient education documents into 8 languages spanning high-resource (Spanish, Chinese, Russian, Vietnamese), medium-resource (Korean, Arabic), and low-resource (Tagalog, Haitian Creole) categories. Two co-primary outcomes were assessed across 702 translation pairs: back-translation semantic fidelity (LaBSE, BLEU) and comparison to professional human translations (COMET, BERTScore). All models achieved high semantic preservation (LaBSE: 0.921–0.987) and approached professional translation quality (COMET: 0.87–0.88). Low-resource languages did not differ significantly from high-resource languages (p = 0.066). Cancer education materials achieved higher fidelity than vaccine documents (p < 0.001). These findings support further investigation of LLMs for extending medical translation access, with patient comprehension validation as a critical next step.

Speaker(s):
Chukwuebuka Anyaegbuna, MD
Stanford

Author(s):
Chukwuebuka Anyaegbuna, MD - Stanford; Eduardo Guerrero, MD - Stanford; Jerry Liu, MD - Stanford; Timothy Keyes, PhD - Stanford; April Liang, MD - Stanford University; Natasha Steele, MD, MPH - Stanford Medicine; Stephen Ma, MD, PhD - Stanford University School of Medicine; Jonathan Chen, MD, PhD - Stanford University Hospital; Kevin Schulman;

Semantic Drift Between Regulatory and Consumer-Facing Drug Information: A Large-Scale Comparative NLP Study

Poster Number: 304
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Natural Language Processing, Data Mining, Data transformation/ETL, Patient Safety, Patient Safety
Programmatic Theme: Consumer Health Informatics

Medication decisions rely on both clinician-facing and consumer-facing drug information, so inconsistencies can affect understanding and risk communication. We quantified semantic drift between DailyMed and MedlinePlus using section-aligned BERT-embeddings. Drift was highest in safety content and significantly greater than use/directions domains. External linkage analyses showed inverse associations between drift and CMS spending, and between drift and drug age. Effects were statistically robust but small, supporting utility of targeting high-drift safety language.

Speaker(s):
Erik Holbrook, MD
Mass General Brigham

Author(s):
Erik Holbrook, MD - Mass General Brigham; Thomas McCoy, MD - Mass General Hospital;

Transformer-Based NLP for Standardizing Officer-Reported Client Needs in a Correctional Learning Health System (LHS)

Poster Number: 305
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Natural Language Processing, Machine Learning, Population Health, Administrative Systems
Programmatic Theme: Public Health Informatics

Unstructured officer reported client needs in Missouri’s Improving Community Treatment Success program limit standardized documentation and system level analysis. A transformer based natural language processing pipeline using DistilBERT and semi supervised learning classified free text assessments into 21 standardized need categories. DistilBERT achieved 92.3 percent accuracy and 91.5 percent F1 score, outperforming traditional models. This approach supports learning health system documentation and enables scalable analytics for improving correctional behavioral health service coordination.

Speaker(s):
Jeffrey Appiagyei, Data Science & Informatics
University of Missouri

Author(s):
Mirna Becevic, PhD - University of Missouri Department of Dermatology; Ashley Givens, Phd - University of Missouri;

Preliminary Analysis of Structured Electronic Health Record Data and Development of an LLM-Enhanced NLP Pipeline for Suicidal Risk Detection

Poster Number: 306
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Natural Language Processing, Public Health, Artificial Intelligence
Programmatic Theme: Clinical Research Informatics

Accurate identification of suicidal ideation (SI) and suicide attempts (SA) in electronic health records (EHRs) is essential for advancing suicide research. This project summarizes procedures for EHR data acquisition, data extraction, and diagnostic code-based identification of SI/SA cases. It also outlines preparatory steps for developing a natural language processing (NLP) pipeline, including the use of a large language model (LLM) component, to support future analysis of clinical notes and improve case identification accuracy.

Speaker(s):
Sarah Arias, PhD
Butler Hospital/Brown University

Author(s):
Ivan Miller, PhD - Butler Hospital/Brown University; Charles Eaton, MD, MPH - Kent Hospital; Richard Jones, ScD - Brown University; Elizabeth Chen, PhD - Brown University;

Using Clinical NLP to Supplement Structured EHR Data for Smoking and Obesity Phenotyping in a Health System

Poster Number: 307
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Natural Language Processing, Information Extraction, Documentation Burden
Programmatic Theme: Clinical Research Informatics

Electronic health records (EHRs) are widely used for research but often rely on incomplete structured data. We evaluated two natural language processing (NLP) tools to extract smoking and obesity phenotypes from 19 million clinical notes across 500,000 patients in a large healthcare system. NLP demonstrated high accuracy and identified substantially more patients than structured EHR fields alone, highlighting the value of clinical notes for improving large-scale phenotyping.

Speaker(s):
Jie Yang, PhD, FACMI, FAMIA
Harvard Medical School

Author(s):
Jie Yang, PhD, FACMI, FAMIA - Harvard Medical School; Bowen Gu, MS - Brigham and Women's Hospital; Haritha Pillai, MS - Brigham and Women's Hospital; Joyce Lii, MS - Brigham and Women's Hospital; David Cronkite, MS - Kaiser Permanente Washington Health Research Institute; Keith Marsolo, PhD - Duke University School of Medicine; Rishi Desai - Brigham and Women's Hospital;

FigSIM: A Dataset for Fine-grained Suicide Severity and Figurative Language in Suicide Memes

Poster Number: 308
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Natural Language Processing, Information Extraction, Public Health, Artificial Intelligence, Population Health
Programmatic Theme: Consumer Health Informatics

Suicide memes are increasingly used on social media to express suicide-related thoughts or comment on suicide-related issues, yet remain poorly understood and potentially harmful. We introduce FigSIM, the first dataset designed for fine-grained analysis of suicide memes. The dataset contains 1049 memes annotated for (1) suicide severity levels, (2) figurative phenomena, and (3) suicide-related content (e.g., method depiction). Baseline experiments show that suicide memes pose unique challenges for detection and moderation.

Speaker(s):
Brian Chapman, PhD
UT Southwestern Medical Center

Author(s):
Liuliu Chen, PhD student - The University of Melbourne; Elise Carrotte, PhD - The University of Melbourne; Brian Chapman, PhD - UT Southwestern Medical Center; Jo Robinson, PhD - The University of Melbourne; Michael Conway, PhD - University of Melbourne;

Regex + LLM Improved Score Extraction from MoCA & MMSEs in Clinical Notes

Poster Number: 309
Presentation Time: 05:00 PM - 06:30 PM

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

This study combined regular expressions with LLMs in a score & date extraction pipeline for MoCAs/MMSEs in notes. No existing research focuses on hybrid approaches, which is optimal for increasing sensitivity while maintaining high PPV. GPT-5.2.1 achieved weighted PPVs of 99.18% (score) & 92.28% (date) and sensitivities of 98.67% (score) & 95.73% (date). We conclude our hybrid approach of a rule-and-LLM-based pipeline is a suitable alternative for score and date extraction.

Speaker(s):
Jared Darrow, B.S., MSBA
Vanderbilt University Medical Center: Department of Bioinformatics

Author(s):
Wei-Qi Wei, M.D., Ph.D, FAMIA - Vanderbilt University Medical Center: Department of Bioinformatics; Jared Darrow, B.S., MSBA - Vanderbilt University Medical Center: Department of Bioinformatics;

Formative Evaluation of Pheno+: A SMART on FHIR Application for Rapid Phenotyping in the Neonatal Intensive Care Unit

Poster Number: 310
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Natural Language Processing, Clinical Decision Support, Controlled Terminologies, Ontologies, Vocabularies, Evaluation, Usability, User-centered Design Methods, Workflow, Information Extraction
Programmatic Theme: Clinical Informatics

Pheno+ is a SMART on FHIR application utilizing natural language processing to generate human phenotype ontology (HPO) terms from clinical notes. This study evaluated Pheno+ via think-aloud interviews and survey data from phenotyping tasks for neonatal intensive care unit patients with suspected genetic disorders. We report HPO term yield, time spent on HPO term curation, usability and trust in the tool, as well as qualitative insights into clinicians’ interpretations of the tool’s HPO term output.

Speaker(s):
Peter Taber, PhD
University of Utah

Author(s):
Peter Taber, PhD - University of Utah; L Weaver, MLIS, PhD - University of Utah; Emerson Lebleu, MS - University of Utah; Mickey Bolyard, MS - University of Utah; Jorie Butler, PhD - University of Utah; Elise Garton, MSc - University of Utah; Tony Disera, MS - University of Utah; Tanner Ellsworth, MD - University of Utah; Paul Estabrooks, PhD - University of Utah; Kensaku Kawamoto, MD, PhD, MHS - University of Utah; Phillip W Warner, MS - University of Utah; Kelsey Simek, MD - University of Utah; Chelsea Solorzano, RN - University of Utah; Martin Tristani-Firouzi, MD - University of Utah; Alistair Ward, PhD - Frameshift Genomics; Douglas Martin, MD, FAMIA - Biomedical Informatics Department, University of Utah; Sabrina Malone Jenkins, MD - University of Utah;

Multi-Class Symptom Abstraction From Conversational Clinical Interviews Using Large Language Models: An Evaluation of Clinical Evidence Scoping

Poster Number: 311
Presentation Time: 05:00 PM - 06:30 PM

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

Large conversational transcripts from diagnostic interviews must be summarized into clinically relevant symptoms for efficient review. We evaluated large language models for multi-class symptom extraction from dementia screening interviews. Using clinician-annotated labels, we compared transcript versus excerpt inputs, prompting strategies, and models. Symptom-specific excerpts improved performance. The best configuration achieved accuracy 0.81 (κ = 0.61), comparable to agreement between human reviewers. LLMs may assist symptom abstraction but require clinician oversight.

Speaker(s):
Arin Nelson, BS
Emory University

Author(s):
Jeanne Powell, PhD - Emory University; Arin Nelson, BS - Emory University; Oz Alon, High School Diploma - Emory University; Nayoung Choi, MS - Emory University; Noah Sturgill, BS - Emory University; Jinho Choi, PhD - Emory University; Abeed Sarker, PhD - Emory University School of Medicine; Andrew Breithaupt, MD - Emory University;

Accelerating Clinical NLP at Scale with a Hybrid Framework with Reduced GPU Demands: A Case Study in Dementia Identification

Poster Number: 312
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Natural Language Processing, Information Extraction, Deep Learning
Working Group: Natural Language Processing Working Group
Programmatic Theme: Clinical Informatics

Most state-of-the-art clinical natural language processing (NLP) solutions are transformer-based and require high computing resources, limiting their accessibility. We present a hybrid NLP framework that combines a rule-based filter, a Support Vector Machine classifier, and a BERT-based model to improve efficiency while preserving accuracy. We used this framework in a dementia cohort identification study of 4.9 million veterans with incident hypertension, analyzing 2.1 billion notes. At the patient level, our method achieved a precision of 0.90, a recall of 0.84, and an F1-score of 0.87. Moreover, this method identified three times more dementia cases than structured data methods. All processing was completed in about two weeks using a single machine with dual A40 GPUs –a 95% reduction in processing time compared with the estimated 40 weeks required for a purely BERT model. Although demonstrated for dementia identification, this framework is task‑agnostic and adaptable to other large‑scale clinical NLP tasks.

Speaker(s):
Jianlin Shi, MD, PhD
The Division of Epidemiology, School of Medicine, University of Utah; VA Salt Lake City Healthcare System

Author(s):
Jianlin Shi, MD, PhD - The Division of Epidemiology, School of Medicine, University of Utah; VA Salt Lake City Healthcare System; Qiwei Gan; Elizabeth Hanchrow, RN, MSN - Veterans Affairs and WIVR; Annie Bowles, MS Biomedical Informatics - VHA Salt Lake City; Johnathan Stanley, Biomedical Informatics - Department of Veterans Affairs; Yizhe Xu, PhD, Mstat - University of Utah; Adam Bress, PharmD, MS - University of Utah; Jordana Cohen, MD, MSCE - University of Pennsylvania; Patrick Alba, MS - United States Department of Veterans Affairs;

Dialogue-Driven Generative AI for Patient Questionnaire Completion: Development Insights and Design Principles

Poster Number: 313
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Patient Engagement and Preferences, Human-computer Interaction, Large Language Models (LLMs)
Programmatic Theme: Clinical Informatics

Collecting patient-reported outcome measures (PROMs) is essential for clinical care and research, yet traditional form-based approaches are often tedious for patients and burdensome for clinicians. We developed a generative AI conversational agent(CA) using GPT-5 to collect back pain data according to the NIH Task Force's Recommended Minimal Dataset. Unlike prior CAs that ask questions one-by-one, our CA engages users in topic-based conversations, allowing multiple data items to be captured in a single exchange. Through iterative development and pilot testing with clinicians and a consumer panel, we identified key design principles for health data collection CAs. These principles extend established clinical decision support design guidelines to conversational interfaces, addressing: flexibility of interaction style, personality calibration, data quality assurance through confidence visualization, patient safety constraints, and interoperability requirements. We present our prompt design methodology and discuss challenges encountered. Our design principles provide a practical framework for developers creating conversational agents for patient questionnaire completion.

Speaker(s):
Mor Peleg, PhD, FACMI, FIAHSI
University of Haifa

Author(s):
David Navarro, PhD, MD - Macquarie University; Mor Peleg, PhD, FACMI, FIAHSI - University of Haifa;

Perceptions of an AI Chatbot Integrated into a Patient Portal

Poster Number: 314
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Patient Engagement and Preferences, Artificial Intelligence, Health Equity
Programmatic Theme: Consumer Health Informatics

Health systems are integrating chatbots into patient portals to help patients interpret health information. Most research examines stand-alone health chatbots outside electronic health records. We surveyed 2,955 active primary care portal users to assess acceptability of a portal-embedded chatbot using the Theoretical Framework of Acceptability. Acceptability was moderate. Older age and portal engagement predicted higher acceptability, while higher disease burden and technophobia predicted lower acceptability. Implementation strategies should address these barriers to support equitable adoption.

Speaker(s):
Philipp Haessner, BS, MBA, MS
University of Florida

Author(s):
Philipp Haessner, BS, MBA, MS - University of Florida; Crystal Romero, MD, MPH - University of Florida; Onyekachi Ike-Okpe, MD - University Of Florida; Nicole Hammer, B.A. - University of Florida - Department of Health Outcomes and Biomedical Informatics; Megan Gregory, Ph.D. - University of Florida;

Detecting Unreported Patient Safety Events and Unmet Care Needs in Nursing Notes Using RAG-GPT-Enabled Chart Review

Poster Number: 315
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Patient Safety, Natural Language Processing, Information Retrieval, Documentation Burden, Information Extraction, Healthcare Quality, Artificial Intelligence, Large Language Models (LLMs)
Programmatic Theme: Clinical Informatics

We developed a RAG-GPT-enabled chart review system to detect unreported falls and care needs from nursing notes. In 222 patients across three hospitals, fall detection achieved F1=0.948 and MCC=0.906. The system identified 94.6% of falls versus 32.3% by self-report, with substantially lower review burden.

Speaker(s):
INSOOK CHO, PhD
Inha University

Author(s):
Hyunchul Park, MBA - Kookmin University;

Frailty Assessment in Hospitalized Heart Failure: Comparing Measurement Approaches Using Electronic Health Records and Patient-Reported Outcomes

Poster Number: 316
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Patient-/Person-Generated Health Data, Real-World Evidence Generation, Transitions of Care, Population Health, Data Standards, Evaluation, Clinical Decision Support
Programmatic Theme: Clinical Research Informatics

We evaluated how frailty classification varies when derived from electronic health records (EHR) and patient-reported outcomes (PROs) in patients hospitalized for heart failure. Across three approaches, frailty prevalence differed substantially, overlap was incomplete, and agreement for categorical frailty classification was low. Results show these approaches are not interchangeable and that data completeness strongly shapes computability, critical for real-world evidence, cross-site portability, and health system implementation.

Speaker(s):
Jung A Kang, Phd, RN, AGACNP-BC, AGCNS-BC
Columbia University

Author(s):
Ruth Masterson Creber, PhD, MSc, RN - Columbia University; Meghan Reading Turchioe, PhD, MPH, RN - Columbia University School of Nursing; Brock Daniels, MD, MPH - Weill Cornell Medicine; Jacky Choi, MPH - Weill Cornell Medicine; Sarah Danziger, MD - Columbia/New York Presbyterian; Stacey Dai, MPH - Columbia University School of Nursing; Melani Ellison, MPH - Columbia University School of Nursing; Yihong Zhao, PhD - Columbia University School of Nursing;

Bookmarking Behavior as a Signal of Health Information Interest: Insights from Health E-Librarian with Personalized Recommender (HELPeR)

Poster Number: 317
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Patient-/Person-Generated Health Data, Patient Engagement and Preferences, Delivering Health Information and Knowledge to the Public
Programmatic Theme: Consumer Health Informatics

This study examined how bookmarking behavior reflects user interest in health information by analyzing user logs from HELPeR, a health recommender system designed for ovarian cancer patients and caregivers. Results showed that patient users read and bookmarked more documents than caregivers, despite being fewer in number. Documents addressing broader cancer-related topics were more frequently bookmarked. Notably, many bookmarks occurred without prior views, suggesting that presentation features (e.g., document titles) may influence bookmarking decisions.

Speaker(s):
Bora Jeong, BSN, RN
University of Pittsburgh

Author(s):
Peter Brusilovsky, PHD - University of Pittsburgh; Young Ji Lee, PhD, MSN , RN, FAAN - University of Pittsburgh; Mohammad Hassany, BS - University of Pittsburgh; Youjia Wang, BSN, RN - University of Pittsburgh School of Nursing; Daqing He, PhD - University of Pittsburgh; Khushboo Thaker, PhD - Crater Labs; Heidi Donovan, PhD - University of Pittsburgh;

Apple’s HealthKit and Related Frameworks in Healthcare Practice and Research: A Scoping Review

Poster Number: 318
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Patient-/Person-Generated Health Data, Mobile Health, Usability, Evaluation, Patient Engagement and Preferences, Informatics Implementation
Programmatic Theme: Clinical Research Informatics

This scoping review mapped the use of Apple’s HealthKit’s Frameworks in clinical and research settings. Following JBI and PRISMA‑ScR guidance, 99 studies were included from 1,926 records. Research included multiple regions, care settings and more than 80 clinical domains. HealthKit and ResearchKit were the predominant frameworks. Usability evaluation was limited and engagement declined over time, although passive data was more complete. Clinical outcomes showed variable signals, calling for standardized evaluation frameworks and broader device inclusivity.

Speaker(s):
Laura Otalora Gonzalez, M.D.
Mayo Clinic

Author(s):
Peyman Nejat, M.D. - Mayo Clinic; Laura Otalora Gonzalez, M.D. - Mayo Clinic; Hugo Botha, M.B., Ch.B. - Mayo Clinic; Vitaly Herasevich, MD, PhD, FCCM, FAMIA - Mayo Clinic;

Hospital Readmission Rates for Suicidal Patients: Preliminary Examination of Demographic Characteristics

Poster Number: 320
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Population Health, Quantitative Methods, Natural Language Processing, Public Health, Information Extraction, Information Retrieval, Qualitative Methods, Knowledge Representation & Information Modeling
Programmatic Theme: Clinical Research Informatics

The current investigation examined suicidal thoughts and behaviors (STB) associated with readmission status and demography of 5,798 (2,899 STB; 2,899 stratified) patients from MIMIC-IV. Binary logistic regression observed readmission (OR = 3.64), younger age (OR=0.26), single marital status (OR =0.38), and Medicaid (OR=1.98) or Medicare (OR = 1.88) as significant STB predictors (AUC=0.84), highlighting the potential of EHRs to identify high-risk populations and inform targeted patient identification.

Speaker(s):
Jung Park, Bachelors
Brown University

Author(s):
Elizabeth Chen, PhD - Brown University; Sarah Arias, PhD - Butler Hospital/Brown University;

A Hybrid Machine Learning Pipeline Utilizing Local LLMs to Characterize GLP-1/GIP Receptor Agonist Adverse Events and Epidemiological Risk within the FAERS Database

Poster Number: 321
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Population Health, Deep Learning, Public Health
Programmatic Theme: Public Health Informatics

Standard pharmacovigilance utilizing FAERS is constrained by non-linear confounding and extensive missing covariate data. We developed a privacy-preserving hybrid pipeline using local Large Language Models to impute missing demographics, enhancing XGBoost prediction of severe adverse outcomes within the 2025 FAERS dataset (N=296,262). Salvaging 45% of missing variables, XGBoost outperformed logistic regression (AUROC 0.86 vs. 0.68). This framework provides high-fidelity epidemiological risk stratification, significantly optimizing GLP-1/GIP safety signal detection.

Speaker(s):
Gilbert Feng, PhD
Cura Diagnostics

Author(s):
Gilbert Feng, PhD - Cura Diagnostics;

Sentinel: A Healthcare-Oriented Training Dynamics Monitoring Framework for Misconduct Detection in Vertical Federated Learning

Poster Number: 322
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Privacy and Security, Machine Learning, Data Mining, Artificial Intelligence
Programmatic Theme: Clinical Research Informatics

Vertical Federated Learning (VFL) enables privacy-preserving collaborative modeling when covariates for the same patients are stored in multiple hospitals, but may introduce risks of model misconduct (e.g., plagiarism, fabrication, and falsification) during distributed training. We present Sentinel, a monitoring framework for detecting abnormal updates in VFL using training dynamics indicators. Our experimental results show that although requiring additional detection time, Sentinel can detect misconduct with perfect precision (1.00) with a reasonable F1-score (0.65).

Speaker(s):
Yan Wang, PhD
Yale University

Author(s):
Yan Wang, PhD - Yale University; Aakriti Adhikari, PhD - Yale University; Tsung-Ting Kuo, PhD, FAMIA - Yale University;

Large-Scale Gold Standard Evaluation of Privacy-Preserving Record Linkage Across Multiple Use Cases in a Statewide Health Information Exchange

Poster Number: 323
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Privacy and Security, Evaluation, Interoperability and Health Information Exchange, Data Sharing, Information Retrieval, Patient Safety
Programmatic Theme: Clinical Research Informatics

This study evaluates privacy-preserving record linkage (PPRL) performance at scale using a gold-standard reference dataset from a statewide health information exchange containing over 100 million registrations. By testing three algorithm configurations across four operational use cases, we demonstrate that thoughtfully calibrated PPRL can approximate the performance of identified cleartext linkages. Specifically, permissive configurations achieved F1-scores approaching 0.97 for clinical and public health linkages.

Speaker(s):
Shaun Grannis, MD, MS, FAAFP, FACMI, FAMIA
Regenstrief Institute/Indiana University

Author(s):
Huiping Xu, PhD - Indiana University; Jarod Baker, MS - Regenstrief Institute; Lauren Lembcke, MS - Regenstrief Institute; John Price - Regenstrief Institute;

Predictive Modeling of Postpartum Contraceptive Behavior: An Expanded Hierarchical Framework for Assessing State-Level and Individual Drivers

Poster Number: 324
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Public Health, Policy, Quantitative Methods
Programmatic Theme: Public Health Informatics

Postpartum birth control (PPBC) reduces adverse outcomes associated with short interpregnancy intervals, yet usage remains uneven across the United States. We analyzed PRAMS Phase 8 data (n = 108,676) from 2016–2022 to identify predictors of postpartum contraceptive use using an expanded covariate framework and multilevel logistic regression. Weighted prevalence of PPBC use peaked at 54% in 2019 before declining to 47% in 2022. Significant geographic variation persists, with state-specific prevalence ranging from 40% to 65%. In the baseline hierarchical model, the intraclass correlation coefficient (ICC) was 0.017, indicating modest clustering by state. After adjusting for demographic, socioeconomic, and clinical factors, the ICC decreased to 0.008, suggesting that much of the between-state variation is explained by individual-level characteristics. The improved adjusted-model fit (AIC: 142,231) indicates that while population composition explains much of the disparity, state-level context still influences PPBC uptake and highlights the need for targeted policy interventions.

Speaker(s):
Amber Tran, MS
Yale University

Author(s):
Joyce Wu, MS - Yale School of Public Health; Mengyao Wang, MS - Yale School of Public Health; Yiru Li, MS - Yale School of Public Health; David Chartash, PhD - Department of Emergency Medicine, Yale University;

Interactive Data Visualization of an Early Detection Algorithm for Acute Respiratory Distress Syndrome for the ARDS QUantification and Evaluation SysTem (ARDS-QUEST) Project

Poster Number: 325
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Public Health, Infectious Diseases and Epidemiology, Information Visualization
Programmatic Theme: Public Health Informatics

The ARDS-QUEST Project utilizes an interactive application designed to visualize spatial and temporal trends, facilitating the early detection of Acute Respiratory Distress Syndrome (ARDS). The system is built on a robust, extensible framework that integrates geospatial modeling, temporal aggregation, and interactive filtering, allowing users to explore changes in ARDS incidence, prevalence, and etiologies across the VA healthcare system.

Speaker(s):
Andrew Yang, BS
VA Boston Healthcare System

Author(s):
Amelia Tarren, MS - VA Boston Healthcare System,; Kellyanne Howarth, BS - VA Boston Healthcare System; Stephen La, BS - VA Boston Healthcare System; Nathanael Fillmore, PhD - VA Boston Healthcare System; Austin Vo, BS; Kaitlin Swinnerton, BS - VA Boston Healthcare System; Rafael Fricks, Ph.D. - Department of Veterans Affairs; Frank Meng, PhD - VA Boston Healthcare System; Judith Strymish, MD - VA Boston HCS; Jussi Saukkonen, MD - VA Boston Healthcare System; Kei-Hoi Cheung, PhD - Yale University; Mihaela Aslan, PhD - VA Connecticut Healthcare System, West Haven; Pradeep Mutalik, MD - VA Connecticut Healthcare System and Yale University School of Medicine; Ryan Ferguson, ScD, MPH - VA Boston Healthcare System; Kathleen Akgun, MD - VA Connecticut Healthcare System; Nhan Do, MD, MS, Clinical Informatics Diplomate - Boston VA HCS;

Automatic Pose Detection for Improving Sports Performance

Poster Number: 326
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Public Health, Machine Learning, Human-computer Interaction
Programmatic Theme: Consumer Health Informatics

Physical activity plays a key role in maintaining a healthy lifestyle. We leverage phone video taken to provide personalized feedback to improve form. We explore disc golf as an example sport, though the approach is general-purpose. We developed an algorithm to automatically identify key moments and body angles during a throw. We compare our approach to a pre-existing approach and find significant improvements in identifying the key moments and, particularly, in identifying body angles at these moments.

Speaker(s):
David Kauchak, PhD
Pomona College

Author(s):
Emmett Levine, BA in progress - Pomona College; Miles Chiang, BA in progress - Pomona College; David Kauchak, PhD - Pomona College;

From Access to Interpretation: Using Repository Discussion Forums to Map Evolving Researcher Needs in Alzheimer’s Multiomics

Poster Number: 327
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Qualitative Methods, User-centered Design Methods, Data Sharing, Usability
Programmatic Theme: Clinical Research Informatics

Effective data repository management requires a deep understanding of evolving user needs to ensure data remains actionable for the research community. We conducted a retrospective thematic analysis of 697 repository forum posts (2016–2025) within the AD Knowledge Portal to identify practical friction points in data reuse. Results reveal that user inquiries have shifted from basic access to complex post-access interpretation and data mapping. These findings suggest that repository stewardship must adapt to prioritize documentation transparency. These findings suggest that user-reported challenges can provide a roadmap for evolving data stewardship practices to better support the complex needs of the research community.

Speaker(s):
Savitha Sangameswaran, PhD
Sage Bionetworks

Author(s):
Zoe Leanza, BS - Sage Bionetworks; Jo Scanlan, BS - Sage Bionetworks; Laura Heath, PhD - Sage Bionetworks; Susheel Varma, PhD MBA FBCS - Sage Bionetworks;

Developing and Validating a GraphRAG-LLM Pipeline for Automated Thematic Analysis of Healthcare Transcripts

Poster Number: 328
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Qualitative Methods, Large Language Models (LLMs), Artificial Intelligence, Information Extraction, Workflow
Programmatic Theme: Clinical Research Informatics

Manual thematic analysis is a bottleneck in healthcare research, often hindered by human bias and labor constraints. This study introduces a pioneering GraphRAG-LLM pipeline that mirrors expert inductive-deductive workflows to analyze complex transcripts. Achieving a 0.959 structural similarity score against human experts, the system captures latent nuances and hierarchical themes that standard AI often misses. Discover how graph-based search strategies can revolutionize qualitative accuracy and clinical insight.

Speaker(s):
Shaowei GUAN, BSc
The Hong Kong Polytechnic University

Author(s):
John Law, MSc - POLARIX AI Limited; Xinyu Feng, Master - The Hong Kong Polytechnic University; Lidan Tian, MSC - the hong kong polytechnic university; Vivian Hui, RN, PhD - The Hong Kong Polytechnic University;

An unsupervised multiomics clustering framework for molecular stratification of clinically distinct phenotypes

Poster Number: 329
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Quantitative Methods, Machine Learning, Informatics Implementation
Programmatic Theme: Translational Bioinformatics

Analyzing multi-omics data is challenging due to high dimensionality and heterogeneous scaling. We present a modality-specific analytical framework that integrates robust MAD normalization, consensus clustering, PERMANOVA validation, and systematic clinical characterization to identify stable molecular subtypes. Applied to placental multi-omics data from 321 pregnancies, the framework revealed distinct modality-specific clustering patterns and clinical associations. miRNA profiles showed the strongest clinical stratification, while transcriptomic clusters were less distinct after normalization, suggesting modality-specific sensitivity to this framework.

Speaker(s):
Aparajita Saha, PhD
University of Washington

Author(s):
Oren Barak, MD - Kaplan Medical Center, Hebrew University of Jerujalem; Yoel Sadovsky, MD - Stanford University; Jennifer Hadlock, MD - Institute for Systems Biology; Samantha N. Piekos, PhD - Perelman School of Medicine, University of Pennsylvania;

Monitoring Evidence Accumulation in Clinical Research: A Large-Scale Evaluation of Trial Sequential Analysis

Poster Number: 330
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Quantitative Methods, Evaluation, Clinical Decision Support, Real-World Evidence Generation, Data Mining, Causal Inference
Programmatic Theme: Clinical Research Informatics

Trial Sequential Analysis (TSA) is widely used in systematic reviews to control false positive findings as evidence accumulates across clinical trials. However, the empirical behavior of TSA in real-world evidence synthesis remains poorly understood. Using a large dataset of meta-analyses derived from the Cochrane Database of Systematic Reviews, we examine information accumulation patterns and boundary behavior under common TSA configurations, revealing discrepancies between theoretical assumptions and observed evidence trajectories.

Speaker(s):
Aiqing Li, PhD (c)
University of Arizona

Author(s):
Aiqing Li, PhD (c) - University of Arizona; Lifeng Lin, PhD - University of Arizona;

Discordance Between HIV Screening and Infection Risk in a Large Safety-Net Health System:

Poster Number: 331
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Real-World Evidence Generation, Infectious Diseases and Epidemiology, Public Health
Programmatic Theme: Public Health Informatics

Significant gaps in HIV screening persist among high-risk populations, despite the widespread availability of effective antiretroviral therapy. Using electronic health record data and neighborhood indicators from a safety-net system, researchers employed multivariable logistic regression to analyze predictors of screening completion and identify discordance with documented diagnoses.

Speaker(s):
Haleigh Kampman, MPH
IUI

Author(s):
Tyler Stepsis, MD - Eskenazi Health; Saitejaswi Cherukupalli, Master of Science - Indiana University; Nancy Olmstead, MSW - Eskenazi Health; Michelle Rose, MBA - Gilead;

Scaling a National OMOP Registry: Informatics Lessons from the ASH RC Sickle Cell Disease Data Hub

Poster Number: 332
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Real-World Evidence Generation, Data Sharing, Interoperability and Health Information Exchange
Programmatic Theme: Clinical Research Informatics

The ASH Research Collaborative (ASH RC) Sickle Cell Disease (SCD) Data Hub is a national, 20-site OMOP-formatted research network designed to support high-quality real-world evidence generation for a complex, underserved patient population. This panel will present the technical architecture, data acquisition processes, and interoperability advantages enabled by the use of OMOP/OHDSI standards. Panelists will describe the Data Hub’s data quality evaluation framework, multi-stakeholder computable phenotype development, and the coordination model that enables consistent data contributions across diverse clinical settings. The session will also address the generalizability of this approach to other specialty societies seeking to develop disease-focused, analytically robust real-world data infrastructures. Attendees will gain insight into the informatics strategies required to ensure data fitness-for-use, reproducible analytics, and scalable evidence generation in specialty-driven clinical research networks.

Speaker(s):
Keith Marsolo, PhD
Duke University

Author(s):
Travis Cleaves, CSM - American Society of Hematology; Emily Semmel, MS - ASH Research Collaborative; Keith Marsolo, PhD - Duke University; Joseph Romano, PhD - University of Pennsylvania;

Reproducible Pipeline for Cohort Identification and Characterization of Immune Checkpoint Inhibitor–Documented Admissions and Immune-Related Adverse Event Signal Candidates in MIMIC-IV Database

Poster Number: 334
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Real-World Evidence Generation, Artificial Intelligence, Knowledge Representation & Information Modeling, Informatics Implementation, Information Retrieval, Data Sharing, Data Mining, Natural Language Processing
Programmatic Theme: Clinical Research Informatics

Immune checkpoint inhibitor adverse events are often under-captured by diagnosis codes alone. We developed a reproducible surveillance pipeline using MIMIC-IV discharge summaries, identifying 1,821 ICI-documented admissions and screening for 42 guideline-derived phenotypes across 14 organ systems. Text-mentioned phenotypes lacked corresponding ICD codes in 43.0% of cases. This publicly available framework, partially validated through rubric-based manual abstraction, supports scalable irAE surveillance and the development of context-aware Natural Language Processing models.

Speaker(s):
Md Muntasir Zitu, PhD
Moffitt Cancer Center

Author(s):
Md Muntasir Zitu, PhD - Moffitt Cancer Center; Dwight Owen, MD - The Ohio State University; Yuxi Zhu, PHD - University Hospitals; Samar Binkheder, PhD - King Saud University;

Patient-Reported Access Barriers and GLP-1 Receptor Agonist Discontinuation

Poster Number: 335
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Surveys and Needs Analysis, Patient Engagement and Preferences, Transitions of Care
Programmatic Theme: Clinical Research Informatics

GLP-1 receptor agonist (GLP-1 RA) discontinuation undermines long-term obesity management. Patient-reported barriers remain understudied. We surveyed 261 patients (response rate 5.3%) who experienced a GLP-1 RA treatment gap at Vanderbilt University Medical Center. Using adjusted logistic regression, prior authorization (PA) delays of ≥8 days were significantly associated with medication discontinuation, while patients who knew how to appeal a denied claim were significantly less likely to discontinue. Targeted interventions to reduce PA delays and improve insurance literacy may prevent avoidable treatment interruptions.

Speaker(s):
Xinmeng Zhang, BS
Vanderbilt University

Author(s):
Xinmeng Zhang, BS - Vanderbilt University; Yubo Feng, MS - Vanderbilt University; Gracie Piantek, BS - Vanderbilt University Medical Center; Chao Yan, PhD - Vanderbilt University Medical Center; Laurie Novak, PhD, MHSA - Vanderbilt University Medical Center Dept of Biomedical Informatics; Gitanjali Srivastava, MD - Vanderbilt University Medical Center; You Chen, PhD - Vanderbilt University Medical Center;

Principles for Developing Medical Education Clinical Simulation Scenarios Involving Artificial Intelligence System Interactions

Poster Number: 336
2026 Symposium LIEAF Presentation

Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Teaching Innovation, Artificial Intelligence, Human-computer Interaction
Programmatic Theme: Academic Informatics / LIEAF

Increasing clinical use of artificial intelligence systems makes it necessary to educate clinicians on their use. This podium abstract presents ongoing work to construct principles based on a scoping review, gap analysis, needs assessment, and modified Delphi panel to guide the development of clinical simulation scenarios for medical education involving interactions with clinical AI systems to address both clinical competencies and AI competencies.

Speaker(s):
Benjamin Collins, MD
Vanderbilt University Medical Center

Author(s):
Benjamin Collins, MD - Vanderbilt University Medical Center;

Data-Driven Nursing Education: Implementing a Multi-Level Data Science Curriculum with EHR-Derived Datasets

Poster Number: 337
2026 Symposium LIEAF Presentation

Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Teaching Innovation, Curriculum Development, Informatics Implementation, Information Visualization
Programmatic Theme: Academic Informatics / LIEAF

Healthcare is generating more data than ever, but many nursing programs do not teach students how to use it. To address this, we developed a scalable, multi-level data science curriculum at the Emory University School of Nursing. The curriculum engages students across undergraduate, graduate, and doctoral programs, progressively building skills in data literacy, applied analysis, and research. Students use Project NeLL, a platform providing safe access to de-identified electronic health record data from over 1 million patients, and work with both patient dashboards and case-based exercises to connect individual patient stories to population-level patterns. Over the past three years, more than 500 students have participated, with measurable gains in Excel and data analysis skills, and more than 15 doctoral projects completed using the datasets. This approach demonstrates how informatics-driven education can prepare nurses to interpret data, support quality improvement, and engage in research, providing a model for data-driven nursing education.

Speaker(s):
Ramya Govindarajan, Director
Emory University

Author(s):
Ramya Govindarajan, Director - Emory University;

Virtual Reality for Health Informatics Education: A Systematic Review to Inform Workforce Training and Virtual Placements

Poster Number: 338
2026 Symposium LIEAF Presentation

Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Teaching Innovation, Human-computer Interaction, Workforce Development, Delivering Health Information and Knowledge to the Public, Real-World Evidence Generation, Usability, Transitions of Care, User-centered Design Methods
Programmatic Theme: Academic Informatics / LIEAF

This systematic review examined how virtual reality (VR) has been implemented in higher education and health professions training and assessed its relevance to Health Informatics competencies. Eighty empirical studies were analyzed following PRISMA guidelines. While VR effectively supports knowledge and skill acquisition, few studies address informatics competencies such as workflow analysis, systems thinking, and privacy. VR-based “virtual placements” may offer scalable experiential learning opportunities to strengthen the informatics workforce.

Speaker(s):
Pranitha Presingu, Master In Health Informatics
Grand Valley State University

Author(s):
Suhila Sawesi, PhD - GVSU;

LLM-Assisted Taxonomy and Temporal Analysis of Provider Questions About HIV in provider-to-provider telehealth

Poster Number: 339
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Telemedicine, Mobile Health, Teaching Innovation, Infectious Diseases & Epidemiology, Large Language Models (LLMs)
Programmatic Theme: Clinical Research Informatics

Ongoing HIV education for providers in rural and non-academic settings is often limited, affecting access to highquality
care. This study analyzed 78 questions from Extension for Community Healthcare Outcomes (ECHO) sessions
to identify clinicians’ learning needs and develop a structured topic classification to improve future tele-mentoring.
Findings showed role-based differences: physicians and pharmacists focused on initiating and optimizing antiretroviral
therapy (ART); nurse practitioners emphasized prevention, adherence, and the “Undetectable = Untransmittable”
(U=U) concept; and allied health professionals, such as physician assistants, asked more about social support and care
navigation. Patterns in medication adherence and ART changes highlighted the value of ECHO data for identifying
clinical trends.
Furthermore, the study evaluated embedding models and large language models (LLMs) to automatically categorize
questions, supporting scalable, role-specific HIV training programs.

Speaker(s):
Amir Erfan Zareei Shams Abadi, PhD in Medical Informatics (Expected [2028])
University Of Missouri Columbia

Author(s):
Amir Erfan Zareei Shams Abadi, PhD in Medical Informatics (Expected [2028]) - University Of Missouri Columbia; Mirna Becevic, PhD - University of Missouri Department of Dermatology; Dima Dandachi, MD - University of Missouri Columbia;

The Uneven Distribution of Patient Portal Messages Across Patients and Physicians

Poster Number: 340
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Telemedicine, Documentation Burden, Patient Engagement and Preferences
Programmatic Theme: Clinical Informatics

Patient-initiated medical advice request (PMAR) messages are a growing source of uncompensated work for
physicians, but how this work is distributed is poorly understood. We characterized the national distribution of portal message volume across specialties and ambulatory visit volume. PMAR volume was highly skewed: a small fraction of patients and physicians accounted for a large share of messages. These results highlight the need for targeted interventions that recognize the uneven burden of portal-based care.

Speaker(s):
A J Holmgren, PhD
University of California, San Francisco

Author(s):
Nate Apathy, PhD - University of Maryland; Ateev Mehrotra, MD MPH - Brown University; Michael Chernew, PhD - Harvard University;

Bridging the Gaps in School-Based Telehealth: A Scoping Review and Evaluation Framework

Poster Number: 341
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Telemedicine, Evaluation, Informatics Implementation, Governance, Public Health, Health Equity
Programmatic Theme: Consumer Health Informatics

School-based telehealth (SBTH) is expanding rapidly as a strategy to improve healthcare access for K-12 students, supported by national policy initiatives from HHS and CMS. However, evaluation approaches remain fragmented and lack standardized metrics. We conducted a PRISMA-ScR scoping review of U.S. SBTH studies (2019–2025) and identified 22 eligible studies. Using insights from the literature and policy guidance, we developed a conceptual framework to support standardized evaluation and inform scalable, evidence-based integration of telehealth in school health systems.

Speaker(s):
Sara AlAjlouny, BPharm, MSc
University of North Carolina at Chapel Hill

Author(s):
Sara AlAjlouny, BPharm, MSc - University of North Carolina at Chapel Hill; Zhaoqiang Zhou, Master of Science - University of North Carolina at Chapel Hill, Chapel Hill, United States; Suguna Kotte, PharmD, PGDRA, MPS-BMHI(Expected Dec 2025) - UNC Chapel Hill; John Jenkins, MD - University of North Carolina at Chapel Hill; David McSwain, MD, MPH - UNC Health; Saif Khairat, PhD, MPH, FACMI, FAMIA - University of North Carolina at Chapel Hill;

Evidence-Based Informatics: A Use Case for Virtual Nursing

Poster Number: 342
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Telemedicine, Workflow, Workforce Development
Programmatic Theme: Clinical Informatics

irtual nursing (vRN) is an emerging care model designed to support bedside nurses through remote clinical activities enabled by telehealth technologies and electronic health record (EHR) access. Despite growing adoption, evidence guiding implementation remains limited. This work describes an evidence-informed informatics framework for virtual nursing, emphasizing workflow alignment, EHR integration, and structured data capture to support clinical decision support, care coordination, and innovation in nursing practice.

Speaker(s):
Brian Douthit, PhD, RN, NI-BC, FAMIA
Department of Biomedical Informatics, Vanderbilt University

Author(s):
Rhonda Day, MSN, RN - Vanderbilt; Catherine Ivory, PhD, NI-BC, NEA-BC, FAAN, FAMIA - Vanderbilt University Medical Center;

RE-AIM Analysis of Mobile Integrated Health

Poster Number: 343
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Transitions of Care, Mobile Health, Qualitative Methods, Telemedicine
Programmatic Theme: Clinical Informatics

We conducted a mixed-methods RE-AIM analysis of Mobile Integrated Health (MIH) with facilitated telehealth for heart failure patients across two urban health systems as part of the MIGHTy-Heart pragmatic RCT. Using EHR data and qualitative interviews, we found MIH reached a diverse population underserved by existing digital health infrastructure. While no significant differences in 30-day readmissions or health status were observed, facilitated telehealth through MIH supported patient trust, engagement, and interdisciplinary care coordination. Structural reimbursement barriers limited sustainability.

Speaker(s):
Ruth Masterson Creber, PhD, MSc, RN
Columbia University

Author(s):
Ruth Masterson Creber, PhD, MSc, RN - Columbia University; Meghan Reading Turchioe, PhD, MPH, RN - Columbia University School of Nursing; Melani Ellison, MPH - Columbia University School of Nursing; Jacky Choi, MPH - Weill Cornell Medicine; Yihong Zhao, PhD - Columbia University; Stacey Dai, MPH - Columbia University School of Nursing; Parag Goyal, MD - Weill Cornell Medicine; Brock Daniels, MD, MPH - Weill Cornell Medicine;

Qualitative Evaluation of Two Ambient Artificial Intelligence (AI) Scribes by Emergency Department Clinicians

Poster Number: 344
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Usability, Qualitative Methods, Artificial Intelligence, Surveys and Needs Analysis
Programmatic Theme: Clinical Informatics

Different ambient AI scribes may offer varying feature sets influencing their usefulness. We qualitatively evaluated two AI scribes in the emergency department, examining work burden, usability, documentation quality, and satisfaction. DAX more effectively reduced work burden, particularly cognitive load and editing time, and aligned better with ED workflow. Both tools showed high adoption and satisfaction. Common suggestions for improvements included greater transcription reliability, billing support, and note template customization.

Speaker(s):
Wendy Yin, MD
UT Southwestern

Author(s):
Robin Higashi, PhD - UT Southwestern Medical Center; Emily Repasky, MA - UTSW Medical Center; Robert Turer, MD - UT Southwestern Medical Center; Justin Rousseau, MD, MMSc - University of Texas Southwestern Medical Center; Janet Webb, MD - UT Southwestern; Amber Salter, PhD - UT Southwestern; Mark Courtney, MD - UT Southwestern; Ling Chu, MD - UT Southwestern; Wendy Chapman, PhD - University of Texas Southwestern; Samuel McDonald, MD - UT Southwestern;

Multi-Method Usability Testing of Nudges to Support Statin Prescribing Incorporating Eye Tracking and Cognitive Load Assessment

Poster Number: 345
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Usability, Clinical Decision Support, Human-computer Interaction, User-centered Design Methods
Programmatic Theme: Clinical Informatics

Many patients who meet guideline criteria for a statin have not been prescribed one. Nudges can be a promising approach for providing clinical decision support that improves guideline-concordant statin prescribing while minimizing provider cognitive load and workflow interruptions. We developed three nudges within encounter, lab result review, and refill request workflows. In usability testing, nudges within the encounter and lab result review directed visual attention and information processing without increasing perceived workload.

Speaker(s):
Angela Mastrianni, PhD
NYU Grossman School of Medicine

Author(s):
Angela Mastrianni, PhD - NYU Grossman School of Medicine; Nicole Redfern, MPH - NYU Langone Health; Priyanka Solanki, MD - NYU; Defne Levine, MPH - NYU Grossman School of Medicine; Nusrat Jahan, MD - NYU Langone Health; Yuhan Cui, MS - NYU Langone; Grant Ketron, BSN-RN, MBA - NYU Langone Health; John Dodson, MD, MPH - NYU Langone Health; Todd Hudson, PhD - NYU Langone Health; Devin Mann, MD - NYU Grossman School of Medicine; Safiya Richardson, MD, MPH - NYU Langone;

PICU Aware: Real-time Situational Awareness Integrating Machine Learning Risk Prediction and Spatial Visualization in the Pediatric Intensive Care Unit

Poster Number: 346
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: User-centered Design Methods, Human-computer Interaction, Critical Care, Artificial Intelligence, Large Language Models (LLMs), Information Visualization, Clinical Decision Support
Programmatic Theme: Clinical Informatics

Clinical deterioration in the PICU remains challenging to anticipate due to misaligned mental models and fragmented data. We employed a human-centered approach (N=75 clinicians) to develop PICU Aware, a sociotechnical intervention bridging ML-driven risk scores with spatial visualization. Utilizing real-time FHIR and Epic APIs, the platform translates the PICU Warning Index (PWIN) – a deterioration prediction, into actionable situational awareness, demonstrating how explainable AI and spatial design unify team-based anticipation of clinical deterioration.

Speaker(s):
Sachin Grover, Data Scientist
The Children's Hospital of Philadelphia

Author(s):
Sachin Grover, Data Scientist - The Children's Hospital of Philadelphia; Sanjiv Mehta, MD, MSCE - Children's Hospital of Philadelphia and University of Pennsylvania School of Medicine; Hannah Stinson, MD - Children's Hospital of Philadelphia and University of Pennsylvania; Eamonn Tweedy, PhD - Children's Hospital of Philadelphia; Megan Bernstein, CNS - Children's Hospital of Philadelphia; Lauren Dinsick, BSN - Children's Hospital of Philadelphia; James Sannino, MS; Akira Nishisaki, MD, MSCE - Children's Hospital of Philadelphia; Fuchiang (Rich) Tsui, PhD, FAMIA, IEEE Senior Member - Children's Hospital of Philadelphia and University of Pennsylvania;

Enabling Large-Scale Decentralized Clinical Trials Through Digital Collaborative Tools

Poster Number: 347
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Workflow, Real-World Evidence Generation, Patient-/Person-Generated Health Data, Mobile Health
Programmatic Theme: Clinical Research Informatics

Decentralized clinical trials (DCTs) are expanding as digital health technologies increasingly facilitate remote recruitment, data collection, and monitoring, yet many DCTs face operational and methodological challenges. We present a case study of a fully remote 6-month trial (n=199) evaluating digital tools for heart failure self-care. We discuss the ecosystem of platforms we used to support trial operations, including advantages and challenges we faced, offering practical guidance for research teams implementing large-scale DCTs.

Speaker(s):
Rachel Tunis, PhD Student
University of Texas at Austin

Author(s):
Rachel Tunis, PhD Student - University of Texas at Austin; Vershanjali Chauhan, MS - University of Texas at Austin; Angelica Rangel, MS - The University of Texas at Austin; Kavita Radhakrishnan, PhD - University of Texas - Austin;

An Extensible Open Pipeline for Biomedical Knowledge Graph Builds

Poster Number: 348
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Workflow, Data transformation/ETL, Controlled Terminologies, Ontologies, Vocabularies, Data Sharing
Programmatic Theme: Translational Bioinformatics

Biomedical knowledge graphs (KGs) support data integration and AI-driven discovery but are difficult to maintain as source data and access conditions change. We developed an extensible, open-source pipeline using Snakemake to rebuild and extend the PrimeKG precision medicine knowledge graph. The modular workflow enables reproducible updates, flexible data-source integration, and new entity types. Demonstrated with ClinVar variant data, the resulting KG contains 90,686 nodes and 4.9 million edges, supporting scalable biomedical knowledge integration.

Speaker(s):
Kazi Rana, PhD Student
Clemson University

Author(s):
Kazi Rana, PhD Student - Clemson University; Aaron Masino, PhD - Clemson University;

Building National AI/ML Capacity in Translational Informatics: Design and Outcomes of the AIM-AHEAD & NCATS Health Data Science Training Program Using the National Clinical Cohort Collaboration Data Enclave

Poster Number: 349
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Workforce Development, Artificial Intelligence, Machine Learning
Programmatic Theme: Translational Bioinformatics

Our dynamic traineeship equips nationwide early-career professionals with the acumen needed to further their careers in AI/ML and healthcare. We have launched three cohorts, providing hands-on experience in AI/ML and health data analysis using real-world health data, engaging 154 trainees; 2024 (n=55), 2025 (n=49), and 2026 (n=50), and their mentors. This presentation highlights our successes, lessons learned, and curation of a scalable model for national AI/ML workforce development programs.

Speaker(s):
Toufeeq Syed, PhD, MS
UT Health Houston

Author(s):
Toufeeq Syed, PhD, MS - UT Health Houston; Aubri S. Hoffman, PhD - Axle Research & Technologies; Robert T. Mallet, PhD - UNT Health Science Center; Legand L. Burge, III, PhD - Howard University; Nathan Hotaling, PhD - Axle Research & Technologies; Suzanne Randolph Cunningham, PhD - The MayaTech Corporation; Shelly Kowalczyk, MSPH - The MayaTech Corporation; Alen Delic, MS - Axle Research & Technologies; Mohadeseh Hashemidehagi, PhD - Axle Research & Technologies; Babajide Sadiq, DrPH - Axle Research & Technologies; Mahbubul Hasan, MS - Axle Research & Technologies; Matthew Owens, PhD - Axle Research & Technologies; Shawn O'Neil, MS - Axle Research & Technologies; Johanna Loomba, ME - University of Virginia; Douglas F. Diuzen, PhD - Axle Research & Technologies; Carolyn Kelley, MS - Axle Research & Technologies; D'Laney Kernan, BA - The University of Texas Health Science Center at Houston; Sarah Popal, MS - The University of Texas Health Science Center at Houston; Erika Cavazos-Juarez, DHI, MBBS, MS, PMP - University of Texas (UTH) Health Science Center at Houston McWilliams School of Biomedical Informatics (SBMI); Jamboor K. Vishwanatha, PhD - UNT Health Science Center;

Custom CSS

Poster Session 3

Description

Enhancing Student Learning and Engagement with AI Tutors in Health Informatics Higher Education

Category

Description

Custom CSS