Building a Foundation of Quality Data for AI for Clinical Decision-Making: The Post-Acute Care InterOperability (PACIO) Project

Poster Number: P01

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: Artificial Intelligence/Machine Learning, Interoperability, Co-production/Co-Design, Cross-organization Partnerships including Public-private Partnerships, Health Information Exchange (HIE), Health IT Standards (USCDI, FHIR®, SMART, etc.)
Primary Track: AI and Care Outcomes
Programmatic Theme: Clinical Decision Support and Analytics

Artificial intelligence (AI) has tremendous potential to improve clinical care and decision-making. However, siloed systems and lack of standardized data across EHRs constrains the ability of AI to generate accurate and useful outputs. This is particularly relevant to in post-acute care (PAC) that often lack interoperability between other PACs and acute care settings, lead swaths of critical data missing from PAC patients records, potentially impacting AI used in their care. To address this issue, the Post Acute Care InterOperability Project (PACIO), to establish a framework for developing Health Level Seven (HL7®) Fast Healthcare Interoperability Resources (FHIR®)* implementation guides (IGs) to facilitate health information exchange between PAC and other entities. PACIO has applied user-centered design and evidence-based approaches to develop IGs, which were then tested in Connectathons with PAC stakeholders. The Connectathons have demonstrated the IGs' technical capabilities, provided opportunities to identify issues that implementers may face, and built community among PAC stakeholders. Next steps include developing new use cases and improving existing IG through Connectathons. By improving interoperability, PACIO has the potential to support an EHR ecosystem that is more complete, incorporating data from acute and PAC settings. This can better inform the development and application of AI to support clinical care.

Speaker:
Uba Backonja, PhD, MS, RN
The MITRE Corporation

Authors:
Uba Backonja, PhD, MS, RN - The MITRE Corporation; Jessica Skopac, PhD/JD/MA; Dave Hill, BS - MITRE; Howard Capon, MPH NRP; Brian Meshell, MS - MITRE; Lorraine Wickiser; Elizabeth Palena Hal, MIS MBA RN - Centers for Medicare & Medicaid Services; Alyssa Ford, MOT, DHSc - Indiana Wesleyan University;

Summarizing Image Dataset to Detect Malignancies using Supervised Deep Learning

Poster Number: P02

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: Care Delivery Models, Adaptive Clinical Decision Support, Artificial Intelligence/Machine Learning, Data Science, Data Visualization
Primary Track: AI and Care Outcomes
Programmatic Theme: Clinical Decision Support and Analytics

In this study, we explore a supervised deep learning approach with transfer learning to analyze 3D imaging scans for malignancy diagnosis. Each scan comprises hundreds of slices, though only a few may depict malignant lesions. Our model aimed to summarize datasets by focusing on these critical slices. However, initial results showed limited performance that suggests further exploration of advanced techniques, such as Reinforcement Learning, to enhance the ability to effectively distinguish malignancies within complex datasets.

Speaker:
Rayyan Khan, PhD
CancerCare Manitoba

Leveraging AI Knowledge-sharing Tools to Enhance Human-System Performance in EHR Environments

Poster Number: P03

2025 Clinical Informatics Conference 25x5 Presentation

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: Artificial Intelligence/Machine Learning, Usability and Measuring User Experience, Documentation Burden, Building Value for Informatics via Education and Training, Clinician Burnout, Workflow Efficiency, Patient Safety, Disruptive and Innovative Technologies
Primary Track: AI and Care Outcomes
Programmatic Theme: Emerging Technology and Technical Infrastructure

As healthcare technology advances, the challenge of addressing usability and workflow inefficiencies within EHR systems becomes increasingly urgent. This poster presents innovative AI-driven knowledge-sharing tool, designed to strengthen usability competencies among VA informaticists. Our AI/ML framework enables dynamic knowledge sharing and real-time feedback, facilitating continuous improvement in user training and support. We will detail how this tool is used to identify and address workflow inefficiencies in the Veterans Health Information Systems and Technology Architecture (VistA) EHR and the new Federal EHR powered by Oracle Health technology. Additionally, we will demonstrate how AI-driven analytics can deliver actionable insights tailored to meet the needs of informaticists. By fostering interoperability and streamlining usability evaluations, our tool supports clinician satisfaction and improves patient care outcomes. This poster invites attendees to explore how AI-driven insights can transform usability training and enhance EHR user experience across the VHA network.

Speaker:
Ashley Ercolino, Ph.D.
Design Interactive

Authors:
Jesse Flint, MS - Design Interactive, Inc.; D’An Knowles Ball, PhD - Design Interactive; Nicole Dorey, PhD - Design Interactive; Logan Gisick; Logan Gisick, PhD - Design Interactive; William Rivera, PhD - Design Interactive; Kay M. Stanney, PhD - Design Interactive;

SNAP Judgments: Using LLMs and Pharmacy Dispense Data to Uncover Disparities in Pediatric AOM Antibiotic Prescribing and Utilization

Poster Number: P04

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: Artificial Intelligence/Machine Learning, Medication Adherence, Social Determinants of Health
Primary Track: AI and Care Outcomes
Programmatic Theme: Informatics-Driven Value-Based Healthcare

This study used an LLM-based system and novel SQL methodology to classify treatment plans and capture dispensing data for pediatric AOM, addressing the understudied area of SNAP utilization. Findings revealed disparities in SNAPs, with non-English speakers and lower SES patients less likely to receive SNAPs, highlighting equity issues.

Speaker:
Anh Vo, MD, MPH
UCSF Benioff Children's Hospitals

Authors:
Jessica Pourian, MD - UCSF; Ben Michaels, PharmD - UCSF; Anh Voh, MD MPH - UCSF; A J Holmgren, PhD - University of California, San Francisco; Augusto Garcia-Agundez, PhD - UCSF; Valerie Flaherman - UCSF;

Promoting Health Equity: A Section 1557 Compliance Framework for CDS Discrimination Risk Stratification

Poster Number: P05

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: Algorithmic bias and impacts on Health Equity, Adaptive Clinical Decision Support, Risk Measurement, Patient Safety, Affordable Care Act (ACA), Ethical, Legal, and Social Issues
Primary Track: AI and Care Outcomes
Programmatic Theme: Clinical Decision Support and Analytics

We present a framework for healthcare institutions to assess and mitigate discrimination risks in clinical decision support (CDS) tools, addressing Section 1557 compliance of the Affordable Care Act. The approach identifies CDS subtypes, stratifies risks, and proposes mitigation strategies. Ten CDS subtypes were identified, with three classified as high-risk. Recommendations include improving awareness, establishing governance, implementing monitoring protocols, and developing specific strategies for high-risk CDS subtypes.

Speaker:
Bethel Mieso, MD
Stanford

Authors:
Kameron Black, DO, MPH - Stanford University; Aydin Zahedivash, MD, MBA - Stanford University; Nicholas Marshall, MD - Stanford; Keith Morse, MD - Stanford University School of Medicine;

Synthesizing Excellence: A Solution for Complete and Reliable Trauma Registry Testing

Poster Number: P06

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: Artificial Intelligence/Machine Learning, Big Data, Data Science, Learning Health System, Clinical Process Automation
Primary Track: AI and Care Outcomes
Programmatic Theme: Emerging Technology and Technical Infrastructure

Description of the Problem or Gap
Measuring combat trauma outcomes allows for medical performance optimization, clinical practice guideline revision and material solution development. Trauma outcomes are currently evaluated by the Department of Defense Trauma Registry (DODTR). The DODTR is currently limited by incomplete data, inconsistent data, illegible data and survivor bias.

Methods: Synthetic data allows for reliable, consistent, and valid representations of actual patient injuries, interventions and outcomes. Given the difficulties in obtaining complete and accurate combat data, synthetic data sets help to close this gap. Using Recurrent Neural Networks and Long Short Term Memory models, a synthetic dataset will be generated to accurately reflect operational medicine trauma patients’ injuries, interventions, and outcomes. Artificial Intelligence and Machine Learning techniques such as Transfer Learning, Gradient Boosting, Random Forrest and Causal Inference will be used to test and validate the synthetic dataset.

Results: Results are pending

Conclusion: Operational Medicine combat data is limited in accuracy, completion and survivor bias. This endeavors aims to investigate Generative AI models to create synthetic patient cases for casualty prediction, logistics forecasting, and novel generation of clinical practice guidelines. The generation of synthetic datasets helps to address these gaps in a meaningful and reliable way.

Speaker:
Darshan Thota, MD
United States Navy

Authors:
Jonathan Stallings, PhD - Joint Trauma System; Jennifer Gurney, MD - Defense Health Agency;

Development and Validation of a Feature-Based Machine Learning for ECG Artifact Detection and Classification

Poster Number: P07

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: Artificial Intelligence/Machine Learning, Clinical informatics organizational models, Data Science
Primary Track: AI and Care Outcomes
Programmatic Theme: Clinical Decision Support and Analytics

This study focuses on enhancing ECG diagnostic accuracy by addressing the challenge of signal artifacts, which can obscure critical information. A machine learning model was developed to classify ECG artifacts using a feature-based approach. Features were extracted from Time, Frequency, Time-frequency, and Decomposition domains, and Recursive Feature Elimination (RFE) was employed to narrow down to 45 key features relevant to artifact detection, such as motion artifacts, baseline wander, muscle tremors, and power line interference. The Light Gradient Boosting Machine (LightGBM) was chosen as the classifier, and model performance was tested internally on the KURIAS ECG database, containing 36,000 records, and validated externally on two databases: CinC2011 (12,000 records) and MIT-BIH NST (800 records). Results demonstrated high accuracy, with the model achieving an F1 score of 89.88% and an MCC of 88.35% in internal testing, and F1 scores of 94.35% and MCCs around 92.70% in external testing. These findings indicate that the model can effectively classify artifacts across various conditions, enhancing ECG reliability and diagnostic accuracy. Future efforts will explore real-time analysis and broader clinical applications, aiming to integrate this solution into routine clinical workflows.

Speaker:
Jose Moon, MS
Korea University

Authors:
Hyung Joon Joo, MD, PhD; JONGHO KIM, Ph.D;

AI Encounter Summarization: Early Insights from Clinician-Users

Poster Number: P08

2025 Clinical Informatics Conference 25x5 Presentation

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: Artificial Intelligence/Machine Learning, EHR Implementation and Optimization, Workflow Efficiency
Primary Track: AI and Care Outcomes
Programmatic Theme: Usability, Efficiency, and User Experience

This session explores early clinician feedback on an Epic AI encounter-summarization tool. Based on survey data, clinicians desire a summarization approach that prioritizes high-value information without compromising patient safety. Key concerns included ensuring an efficient workflow and accuracy in AI outputs. These findings underscore the importance of iterative tool development to balance conciseness and completeness to enhance clinical efficiency and care quality. Attendees will gain insights into optimizing AI summarization tools for diverse clinical settings.

Speaker:
Jared Silberlust, MD MPH
NYU Langone Health

Assessing the Reliability of Large Language Models in Assisting with Qualitative Analysis for Healthcare Informatics Research

Poster Number: P09

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: Artificial Intelligence/Machine Learning, Disruptive and Innovative Technologies, Precision Health and Genomics
Primary Track: AI and Care Outcomes
Programmatic Theme: Emerging Technology and Technical Infrastructure

This study explored using large language models (LLMs) to expedite thematic analysis in healthcare informatics research. Four LLMs (Gemini-1.5-pro, GPT-4o, Claude-3.5-sonnet, and llama-3.1-8b-instruct) analyzed clinician interviews evaluating pharmacogenomics module using a standardized prompt. Gemini and Claude showed 80% consistency with human-generated themes, also revealing novel insights. While significantly faster, LLMs lacked the contextual interpretation depth of human coders. LLMs show promise but currently serve best as complementary tools for healthcare informatics analysis.

Speaker:
Je-Won Hong, PharmD
University of Florida

Author:
Khoa Nguyen, Pharm.D - University of Florida;

Analyzing Clinician Responses to Electronic Health Records Alerts with Large Language Models

Poster Number: P10

2025 Clinical Informatics Conference 25x5 Presentation

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: Adaptive Clinical Decision Support, Artificial Intelligence/Machine Learning, EHR Implementation and Optimization
Primary Track: AI and Care Outcomes
Programmatic Theme: Clinical Decision Support and Analytics

Electronic health record alerts offer summarized computerized knowledge to aid clinical decision-making. However, inappropriate alerts contribute to alert fatigue. Large language models offer the ability to relatively accurately categorize and determine the appropriateness of free-text clinician rationale for bypassing alerts that warn of duplicate chest X-ray orders. This analysis can be used to improve alert triggers to reduce inappropriate alerts, or to inform and influence clinical practice to minimize the bypass of appropriate alerts.

Speaker:
Michael Phillipi, BA
UCI

Authors:
Michael Phillipi, BA - UCI; Shawn Sun, MD - University of California, Irvine; Jeanette Meraz, BS - University of California, Irvine; Gavin Shu, BS - University of California, Irvine; Hannah Cho, BS - University of California, Irvine; Jadyn Lontoc, BS - University of California, Irvine; Justin Glavis-Bloom, MD - University of California, Irvine; Roozbeh Houshyar, MD - University of California, Irvine;

Development of a Deep Learning Model for Intussusception Using Point-of-Care Ultrasound

Poster Number: P11

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: Artificial Intelligence/Machine Learning, Disruptive and Innovative Technologies, Adaptive Clinical Decision Support
Primary Track: AI and Care Outcomes
Programmatic Theme: Emerging Technology and Technical Infrastructure

Introduction: Intussusception is a pediatric emergency frequently with delays. We developed deep learning models for detection on POCUS images.

Methods: Images trained using various machine learning and classification approaches.

Results: 1,582 intussusception and 1,965 normal images were converted. Fine-tuning models performed better than machine, ensemble and transfer learning. Threshold-based classification resulted in greatest predictive performance.

Conclusions: Feasibility of developing deep learning model for intussusception using a smaller dataset of POCUS images compared to previous studies.

Speaker:
Brian Lefchak, MD, MPH
Hennepin Healthcare

Authors:
Anand Thyagachadnran, PhD - Department of Computer Science & Engineering, Indian Institute of Technology Madras; Brian Lefchak, MD, MPH - Hennepin Healthcare; Hema Murthy, PhD - Department of Computer Science & Engineering, Indian Institute of Technology Madras; Kelly Bergmann, DO, MS - Department of Pediatric Emergency Medicine, Children’s Minnesota; Manu Madhok, MD, MPH - Department of Pediatric Emergency Medicine, Children’s Minnesota;

Predicting In-Hospital Mortality following Blunt Splenic Injury

Poster Number: P12

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: Artificial Intelligence/Machine Learning, Data Science, Big Data
Primary Track: AI and Care Outcomes
Programmatic Theme: Clinical Decision Support and Analytics

Research with a North American Trauma Database to examine risk factors for in-hospital mortality in blunt splenic injury patients demonstrated that splenic injury severity and its management strategy were not the prime determinants of mortality. As compared to XGBoost and 5-layer neural networks, Generalized Additive Models (GAMs) were interpretable with comparable performance.

Speaker:
Randeep Jawa, MD
Stony Brook Medicine

Authors:
Randeep Jawa, MD - Stony Brook Medicine; James Vosswinkel, MD - Stony Brook University; Janos Hajagos, Ph.D. - Stony Brook University;

Explainable LLM Classification (ELC) for Discharge Disposition Prediction

Poster Number: P13

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: Artificial Intelligence/Machine Learning, Adaptive Clinical Decision Support, Data Science, Coordination Across the Continuum of Care, Risk Measurement, Big Data
Primary Track: AI and Care Outcomes
Programmatic Theme: Clinical Decision Support and Analytics

Discharge disposition planning to Skilled Nursing Facilities (SNFs) for hospitalized patients is an important process. A novel method called Explainable LLM Classification (ELC) was developed, where GPT4-generated summaries of unstructured history and physical notes are used as both input data to discriminative LLMs and as an interpretable output for end-users. ELC enhances the predictive performance of discriminative LLMs above truncated text alone, and the best-performing fine-tuned LLM, NYUTron, vastly outperformed human predictions

Speaker:
Ryan Crowley, BS, Mphil
NYU Grossman School of Medicine

Authors:
Ryan Crowley, BS, Mphil - NYU Grossman School of Medicine; Chloe Pariente, MBAN - NYU Langone Health; Kevin Eaton, MD - NYU Langone Health; Yindalon Aphinyanaphongs, MD - NYU Langone Health; William Small, MD, MBA - NYU Langone Health;

A Deep Learning Approach to Classifying Catatonia Diagnosis from Clinical Text

Poster Number: P14

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: Artificial Intelligence/Machine Learning, Adaptive Clinical Decision Support, Risk Measurement
Primary Track: AI and Care Outcomes
Programmatic Theme: Clinical Decision Support and Analytics

Catatonia is a treatable condition that is underrecognized in the hospital setting. No HIT tools are available to help clinicians in considering targeted screening for the diagnosis. We have built a machine learning model that accurately classifies the diagnosis of catatonia from a subset of MIMIC-IV clinical text with high precision. Further work is needed in data collection, local training, and explainability analysis but initial results are encouraging for possible downstream CDS applications.

Speaker:
Jonathan Chastain, DO
University of Arkansas for Medical Sciences

Authors:
Jonathan Chastain, DO - University of Arkansas for Medical Sciences; Jonathan Bona, PhD - University of Arkansas for Medical Sciences (UAMS);

Dose-volume histograms guided deep dose predictions

Poster Number: P15

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: Artificial Intelligence/Machine Learning, Workflow Efficiency, Clinical Process Automation, Adaptive Clinical Decision Support, Digital Therapeutics
Working Group: Clinical Decision Support Working Group
Primary Track: AI and Care Outcomes
Programmatic Theme: Usability, Efficiency, and User Experience

In radiotherapy treatment planning, we propose a deep learning framework that integrates target Dose-Volume Histograms (DVHs) to improve dose prediction accuracy and enable interactive adjustments within the treatment planning system.
Our model uses DVHs with Direct Affine Feature Transforms (DAFT) to adapt dose distributions based on dosimetrist-defined DVHs.
The DVH-guided model outperformed traditional approaches, facilitating a workflow where dosimetrists design target DVHs, streamlining plan customization for optimized, patient-specific treatments.

Speaker:
Paul Dubois, Master
TheraPanacea

Post-Processing Algorithmic Bias Mitigation at NYC Health + Hospitals

Poster Number: P16

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: Algorithmic bias and impacts on Health Equity, Educating on Self Service Analytics, Diversity, Equity and Inclusion, Population Health, HIT Safety/EHR Unintended Consequences, Adaptive Clinical Decision Support, Artificial Intelligence/Machine Learning
Primary Track: AI and Care Outcomes
Programmatic Theme: Emerging Technology and Technical Infrastructure

Algorithmic bias in healthcare occurs when model performance varies by protected class, leading to biased treatment. We identified bias in an Epic asthma model at a large urban safety net system. Threshold adjustment and reject option classification post-processing methods both reduced bias meaningfully, with custom threshold adjustment in RStudio out-performing open-source python methods. Our findings show the promise of post-processing to lower the barrier-to-entry for mitigating bias in binary predictive models for other healthcare systems.

Speaker:
Silas Lee, MS
NYC Health and Hospitals

Authors:
Shaina Mackin, MPH - NYC Health + Hospitals; Silas Lee, MS - NYC Health and Hospitals; Abe Dickenson, BS - Epic; Vincent Major, PhD - NYU Grossman School of Medicine; Rumi Chunara, PhD - NYU Center for Health Data Science; Remle Newton-Dame, MPH - NYC Health + Hospitals;

The Association of Age and Sex on Revision Microdiscectomy and Subsequent Fusion Following Lumbar Microdiscectomy

Poster Number: P17

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: Data Visualization, Big Data, Adaptive Clinical Decision Support, Data Science, Risk Measurement
Primary Track: AI and Care Outcomes
Programmatic Theme: Informatics-Driven Value-Based Healthcare

A retrospective analysis of insurance claims and electronic health records identified risk factors for revision lumbar microdiscectomy and subsequent fusion. The findings revealed a significantly higher risk of lumbar fusion in women aged 40-59 with at least one comorbidities. This suggests a potential for improved patient outcomes through targeted interventions and preventative strategies for this high-risk group.

Speaker:
Charlotte Drury-Gworek, MSN
Highmark Health

Authors:
Charlotte Drury-Gworek, MSN - Highmark Health; Shannon Richards, MSN - Highmark Health; Tyson S. Barrett, PhD - Highmark Health; Dallas E. Kramer, MD - Allegheny Health; Keith LeJeune, PhD - Highmark Health/Allegheny Health Network; Lara Massie, MD - Highmark Health;

Validating a Proprietary No-Show Predictive Model in a Large Pediatric Primary Care Network

Poster Number: P18

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: Artificial Intelligence/Machine Learning, Algorithmic bias and impacts on Health Equity, Data Visualization
Primary Track: AI and Care Outcomes
Programmatic Theme: Clinical Decision Support and Analytics

This study evaluates the performance of a proprietary naive Bayes No-Show prediction algorithm in a large pediatric primary care network. Despite widespread use of such proprietary models, validation remains limited. Findings indicate a poor calibration, especially within the 31-70% prediction range, and an area under the curve (AUC) of 0.68, suggesting limited efficacy for predicting No-Shows in the local context of this pediatric primary care network.

Speaker:
Eleanor Verhagen, MS
Boston Children's Hospital

Authors:
Emily Trudell Correa, MPH, MS - Boston Children's Hospital; Naveed Rabbani, MD - Boston Children's Hospital; Louis Vernacchio, MD, MSc - Boston Children's Hospital; Jonathan Hatoun, MD, MPH, MS - Boston Children's Hospital;

Using a Word Embedding Approach to Detect Negative Provider Sentiment in Clinical Notes

Poster Number: P19

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: Artificial Intelligence/Machine Learning, Bridging Analytics, Bedside Care, Clinical Documentation, and Education, Diversity, Equity and Inclusion
Primary Track: AI and Care Outcomes
Programmatic Theme: Clinical Decision Support and Analytics

Negative provider sentiment toward patients can negatively impact patient care, with some patients receiving lower quality care due to factors such as substance use disorder. Negative sentiment in clinical notes can both indicate negative sentiment toward patients from a given provider, and also propagate this negative sentiment, affecting other providers' decision making. We have developed an NLP model to quantify language containing negative sentiment in clinical notes, an important step in understanding the relationship of the language found in notes to clinical outcomes and mitigating its impact.

Speaker:
Patrick Wedgeworth, MD, MISM
University of Washington

Authors:
Priscilla Lui, PhD - University of Washington; Aishwarya Raj, Student - University of Washington;

Preliminary Findings From A Comparative Study of GPT and Neuro-symbolic Models in Salt Content Information Delivery

Poster Number: P20

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: Artificial Intelligence/Machine Learning, Algorithmic bias and impacts on Health Equity, Driving Digital Equity, Social Determinants of Health
Primary Track: AI and Care Outcomes
Programmatic Theme: Usability, Efficiency, and User Experience

Introduction: African American patients (AA) with heart failure (HF) have poorer prognoses, exacerbated by limited access to customized self-care resources (Nayak, Hicks, and Morris 2020). Our previous work focused on developing dialogue agents (DAs) for personalized, accurate self-care guidance (Tayal et al. 2024). The second iteration of our DA, HFChat, uses a neuro-symbolic approach and USDA data to provide salt content information in foods, adjusting for cooking method, portion size, home-cooked vs. restaurant meals, and added ingredients. We compare this HFChat iteration with a custom GPT-based DA built on OpenAI’s infrastructure utilizing the same data inputs to assess the feasibility of an LLM-based solution.
Methods: Nineteen AA participants with HF were recruited. Each participant asked both ChatGPT and HFChat five food-related questions. Both systems requested clarifying information to personalize answers. Demographic, health literacy, and digital literacy data, along with user experience surveys, were also collected.
Results: The average age of the cohort was ~60 years, with 7/19 participants being women. HFChat was preferred by 10/19 participants; preferences varied by digital literacy (see graph).
Conclusion: Preliminary findings show that participants with higher digital literacy preferred HFChat, while those with lower digital literacy favored ChatGPT. Individuals with lower digital literacy may lack the skills to critically assess health information.(Yuen et al. 2024). This is especially relevant given ChatGPT’s potential for hallucinations (Siontis et al. 2024). Additionally, DAs built on external infrastructure like OpenAI cannot be fully controlled making consistency and reliability of information difficult to guarantee.

Speaker:
Devika Salunke, BPharm, MS, MSc
University of Illinois

Authors:
Devika Salunke, BPharm, MS, MSc - University of Illinois; Anuja Tayal, BTech - University of Illinois Chicago; Barbara Di Eugenio, PhD - University of Illinois Chicago; Paula Allen-Meares, PhD - University of Illinois Chicago; Carolyn Dickens, PhD; Olga Garcia-Bedoya, MD - University of Illinois Chicago; Eulalia Abril, PhD - University of Illinois Chicago; Andrew Boyd, MD - University of Illinois at Chicago;

Leveraging Generative AI in Clinical Decision Support: A Case Report on Using OpenEvidence for Diagnosing Treatment-Induced Neuropathy in Diabetes

Poster Number: P21

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: Artificial Intelligence/Machine Learning, Workflow Efficiency, SAFER guidelines, Patient Safety
Primary Track: AI and Care Outcomes
Programmatic Theme: Informatics-Driven Value-Based Healthcare

This case report demonstrates the application of OpenEvidence, a generative AI tool, in diagnosing Treatment-Induced Neuropathy of Diabetes (TIND). A 74-year-old male presented with progressive neuropathy following rapid correction of serum glucose. Initial extensive investigation did not identify a cause. Using OpenEvidence, clinician accessed synthesized evidence and prioritized TIND as a differential diagnosis enabling evidence-based management. This case illustrates the role of AI in improving clinical decision-making and patient outcomes in complex primary care cases.

Speaker:
Toyosi Akinbami, MD
Oregon Health and Science University

Leveraging Large Language Models to Predict Mental Status in Patients with Dementia Using Electronic Health Records

Poster Number: P22

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: Artificial Intelligence/Machine Learning, Big Data, Adaptive Clinical Decision Support
Primary Track: AI and Care Outcomes
Programmatic Theme: Emerging Technology and Technical Infrastructure

Elderly individuals living with dementia are particularly vulnerable to mental health challenges. This study employs Large Language Models (LLMs) to analyze Electronic Health Records (EHRs) and predict patients' mental status. We evaluated the performance of four LLMs combined with three prompting strategies in assessing the mental status of individuals with dementia.

Speaker:
Jie Yang, PhD, FAMIA
Brigham and Women's Hospital/Harvard Medical School

Authors:
Jiageng Wu, MS - Brigham and Women’s Hospital and Harvard Medical School; Richard Wyss, PhD - Brigham and Women’s Hospital and Harvard Medical School; Kueiyu Joshua Lin, ScD - Brigham and Women’s Hospital and Harvard Medical School; Jie Yang, PhD - Brigham and Women’s Hospital and Harvard Medical School;

Bridging AI and Clinical Guidelines: An integrated clinical and AI stratification framework for Diabetes Management

Poster Number: P23

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: Artificial Intelligence/Machine Learning, Care Delivery Models, Population Health
Primary Track: AI and Care Outcomes
Programmatic Theme: Informatics-Driven Value-Based Healthcare

A significant gap exists between the predictive capabilities of AI-driven diagnostics and the practical application of clinical guidelines in chronic disease management. Current systems often operate in isolation, failing to integrate AI insights with guideline-based risk stratification to provide actionable, patient-centered care. The Integrated Clinical-AI Stratification Framework by bridges this divide by combining structured and unstructured EHR data with clinical guidelines. Patients are dynamically assigned to clinically sound risk groups, validated by healthcare professionals, with tailored interventions aligned with international standards. From early screening to complication prevention, the integrated clinical and AI stratification enables proactive, precision-targeted care. Ongoing validation efforts with national health data in Saudi Arabia demonstrated its potential to transform diabetes management.

Speaker:
Mohammad Ghosheh, M.D
iO Health

Designing Clinician Personas to Enhance Technology Adoption

Poster Number: P24

2025 Clinical Informatics Conference 25x5 Presentation

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: Ambient documentation, Clinician Burnout, Documentation Burden
Primary Track: Documentation Burden, Clinician Well-Being and Care Outcomes
Programmatic Theme: Usability, Efficiency, and User Experience

Clinician documentation burden continues to challenge healthcare systems, contributing to burnout and inefficiencies. Despite the availability of various documentation tools—such as note templates, speech-to-text dictation, scribes, and ambient scribes—their adoption often falters due to a lack of customization for diverse clinician needs. This study introduces a persona-based approach to optimize technology implementation for reducing documentation burden. Personas, fictional characters representing key clinician profiles, were developed through user interviews, and input from frontline clinicians. The resulting four personas—Cautious Carl, Traditional Tina, Efficient Edward, and Innovative Indira—illustrate the spectrum of clinician preferences based on technology comfort, documentation habits, and specialties. EHR coaches (N=21) validated the utility of these personas, with 100% agreeing that personas facilitate the development of tailored tools and training. Coaches also provided recommendations for tools matched to each persona, such as human scribes for Cautious Carl and ambient AI for Innovative Indira. These findings underscore the importance of user-centered design in healthcare technology, offering a scalable, needs-based strategy to enhance clinician satisfaction and efficiency. Future work includes validating this approach in broader healthcare settings to further refine its applicability and impact.

Speaker:
Charumathi Raghu Subramanian, MD
UCSF

Author:
Matthew Sakumoto, MD - Sutter Health;

Ambient AI Scribes: Surveying and Quantifying Education and Care Quality Concerns

Poster Number: P25

2025 Clinical Informatics Conference 25x5 Presentation

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: Bridging Analytics, Bedside Care, Clinical Documentation, and Education, Ambient documentation, Artificial Intelligence/Machine Learning
Primary Track: Documentation Burden, Clinician Well-Being and Care Outcomes
Programmatic Theme: Emerging Technology and Technical Infrastructure

Our study examines concerns about AI ambient documentation tools among medical students, residents, fellows, and attending doctors. By conducting a cross-sectional survey at UCI Health, the study discovered significant gender differences related to AI’s impact on medical training milestones, privacy, and documentation accuracy. Our work indicates the need for further investigation to explore AI tools and to integrate these insights into both AI-ambient documentation tools and AI training curriculum for medical trainees.

Speaker:
Yawen Guo, MISM
University of California - Irvine

Authors:
Di Hu, Master of Science in Information Systems - University of California - Irvine; Yawen Guo, MISM - University of California - Irvine; Shirin Salehi, BS - University of California, Irvine, School of Medicine; Kai Zheng, PhD - University of California, Irvine; Emilie Chow, MD - University of California, Irvine;

Improving Provider Documentation Using Pediatric AutoDx

Poster Number: P26

2025 Clinical Informatics Conference 25x5 Presentation

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: Workflow Efficiency, EHR Implementation and Optimization, Change Management
Primary Track: Documentation Burden, Clinician Well-Being and Care Outcomes
Programmatic Theme: Usability, Efficiency, and User Experience

Provider documentation is essential to hospital operations and clinical communication. Coding queries account for a non-insignificant amount of provider time and effort, increasing provider frustration and dissatisfaction with practice. We aimed to decrease applicable coding queries by 50% and improve ease of practice for 25% of pediatric inpatient providers in Comer Children’s Hospital by June 30, 2024. We adapted the existing AutoDx tool used at our institution for adults to meet the unique needs of the pediatric population. This involved creating pediatric-specific rules, optimizing lookback windows, and accounting for documentation variations. The system was then deployed to inpatient pediatric primary services. The impact of the intervention was assessed by surveying providers before and after implementation, measuring coding queries before and after implementation, and conducting PDSA cycles to improve usage rates of the tool. Pediatric AutoDx implementation resulted in a 58% decrease in coding queries for the relevant conditions. The most significant gains were in rules evaluating laboratory results, such as hemoglobin and electrolyte values. Nearly half of providers agreed that AutoDx improved ease of practice. AutoDx usage steadily increased and then leveled off during the intervention period, suggesting consistent adoption and efficacy of multiple PDSA cycles to improve usage.

Speaker:
Kevin Smith, MD
University of Chicago Medicine

Authors:
Riley Boland, MD - University of Chicago; Matthew Cerasale, MD, MPH - University of Chicago; Cheng-Kai Kao, MD - University of Chicago;

Communicating Genetic Test Results to Clinicians and Patients

Poster Number: P27

2025 Clinical Informatics Conference 25x5 Presentation

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: EHR Implementation and Optimization, Precision Health and Genomics, Bridging Analytics, Bedside Care, Clinical Documentation, and Education
Primary Track: Documentation Burden, Clinician Well-Being and Care Outcomes
Programmatic Theme: Usability, Efficiency, and User Experience

Communicating genetic test results to clinicians and patients via EHR presents challenges. Conveying the complexity of “negative” and “positive” test results for conditions with varied inheritance patterns and risk profiles requires different approaches in the absence of a standardized reporting system. We convened a multidisciplinary panel to construct genomic indicators for three Tier 1 conditions. We present our model based on a well-considered approach for clinicians and patients to better understand the significance of results.

Speaker:
Sophie Cameron, MBChB, MS
Mayo Clinic

Authors:
Pedro Caraballo, MD - Mayo Clinic; Joseph Sutton, MS (CIS) - Mayo Clinic; Jennifer Kemppainen, MS, CGC - Mayo Clinic; Sophie Cameron, MBChB, MS - Mayo Clinic;

Creation of a Novel, Multicenter Neonatal Clinical Decision Support Rounding Tool

Poster Number: P28

2025 Clinical Informatics Conference 25x5 Presentation

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: EHR Implementation and Optimization, Patient Safety, Adaptive Clinical Decision Support, Usability and Measuring User Experience, Workflow Efficiency, Quality Measures and eCQMs / Quality Improvement
Primary Track: Documentation Burden, Clinician Well-Being and Care Outcomes
Programmatic Theme: Clinical Decision Support and Analytics

There is a well-established body of research associating checklists with improved safety and clinical outcomes. In response to provider feedback, our team was able to develop an integrated EHR clinical decision support rounding checklist tool that supports quality improvement and patient safety across four Penn Medicine neonatal intensive care units. Our work represents the first multicenter application of this framework to a more generalizable neonatal care population.

Speaker:
Alex Ruan, MD
Children's Hospital of Philadelphia

Authors:
Osvaldo Mercado, MD - Children's Hospital of Philadelphia; Deborah Welsh, BSN, RNC-OB - University of Pennsylvania Health System; Leah Carr, MD - Children's Hospital of Philadelphia;

A Patient-Driven versus Staff-Driven Registration Process for Noona in the Oncology Setting

Poster Number: P29

2025 Clinical Informatics Conference 25x5 Presentation

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: Documentation Burden, Clinician Burnout, Consumer and Patient Engagement, Innovation in Digital Care
Primary Track: Documentation Burden, Clinician Well-Being and Care Outcomes
Programmatic Theme: Usability, Efficiency, and User Experience

Noona is a patient facing app that is used by our organization to connect patients to their electronic health record and communicate with their care team. Registration is done by front-line staff but with competing priorities in the clinic setting, registration rates have remained low regardless of the organizational campaign efforts. Creating an alternative solution for patients to register became a required necessity.

Speaker:
Barbara Kitzan, Nurse
CancerCare Manitoba

Authors:
Kathleen Decker, PhD - CancerCare Manitoba; Marshall Pitz, MD. MPH - CancerCare Manitoba; Pascal Lambert, MSc - CancerCare Manitoba;

Multi-Disciplinary Workflow for Diabetic Retinopathy Screenings and Referrals: A Simplified Approach

Poster Number: P30

2025 Clinical Informatics Conference 25x5 Presentation

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: Usability and Measuring User Experience, Workflow Efficiency, EHR Implementation and Optimization
Primary Track: Documentation Burden, Clinician Well-Being and Care Outcomes
Programmatic Theme: Usability, Efficiency, and User Experience

Informatics tools can be used to enhance diabetic retinopathy (DR) screenings in primary care settings, as we implemented a simplified screening interpretation form and clinical decision support in the electronic health record to improve provider efficiency and patient follow-up with ophthalmology. We provide preliminary data of our screening results, relaying insights into implementation and challenges in workflow integration, ultimately setting a foundation for improved screening accessibility and care coordination in primary care.

Speaker:
Kiran Malhotra, M.D.
NYU Langone Health

Authors:
Samyuktha Guttha, M.D. - NYU Langone Health; Kristin Hanselmann, B.S. - Epic; Lauren Golden, M.D. - NYU Langone Health; Elisabeth Cohen, M.D. - NYU Langone Health; Kathryn Colby, MD, PhD - NYU Langone Health; Adam Szerencsy, DO - NYU Langone Health;

Streamlining Drug Administration: The Role of Real-Time Data Integration

Poster Number: P31

2025 Clinical Informatics Conference 25x5 Presentation

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: Workflow Efficiency, Adaptive Clinical Decision Support, Data Visualization
Primary Track: Documentation Burden, Clinician Well-Being and Care Outcomes
Programmatic Theme: Usability, Efficiency, and User Experience

The integration of real-time data retrieval for laboratory results and vital signs into the In-Patient Medication Order Entry (IPMOE) system is designed to enhance medication safety and workflow efficiency in Hong Kong's public hospitals. By streamlining access to critical patient information, nurses can make informed medication decisions without navigating multiple systems. This initiative is expected to reduce application-switching time by 500 hours daily, resulting in significant annual savings and improved patient care through timely assessments during drug administration.

Speaker:
Sin Ting Leung, Bachelor of Nursing
Hospital Authority, Hong Kong

How Effective are Sepsis-Related Best Practice Advisories?: A Sabercaremetrics Approach

Poster Number: P32

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: Adaptive Clinical Decision Support, Bridging Analytics, Bedside Care, Clinical Documentation, and Education, Communication Strategies
Primary Track: Documentation Burden, Clinician Well-Being and Care Outcomes
Programmatic Theme: Clinical Decision Support and Analytics

This study evaluates the impact of sepsis best practice advisories (BPAs) on the actions of providers caring for patients on hospital wards and step-down units. Our results support the use of general EHR-based clinical activity data following BPA firing over the use of in-BPA action data to evaluate the effectiveness of BPAs.

Speaker:
Colleen Flanagan, MD
UCSF

Enhancing Pediatric Headache Management in Primary Care Through a Dynamic EHR-Integrated Clinical Decision Support Tool

Poster Number: P33

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: Adaptive Clinical Decision Support, EHR Implementation and Optimization, Workflow Efficiency, Usability and Measuring User Experience, Documentation Burden, Quality Measures and eCQMs / Quality Improvement
Primary Track: Documentation Burden, Clinician Well-Being and Care Outcomes
Programmatic Theme: Clinical Decision Support and Analytics

In CHOP’s Primary Care Network, an EHR-integrated clinical decision support tool was developed to guide providers in managing pediatric headaches and minimizing unnecessary referrals. This tool builds off of a text macro to be more interactive, dynamic, and easily integrated into provider workflow. Following pilot testing, this project will discuss the CDS development and outcomes including referral patterns, patient outcomes and provider satisfaction.

Speaker:
Loukya Kanakamedala, DO
CHOP

Authors:
Loukya Kanakamedala, DO - CHOP; Alex Ruan, MD - Children's Hospital of Philadelphia; Greg Lawton, MD - Children's Hospital of Philadelphia;

Engaging Our Clinical Nurses: Fundamental Feedback Process Driving Groundbreaking, Technology-Driven EHR Solutions

Poster Number: P34

2025 Clinical Informatics Conference 25x5 Presentation

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: Documentation Burden, Innovation in Digital Care, Clinician Burnout, Workflow Efficiency, Artificial Intelligence/Machine Learning
Primary Track: Documentation Burden, Clinician Well-Being and Care Outcomes
Programmatic Theme: Usability, Efficiency, and User Experience

Literature links EHR usability with clinician burden and burnout, with notable effects on nursing. In a nationwide survey of 146 inpatient registered nurses, "human-computer interaction" emerged as the primary area for EHR improvement. Findings will inform AI-driven strategies to enhance EHR usability and reduce documentation workload. Effective EHR solutions necessitate direct nursing input to ensure alignment with clinical priorities— a goal that non-nurse collaborators alone cannot achieve. Ongoing feedback will further refine these solutions.

Speaker:
Lisa Johnson
University of Florida College of Nursing

Authors:
Lisa Johnson - University of Florida College of Nursing; Tamara Macieira, PhD, RN - University of Florida; Olatunde Madandola, MPH, RN - University of Florida, College of Nursing; Karen Priola, MSCIS - University of Florida College of Nursing; Gail Keenan, PhD, RN, FAAN - University of Florida, College of Nursing;

Leveraging large language models for complex care planning in patients with multiple chronic conditions

Poster Number: P35

2025 Clinical Informatics Conference 25x5 Presentation

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: Artificial Intelligence/Machine Learning, Documentation Burden, Patient Safety, Clinician Burnout
Primary Track: Documentation Burden, Clinician Well-Being and Care Outcomes
Programmatic Theme: Informatics-Driven Value-Based Healthcare

Care planning for patients with multiple chronic conditions (MCCs) is data-intensive, time-consuming, and requires experienced care teams. Due to these complexities, discrepancies in care plans during transitions can occur and result in medical errors, patient harm, and clinician burnout. Large Language Models (LLMs) can potentially alleviate this burden by generating person-centered care summaries. We conducted a comparison study between physician participants and a LLM to explore the ability of a LLM to generate accurate clinical summaries for recently discharged patients using the MIMIC-III database.

Six physicians from OHSU created synopses of 12–16 discharge summaries each. The LLM generated synopses for the same summaries. Both physicians and the LLM were prompted to generate summaries following the DIKW framework, which prompts to identify key data, contextualize information, integrate evidence-based guidelines, and incorporate social and behavioral needs. Five physicians then blindly rated all synopses based on four criteria: Accuracy, Synthesis, Succinctness, and Usefulness using a 5-point Likert scale.

Results showed physicians outperformed the LLM in areas of accuracy of knowledge synopses, synthesis of data, succinctness of information, knowledge, and wisdom, and the usefulness of data and information. Conversely, the LLM demonstrated higher performance in the accuracy, synthesis, and usefulness of wisdom, and succinctness of data.

Our results suggest the LLM has strengths in generating accurate, harmonized, and useful wisdom synopses. This suggests that LLMs could aid in producing person-centered care plans, helping to reduce the burden of complex care planning on clinicians.

Speaker:
Emma Young, MS
Oregon Health & Science University

Authors:
Jean Sabile, MD - Oregon Health & Science University; LeAnn Michaels, BS - Oregon Health & Science University; Nicole Weiskopf, PhD - Oregon Health & Science University; Steven Bedrick - Oregon Health & Science University; David Dorr, MD, MS, FACMI, FAMIA, FIAHSI - Oregon Health & Science University;

Decision Support Tools Decrease Heart Failure Exacerbation Admissions

Poster Number: P36

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: Bridging Analytics, Bedside Care, Clinical Documentation, and Education, Clinical Process Automation, Quality Measures and eCQMs / Quality Improvement
Primary Track: Documentation Burden, Clinician Well-Being and Care Outcomes
Programmatic Theme: Clinical Decision Support and Analytics

We deployed and trained emergency physicians to use calculators for the Emergency Heart Failure Mortality Risk Grade tool and the Diuretic Optimization Strategies Evaluation using a third-party package from Terumo ePrism™ in our electronic health record. It included partial automatic import of clinical data into the form for ease of use. Monthly baseline admission rates decreased from 91.5% (95%CI: 85.0-97.9%)(n=305) to 84.3% (p<0.001)(n=32).

Speaker:
Jeffrey Nielson, MD
Kettering Health and Ohio University

Authors:
Alex Mueller, MD - Kettering Health; Margaret Soulen, MD - Kettering Health; Jeffrey Nielson, MD, MS - Kettering Health; Josephine Randazzo, DO - Kettering Health; Roy Johnson, MD - Kettering Health;

A Provider EHR Efficiency Training to Reduce Burnout and Turnover

Poster Number: P37

2025 Clinical Informatics Conference 25x5 Presentation

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: EHR Implementation and Optimization, Clinician Burnout, Building Value for Informatics via Education and Training, Workflow Efficiency, Documentation Burden
Primary Track: Documentation Burden, Clinician Well-Being and Care Outcomes
Programmatic Theme: Usability, Efficiency, and User Experience

Developed in 2019, the Super Thrive program is a 3-day off-site EHR efficiency training for ambulatory providers. Our data indicates that Super Thrive is successful in meeting our providers’ EHR educational and efficiency needs. Additionally, we have noted a decrease in attendee turnover rates compared to organizational rates and an anticipated reduction in provider burnout levels. The calculated cost savings due to the Super Thrive program is significant for our organization.

Speaker:
Gillian Piro, MD
Legacy Health

Authors:
Samantha Fogel, BS in Criminal Justice - Legacy Health; Kelley Aurand - Legacy Health;

The AI-Generated Documentation in Healthcare: A Pilot Study on Physician Well-being and Productivity.

Poster Number: P38

2025 Clinical Informatics Conference 25x5 Presentation

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: Workflow Efficiency, Artificial Intelligence/Machine Learning, Clinician Burnout
Working Group: AMIA Clinical Informatics Fellows (ACIF) Working Group
Primary Track: Documentation Burden, Clinician Well-Being and Care Outcomes
Programmatic Theme: Usability, Efficiency, and User Experience

This poster examines the impact of AI-generated documentation on practitioners' efficiency and satisfaction. We surveyed practitioners to assess readiness for adoption and decided on licensing needs.
After 90 days of implementation, we conducted a pilot study to evaluate satisfaction levels; we also used signals to track efficiency changes before and after adoption, focusing on patient numbers, time on the system, and pajamas' time. We aim to offer insights into AI-EHR’s potential benefits in clinical settings.

Speaker:
Mohammad Mohammad, Dr
Waco Family Medicine

Author:
Kelly Nieves, MD - Waco Family Medicine;

Characterization of Abnormal Involuntary Movement Scale (AIMS) Utilization for the Screening of Tardive Dyskinesia and Risk Modeling in Veterans with Severe Mental Illness

Poster Number: P39

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: Artificial Intelligence/Machine Learning, Precision Health and Genomics, Data Science
Primary Track: AI and Care Outcomes
Programmatic Theme: Clinical Decision Support and Analytics

Objective: This study aims to enable clinicians to leverage patients' antipsychotic prescriptions, demographic, and screening records to predict the risk of developing abnormal involuntary movements (AIMS) using electronic health records (EHR).

Description: Tardive dyskinesia (TD) is a debilitating movement disorder caused by long-term antipsychotic use, severely affecting patients' quality of life. The Abnormal Involuntary Movement Scale (AIMS) is the standard tool for TD surveillance in antipsychotic-treated patients. Despite known risk factors such as antipsychotic burden, age, and comorbidities, personalized TD risk prediction remains a challenge.

Methods: The study analyzed EHR data from the VA Cooperative Studies Program #572, covering nearly 20 years and including 7089 veterans with schizophrenia or bipolar disorder on antipsychotics. A random forest model was trained on data from 3974 patients with complete records, using chlorpromazine equivalents, demographic data, and diagnosis information. The model's performance was validated using a 10-fold cross-validation and tested on a separate dataset.

Results: The model achieved a precision-recall AUC of 0.755 and a ROC AUC of 0.726, demonstrating moderate predictive performance.

Conclusion: This study highlights the potential of EHR-based predictive models to identify high-risk TD patients, aiding in early intervention and improved patient outcomes. Future work will integrate genetic data to enhanced prediction accuracy.

Speaker:
Conner Polet, MD MBA
NYU Langone Health

Authors:
Craig Tenner, MD - VA New York Harbor Healthcare System / NYU School of Medicine; Jay Pendse, MD, PhD - VHA/NYU; Tim Bigdeli, PhD - VA New York Harbor Health System;

Using On-Demand Reporting to Enhance Timely Identification of Pediatric Acute Pancreatitis Patients for a Quality Improvement Initiative

Poster Number: P40

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: Quality Measures and eCQMs / Quality Improvement, Bridging Analytics, Bedside Care, Clinical Documentation, and Education, Population Health
Primary Track: Documentation Burden, Clinician Well-Being and Care Outcomes
Programmatic Theme: Informatics-Driven Value-Based Healthcare

Acute Pancreatitis (AP) is rare in children but increasing in incidence with risks for complications like diabetes and recurrent pancreatitis. Timely identification and follow-up with Pediatric Gastroenterologist is often challenging due to short hospital stays and care fragmentation. Our Quality Improvement initiative utilized EHR tools to achieve on-demand and timely identification of AP patients before discharge. This increased GI team involvement from 68% to 84%, improving patient follow-up and monitoring.

Speaker:
John Grisham, MD
Nationwide Children's Hospital

Authors:
John Grisham, MD - Nationwide Children's Hospital; Jennifer Lee, MD - Nationwide Children's Hospital;

From Collection to Acknowledgement: Closed-loop Specimen Tracking Solution to Promote Healthcare Operational Excellence and Sustainability

Poster Number: P41

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: Patient Safety, Workflow Efficiency, Interoperability, Documentation Burden
Primary Track: Documentation Burden, Clinician Well-Being and Care Outcomes
Programmatic Theme: Organizational Challenges

Secure pre-laboratory specimen delivery is crucial to ensure patient safety and patient outcome. To prevent irretrievable specimen loss, a closed-loop specimen tracking IT solution was developed and piloted in the Hong Kong Hospital Authority. The product streamlined the workflow from specimen collection to laboratory acknowledgement. It also showed high effectiveness on improving working efficiency, enhancing patient safety and maintaining environmental sustainability.

Speaker:
Wing Tung Ho, Bachelor of Nursing
Hong Kong Hospital Authority

Coordn8: A Transformative Tool to Optimizing External Medical Record Processing

Poster Number: P42

2025 Clinical Informatics Conference 25x5 Presentation

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: Clinical Process Automation, Documentation Burden, Artificial Intelligence/Machine Learning
Primary Track: Documentation Burden, Clinician Well-Being and Care Outcomes
Programmatic Theme: Emerging Technology and Technical Infrastructure

Acquiring and processing medical records is a challenge in healthcare, often delaying patient care. Staff face manual, error-prone tasks, such as obtaining patient consent and processing faxed records. The Center for Health Care Transformation at Penn Medicine partnered with ambulatory practices to address inefficiencies. The result was coordn8, a platform using automation, OCR, and NLP to streamline consent and record processing, improving efficiency and staff satisfaction.

Speaker:
Jency Daniel, DNP, MSN, RN
Penn Medicine

Authors:
Bhavana Kunisetty, BA - Penn Medicine Center for Health Care Transformation and Innovation; Jency Daniel, MSN, DNP - Penn Medicine Center for Health Care Transformation and Innovation; Yevgeniy Gitelman, MD - Penn Medicine Center for Health Care Transformation and Innovation; Gideon Whitehead, MS - Penn Medicine Center for Health Care Transformation and Innovation; Laura Kavanaugh, MHCI - Penn Medicine Center for Health Care Transformation and Innovation; Steve Honeywell, MBA - Penn Medicine Center for Health Care Transformation and Innovation; Lauren Hahn, MBA - Penn Medicine Center for Health Care Transformation and Innovation; Kathleen Lee, MD - Penn Medicine Center for Health Care Transformation and Innovation; Emeka Anyanwu, MD, MScBMI - Penn Medicine Center for Health Care Transformation and Innovation;

Data Visualization of Oral Anticoagulation Trends in Atrial Fibrillation Patients

Poster Number: P43

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: Data Visualization, Big Data, Medication Adherence
Primary Track: Documentation Burden, Clinician Well-Being and Care Outcomes
Programmatic Theme: Clinical Decision Support and Analytics

Clinical practice guidelines changed from prescribing Warfarin to direct oral anticoagulation (DOAC) for atrial fibrillation patients. To evaluate whether prescribing has followed guidelines, we identified Warfarin and DOAC use from prescription claims for members with atrial fibrillation from 2010-2022. Data showed that any OAC use increased, with Warfarin decreasing and DOAC use increasing. DOAC use outpaced Warfarin after 2017. There was variation in when DOACs were preferentially prescribed, and overall OAC use in patient subgroups.

Speaker:
Shannon Richards, MSN
Highmark Health

Authors:
Shannon Richards, MSN - Highmark Health; Charlotte Drury-Gworek, MSN - Highmark Health; Jillian Rung, PhD - Highmark Health; Brent Williams, PhD - Allegheny Health Network;

Association between Health Equity Scoring and Electronic Health Record Enhancement Deployment Speed

Poster Number: P45

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: EHR Implementation and Optimization, Driving Digital Equity, Workflow Efficiency, EHR Implementation and Optimization
Primary Track: Innovation in Driving Digital Equity and Social Care
Programmatic Theme: Clinical Decision Support and Analytics

Background
In a rapidly evolving clinical and operational environment, the volume of Electronic Health Record (EHR) enhancement requests often exceeds capacity, necessitating scoring for prioritization. At UCSF, Physician Informaticists evaluate enhancement requests using an overall impact score (OIS) and a Health Equity Impact Score (HEIS). It is unknown if these scores are associated with faster enhancement deployment.

Objectives
To quantify the effect of the OIS and HEIS on EHR enhancement deployment speed.

Methods
At UCSF, Physician Informaticists score EHR enhancement requests for impact on outcomes, efficiency, feasibility, and operations. Starting 2019, though not directly incorporated into the OIS, HEIS is used to subjectively categorize these requests based on health equity impact. We studied the association between these scores and the time from initiation of a request to deployment.

Results
Between October 2018 and September 2024, 318 EHR enhancement requests were scored for overall impact. After HEIS was introduced, 209 requests were additionally scored as Beneficial or Neutral. Most Beneficial requests were made by the Emergency Department and aimed to improve the care for vulnerable or socially disadvantaged populations.
The average time from request to deployment was 305.9 days. Higher OIS was associated with a modest but significantly decreased time (2.5 days per increment in score) to deployment. There was no association between HEIS and the time to EHR enhancement deployment.

Conclusions
Though higher OIS is associated with quicker EHR enhancement deployment, HEIS is not. Prioritizing requests that advance health equity may require strategies beyond this standalone score.

Speaker:
Sristi Sharma, M.D., M.P.H.
UCSF

Authors:
Anoop Muniyappa, MD, MS - UCSF; Susan Chim, DHA, MAT - University of CA, San Francisco; Christy Sedore, BS - UCSF; Michael Lang, MD, MPH - UCSF Health; Raman Khanna, MD, MAS - University of California, San Francisco;

Standardizing Pre-exposure Prophylaxis for HIV (PrEP) Management in Primary Care

Poster Number: P46

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: Workflow Efficiency, Driving Digital Equity, Driving Digital Equity, Clinician Burnout, Usability and Measuring User Experience
Primary Track: Innovation in Driving Digital Equity and Social Care
Programmatic Theme: Clinical Decision Support and Analytics

This implementation study explores barriers to guideline-directed PrEP management in primary care and the implementation of EHR-based clinical decision support (CDS) tools. At a large academic health center, redesigned workflows—including alerts, an updated order panel, and a patient-facing tracker—enhanced PrEP prescribing efficiency and patient engagement. Results demonstrate increased tool adoption, highlighting the scalability of this approach to improve access for underserved populations via telemedicine.

Speaker:
Kevin Truong, MD, MBA
UCLA

Authors:
Hazel Oza, BS - UCLA Health IT; David Gomez, MBA - UCLA Health IT; Ashwin Buchipudi, BS - UCLA Health IT; Annapoorna Chirra, MD - UCLA Health IT; Lawrence Dardick, MD - UCLA;

Case Study on the Role of Clinical Informaticists in Ensuring Data Quality and Reusability for Shared Data Sets

Poster Number: P47

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: Health IT Standards (USCDI, FHIR®, SMART, etc.), Quality Measures and eCQMs / Quality Improvement, Coordination Across the Continuum of Care
Primary Track: Leadership and Governance
Programmatic Theme: Informatics-Driven Value-Based Healthcare

Medical researchers publish deidentified data sets and code for reproducibility, a requirement for NIH grants since 2023. Data sets, primarily created for specific studies, often lack reusability. We contrast three data sets where the involvement of clinical informaticists earlier in the data set creation process improved data quality and reusability.

Speaker:
Divya Kapoor, MD


Author:
Senthil Nachimuthu, MD, PhD, FAMIA - University of Utah School of Medicine;

Nine Steps for a Smooth and Effective EHR Transition: An AMA STEPS Forward Toolkit for Healthcare Leaders

Poster Number: P48

2025 Clinical Informatics Conference 25x5 Presentation

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: Change Management, EHR Implementation and Optimization, Clinician Burnout
Primary Track: Leadership and Governance
Programmatic Theme: Organizational Challenges

Introduction: Transitioning to a new electronic health record (EHR) is a complex and costly endeavor that impacts nearly every aspect of a healthcare organization. The shift demands a proactive, strategic approach to mitigate disruptions in patient care, support physician morale, and enhance overall system effectiveness. Most EHR transitions are guided by EHR vendors, and limited empirical data is accessible to guide healthcare leaders preparing to undergo EHR transitions. We worked with the AMA STEPS Forward initiative to develop 9 steps that support EHR transitions.

Methods: Our framework draws from prior EHR implementation studies, practical lessons from health systems undergoing transitions, and personal experiences to compile best practices for each step.

Nine Steps for a Smooth EHR Transition:
1. Create an EHR Transition Team
2. Communicate with End-Users
3. Decide on Implementation Approach
4. Understand the Current EHR State
5. Consider Customization Needs
6. Anticipate Challenges
7. Offer Training and Support
8. Mitigate Other Workplace Stressors
9. Gather Feedback and Optimize
Conclusion: This nine-step framework provides actionable guidance for organizations undertaking EHR transitions, with a focus on minimizing disruptions, fostering clinician engagement, and ensuring patient safety. By following these steps, healthcare organizations can create a foundation for successful EHR implementation and ongoing optimization.

Speaker:
Seppo Rinne, MD, PhD
VA Bedford Healthcare System

Author:
Dominis Matt, BS - American Medical Association;

Designing EHR system usability for the sleepy shift worker with high work demands? - EHR usability trajectories in association with work demands, work schedules and shift work sleep disorder in health professionals

Poster Number: P50

2025 Clinical Informatics Conference 25x5 Presentation

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: Usability and Measuring User Experience, Clinician Burnout, Documentation Burden, Disruptive and Innovative Technologies
Primary Track: Documentation Burden, Clinician Well-Being and Care Outcomes
Programmatic Theme: Usability, Efficiency, and User Experience

Healthcare needs supportive technology. The aim of this study is to assess how health professionals (n = 4160) at a Scandinavian hospital evaluated the EHR usability before and after adoption of an American EHR system, and to describe the association with work-related factors. Preliminary analyses show a significant drop in EHR usability from the old to the new system that stays low 6 and 14 months after Go-live, irrespective of work demands. Analyses on work schedules and shift work sleep disorder will be performed.

Speaker:
Signe Lohmann-Lafrenz, MD
ISM, NTNU, Trondheim, Norway

Automated Derivation of Complication after Pediatric Cardiothoracic Surgery from Physician Notes

Poster Number: P51

2025 Clinical Informatics Conference 25x5 Presentation

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: Artificial Intelligence/Machine Learning, Documentation Burden, Data Science
Primary Track: AI and Care Outcomes
Programmatic Theme: Clinical Decision Support and Analytics

Identifying surgical complications efficiently is critical for monitoring pediatric cardiovascular surgery outcomes. This study built an Artificial Intelligence/Machine Learning-based model to automate the complication extraction process and successfully captured the first targeted surgical complication “Sternum left open, Planned” from unstructured physician case notes. The model reduces resources needed to identify complications by automating data extraction.

Speaker:
Jiayu Dai, MS
University of Nebraska Omaha

Enhancing EHR Utilization and Experience

Poster Number: P52

2025 Clinical Informatics Conference 25x5 Presentation

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: EHR Implementation and Optimization, Workflow Efficiency, Usability and Measuring User Experience
Primary Track: Documentation Burden, Clinician Well-Being and Care Outcomes
Programmatic Theme: Usability, Efficiency, and User Experience

This session will discuss UTHealth Houston's (UTHH) review of EHR functionality in fall 2023, identifying underutilized opportunities and missing support components. It will cover the creation of the Clinic Operations Systems Support Liaisons team to provide at-the-elbow support, improve Epic workflows, and enhance user experience. The session will also outline the team's responsibilities, governance structure, and the goal of building relationships with department and clinic leadership to standardize best practices.

Speaker:
olasunkanmi adeyinka, MD
ut health

Authors:
olasunkanmi adeyinka, MD - ut health; Lindy Anderson-Papke, MHA - UTHealth Houston;

Evaluating Postpartum Hemorrhage: A Fault Tree Analysis Approach

Poster Number: P53

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: Bridging Analytics, Bedside Care, Clinical Documentation, and Education, Change Management, Clinical Content and IT Project Governance, Clinician Burnout, Building Value for Informatics via Education and Training
Primary Track: Documentation Burden, Clinician Well-Being and Care Outcomes
Programmatic Theme: Informatics-Driven Value-Based Healthcare

Fault Tree Analysis (FTA) is a graphical representation of system errors that contribute to adverse outcomes. This approach examines system failures without assigning blame to individuals. This systematic approach will decrease bias and provide a more objective understanding of contributing factors to adverse events. These changes in perspective will encourage behavior modifications among physicians, leading to improvements in patient safety. A Fault Tree Analysis was created to assess postpartum hemorrhage at Kettering Health.

Speaker:
Jennifer Glance, DO
Kettering Health

Author:
Jeffrey Nielson, MD, MS - Kettering Health;

In-hospital Patient Portal use in Admitted Surgical Patients

Poster Number: P54

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: 21st Century Cures (including considerations for special populations such as adolescents), Telemedicine and Telehealth including mHealth, App’s etc, Secure Communication, Innovation in Digital Care, Social Determinants of Health
Primary Track: Innovation in Driving Digital Equity and Social Care
Programmatic Theme: Emerging Technology and Technical Infrastructure

In surgical patients, in-hospital patient portal use is common and increasing, offering opportunities for improved transparency and communication. However, portal use disparities persist in age, race, insurance status, and surgical specialty.

Speaker:
Andrew Bain, MD
University of Texas Southwestern Medical Center

Clinical Validation of a Machine Learning Model Prior to Implementation: Role, Impact, and Lessons Learned

Poster Number: P55

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: Artificial Intelligence/Machine Learning, Adaptive Clinical Decision Support, Quality Measures and eCQMs / Quality Improvement
Primary Track: AI and Care Outcomes
Programmatic Theme: Clinical Decision Support and Analytics

Introduction: Palliative Care (PC) consults often rely on clinicians' initiative to identify and refer appropriate patients. To support this process, we developed a Machine Learning (ML) model that predicts six-month post-hospitalization mortality, prompting clinicians to consider PC consultations for high-risk patients, serving as a Clinical Decision Support (CDS) tool.

Method: Our LightGBM-based ML model was trained on data from Rush University Medical Center, incorporating 173 variables, including demographics, laboratory results, healthcare utilization, ICD-10 codes, and more, among patients aged 18 and older. The model calculates the six-month mortality risk (0–100%) post-hospitalization for inpatients. A two-phase silent validation pilot preceded implementation to ensure technical feasibility, operational impact, and alignment with clinical needs. Phase 1 verified the model’s technical deployment and assessed the feasibility of various risk thresholds. Phase 2 focused on clinical validation, during which three blinded reviewers evaluated the appropriateness of PC consults for flagged patients.

Results: The initial threshold of 0.3 generated excessive triggers with lower clinical concordance, prompting an adjustment to 0.41. During the updated threshold pilot, 137 patients were flagged over 23 days, resulting in a median of 6 triggers per day. Concordance rates reached 68.6% for inpatient consults and 98.5% for outpatient consults. Notably, 86.4% of triggers occurred within three days of admission.

Discussion/Conclusion: This study highlights the importance of pre-implementation validation for refining the model and ensuring operational readiness. This systematic process enhances patient-centered care by enabling the early identification of high-risk patients and facilitating timely palliative care consultations.

Speaker:
Kyunghoon Rhee, MD
Rush University Medical Center

Authors:
Kyunghoon Rhee, MD - Rush University Medical Center; Ajeet Singh, MD MPH - Rush University Medical Center; Vaishvik Chaudhari, MS - Rush University Medical Center; Mia McClintic, BS - Rush University Medical Center; Elaine Chen, MD - Rush University Medical Center; Juan C Rojas, MD, MS - Rush University Medical Center;

Human-Centered Approach to Audiology Workflow Optimization

Poster Number: P56

2025 Clinical Informatics Conference 25x5 Presentation

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: Workflow Efficiency, Usability and Measuring User Experience, Human Factors Testing
Primary Track: Documentation Burden, Clinician Well-Being and Care Outcomes
Programmatic Theme: Usability, Efficiency, and User Experience

This case study highlights a repeatable and scalable human factors workflow optimization process. We applied this process to audiology workflows within the Veterans Health Administration’s Federal EHR. Technology and non-technology interventions were identified to address workflow inefficiencies and pain points. Integral to the success of the project were steps taken to engage internal and external stakeholders as well as end users throughout the entire study process.

Speaker:
Janey Barnes, PhD
Ben Allegretti Consulting

Authors:
Jennifer Clark, AuD, MHA - Department of Veterans Affairs; Kyle Maddox, MS - Department of Veterans Affairs; Kendyl Stubleski, Master's of Health Informatics - Department of Veteran Affairs; Janey Barnes, PhD - Ben Allegretti Consulting;

Electronic Clinical Decision Support System Guided Blood Culture Stewardship in Emergency Departments: Response to the National Blood Culture Media Shortage

Poster Number: P57

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: Adaptive Clinical Decision Support, Clinical informatics organizational models, Care Delivery Models
Primary Track: Documentation Burden, Clinician Well-Being and Care Outcomes
Programmatic Theme: Clinical Decision Support and Analytics

This study examined the effect of implementation of electronic clinical decision support system (CDSS)-guided blood culture stewardship across 7 emergency departments during the national shortage of blood culture bottles. An analysis of 13,166 cultures from June to August 2024 showed a significant reduction in volume and increased diagnostic yield after CDSS implementation, with no effect on length of stay or mortality. These findings highlight the effectiveness of CDSS in optimizing resource utilization.

Speaker:
Ankit Sakhuja, MBBS, MS
Icahn School of Medicine at Mount Sinai

Hypercoagulability Testing Optimization using Clinical Decision Support

Poster Number: P58

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: Adaptive Clinical Decision Support, EHR Implementation and Optimization, Quality Measures and eCQMs / Quality Improvement
Primary Track: Documentation Burden, Clinician Well-Being and Care Outcomes
Programmatic Theme: Clinical Decision Support and Analytics

Thrombophilia testing practices exhibit significant variability across medical specialties, leading to discrepancies in the perception of tests associated with unprovoked thrombosis and their influence on clinical decision-making. Substantial expenses related to inappropriate laboratory testing can reach up to $1 million annually. Implementing clinical decision support (CDS) has shown promise in reducing unnecessary testing and alleviating patient financial burdens. At Stony Brook University Hospital, outdated and conflicting PowerPlans within the electronic health record exacerbated these challenges.

To address this, we analyzed 4,738 thrombophilia orders before and 2,290 after implementing the redesigned PowerPlans. Results demonstrated a reduction in inappropriate testing rates, from 18.8±5.6% pre-implementation to 12.2±3.4% post-implementation (p = 0.00001). The intervention is estimated to yield annual savings of $28,608, with monthly reductions in patient costs of $2,383.97.

Our findings suggest that multi-disciplinary teams can enhance the effectiveness of thrombophilia testing, optimize care, and lower costs through targeted redesign of clinical decision support tools.

Speaker:
Lyncean Ung, D.O.
Stony Brook Medicine

Authors:
Lyncean Ung, D.O. - Stony Brook Medicine; Rachel Wong, MD, MPH, MBA, MS - Stony Brook University, School of Medicine; Veena Lingam, MBBS - Moffitt Cancer Center-USF; Suguna Raju, MD - Stony Brook Medicine; Luke Li, MD - Stony Brook Medicine; Theodore Gabig, MD - Stony Brook Medicine; Lisa Senzel, MD - Stony Brook Medicine; Michael Guido, MD - Stony Brook Medicine;

The Effect of Ambient AI on Clinician Burnout at University of Iowa Health Care

Poster Number: P59

2025 Clinical Informatics Conference 25x5 Presentation

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: Ambient documentation, Clinician Burnout, Documentation Burden, Learning Health System
Primary Track: Documentation Burden, Clinician Well-Being and Care Outcomes
Programmatic Theme: Usability, Efficiency, and User Experience

- Definition of Problem or Gap
Healthcare burnout is a significant challenge. Electronic health records (EHRs) were meant to simplify but have instead increased workloads and stress, leading to higher burnout rates. EHRs contribute to information overload and work disruptions, reducing meaningful patient interactions and causing dissatisfaction. Studies show that for every hour spent with patients, physicians need two hours for electronic documentation and desk work, extending the workload beyond regular hours and increasing burnout.

- What did you use to address the problem or gap?
Ambient AI documentation software was launched at UIHC after a successful pilot in May 2024. The pilot showed a reduction in burnout from 69% to 43%. With our enterprise implementation, the Stanford Professional Fulfillment Index (PFI) measured burnout reductions by surveying users when they started using Ambient AI and again after 30 days.

- Results
Out of 576 initial PFI respondents, 284 completed 30-day follow-ups. Overall burnout decreased from 49% to 33% for AI users but remained unchanged for non-users (45% vs 47%). Scores improved in overall burnout (3.3 to 2.5), work exhaustion (4.2 to 3.4), and interpersonal disengagement (2.7 to 1.9) among AI users. The highest users of Ambient AI showed the greatest improvements.

- Discussion
Ambient AI led to significant decreases in burnout symptoms like work exhaustion and interpersonal disengagement, improving clinicians' experiences with patient care.

- Conclusions
Ambient AI at the University of Iowa Health Care reduced clinician burnout significantly, showcasing that well-implemented AI can serve as a powerful tool to address burnout in healthcare.

Speaker:
Jason Misurac, MD, MS
University of Iowa - Dept. of Pediatrics

Authors:
Jason Misurac, MD, MS - University of Iowa - Dept. of Pediatrics; James Blum, MD - University of Iowa;

Evaluating the Impact of AI Scribe Integration in Primary Care and Subspecialty Clinics: A Mixed-Methods Pilot to Address Documentation Burden and Burnout

Poster Number: P60

2025 Clinical Informatics Conference 25x5 Presentation

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: Ambient documentation, Clinician Burnout, Documentation Burden, Workflow Efficiency, Clinician Burnout, Human Factors Testing, Artificial Intelligence/Machine Learning, Usability and Measuring User Experience
Primary Track: Documentation Burden, Clinician Well-Being and Care Outcomes
Programmatic Theme: Usability, Efficiency, and User Experience

What might the attendee be able to do after being in your session?
Understand the potential of AI scribe tools to reduce documentation burden, improve workflows, and identify implementation challenges.

Description of the Problem
EHR documentation contributes to clinician burnout (1,2). AI scribes show promise for alleviating this burden (3,4), but their utility in diverse clinical settings requires further evaluation.
Methods: What did you do to address the problem or gap?
A mixed-methods pilot involving 54 clinicians in two phases assessed a non-integrated and Epic-integrated AI scribe tool using surveys, interviews, and EHR data.

Results:
Interviews showed several positive themes (Table 1). Survey results showed increased satisfaction (p=0.004) and reduced documentation time (p=0.001) (Table 2). Burnout symptoms decreased in Phase 1 (from 3/20 to 0/14 respondents). While editing challenges persisted, feedback for complex visits was positive. Satisfaction was higher in Phase 2, linked to Epic integration and tool improvements.

Discussion of Results
The tool's utility varied; it was valuable for longer encounters but raised concerns about editing time. Qualitative feedback highlighted enthusiasm, particularly with the integrated tool. Metrics for efficiency or burnout were limited due to sample size but suggest promising trends.

Conclusion
AI scribe tools show potential for reducing documentation burden, though their impact varies by provider. More research is needed to refine implementation strategies and measure outcomes across diverse clinical settings.
Attendee’s Take-away Tool
AI scribes can improve documentation workflows, particularly for complex visits, offering a promising solution for reducing clinician burnout.

Speaker:
Leopold Arko, MD, MS
University of Minnesota

Authors:
Carly Hudelson, MD MSc - University of Minnesota; Melissa Gunderson, MD - University of Minnesota; Deborah Pestka, PharmD, PhD; Tori Christiaansen, MD - Fairview Health System; Sameer Badlani, MD, FACP - M Health Fairview; Rebecca Markowitz, MD - M Health Fairview; Genevieve Melton-Meaux, MD, PhD - University of Minnesota;

Differences in Ambulatory Physician EHR Use by Telemedicine Intensity

Poster Number: P61

2025 Clinical Informatics Conference 25x5 Presentation

Presentation Time: 04:00 PM - 05:30 PM

Abstract Keywords: Telemedicine and Telehealth including mHealth, App’s etc, Care Delivery Models, Workflow Efficiency, Usability and Measuring User Experience, Documentation Burden, Data Science
Primary Track: Documentation Burden, Clinician Well-Being and Care Outcomes
Programmatic Theme: Informatics-Driven Value-Based Healthcare

This study investigates the impact of telemedicine on ambulatory physician EHR use across two academic medical centers, UCSF and WashU, in the post-COVID onset period. Using EHR audit log data, it analyzes how telemedicine affects physicians' EHR use, examining frequency-based (e.g., information seeking, patient communication) and time-based (e.g., EHR time, documentation time) measures across workdays. The findings reveal significant differences in EHR use patterns, influenced by institutional telemedicine infrastructure, highlighting the need for adaptable, site-specific telemedicine workflows.

Speaker:
A J Holmgren, PhD
University of California, San Francisco

Authors:
Julia Adler-Milstein, PhD - UCSF School of Medicine; Robert Thombley, BS - University of California, San Francisco; Sunny Lou, MD, PhD - Washington University, St. Louis; Elise Eiden, MS - Washington University in St. Louis; Thomas Kannampallil, PhD - Washington University School of Medicine; A J Holmgren, PhD - University of California, San Francisco;