Golden Rule for AI Model Inputs of Nurse-Generated Data: Do Not Assume

Presentation Type: Podium Abstract
Presentation Time: 09:45 AM - 09:57 AM

Abstract Keywords: Knowledge Representation and Information Modeling, Workflow, Artificial Intelligence
Programmatic Theme: Clinical Informatics

Safe and trustworthy AI requires moving beyond surface-level interpretations of nurse-generated data to a precept of “Do Not Assume”. We will present 12 heuristics with detailed examples to elucidate common assumptions that, if left unchecked, can undermine the validity and clinical relevance of model outputs that use data documented by nurses. We invite a robust discussion to inform a community-driven knowledge base for the AI community in creating technically robust, clinically relevant and trustworthy AI.

Speaker(s):
Sarah Rossetti, RN, PhD, FAAN, FACMI, FAMIA, FIAHSI
Columbia University Irving Medical Center
Author(s):
Sarah Rossetti, RN, PhD, FAAN, FACMI, FAMIA, FIAHSI - Columbia University Irving Medical Center; Rachel Lee, PhD, RN - Columbia University; Kenrick Cato, PhD, RN, CPHIMS, FAAN, FACMI - University of Pennsylvania/ Children's Hospital of Philadelphia; David Albers, PhD - University of Colorado, Department of Biomedical Informatics; Shalmali Joshi, PhD - Columbia University; Varsha Varkhedi, Bachelor - Columbia University; Amy Finnegan, PhD - Independent Consultant; Pinyue Vicky Wang, Bachelor's Degree - Columbia University Department of Biomedical Informatics; Carolyn Scheese, DNP, RN - Biomedical Informatics University of Utah; Victoria Tiase, PhD, RN, NI-BC, FAMIA, FAAN, FNAP, ACHIP - University of Utah;

---

Improving Prospective Weight Entry Error Detection using Deep Learning Methods

Presentation Type: Paper - Student

Click to View Presentation

Presentation Time: 09:57 AM - 10:09 AM

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

Weight entry errors can cause significant patient harm in pediatrics due to pervasive weight-based dosing practices. Existing studies primarily utilize rule-based methods, achieving an Area Under the Receiver Operating Characteristic Curve (AUROC) ranging from 0.546 to 0.620. In this study, a two-module method employing bi-directional Long Short-Term Memory (bi-LSTM) with the Multi-Head Attention Mechanism for the prospective detection of anomalous weight entries was implemented to improve prospective error detection. The proposed method consists of a predictor and a classifier module. The predictor module learns the normal pattern of weight changes, and the classifier module identifies anomalous weight entries. The performance of both modules was evaluated, exhibiting a clear superiority over other methods in distinguishing normal from anomalous data points. The proposed approach achieved an AUROC of 0.977 and a precision of 8.98% when calibrated for 90% sensitivity, close to double the previous results, significantly outperforming other methods.

Speaker(s):
Han-Nieh Shih, M.Sc.
University of Cincinnati
Author(s):
Raj Bhatnagar, Ph.D. - University of Cincinnati; Vikram Ravindra, Ph.D. - University of Cincinnati; Justin Zhan, Ph.D. - University of Cincinnati; Stephen Spooner, MD, FAAP - Cincinnati Children's Hospital; Danny Wu, PhD - University of North Carolina at Chapel Hill;

---

An Integrated Informatics Platform for Consent-Aware Clinical Phenotyping and High-Throughput Specimen Interception

Presentation Type: Podium Abstract

Click to View Presentation

Presentation Time: 10:09 AM - 10:21 AM

Abstract Keywords: Informatics Implementation, Knowledge Representation and Information Modeling, Policy, Administrative Systems, Real-World Evidence Generation, Workflow, Evaluation, Data transformation/ETL
Programmatic Theme: Clinical Research Informatics

Prospective biospecimen acquisition requires navigating serial dependencies across protocol development, consent, recruitment, collection, and data delivery. We present two informatics components implemented at UC Davis Health: the Universal Consent Registry, which captures affirmative EHR-persisted consent for deidentified remnant biospecimen and data re-use from active health-system clinical patients; and Bioexplorer, a collection lifecycle automation platform that manages investigator-defined biospecimen collection protocols through OMOP-based phenotype screening, pathology-oriented biospecimen identification and re-alliquoting, and delivers de-identified historical patient data and collected biospecimens and processing metadata to secure research environments. The platform utilizes operational clinical infrastructure for all workflows, and does not require dedicated research coordinators or other staff.

Speaker(s):
Nicholas Anderson, PhD
University of California, Davis
Author(s):
Albert Riedl, M.S. - UC Davis Health System; Cole Torvick, BS - UC Davis; Shehnaz Hussain, PhD - UC Davis; Leslie Solis, MS - UC Davis;

---

Identifying Disparities in Electronic Health Record Inbox Workload and Time Burden Among Primary Care Providers

Presentation Type: Podium Abstract

Click to View Presentation

2026 Symposium Health Equity Presentation

2026 Symposium 25x5 Presentation

Presentation Time: 10:21 AM - 10:33 AM

Abstract Keywords: Diversity, Equity, Inclusion, and Accessibility (DEIA), Evaluation, Documentation Burden
Programmatic Theme: Clinical Informatics

Electronic health record inbox burden has grown substantially, contributing significantly to provider burnout. We examined the association between daily inbox message volume and time spent among primary care providers, adjusting for provider, patient, and patient-level characteristics. Message volume was the strongest predictor of time spent (p<0.001), followed by practice type (p<0.001), and provider type (p=0.002). Findings suggest that inbox burden and time spent are largely driven by different operational workflows.

Speaker(s):
Nicole Redfern, MPH
NYU Langone Health
Author(s):
Safiya Richardson, MD, MPH - NYU Langone; Valy Fontil, Medical Doctor - NYU Langone Health; Ting-Fang Lee, PhD - NYU Langone Health; Samantha Matiychenko, BS - NYU Langone Health;

---

Who Benefits Most from AI Scribes: Results from the Multisite Ambient Clinical Documentation Collaborative

Presentation Type: Podium Abstract
2026 Symposium 25x5 Presentation

Presentation Time: 10:33 AM - 10:45 AM

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

Time spent in the electronic health record (EHR) contributes to clinician burnout, and artificial intelligence (AI) scribes are increasingly being adopted to reduce documentation burden. In this multisite study across six U.S. health systems, we evaluate the impact of AI scribe adoption on clinician EHR time, documentation workload, and productivity. Results reveal substantial variation in benefits by clinician role, specialty, documentation style, and level of AI scribe utilization.

Speaker(s):
Lisa Rotenstein, MD, MBA, MSc
UCSF
Author(s):
Lisa Rotenstein, MD MBA MSc - University of California at San Francisco (UCSF); Reema Dbouk, MD - Emory University School of Medicine; Scott MacDonald, MD - UC Davis Health; Brian Williams, MD - Yale New Haven Health; Lee Schwamm, MD - Yale School of Medicine; Sara Murray, MD, MAS - UCSF; Maria Byron, MD - University of California, San Francisco; Jacqueline You, MD, MBI - Mass General Brigham; Adam Landman, MD - Mass General Brigham; Karandeep Singh, MD, MMSc - University of California, San Diego; Ming Tai-Seale, PhD, MPH - UCSD; Katherine Lawrence, MD MPH - New York University Grossman School of Medicine; Edward Melnick, MD, MHS - Yale New Haven Health; David Bates, MD - Mass General Brigham; Harvard University; Julia Adler-Milstein, PhD, FACMI - UCSF School of Medicine; Rebecca Mishuris, MD, MS, MPH - Mass General Brigham; A J Holmgren, PhD - University of California, San Francisco;

---

Leveraging Large Language Models to Analyze Nursing Free-Text Comments to Understand What Matters to Older Adults

Presentation Type: Podium Abstract

Click to View Presentation

Presentation Time: 10:45 AM - 10:57 AM

Abstract Keywords: Large Language Models (LLMs), Clinical Decision Support, Patient Engagement and Preferences
Programmatic Theme: Clinical Informatics

Nurses routinely document older adults' goals in free-text EHR comments, yet this information remains unstructured and underutilized. We developed a What Matters taxonomy and applied a GPT-4o few-shot classification pipeline to 5,700 nursing flowsheet comments validated against 300 clinician-coded comments. Category accuracy reached 91.3% (κ=0.80); Subcategory accuracy was 77.8% (κ=0.72). Most comments were Comfort Focused (75.2%). LLMs can scalably transform unstructured What Matters documentation into structured data to support Age-Friendly care measurement.

Speaker(s):
Mollie Hobensack, PhD, RN, NI-BC
Vanderbilt University Medical Center
Author(s):
Mollie Hobensack, PhD, RN, NI-BC - Vanderbilt University Medical Center; Sean Huang, MD - Vanderbilt University; James Powers, MD - Vanderbilt; Adam Wright, PhD - Vanderbilt University Medical Center;

---