Family Context, Reproductive Experiences, and Digital Engagement in Relation to Women’s Genetic Health Knowledge, Perceptions, and Communication

Presentation Type: Paper - Student

Click to View Presentation

Presentation Time: 08:00 AM - 08:12 AM

Abstract Keywords: Human-computer Interaction, Patient-/Person-Generated Health Data, Patient Engagement and Preferences, Surveys and Needs Analysis, Quantitative Methods
Programmatic Theme: Consumer Health Informatics

Genetic factors shape many women’s health conditions across the life course, yet how women perceive genetic influences within family and digital contexts remains underexplored. We conducted a cross-sectional survey of 249 U.S. women examining associations among family structure, reproductive experiences, digital health engagement, and three perception outcomes: self-reported genetic knowledge, perceived importance of family health history, and perceived genetic impact. We found that larger family networks were associated with higher perceived importance of family history, while childbirth experience was associated with stronger perceived genetic impact. Digital engagement, particularly technology breadth and generative AI use, was consistently associated with higher perceived importance and impact. Age and family size were associated with distinct communication patterns across women’s health topics. These findings highlight the role of digital engagement and life-course context in shaping genetic health salience, with implications for informatics tools supporting family-based risk communication.

Speaker(s):
Ziqi Yang, M.S.
University of California, Irvine
Author(s):
Ziqi Yang, M.S. - University of California, Irvine; Weijun Li, M.S. - University of California, Irvine; Di Hu, Master of Science in Information Systems - University of California - Irvine; Daniel Epstein, Ph.D. - University of California, Irvine; Yunan Chen, PhD - University of California, Irvine;

---

Analyzing Liver Transplant Selection Meetings with Large Language Models: Topic and Sentiment Variations across Patient Groups

Presentation Type: Paper - Student

Click to View Presentation

2026 Symposium Health Equity Presentation

Presentation Time: 08:12 AM - 08:24 AM

Abstract Keywords: Large Language Models (LLMs), Health Equity, Information Extraction, Natural Language Processing
Programmatic Theme: Clinical Research Informatics

Prior studies suggest potential bias in liver transplant (LT) candidate selection, particularly for patients with alcohol-associated liver disease (ALD) or metabolic dysfunction-related steatotic liver disease (MASLD). Deidentified transcripts from 27 selection meetings were analyzed using gpt-oss-120b hosted in a HIPAA-compliant environment to examine variations across liver disease etiologies. For each patient discussion, we identified topics, assigned decision-weights, and assessed sentiment for the initial introduction and overall discussion. Cross-group comparisons revealed significant differences in selection practice by etiology. Discussions on patients with ALD more often emphasized insight, adherence behaviors, and addiction-related features compared to MASLD or other liver disease. While initial introduction sentiment was predominantly neutral across all etiologies, 35.8% of overall discussion sentiments were negative for patients with ALD. These findings indicate disproportionate topic weights potentially compromising objective clinical assessment in LT candidate selection. This pilot study lays the groundwork for larger-scale analysis using large language models for transcript data.

Speaker(s):
Sue Hyon Kim, MSN, RN
University of Pennsylvania
Author(s):
Sue Hyon Kim, MSN, RN - University of Pennsylvania; Danielle Mowery, PhD, MS, MS, FAMIA - University of Pennsylvania; Jacqueline Theroux, MPH - University of Pennsylvania; Abbie Jin, BS - University of Pennsylvania; Michael Lucey, MD - University of Wisconsin-Madison; Esra Alagoz, PhD - University of Wisconsin-Madison; Marina Serper, MD, MS - University of Pennsylvania;

---

When Annotators Differ In Expertise Level: Human Rater Domain Shift In Human-In-The-Loop Autism Classification

Presentation Type: Paper - Student

Click to View Presentation

Presentation Time: 08:24 AM - 08:36 AM

Abstract Keywords: Artificial Intelligence, Machine Learning, Mobile Health, Quantitative Methods
Programmatic Theme: Consumer Health Informatics

Human-in-the-loop artificial intelligence (AI) systems for autism classification use behavioral features annotated
from short home videos as inputs to machine learning models. The human raters in the loop can vary in expertise
with a tradeoff: while non-expert annotations enable scalable data collection, clinicians are the gold standard. We
study domain shift from training human-in-the-loop AI models on clinicians and evaluating them on non-clinicians,
and vice versa. Using a curated video-based behavioral screening dataset with clinicians and non-clinician ratings,
we applied exploratory factor analysis characterizing latent behavioral structure within each annotator group and
examined the alignment between clinician- and non-clinician-derived latent representations. We find that both
groups exhibit similar latent behavioral structure, but clinician-derived latent representations show stronger
within-domain diagnostic separability than non-clinician representations. These findings suggest that annotator
variability primarily manifests as differences in rating calibration rather than underlying behavioral constructs,
which can propagate into downstream predictive models.

Speaker(s):
Aditi Jaiswal, MS
University of Hawaii at Manoa
Author(s):
Aditi Jaiswal, MS - University of Hawaii at Manoa, Department of Information and Computer Sciences; Saimourya Surabhi, masters - Stanford university, Department of Pediatrics, Division of Clinical Informatics; Kaitlyn Dunlap, MRes - Stanford University, Department of Pediatrics, Division of Clinical Informatics; Mohammadmahdi Honarmand - Stanford University, Department of Mechanical Engineering; Parnian Azizian - Stanford University, Department of Mechanical Engineering; Katy Tarrit, PhD - University of Hawai'i at Manoa, Department of Information and Computer Sciences; Dennis Wall, PhD, FACMI - Stanford University, Stanford University, Department of Pediatrics, Division of Clinical Informatics and Department of Biomedical Data Science ; Peter Washington, PhD - University of California, San Francisco, Department of Medicine, Division of Clinical Informatics and Digital Transformation;

---

Validating a Deep Learning Algorithm to Identify Patients with Glaucoma using Systemic Electronic Health Records

Presentation Type: Paper - Student

Click to View Presentation

Presentation Time: 08:36 AM - 08:48 AM

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

We evaluated whether a glaucoma risk assessment (GRA) model trained on All of Us national data can identify patients at high probability of glaucoma using only systemic electronic health records (EHR) at an independent institution. In this cross‑sectional study, 20,636 Stanford patients seen from November 2013 to January 2024 were included (15% with glaucoma). A pretrained GRA model was fine‑tuned on the Stanford cohort and tested on a held‑out set using demographics, systemic diagnoses, medications, laboratory results, and physical examination measurements as inputs. The best model achieved AUROC 0.894 and PPV 0.796. Calibration was consistent with clinical risk: the highest prediction decile showed the greatest glaucoma diagnosis rate (84.3%) and treatment rate (68.2%). Performance improved with more trainable layers up to 15 and with additional data, plateauing near 80% of the cohort. An EHR‑only GRA model may enable scalable and accessible pre‑screening without specialized imaging.

Speaker(s):
John Xiang, BA
Stanford
Author(s):
John Xiang, BA - Stanford; Rohith Ravindranath, MS - Stanford; Sophia Wang, MD, MS - Stanford University;

---

Are Multimodal LLMs Ready for Clinical Dermatology? A Real-World Evaluation in Dermatology

Presentation Type: Podium Abstract
Presentation Time: 08:48 AM - 09:00 AM

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

MLLMs perform well on curated benchmarks but remain unvalidated in real-world dermatology. Evaluating four models across 5,811 hospital cases (46,405 images) demonstrated consistently poor diagnostic accuracy. Adding clinical context improved top-3 accuracy from 13.4% to 28.8%, yet expert review identified incomplete visual feature extraction — not factual errors — as the primary failure mode. Current benchmarks therefore overestimate clinical readiness. Advancing safe deployment requires targeted improvements in visual grounding and robustness to varied clinical contexts.

Speaker(s):
Morten Lee, BS
Yale School of Medicine
Author(s):
Ruoyi Jiang, MD/PhD - Yale School of Medicine, Department of Dermatology; Hyunjae Kim, Ph.D. - Yale University; Morten Lee, BS - Yale School of Medicine; Margaret MacGibeny, MD/PhD - Yale School of Medicine; Ailish Hanly, MD - Yale School of Medicine; Shanin Chowdhury, BS - Yale School of Medicine; Angela Sadlowski, BS - Yale School of Medicine; Jeffrey Gehlhausen, MD/PhD - Yale School of Medicine; Qingyu Chen, PhD - Yale University;

---

Developmental Challenges in Building a Patient-Reported Symptom-Based Risk Stratification System for Suspected Head and Neck Cancer: Early Qualitative Findings

Presentation Type: Podium Abstract

Click to View Presentation

2026 Symposium LIEAF Presentation

Presentation Time: 09:00 AM - 09:12 AM

Abstract Keywords: Governance, Human-computer Interaction, Patient Safety, Qualitative Methods, Workflow, Qualitative Methods
Programmatic Theme: Academic Informatics / LIEAF

Developing regulated digital health technologies involves organisational and regulatory challenges that are often overlooked. This qualitative case study examines stakeholder experiences developing the SYNC (Symptom iNput Clinical) system, a patient-reported symptom questionnaire supporting triage for suspected head and neck cancer. Early findings highlight governance complexity, socio-technical misalignment, and collaborative adaptations. Using Actor-Network Theory, we show how Software as a Medical Device regulations and NHS information governance policies shaped development.

Speaker(s):
CHINASA REBECCA Odo, PhD
University of Bradford
Author(s):
Chinasa Odo, PhD - University of Bradford; Nikki Rousseau, PhD - Leeds Institute of Clinical Trials Research, University of Leeds, Leeds, UK; Joanne Patterson, PhD - University of Liverpool, Liverpool, UK; Vinidh Paleri, MS, FRCS - The Royal Marsden NHS Foundation Trust, London, UK; Rebecca Randell, PhD - University of Bradford;

---