Identifying Family Relationships from Electronic Health Records: A Machine Learning Approach

Presentation Type: Podium Abstract

Click to View Presentation

Presentation Time: 03:30 PM - 03:42 PM

Abstract Keywords: Machine Learning, Interoperability and Health Information Exchange, Data Mining, Population Health, Quantitative Methods, Data transformation/ETL, Public Health, Chronic Care Management
Programmatic Theme: Clinical Research Informatics

Family health history predicts disease risk, but EHRs lack reliable family linkage. We developed five Random Forest classifiers to identify mother-child, father-child, sibling, twin, and partner relationships using only standard demographic fields. Validated on 1.5 million record pairs per model using birth certificate data from 45 million Indiana health records, models achieved F1-scores of 0.94-1.00. This framework enables population-scale hereditary disease surveillance without proprietary identifiers.

Speaker(s):
Abhinav Pundir, Master of Science
University of Colorado Denver
Author(s):
Abhinav Pundir, Master of Science - University of Colorado Denver; Andrew Hill, MS - Colorado School of Public Health, University of Colorado Denver, Denver, CO; Michael G. Kahn, PhD,MD - University of Colorado Anschutz Medical Campus, Aurora, CO; Shaun Grannis, MD, MS, FAAFP, FACMI, FAMIA - Regenstrief Institute/Indiana University; Titus Schleyer, DMD, MBA, PhD - Regenstrief Institute; Lisa Schilling, MD, MSPH - U of Colorado Anschutz; Toan Ong, PhD - University of Colorado Denver;

---

Using Large Language Models to Characterize Demographic Differences in Patient Portal Message Content in Primary Care

Presentation Type: Podium Abstract

Click to View Presentation

2026 Symposium Health Equity Presentation

Presentation Time: 03:42 PM - 03:54 PM

Abstract Keywords: Large Language Models (LLMs), Patient-/Person-Generated Health Data, Health Equity, Patient Engagement and Preferences
Programmatic Theme: Clinical Informatics

Patient portal messaging is a major channel of communication between patients and primary care teams, yet the content of these messages has been difficult to study at scale. Using a large language model, we categorized 170,250 patient-initiated message threads and examined differences across demographic groups. Message content differed significantly by race and ethnicity, insurance, age, and language, demonstrating how AI can reveal patterns in patient-generated health communication.

Speaker(s):
Melanie Molina, MD, MAS
University of California, San Francisco
Author(s):
Melanie Molina, MD, MAS - University of California, San Francisco; Sienna Kurland, MD MPH - University of California, San Francisco; Elisa Danthinne, BS - University of California San Francisco; Madhumita Sushil, PhD - UCSF; Mitchell Tang, PhD - Columbia University; Christopher Toretsky, MPH - University of California, San Francisco; Jade Christey, BA - UCSF; Robert Thombley, MS - University of California, San Francisco; Michael Barnett, MD - Harvard University; Lisa Rotenstein, MD, MBA, MSc - UCSF;

---

Characterizing Variation in Patient Provider Messaging Workflows Across Patient Groups

Presentation Type: Podium Abstract

Click to View Presentation

Presentation Time: 03:54 PM - 04:06 PM

Abstract Keywords: Telemedicine, Patient Engagement and Preferences, Fairness and elimination of bias, Administrative Systems
Programmatic Theme: Clinical Informatics

With an increased use of secure electronic messaging between patients and their primary care teams, it is critical to understand patterns of interaction downstream of an initial message and how these patterns vary by patient characteristics to assess and improve health system equity. Across a sample of 38,060 patients who exchanged one of 495,531 messages, we found significant differences in the prevalence of responses to patient messages by patient race/ethnicity and insurance.

Speaker(s):
Lisa Rotenstein, MD, MBA, MSc
UCSF
Author(s):
Lisa Rotenstein, MD, MBA, MSc - UCSF; Melanie Molina, MD, MAS - University of California, San Francisco; Sienna Kurland, MD - University of California, San Francisco; Mitchell Tang, PhD - Columbia University Mailman School of Public Health; Christopher Toretsky, MPH - University of California, San Francisco; Jade Christey, BA - UCSF; Robert Thombley, MS - University of California, San Francisco; Michael Barnett, MD - Harvard University;

---

The Role of Patient Portal Messaging to Coordinate and Receive Care for Older Adults with Dementia

Presentation Type: Podium Abstract
Presentation Time: 04:06 PM - 04:18 PM

Abstract Keywords: Patient Engagement and Preferences, Natural Language Processing, Transitions of Care
Programmatic Theme: Clinical Informatics

Patient portal messaging is widely used to coordinate and deliver care, yet little is known about how it is used by patients with dementia and their care partners. Using a validated large language model pipeline, we analyzed messages from patients with dementia and matched controls to classify information needs and authorship. Messages most often concerned medications, symptom management, and scheduling, and were primarily initiated by care partners, highlighting their central role in coordinating dementia care.

Speaker(s):
Bryan Steitz, PhD
Vanderbilt University Medical Center
Author(s):
Emily Morrow, PhD - Vanderbilt University Medical Center; Leslie Gaynor, PhD - Vanderbilt University Medical Center; Raymond Romano, PhD, MPH - Vanderbilt University Medical Center; Adam Wright, PhD - Vanderbilt University Medical Center;

---

Comparing Human and Large Language Model Responses to Patients’ Online Questions: Towards Multi-dimensional Patient-centered Support

Presentation Type: Paper - Student

Click to View Presentation

Presentation Time: 04:18 PM - 04:30 PM

Abstract Keywords: Large Language Models (LLMs), Natural Language Processing, Artificial Intelligence, Patient-/Person-Generated Health Data, Health Equity, Evaluation, Qualitative Methods, Quantitative Methods
Working Group: Consumer Health Informatics Working Group
Programmatic Theme: Consumer Health Informatics

Patients and caregivers seek informational and emotional support throughout medical care, especially when interpreting unfamiliar laboratory test results. Although resources such as patient portals and online health communities (OHCs) help address questions, gaps remain. The emergence of large language models (LLMs) offers the potential to be a complementary source of support to assist patients and caregivers in understanding and using their test results. The objective of our study is to empirically compare LLM responses to patients’ online questions containing their laboratory test results to responses written by peers in an OHC. We compared the 519 peer replies to 122 laboratory test-related posts from an OHC to 488 responses generated from four LLMs using mixed computational and qualitative methods. LLMs frequently provided clear explanations of medical terminology and structured interpretations of numeric results but were longer and less readable. Peers offered more personalized, context-specific emotional support. Overall, LLMs have the potential to complement peer responses in OHC, but require greater emotional depth, reasoning transparency, and alignment with community norms.

Speaker(s):
Md Alomgeer Hussein, MS
Unversity of Maryland Baltimore County
Author(s):
Md Alomgeer Hussein, MS - Unversity of Maryland Baltimore County; Rajmi Doshi, MS - University of Maryland Baltimore County; Lu He, PhD - University of Wisconsin-Milwaukee; Tera Reynolds, PhD, MPH, MS - University of Maryland Baltimore County;

---

Health Information Breadth and Self-Medication: Evidence from a National Consumer Survey

Presentation Type: Podium Abstract

Click to View Presentation

Presentation Time: 04:30 PM - 04:42 PM

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

Self-medication with over-the-counter (OTC) medicines and herbal supplements is increasingly common, yet the role of health information seeking in these behaviors remains unclear. Using data from the 2021 National Consumer Survey on the Medication Experience & Pharmacists’ Roles (n=1,521), we examined whether the breadth of health information sources predicts daily self-medication. Each additional information source significantly increased the probability of herbal/supplement and OTC use. These findings suggest broader access to information influences consumer self-care behaviors.

Speaker(s):
Nurudeen Showole, Health Informatics and Bioinformatics
Grand Valley State University
Author(s):
Suhila Sawesi, PhD - GVSU;

---