Developing a preliminary research agenda for climate, health and informatics

Poster Number: P01
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Personal Health Informatics, User-centered Design Methods, Human-computer Interaction, Usability, Surveys and Needs Analysis, Behavioral Change, Mobile Health, Self-care/Management/Monitoring
Working Group: Climate, Health and Informatics Working Group
Primary Track: Applications
Programmatic Theme: Academic Informatics / LIEAF

Climate change is an escalating global health crisis, with human-induced activities causing extreme weather, rising sea levels, and temperature fluctuations. These environmental changes heighten health risks, such as heat-related illnesses and poor air quality, disproportionately impacting vulnerable groups, including children, the elderly, and those with chronic conditions or without homes. In response, the field of biomedical informatics emerges as a pivotal force in assessing and addressing the impacts of climate change on health through the strategic application of data and health-related knowledge. The American Medical Informatics Association (AMIA) convened a mini-summit during the 2023 Annual Symposium in New Orleans, focusing on identifying critical research areas for the informatics community to aid healthcare in both mitigating and adapting to climate change. Employing an affinity diagramming method under IRB Protocol #21183, the summit engaged 50 experts from the technology, climate change, population health, and public policy sectors to collaboratively explore and synthesize diverse perspectives on this urgent issue. The findings, summarized in this poster, underscore the significant potential of informatics to tackle the health challenges presented by climate change. This research agenda advocates for a proactive and evidence-based approach, emphasizing the crucial role of biomedical informatics in fostering a healthcare system that is not only resilient to climate change but also actively contributes to its mitigation. The outcomes of this summit highlight the imperative for interdisciplinary collaboration and innovation in developing strategies that address the far-reaching health implications of climate change.

Speaker(s):
Beenish Chaudhry, PhD
University of Louisiana at Lafayette

Author(s):
Titus Schleyer, DMD, PhD - Regenstrief Institute; Manijeh Berenji; Michael Zaroukian, MD, PhD, MACP, FHIMSS, ABPM-CI - Self-employed; Suzanne Tamang, PhD - Stanford University; Chethan Sarabu, MD - Stanford Medicine;

Aligning Health Information Management and Informatics Cybersecurity Training with Healthcare Advancements

Poster Number: P02
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Education and Training, Privacy and Security, Mobile Health
Primary Track: Foundations
Programmatic Theme: Academic Informatics / LIEAF

Ensuring up-to-date cybersecurity education for graduates from Health Information Management (HIM) and Health Informatics (HI) programs plays an important role for them to perform their roles effectively and safeguard healthcare systems. This work looks at the cybersecurity curriculum in a typical HIM program and analyze the limitations. We identify that gaps persist, especially concerning emerging technologies like mHealth and genetic testing. We describe our hybrid approach comprising of asynchronous viewing of short videos and in-person discussions designed to incorporate timely security education into existing curriculum.

Speaker(s):
Huanmei Wu, FAMIA, PhD
Temple University

Author(s):
Kesa Bond, Ph.D. - Temple University; Bari Dzomba, Ph.D. - Temple University; Chiu Tan, PhD - Temple University;

Topic Analysis of the Global Clinical Trials using Large Language Model

Poster Number: P03
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Natural Language Processing, Information Retrieval, Information Visualization
Primary Track: Applications
Programmatic Theme: Academic Informatics / LIEAF

Clinical trials have long been a cornerstone in advancing medical knowledge, driving innovation, and patient care worldwide. The landscape of clinical trials has evolved significantly, witnessing an exponential increase in the volume and diversity of studies. Understanding the topics and underlying patterns within this vast clinical trial data is crucial for various stakeholders. Traditionally, studies in this domain have focused on leveraging domain-specific categorizations or statistical metrics, such as nationality or geographical distribution, to analyze the trial data. While these approaches have provided valuable insights into specific therapeutic areas of clinical research, the broader landscape of diverse clinical trial topics is less studied. In this ongoing work, we propose a Large Language Model (LLM)-based approach for conducting topic analysis of global clinical trials by leveraging the semantic embeddings to capture the underlying topics and relationships within the entire clinical trial data.

Speaker(s):
Zhiyuan Cao, B.Eng.
Yale University

Author(s):
Zhiyuan Cao, B.Eng. - Yale University; Qinhan Hu; Dingwei Zhan; Siyan (Amy) Guo; Huan He, Ph.D. - Yale University; Hua Xu, Ph.D - Yale University;

Artificial Intelligence in Research Administration: A Review and Standardized Rating of Generative AI Applications for Research Protocols

Poster Number: P04
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Information Retrieval, Usability, Workflow
Primary Track: Applications
Programmatic Theme: Academic Informatics / LIEAF

This study explores the potential of Generative Artificial Intelligence (Gen AI) in health informatics, specifically focusing on its application for generating Institutional Review Board (IRB) research documents. Through a systematic review of custom GPT models and collaboration with IRB protocol authors, we identified limitations and proposed an ideal workflow. Results indicate the need for improved AI tools to alleviate administrative burdens in research, with insights into platform comparisons between OpenAI and Microsoft.

Speaker(s):
JaeEun Kwon, Master of Public Policy
NYU Langone Health

Author(s):
JaeEun Kwon, Master of Public Policy - NYU Langone Health; Katerina Andreadis, MS - NYU Grossman School of Medicine; Devin Mann, MD - NYU Grossman School of Medicine;

Utilizing large Language Models (LLM) to Optimize Domain-Specific Natural Language Processing (NLP) for Identifying Patients with No Reason for Not Prescribing ACEI/ARB in Chronic Kidney Disease (CKD) Management

Poster Number: P05
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Natural Language Processing, Large Language Models (LLMs), Information Extraction
Primary Track: Applications
Programmatic Theme: Academic Informatics / LIEAF

This study developed a framework utilizing nurse-annotated examples and LLM-generated training data to improve NLP models for identifying reasons for not prescribing ACEI/ARB in CKD patients. Combining nurse-annotated and LLM-generated data resulted in the best model performance, addressing the challenge of data annotation in healthcare NLP applications. This approach is important for developing accurate, efficient, and small NLP models tailored to specific healthcare domains while reducing the burden of manual data annotation.

Speaker(s):
Mohammed Al-Garadi, PhD
VUMC

Author(s):
Mohammed Al-Garadi, PhD - VUMC; Michael Matheny, MD, MS, MPH, FACMI, FAMIA - Vanderbilt University Medical Center; Sankar Navaneethan, M.D - Section of Nephrology, Michael E. DeBakey Veterans Affairs Medical Center, Houston, TX; Richard Noriega, BSC - VUMC; Dax Westerman, MS - Vanderbilt University Medical Center Department of Biomedical Informatics; Parker Gregg, M.D - Section of Nephrology, Michael E. DeBakey Veterans Affairs Medical Center, Houston, TX; Sheena Wydermyer, M.D - Section of Nephrology, Michael E. DeBakey Veterans Affairs Medical Center, Houston, TX; Jennifer Arney, PhD - Michael E. DeBakey Veterans Affairs Medical Center, Houston; Salim Virani, M.D - Michael E. DeBakey Veterans Affairs Medical Center, Houston, TX; Michael Herrera, BSC - Michael E. DeBakey Veterans Affairs Medical Center, Houston, TX; Peter Richardson, BCS - Michael E. DeBakey Veterans Affairs Medical Center, Houston, TX; Jill Whitaker, Department of Biomedical Informatics, Vanderbilt University Medical Center, Nashville, TN - Department of Biomedical Informatics, Vanderbilt University Medical Center, Nashville, TN; Tina French, RN, CPHQ - Department of Biomedical Informatics, Vanderbilt University Medical Center, Nashville, TN,; Lisa Roddy, RN - VUMC; Glenn Gobbel - VA Tennessee Valley Healthcare System;

Empowering Healthcare Professionals with Data Science: Insights from Botswana Workshops

Poster Number: P06
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Education and Training, Curriculum Development, Global Health
Primary Track: Applications
Programmatic Theme: Academic Informatics / LIEAF

Botswana’s healthcare professionals generate vast amounts of data but often lack the skills to utilize the data for decision-making and service delivery. To address this, we created a series of foundational data science workshops. ~90 healthcare professionals were trained, focusing on data science fundamentals and practical Python application. Participants reported significant gains in data handling skills. Insights gained will inform future workshops, help reach more healthcare professionals, and contribute to data-driven healthcare in Botswana.

Speaker(s):
Badisa Mosesane, Bachelors
CHOP

Author(s):
J. Grey Faulkenberry, MD, MPH - Children's Hospital of Philadelphia;

Prototype for Automated Case Logging in a Diagnostic Radiology Residency

Poster Number: P07
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Education and Training, Documentation Burden, Informatics Implementation
Primary Track: Applications
Programmatic Theme: Academic Informatics / LIEAF

Radiology residents are required to complete case logs, leading to a significant documentation burden when done manually. This project introduces a partially automated logging prototype that seamlessly integrates with the current infrastructure. Several considerations were taken into account when selecting the implementation approach, in order to balance convenience and effectiveness. Testing against manually generated case logs revealed comparable accuracy and enhanced efficiency. Future goals include further optimization of automation and more extensive accuracy testing.

Speaker(s):
Gabriel Frattallone-Llado, MD
University of Puerto Rico

Author(s):
Gabriel Frattallone-Llado, MD - University of Puerto Rico; Ricardo Albino-Camacho, BS - University of Puerto Rico;

Enhancing Public Health Surveillance in King County: Applying Data Modernization to Social Determinants of Health

Poster Number: P08
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Population Health, Delivering Health Information and Knowledge to the Public, Health Equity, Standards
Primary Track: Foundations
Programmatic Theme: Academic Informatics / LIEAF

The Centers for Disease Control and Prevention (CDC) Data Modernization Initiative (DMI) aims to make the collection, sharing, and analysis of health data more efficient and impactful. However, there is limited literature available on incorporating social determinant of health (SDOH) into state and national DMI efforts. This study explore ongoing challenges in modernizing SDOH data collection and analysis, underscoring the need for standardized practices to address them effectively.

Speaker(s):
Faisal Yaseen, PhD student in Biomedical and Health Informatics
UW, Seattle

Author(s):
Faisal Yaseen, PhD student in Biomedical and Health Informatics - UW, Seattle; Luna Li, BS - University of Washington, Seattle, WA, USA; Raina Langevin, PhD - University of Washington;

Comparison of Computational Phenotyping Approaches in Identifying Patients with Dementia in the Emergency Department

Poster Number: P09
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Aging in Place, Clinical Decision Support, Computational Biology, Chronic Care Management, Real-World Evidence Generation
Primary Track: Applications
Programmatic Theme: Clinical Informatics

We compared 20 computational phenotyping algorithms using electronic health record data to identify dementia in patients aged 65+ in the emergency department (ED). Results showed variability in dementia prevalence (0.3% to 8.5%) across algorithms, with a higher prevalence in older patients and females. Computational phenotyping holds promise for enhancing dementia recognition in EDs and improving care quality.

Speaker(s):
Inessa Cohen, MPH
Yale University

Author(s):
Inessa Cohen, MPH - Yale University; Richard Taylor, MD MHS - Yale University; Aidan Gilson - Yale School of Medicine; Isaac Faustino, MS - Yale University; Natalia Festa, MD, MHS - Yale University; James Lai, MD, MHS - New York University; Phillip Magidson, MD, MPH - Johns Hopkins University; Adam Mecca, MD, PhD - Yale University; Debra Tomasino, MA - NYU Grossman School of Medicine; Haipeng Xue, MS - Yale University; Ula Hwang, MD, MPH - NYU Grossman School of Medicine;

eVisit, then what? A Patient-level Analysis of Care Utilization and Messages

Poster Number: P10
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Bioinformatics, Change Management, Chronic Care Management
Primary Track: Applications
Programmatic Theme: Clinical Informatics

eVisits comprise non-face-to-face patient-initiated communications through an online portal, and have reduced overall provider message burden. At our institution, we captured patient medical advice requests, calls, and care utilization patterns within 60 days of an eVisit. Completion of an eVisit did not change patient portal messaging or increase utilization, suggesting that this tool may be effective means of reducing message-related burden without discouraging subsequent individual patient-provider interactions.

Speaker(s):
Thomas Ueland, BS
Vanderbilt University School of Medicine

Author(s):
Thomas Ueland, BS - Vanderbilt University School of Medicine; Jacob Franklin, M.D. - VUMC; Allison McCoy, PhD, ACHIP, FACMI, FAMIA - Vanderbilt University Medical Center; Sara Horst, MD MPH;

Identifying Adverse Drug Events Using Ensemble Models

Poster Number: P11
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Natural Language Processing, Large Language Models (LLMs), Deep Learning, Information Extraction, Machine Learning
Primary Track: Applications
Programmatic Theme: Clinical Informatics

Electronic Health Records (EHR) contain descriptions of a patient's health and care outcomes. Often, a patient's reaction response to a medication is recorded in unstructured narratives, including adverse drug events (ADE) that indicate an unintended response to the medication prescribed. ADE information can help improve clinical decision-support systems, leading to better outcomes and improved patient safety. In this work, we build discrete machine learning-based and transformer-based Natural Language Processing systems on a corpus of notes from MIMIC-III to classify sentences containing ADEs. Our preliminary investigation shows that our systems have complementary strengths and ensembling them leads to more effective identification of ADEs.

Speaker(s):
Jacob Sheikh, BS
George Mason University

Author(s):
Jacob Sheikh, BS - George Mason University; Jordan Phillips, BS - George Mason University; Faraaz Rahman, Bachelors - George Mason University; Giridhar Kaushik Ramachandran, Student - George Mason Univeristy; Ozlem Uzuner, PhD - George Mason University;

Predicting Early Acute Kidney Injury in a Novel Multicenter Pediatric Critical Care Dataset

Poster Number: P12
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Machine Learning, Pediatrics, Critical Care
Primary Track: Applications

Acute kidney injury (AKI) is associated with higher morbidity and mortality in critically ill children, and often is detected too late for proactive treatment. We externally validated a machine learning AKI prediction model trained on single-center data, observed poor performance, and developed a new, optimal model with performance similar to the originally reported single-center model on a novel multicenter dataset using features from up to 12 hours after admission to the pediatric ICU.

Speaker(s):
Stephen Drury, MS
University of Rochester Health Lab

Author(s):
Stephen Drury, MS - University of Rochester Health Lab; Alex Clark; L. Nelson Sanchez-Pinto, MD, MBI, FAMIA - Northwestern University - Feinberg School of Medicine; Adam Dziorny, MD, PhD - University of Rochester, School of Medicine and Dentistry;

A Tailored Informatics Solution for Capturing Pediatric Anesthesia Events

Poster Number: P13
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Healthcare Quality, Patient Safety, Surgery, Pediatrics
Primary Track: Applications
Programmatic Theme: Clinical Informatics

Reliable capture of adverse pediatric anesthesia events is critical to the quality and safety mission of children’s surgical programs. We hypothesized that such events may be systematically under-reported, and deployed a customized, EHR solution to improve their capture. Monthly reporting derived from this new data interface over 16 months identified 79 potential events, four of which were confirmed as meeting adverse event criteria, highlighting the reliability and efficacy of this informatics initiative.

Speaker(s):
Bridget Toy, MHA, BSN, RN
Hassenfeld Children's Hospital at NYU Langone Health

Author(s):
Bridget Toy, MHA, BSN, RN - Hassenfeld Children's Hospital at NYU Langone Health; Ponney R. Palanisamy, MD - Hassenfeld Children's Hospital at NYU Langone Health; Gordana Stjepanovic, MD - Hassenfeld Children's Hospital at NYU Langone Health; Mara Karamitopoulos, MD, MPH, FAAOS - Hassenfeld Children's Hospital at NYU Langone Health; Scott M. Rickert, MD - Hassenfeld Children's Hospital at NYU Langone Health; Nilufar Tursunova, MD - NYU Langone Health; Germaine Cuff, PhD, BSN - NYU Langone Health; Jason C. Fisher, MD, FACS, FAAP, FAMIA - Hassenfeld Children's Hospital at NYU Langone Health;

Harnessing Machine Learning for the Selection of Empiric Antibiotics in Urinary Tract Infections

Poster Number: P14
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Machine Learning, Infectious Diseases and Epidemiology, Health Equity
Primary Track: Applications
Programmatic Theme: Clinical Informatics

To improve empiric decision making for antibiotic treatments, we developed machine learning models to predict the presence and type of antimicrobial resistant bacteria in urine cultures during the initial encounter. Using initial encounter information such as demographic, geographic, and socioeconomic data, we offer knowledge to clinicians to reduce the prescribing of unnecessary antibiotics.

Speaker(s):
Lauren Cooper, MS
University of Texas Southwestern Medical Center

Author(s):
Lauren Cooper, MS - University of Texas Southwestern Medical Center; Alaina Beauchamp, PhD MPH - University of Texas Southwestern Medical Center; Tanvi Ingle, BS - UT Southwestern Medical Center; Abdi Wakene, BS - University of Texas Southwestern Medical Center; Marlon Diaz - UT Southwestern; Chaitanya Katterpalli, MS - Texas Health Resources; Tony Keller, BA - Texas Health Resources; Clark Walker, MPH - Texas Health Resources; Alexander Radunsky, ScD MPH - University of Texas Southwestern Medical Center; Zachary Most, MD - University of Texas Southwestern Medical Center; john Hanna, MD - ECU Health; Trish Perl, MD - University of Texas Southwestern Medical Center; Christoph Lehmann, MD, FAAP, FACMI, FIAHSI - UT Southwestern; Richard Medford, MD - East Carolina University;

Foundational Shifts That Make CDSS with AI Capabilities Different

Poster Number: P15
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Clinical Decision Support, Clinical Guidelines, Evaluation
Primary Track: Foundations
Programmatic Theme: Clinical Informatics

By sharing and describing a list of "foundational shifts" related datasets and data analytic approaches, this poster provides a clear articulation of some key differences between Clinical Decision Support Systems (CDSS) developed before and after the use of big data, machine learning, and related advanced AI capabilities. The added clarity about CDSS brings new opportunities to improve CDSS taxonomies and AI-CDSS evaluation.

Speaker(s):
Kashmira Sawant, Masters in Health Informatics
Tempus Labs

Author(s):
Allen Flynn, PharmD, PhD - University of Michigan; Blackford Middleton, MD, MPH, MSc - Middleton Informatics;

Cross-cultural validation of HIV-related infographics

Poster Number: P16
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Information Visualization, Nursing Informatics, Delivering Health Information and Knowledge to the Public, Self-care/Management/Monitoring, Patient Engagement and Preferences
Primary Track: Applications
Programmatic Theme: Clinical Informatics

Infographics are an informatics-based strategy that can help improve clinical communication between patients and their providers. Thus, our team created a set of HIV-related infographics to enhance clinician-patient communication with Latino persons with HIV (PWH) and their providers. After two successful pilot studies with Latino PWH, we aimed to explore non-Latino PWH’s perceptions of infographics, including verifying their meaningfulness and acceptability, and to establish what, if any, changes need to be made prior to widespread scale-up. To complete these aims, we conducted a mixed-methods study with 50 participants who self-identified as non-Latino who completed quantitative surveys and brief, in-depth qualitative interviews after viewing our set of infographics with a study team member. Quantitative surveys were analyzed with descriptive statistics and interviews were analyzed using traditional qualitative content analysis. Findings from both the surveys and interviews indicated promising generalizability of infographics since they were viewed as acceptable and meaningful by non-Latino PWH, who would like to see HIV-related infographics widely distributed and used in future clinic visits.

Speaker(s):
Samantha Stonbraker, PhD, MPH, RN
University of Colorado Anschutz Medical Campus

Author(s):
Stefanie Mayorga, BA - University of Colorado College of Nursing; Kellie Hawkins, MD, MPH - Denver Health and Hospital Authority; Edward Gardner, MD - Denver Health and Hospital Authority; Kenrick Cato, PhD, RN, CPHIMS, FAAN - University of Pennsylvania/ Children's Hospital of Philadelphia; Adriana Arcia, PhD, RN, FAAN - University of San Diego;

Models for Detection of Colorectal Cancer in the Veterans Affairs New England Healthcare System

Poster Number: P17
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Clinical Decision Support, Machine Learning, Real-World Evidence Generation
Primary Track: Applications
Programmatic Theme: Clinical Informatics

We constructed decision tree and several logistic regression models using electronic health record data from Veterans Affairs Veterans Integrated Service Network (VISN) 1 between 1/1/2006 and 12/31/2016. On our testing data, our decision tree model achieved a sensitivity of 0.57, a specificity of 0.72, and a PPV of 0.005. The best performing logistic regression, the LASSO regularization, achieved a sensitivity of 0.64, a specificity of 0.58, and a PPV of 0.005.

Speaker(s):
Kaelyn Nannini, MS
VA Boston Healthcare System

Author(s):
Robin Baidya; Theodore Feldman, PhD - VA Boston Healthcare System; David Winski - Durham VA/Duke University; Kaelyn Nannini, MS - US Department of Veterans Affairs; Nathanael Fillmore, PhD - VA Boston Healthcare System; Nhan Do, MD, MS, Clinical Informatics Diplomate - Boston VA HCS; Mary Brophy, MD; Brian Sullivan, MD - Durham VAMC/Duke University School of Medicine; Jason Dominitz, MD - VA Puget Sound Healthcare System/University of Washington, School of Medicine;

Evaluating the Accuracy, Completeness, and Consistency of ChatGPT's Factual Medical Knowledge about Breast Cancer

Poster Number: P19
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Large Language Models (LLMs), Evaluation, Information Retrieval
Primary Track: Applications
Programmatic Theme: Clinical Informatics

We used Zero-shot Persona Pattern prompts to assess ChatGPT (GPT-4)’s factual knowledge about breast cancer. Across the two versions of ChatGPT responses to specific questions representing 12 clinical domains, we did not find incorrect answers. The completeness was greater than 80% for 11 clinical domains in version 1 and for 7 clinical domains in version 2. The responses between version 1 and version 2 showed inconsistency ranging from 13% to 44% in 9 clinical domains.

Speaker(s):
Vivienne Zhu, MD, MS
Elevance Health, Inc

Author(s):
J. Marc Overhage, MD, PhD - Elevance Health;

The Watcher Program: Implementing a Situational Awareness and Team Communication Tool to Promote Early Recognition of Patient Deterioration

Poster Number: P20
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Informatics Implementation, Patient Safety, Pediatrics, Critical Care
Primary Track: Applications
Programmatic Theme: Clinical Informatics

Hospitalized pediatric hematology/oncology patients are at risk for clinical deterioration, and delayed recognition of clinical deterioration may result in unsafe transfers to the intensive care unit or other adverse events. We developed an evidence-based Watcher program integrated into the Epic Electronic Health Record that promotes situational awareness, team communication, and a shared mental model for patients at risk for clinical deterioration. This approach is feasible, acceptable to users, and is being expanded to additional wards.

Speaker(s):
Wayne Liang, MD MS FAMIA
Children's Healthcare of Atlanta & Emory University

Author(s):
Temima Oratz, BSN, RN, CPHON, EBP-C - Children's Healthcare of Atlanta; Edwin Ray, RN - Children's Healthcare of Atlanta; Joy Barker, MSN, CPNP, CPPS, CCRN-K - Children's Healthcare of Atlanta; Abby Fitzpatrick, BSN, RN, CPHON - Children's Healthcare of Atlanta; Swaminathan Kandaswamy, PhD - Emory University School of Medicine; Wayne Liang, MD MS FAMIA - Children's Healthcare of Atlanta & Emory University;

Using a pathology driven rules engine CDS tool to improve concordance between documented follow up recommendations in the EHR after colonoscopy polypectomy and care guidelines

Poster Number: P21
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Clinical Decision Support, Clinical Guidelines, Documentation Burden, Rule-based artificial intelligence
Primary Track: Applications
Programmatic Theme: Clinical Informatics

Failure to document colonoscopy follow up needs post-polypectomy can lead to delayed detection of colorectal cancer (CRC). Automating the update of a centralized follow-up date in the electronic health record (EHR) based on pathology results may increase the number of patients with guideline concordant CRC follow-up screening. As part of an operational quality improvement initiative, we developed an innovative rules-based clinical decision support tool to automate updates to the patient’s Health Maintenance (HM) follow-up date for CRC screening based on pathology results with the objective of improving concordance with guideline recommendations while reducing provider documentation burden. This study was an organizational-wide prospective pre-post design that compared standard care to the effect of an automated, EHR-based rules-engine CDS tool on the follow-up frequency documented in the patient’s EHR (i.e. HM section). Primary outcome measure is change in follow-up screening interval. Study population included 10,024 standard care and 19,184 intervention participants. The proportion of patients with a 10-year default follow-up frequency significantly decreased (87.4% to 41.6%, p<0.001). Failure to update follow-up dates for recommended screening can lead to missed colonoscopies and delayed detection of CRC. The implementation of an automated rules-engine-based CDS tool has the potential to increase the accuracy of colonoscopy follow-up screening dates recorded in patient EHR HM activities without increasing clinician documentation burden. The results of this study emphasize the need for more automated and integrated solutions for updating and maintaining EHR HM activities.

Speaker(s):
Adam Szerencsy, DO
NYU Langone Health

Author(s):
Elizabeth Stevens, PhD, MPH - NYU Grossman School of Medicine; Arielle Nagler, MD - NYU Grossman School of Medicine; Casey Monina, RN - NYU Langone MCIT; JaeEun Kwon, MPP - NYU Grossman School of Medicine; Amanda Olesen Wickline, None - NYU Langone MCIT; Gary Kalkut, MD - NYU Grossman School of Medicine; Dave Ranson, none - NYU Langone MCIT; Seth Gross, MD - NYU Grossman School of Medicine; Aasma Shaukat, MD, MPH - NYU Grossman School of Medicine; Adam Szerencsy, DO - NYU Langone Health;

MACHINE LEARNING-BASED PREDICTION OF NO-SHOW TO TELEMEDICINE ENCOUNTERS IN PERU

Poster Number: P22
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Telemedicine, Machine Learning, Global Health, Diversity, Equity, Inclusion, Accessibility, and Health Equity
Primary Track: Applications
Programmatic Theme: Clinical Informatics

This study evaluates the effectiveness of machine learning models in predicting no-shows to telemedicine appointments in Peru. Using data from the "Teleatiendo" platform, we explored models like XGBoost, and Random Forest. Results indicate that cost sensitive XGBoost balance well between identifying no-shows and shows, with the Balanced Random Forest model showing the highest recall. The findings highlight the potential of ML in improving telemedicine appointment adherence in resource-limited settings.

Speaker(s):
Christian Reategui Rivera, PhD student
University of Utah

Author(s):
Wanting Cui, Masters - University of Utah; Stefan Escobar Agreda, MD - Ministry oh Health (MINSA - Perú); Leonardo Rojas-Mezarina, MD - Telehealth Unit - Universidad Nacional Mayor de San Marcos; Joseph Finkelstein, MD, PhD - University of Utah;

Identifying Caregiver Depression Among Those Supporting Dementia Patients Through Natural Language Processing (NLP)

Poster Number: P23
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Natural Language Processing, Surveys and Needs Analysis, Informatics Implementation
Primary Track: Applications
Programmatic Theme: Clinical Informatics

This study aims to detect depression symptoms in dementia caregivers by using Natural Language Processing (NLP) to generate depression scores from open-ended responses to a developed questionnaire. The accuracy of NLP-generated scores is assessed by comparing them to survey scores of Patient Health Questionnaire-8 and the Zarit Burden Scale 3. A successful outcome may establish a more accurate and efficient way of identifying depression symptoms through caregivers’ open-ended responses, in adjunct to existing surveys in healthcare.

Speaker(s):
Kruthika Gaddam, Masters
IUPUI

Author(s):

Impact of Longitudinal Trends in Predicting Pediatric Clinical Deterioration

Poster Number: P24
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Machine Learning, Deep Learning, Pediatrics, Patient Safety
Primary Track: Applications
Programmatic Theme: Clinical Informatics

Clinical deterioration among pediatric patients is associated with poor health outcomes. We trained LSTM models that predict which pediatric patients are at an imminent risk of experiencing clinical deterioration. The performance of our constructed LSTM models is compared with that achieved by our previously published gradient boosted machine models, called pCART, on the same longitudinal datasets. While our LSTM models did not outperform pCART, training an LSTM on pCART outputs lead to more generalizable performance.

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

Author(s):
Sierra Strutz, PhD Student in Biomedical Data Science - University of Wisconsin - Madison; Kyle Carey, MPH - University of Chicago; Fereshteh S. Bashiri, PhD - University of Wisconsin - Madison; Priti Jani, MD, MPH - University of Chicago; Emily Gilbert, MD - Loyola University; Majid Afshar, MD, MSCR - University of Wisconsin - Madison; Matthew Churpek, MD, MPH, PhD - University of Wisconsin-Madison; Anoop Mayampurath, PhD - University of Wisconsin - Madison;

Enhancing Health System Implementation: A Contextual Approach Through Utilization Dashboards

Poster Number: P25
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Information Visualization, Data Sharing, Usability
Primary Track: Foundations

Through a retrospective cohort study covering millions of patients across the Southeast United States, a utilization dashboard was created to visualize patient demographics, chronic conditions, and care location data. This tool enables stakeholders to identify regional trends, and plan feasibility studies, underscoring a commitment to improving healthcare delivery through contextualized interventions.

Speaker(s):
Lauren Witek, MStat
Atrium Health Wake Forest Baptist

Author(s):
Richa Bundy, MPH - Wake Forest Baptist Health; Corey Obermiller, MAS Applied Statistics - Wake Forest School of Medicine; Adam Moses, MHA - Wake Forest Baptist Medical Center; Nicholas Pajewski, PhD - Wake Forest School of Medicine; Ajay Dharod, MD - Wake Forest University School of Medicine;

Adaptive Data Extractor for Wearable Health Data Integration

Poster Number: P26
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Patient / Person Generated Health Data (Patient Reported Outcomes), Machine Learning, Workflow
Primary Track: Applications

With growing interest in leveraging wearable health data, standardized pipelines are developed to process such data efficiently. However, these tools depend on a rigid input schema, making it challenging to adapt evolving data schemas of wearable devices, thus limiting their scalability. In response, this study explored the feasibility of building an adaptive wearable health data extractor to facilitate data pre-processing and integration, irrespective of changes in data labels and formats.

Speaker(s):
Sunghoon Kang, B.A.
Seoul National University

Author(s):
Hyeoneui Kim, PhD - Seoul National University;

Assessing ChatGPT Responses to Alzheimer’s Disease Myths

Poster Number: P27
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Advanced Disease, Internal Medicine or Medical Subspecialty, Large Language Models (LLMs), Aging in Place
Primary Track: Applications
Programmatic Theme: Clinical Informatics

Alzheimer's disease is prone to myths and statements that exhibit varying degrees of accuracy, inaccuracy, and misinformation. This study assesses ChatGPT's ability to identify and address AD myths with reliable information. Geriatricians acknowledged the potential value of ChatGPT in mitigating misinformation.

Speaker(s):
Sean Huang, MD
Vanderbilt University

Author(s):
Qingyuan Song, Master of Engineering - Vanderbilt University; Kimberly Beiting, MD - Vanderbilt University Medical Center; Maria Duggan, MD - Vanderbilt University Medical Center; Kristin Hines, MD - Vanderbilt University Medical Center; Harvey Murff, MD - Vanderbilt University Medical Center; Vania Leung, MD - University of Illinois at Chicago; James Powers, MD - Vanderbilt University Medical Center; T.S. Harvey, Ph.D. - Vanderbilt University; Bradley Malin, PhD - Vanderbilt University Medical Center; Zhijun Yin, Ph.D. - Vanderbilt University Medical Center;

Design and Testing of a Cloud-hosted, Standards-Based CDS System for Sexually Transmitted Infection

Poster Number: P28
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Clinical Decision Support, Standards, User-centered Design Methods
Primary Track: Applications
Programmatic Theme: Clinical Informatics

To facilitate adherence to guidelines for treatment of sexually transmitted infections, the Center for Disease Control and Prevention (CDC) partnered with the Public Health Informatics Institute to pilot a cloud-hosted clinical decision support system launched simultaneously in multiple institutions. The design and logic development required multiple meetings and testing sessions prior to going live in patient care. Such a system is a feasible but resource-intensive method for some select cases for clinical decision support.

Speaker(s):
Edna Shenvi, MD, MAS
Elimu Informatics Inc.

Author(s):
Edna Shenvi, MD, MAS - Elimu Informatics Inc.; Aziz Boxwala, MD, PhD - Elimu Informatics; Charisse LaVell, MPH - Public Health Informatics Institute; Ritche Hao, MD - Yale School of Medicine; Carlos Paredes, RN, BSN, AS - Yale New Haven Health; Sanjat Kanjilal, MD MPH; Sarah Shaw, MPH, PMP - Public Health Informatics Institute; Alejandro Perez, MPH - Centers for Disease Control and Prevention; Gema Dumitru, MD, MPH - Centers for Disease Control and Prevention; Jeanne Ocampo, BS - Centers for Disease Control; Saugat Karki, MD, MS - US Centers for Disease Control and Prevention; Andre Berro;

Enhancing Prediction of Systolic Heart Failure Outcomes through Genetic Algorithm-Optimized Laboratory Tests

Poster Number: P29
Presentation Time: 05:00 PM - 06:30 PM

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

This research sought to advance in-hospital mortality predictions for systolic heart failure patients by refining the utilization of laboratory test data. Employing the MIMIC-IV database, we applied a Genetic Algorithm (GA) to optimize feature selection and testing intervals, which significantly improved the model’s AUC from 0.70 to 0.78. Our findings highlight the importance of targeted feature selection and timing optimization in enhancing prediction model performance and improving clinical decision-making and patient outcomes.

Speaker(s):
Bhumi Patel, Ph.D. in Health Informatics
George Mason University

Author(s):
Lemba Priscille Ngana, PhD Health Services Research (Knowledge Discovery and Health Informatics Concentration) - George Mason University; Janusz Wojtusiak, PhD - George Mason University;

An Ontology for the Indigenous Determinants of Health

Poster Number: P30
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Controlled Terminologies, Ontologies, and Vocabularies, Diversity, Equity, Inclusion, Accessibility, and Health Equity, Health Equity
Primary Track: Foundations
Programmatic Theme: Clinical Informatics

Indigenous communities continue to face health disparities due to risk factors, limited access to healthcare, and historical trauma. The Indigenous Determinants of Health (IDoH) recognize autonomous initiatives by communities for resilient well-being. Developing an ontology for IDoH aims to standardize knowledge management, categorizing factors influencing health outcomes in Indigenous communities and support effective health governance and policy-making across Tribal, state, and federal levels.

Speaker(s):
Danner Peter, MPH
University of Washington

Author(s):
Peter Tarczy-Hornoch, MD - University of Washington; Danner Peter, MPH - University of Washington; Neil Abernathy, PhD - University of Washington; Oliver Bear Don't Walk, PhD - University of Washington;

Lessons Learned for Co-creation Workshops: Collaborating with Clinicians and Patients to Design an EHR-connected Chat Tool

Poster Number: P31
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: User-centered Design Methods, Human-computer Interaction, Mobile Health, Patient Engagement and Preferences
Primary Track: Applications
Programmatic Theme: Clinical Informatics

Co-creation workshops provide key information about end-user needs and need to be well prepared to make the most of each participant's expertise and time. We created clinical scenarios, which were adapted to the participants’ expertise, with planned overlap in the questions asked. This allowed participants to compare their results among groups, with context-rich discussions, and to avoid perceived redundancies in the final restitution phase.

Speaker(s):
Katherine Blondon, MD, PhD
Hopitaux Universitaires Geneve

Author(s):
Arnaud Ricci, MSc - University Hospitals of Geneva; Frederic ehrler, PhD - Division of medical information sciences;

Evaluating GPT-4 Generated Draft Replies to Patient Messages for Differences in Patient-centeredness and Positive Affect by Patient Race

Poster Number: P32
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Large Language Models (LLMs), Health Equity, Patient Engagement and Preferences
Primary Track: Applications
Programmatic Theme: Clinical Informatics

Patient-provider communication via messaging has surged, resulting in increased provider burnout. AI tools, like Epic-integrated GPT-4 which generates drafts to patient messages, are a possibly remedy. Concerns exist that models may perpetuate disparities, but have not been studied in this context. Computational linguistics techniques performed on Epic-integrated GPT-4 draft replies revealed disparities in communication styles by patient race. Drafts for Black and Asian patients were more positive but less patient-centered when compared to White counterparts.

Speaker(s):
Amelia Shunk, MMCi
NYU Grossman School of Medicine

Author(s):
William Small, MD, MBA - New York University Langone Medical Center; Yuhan Cui, MS - NYU Grossman School of Medicine; Chelsea Twan, MS - NYU Grossman School of Medicine; Safiya Richardson, MD, MPH - NYU Grossman School of Medicine; Soumik Mandal, PhD - NYU Langone; Yindalon Aphinyanaphongs, MD, PhD - NYU Langone;

Evolution of Genomic Indicators for Pharmacogenomics: Retrospective Analysis and Implications for Knowledge Management

Poster Number: P33
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Clinical Decision Support, Precision Medicine, Pharmacogenomics, Change Management, Informatics Implementation, Evaluation
Primary Track: Applications
Programmatic Theme: Clinical Informatics

Pharmacogenomics (PGx) incorporates patient genetic data into pharmacotherapy guidelines to improve patient outcomes. Clinical decision support (CDS) systems rely on underlying knowledge bases, information models, and encoded rule logic to implement clinical guidelines. Mayo Clinic has implemented Epic's Genomic Indicators (GI) activity to interpret PGx test results into criteria for CDS rules execution, however, changes in PGx knowledge and result reporting standards necessitate continual maintenance of CDS rule logic and GIs attached to patient records.

We reviewed nearly 7-years of GI implementation, identifying 158 unique indicators and 15 events resulting in changes to one or more GIs. Several events incorporated new PGx knowledge, including implementing new gene-drug pairings and updating genotype-phenotype specifications, such as utilizing haplotype enzyme activity score data to quantitatively assign phenotypes. These changes resulted in 25 new GIs and updates to 12 indicator names. The incorporation of phenotype results from a large multi-gene panel resulted in the creation of 29 new indicators across eight gene-drug pairings. Upon later review, 12 of the 29 GIs were removed or merged with previously established GIs due to the use of non-standardized nomenclature and classifications, necessitating verification of GI assignments for all patients associated with the discontinued indicators.

This study highlights the importance of standardized nomenclature for CDS documentation and limitations associated with phenotype-only genetic test result reporting. The use of discrete, variant or haplotype level result reporting facilitates validation and reinterpretation of patient genetic results which is not possible when only phenotype level results are reported.

Speaker(s):
Robert Freimuth, PhD
Mayo Clinic

Author(s):
Sarah Senum, MS - Mayo Clinic; Jessica Wright, PharmD, RPh - Mayo Clinic;

Broadening Patient Treatment Options: Large Language Models Pipeline Improves Cancer Patient Matching to Clinical Trials in Day-To-Day Practice

Poster Number: P34
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Precision Medicine, Large Language Models (LLMs), Information Extraction, Data Standards, Informatics Implementation
Primary Track: Applications
Programmatic Theme: Clinical Informatics

Most modern cancer trials are targeted towards a patient’s molecular profile. These trials may provide longer survival and better quality of life than standard treatments, especially for patients whose cancer advances after initial treatment. Despite availability of these novel treatments, clinicians are poorly equipped to match patients whose cancer progresses to the available treatment options in a timely manner. One of the major challenges in automation of the matching process is recognition of cancer progression in clinical notes. Addition of Large Language Models to the arsenal of tools extracting information from text revolutionizes capabilities in this domain. At the NYU Perlmutter Cancer Center, we successfully implemented an LLM-based pipeline that alerts thoracic oncologists about available clinical trial options based on patient’s molecular profile one day after radiology reports cancer progression. In this manuscript, we intend to describe how combination of novel informatics methods such as AI and good informatics practices such as data standardization help make real impact in clinical care.

Speaker(s):
Rimma Belenkaya, MS, MA
NYU Langone

Author(s):
Ferris Hussein, Computer Science - NYU Langone Health; Abraham Chachoua, MD - NYU Langone; Vamsidhar Velcheti, MD - NYU Langone; William H. Moore, MD - NYU Langone; Kanan Shah, MD - NYU Langone;

Quantifying the Effect of Value Set Choice on Clinical Quality Measures and Clinical Decision Support Accuracy

Poster Number: P35
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Clinical Decision Support, Terminology Systems, Information Retrieval
Primary Track: Applications
Programmatic Theme: Clinical Informatics

Clinical quality measurement and clinical decision support rely on accurate value sets. We considered prescription of beta blockers after hospital discharge for secondary prevention of myocardial infarction using 23 myocardial infarction value sets and 10 beta blocker value sets. The number of eligible patients ranges from 1,229 to 38,656 and the proportion with beta blockers ranged from 25.7% to 77.3%. Careful selection, creation, and maintenance of value sets is essential for accurate CQM and CDS.

Speaker(s):
Adam Wright, PhD
Vanderbilt University Medical Center

Author(s):
Donald Sengstack, MS; Laura Zahn, MS; Elise Russo - Vanderbilt University Medical Center; Dean Sittig, PhD - University of Texas Health Science Center at Houston;

Comparing Deep Learning Approaches for Predicting Clinical Deterioration Using Chest X-Rays

Poster Number: P36
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Critical Care, Deep Learning, Imaging Informatics
Primary Track: Applications
Programmatic Theme: Clinical Informatics

This study compares different computer vision deep learning algorithms for the early detection of clinical deterioration using chest X-ray images. The study utilizes a large dataset of chest X-rays from hospitalized adult patients with corresponding deterioration labels. We compared different deep learning models trained on the data with different data augmentation methods and achieved an area under the receiver operating curve (AUROC) of 0.72, outperforming the eCART score (AUROC of 0.668) in the validation dataset. The results indicate that deep learning models with chest X-ray images have the potential to improve clinical deterioration risk assessment beyond structural data alone.

Speaker(s):
Mahmudur Rahman, PhD
University of Wisconsin-Madison

Author(s):
Mahmudur Rahman, PhD - University of Wisconsin-Madison; Jifan Gao, MS - University of Wisconsin-Madison; Kyle Carey, MPH - University of Chicago; Dana Edelson, MD, MS - University of Chicago; Askar Afshar, MS - University of Wisconsin-Madison; John Garrett, PhD - University of Wisconsin-Madison; Guanhua Chen, PhD - University of Wisconsin-Madison; Majid Afshar, MD, MSCR - University of Wisconsin - Madison; Matthew Churpek, MD, MPH, PhD - University of Wisconsin-Madison;

Increasing Chlorhexidine Gluconate Bath Orders for Eligible Patients in a Neonatal Intensive Care Unit

Poster Number: P37
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Clinical Decision Support, Pediatrics, Workflow
Primary Track: Applications
Programmatic Theme: Clinical Informatics

Daily chlorhexidine gluconate (CHG) baths are a standard of care for patients with central lines and Foleys. However, neonates are not always eligible for CHG due to gestational or chronological age limitations. The complex decision algorithms that determine which patients are eligible to receive CHG baths can lead to missed orders, confusion for nurses, and thus poor compliance. Clinical decision support in the electronic health record can be used to help identify eligible patients increasing CHG orders and hygiene compliance documentation.

Speaker(s):
Sarah Thompson, MSHIMI, BSN, RN
Children's Healthcare of Atlanta

Author(s):
Kacey Nation, MSN, RNC-NIC - Children's Healthcare Of Atlanta; Kacey Church, RN - Children's Healthcare of Atlanta; Kristin Carnall, MSN. RN - Children's Healthcare of Atlanta; Caitlin Pugh, RN - Children's Healthcare of Atlanta; Brenda Pointdexter, MD - Children's Healthcare of Atlanta; Shannon Hamrick, MD; Swaminathan Kandaswamy, PhD - Emory University School of Medicine; Anthony Piazza, MD - Children's Healthcare of Atlanta;

Natural Language Processing Pitfalls in Patient Phenotype Extraction

Poster Number: P38
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Natural Language Processing, Terminology Systems, Standards
Primary Track: Applications
Programmatic Theme: Clinical Informatics

Whole Exome Sequencing (WES) is a powerful genomic technique that links genetic variations to patient phenotypes to support precision medicine. However, manual abstraction of patient phenotypes for providers is labor-intensive and lack integration with Electronic Health Records (EHR). Our study evaluated an enterprise Natural Language Processing (NLP) application to extract standard Human Phenotype Ontology (HPO) concepts, revealing “semantic” challenges in NLP detection process . Our findings underscore the need for NLP optimization and standardization in genomic workflows to enhance data accuracy and streamline structured data capture and usage processes.

Speaker(s):
Sina Madani, MD, PhD
Vanderbilt University Medical Center

Author(s):
Sina Madani, MD, PhD - Vanderbilt University Medical Center; Asli Ozdas Weitkamp, PhD - Vanderbilt University Medical Center; Dario Giuse, DrIng - Vanderbilt University Medical Center;

Automating Procedure Billing Using A ChatGPT Decision Tree Query Architecture

Poster Number: P39
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Documentation Burden, Large Language Models (LLMs), Surgery, Workflow, Standards
Primary Track: Applications
Programmatic Theme: Clinical Informatics

This study assesses ChatGPT's ability to autonomously assign CPT codes to hand surgery operative notes, using a direct comparison between a single-query and a refined two-step query model. Analyzing 150 cases, the two-step model showed improved accuracy (74.3%) over the single-query model (52.0%). The two-step model demonstrates the potential for decision tree architectures for expanding and enhancing model performance. Further refinement is needed for precise code differentiation, particularly in complex cases like distal radius fractures.

Speaker(s):
Patrick Donohue, BS
Washington University in Saint Louis

Author(s):
Patrick Donohue, BS - Washington University in Saint Louis; Emilie Zoldos, MD - Washington University in Saint Louis; Christopher Dy, MD, MPH - Washington University in Saint Louis; Adam Wilcox, PhD - Washington University in St. Louis;

Utilizing Large Language Models for Extracting Abstractions in Muscular Dystrophy from Clinical Notes

Poster Number: P40
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Large Language Models (LLMs), Natural Language Processing, Information Extraction
Working Group: AMIA Clinical Informatics Fellows (ACIF) Working Group
Primary Track: Applications
Programmatic Theme: Clinical Informatics

This study evaluates large language models (LLMs) for extracting data on Duchenne and limb girdle muscular dystrophies from clinical notes. Analyzing notes from 44 patients, the approach shows potential, especially in identifying symptoms with the llama2-7b model achieving an F1 score of 0.499 in symptom detection. The findings highlight the need for model tuning and further exploration of NLP techniques to improve data extraction and tracking of disease progression.

Speaker(s):
Geetanjali Rajamani, BS
University of Minnesota Medical School

Author(s):
Huixue Zhou; Rui Zhang, PhD, FAMIA - University of Minnesota, Twin Cities; Peter Kang, PhD - University of Minnesota;

Charting the Future: Enhancing Medical Record Reviews with Large Language Models

Poster Number: P41
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Large Language Models (LLMs), Real-World Evidence Generation, Natural Language Processing
Primary Track: Applications

Chart reviews are often performed for several purposes ranging from validation studies to care assessments to training models for machine learning, and more. The manual chart review methodology is usually time consuming, costly, and prone to human error. Natural Language Processing (NLP) techniques can already assist with analyzing the human language found in clinical notes, but typically require a significant overhead to set up for a specific use case. With the introduction of large language models (LLMs), there is a Retrieval-Augmented Generation (RAG) process that can be leveraged in conjunction with existing methods to create a more accurate search to assist in chart reviews. At Mass General Brigham, we have developed a robust RAG pipeline to accommodate the multitude of use-case-dependent configurations for clinical chart review. The overall goals were to 1) build an adaptable framework enabling LLMs to assist in chart review, 2) use locally hosted LLMs so that no protected health information leaves the organization’s premises, and 3) generalize the workflow to “future-proof”, as much as possible, the compatibility with tomorrow’s next-generation language models.

Speaker(s):
Nich Wattanasin
Mass General Brigham

Author(s):
Martin Rees, BSc - Mass General Brigham; Victor Castro, MS - Mass General Brigham; Taowei Wang, PhD - Mass General Brigham; Heekyong Park, PhD - Mass General Brigham; Allan Harris, Bsc. - Mass General Brigham; Barbara Benoit, Bsc - Mass General Brigham; Vivian Gainer, MS - Mass General Brigham; Christopher Herrick, MBA - Mass General Brigham; Shawn Murphy, MD, Ph.D. - Massachusetts General Hospital;

Leveraging the Electronic Health Record Patient Portal to Identify Unmet Health-Related Social Needs Among Adolescents and Young Adults with Cancer

Poster Number: P42
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Diversity, Equity, Inclusion, Accessibility, and Health Equity, Health Equity, Informatics Implementation, Qualitative Methods, Surveys and Needs Analysis
Primary Track: Applications
Programmatic Theme: Clinical Informatics

Utilizing EHR portal screenings for adverse social determinants of health enables easier identification of health-related social needs among adolescents and young adults with cancer. Screenings were implemented in two oncology clinics in New York City and data captured were used to match potential AYAs to intervention studies in hopes of delivering more equitable care. Additionally, qualitative data from semi-structured interviews were collected to understand perspectives of the screening procedures and how to improve the procedures.

Speaker(s):
Marcela Algave, BSN, RN
Columbia University

Author(s):
Marcela Algave, BSN, RN - Columbia University; David DeStephano, MPH - Herbert Irving Comprehensive Cancer Center at Columbia University; Rhea Khurana, BS - Columbia University School of Nursing; Kathryn Valera, MSW - Comprehensive Cancer Center at Columbia University; Claire Sathe, MD JD - Herbert Irving Comprehensive Cancer Center at Columbia University; Justine Kahn, MD MS - Herbert Irving Comprehensive Cancer Center at Columbia University; Dawn Hershman, MD MS - Herbert Irving Comprehensive Cancer Center at Columbia University; Melissa Beauchemin, PhD, RN, MS, CPNP - Columbia University;

TAPESTRY: Temporal Analytics Platform for Exploratory Scientific and Translational Research inquirY

Poster Number: P43
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Informatics Implementation, Information Visualization, User-centered Design Methods
Primary Track: Applications
Programmatic Theme: Clinical Informatics

TAPESTRY is a modular, generalizable toolkit designed for clinical scientists and informatics engineers to rapidly develop temporally oriented patient datasets. This allows them to study disease progression, treatment efficacy, and longitudinal patient care. The platform achieves temporal orientation by adopting a universal patient:time key, allowing the inclusion of patient data from any system. Each TAPESTRY instance is envisioned as a component in a larger "data fabric," offering a scalable ability to combine and view longitudinal clinical records, each specialized to a specific domain. The TAPESTRY interface harmonizes clinical experts and developers, providing a continuous feedback loop through embedded annotation workflows. This toolbox-driven approach can provide a new, technically supported platform to enhance collaboration and expertise sharing throughout an institution. TAPESTRY has been implemented in 3 clinical domains: ICU, Cardiology, and Cancer, providing a richer ability to stratify patient risk by shifting focus to time at risk. The platform is an effective tool for fine-grained cohort identification and selection, disease state change tracking, and rich comparative analyses. Plans are underway to extend the platform with greater group identification and group comparative analytics capabilities via the implementation of advanced techniques for event detection, forecasting, clustering, and phenotyping of time series profiles.

Speaker(s):
Albert Riedl, M.S.
UC Davis Health System

Author(s):
Nicholas Anderson, PhD - University of California, Davis;

Percentage Agreement Between Individual LLMs and Certified Human Coder in Extraction of ICD-10-CM Codes

Poster Number: P44
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Large Language Models (LLMs), Deep Learning, Human-computer Interaction, Information Extraction
Primary Track: Foundations
Programmatic Theme: Clinical Informatics

The International Classification of Diseases- tenth revision – clinical modification (ICD-10-CM) is a complex code system. This study assessed the capability of ChatGPT 3.5, ChatGPT 4, Claude 2.1 and Claude 3 to extract these codes from unstructured clinical notes in comparison to a certified human coder. Based on our findings, we observed minimal to no agreement between LLMs and the certified human coder.

Speaker(s):
Megan McDougal, MS
West Virginia University

Ashley Simmons, MBA
West Virginia University

Author(s):
Ankit Sakhuja, MBBS, MS - Icahn School of Medicine at Mount Sinai; Kullaya Takkavatakarn, MD - Icahn School of Medicine at Mount Sinai;

Delinking Cephalosporin Cross Sensitivity Alerts in Patients with PAL

Poster Number: P45
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Clinical Decision Support, Drug Discoveries, Repurposing, and Side-effect, Clinical Guidelines
Primary Track: Applications

This retrospective study assesses the impact of a clinical decision support (CDS) alert recommending cephalosporin avoidance for patients with penicillin allergy labels (PALs). Using R Studio, we analyzed 13560 alerts from 07/01/2021-07/19/2023, 63.7% of which were overridden. Patients receiving cephalosporins despite PALs exhibited significantly better outcomes compared to patients given alternatives. These findings challenge the efficacy of the alert and suggest potential for improvements to optimize patient care.

Speaker(s):
Megan Wang, Undergraduate Student
Vanderbilt University

Author(s):
Allison McCoy, PhD, ACHIP, FACMI, FAMIA - Vanderbilt University Medical Center; Adam Wright, PhD - Vanderbilt University Medical Center; Cosby Stone, MD, MPH - Vanderbilt University Medical Center;

The CONCERN Implementation Toolkit: Standardizing, Scaling, and Spreading an AI-Based Clinical Decision Support System

Poster Number: P46
Presentation Time: 05:00 PM - 06:30 PM

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

CONCERN is an AI-based early warning system that uses nursing data to predict patient deterioration. CONCERN has been implemented and studied at two clinical sites, with planned implementations at three additional locations. The CONCERN Implementation Toolkit (CIT) is a collection of resources that provide guidance, support and structure needed for successfully scaling and spreading CONCERN.

Speaker(s):
Jennifer Withall, PhD
Columbia University Department of Biomedical Informatics

Author(s):
Kenrick Cato, PhD, RN, CPHIMS, FAAN - University of Pennsylvania/ Children's Hospital of Philadelphia; PATRICIA C DYKES, PhD, MA, RN - Brigham and Women's Hospital; Sandy Cho, MPH IN-BC - Newton-Wellesley Hospital; Catherine Ivory, PhD, NI-BC, NEA-BC, FAAN - Vanderbilt University Medical Center; Po-Yin Yen, PhD, RN, FAMIA, FAAN - Washington University in St. Louis; Rachel Lee, PhD, RN - Columbia University; Brian Douthit, PhD, RN, NI-BC - Department of Biomedical Informatics, Vanderbilt University; Temiloluwa Daramola; Graham Lowenthal, BA - Brigham and Women's Hospital; David Albers, PhD - University of Colorado, Department of Biomedical Informatics; Syed Mohtashim Abbas Bokhari; Sarah Rossetti, RN, PhD - Columbia University Department of Biomedical Informatics;

InBasket Utilization for Primary and Subspecialty Providers with Generative AI Implementation

Poster Number: P47
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Informatics Implementation, Workflow, Natural Language Processing, User-centered Design Methods, Human-computer Interaction, Social Media and Connected Health, Administrative Systems
Primary Track: Applications
Programmatic Theme: Clinical Informatics

We present preliminary data from University of California Irvine Health on the implementation of generative AI into InBasket, resulting in decreased provider time for both primary care and specialty care providers in responding to patient messages. With a specialty-agnostic AI prompt, primary care providers more frequently utilized the generative AI feature as compared to specialty care providers. There may be benefit to engineer AI prompts customized to each specialty to increase specialty provider utilization.

Speaker(s):
Kenneth Leung, MD, MS
UCI Health

Author(s):
Kenneth Leung, MD, MS - UCI Health; Danielle Perret, MD - UC Irvine Health; Emilie Chow, MD - University of California, Irvine; Charles Gilman, MBA - UC Irvine Health; Jennifer Botelho, RN, MSEd - UC Irvine Health; Deepti Pandita, MD, FACP, FAMIA - University of California Irvine;

Examining the impact of patient and room features on risk of hospital onset infection

Poster Number: P48
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Infectious Diseases and Epidemiology, Patient Safety, Education and Training
Primary Track: Foundations
Programmatic Theme: Clinical Informatics

Environmental factors including room surfaces and room features can contribute to the transmission of Clostridioides Difficile (C.diff) infection. Patients with similar conditions are often co-located within hospital units. These complex factors have potential to impact the risk of HO-CDI and necessitate exploration. In this study we wish to explore the association between room features and risk of HO-CDI, when accounting for room assignment factors.We conducted a nested case control study using retrospective EHR data of patients hospitalized at our center between January 2019 to April 2021. Cases were matched to controls (1:4 ratio) using incidence density sampling without replacement. In the final model, adjustments were made patient factors, previous diagnosis of C.diff, age and number of room transfers.Conditional multivariate logistic regression was employed to explore the association between the key variables. Protective effects were seen for rooms with cubical curtain near patient (OR=0.705, 95% CI=0.549-0.906) compared to those near door, rooms with separate shower units (OR=0.674, 95% CI=0.528-0.860) over those with a shower pan unit, and rooms with wall mounted toilets (OR=0.749, 94% CI=0.592-0.950) compared with floor mounted toilets. Moreover, rooms with manual paper towel dispensers had increased risk of infection (OR=1.334, 95% CI=1.053-1.691) over those with automatic towel dispensers.Results suggest possible association between specific room features and HO-CDI which should be further investigated with techniques such as environmental sampling. These results may inform room allocation practices and cleaning protocols to disinfect high-risk surfaces.

Speaker(s):
Priti Singh
The Ohio State University

Author(s):

Quality Metric Dashboard Implementation: A Case Report from a VA Inpatient Ward

Poster Number: P49
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Healthcare Quality, Patient Safety, Human-computer Interaction, Informatics Implementation, Qualitative Methods, Usability, Workflow
Primary Track: Applications
Programmatic Theme: Clinical Informatics

We evaluated the implementation of a novel quality metric dashboard in a single inpatient unit at a VA hospital. We identified barriers and facilitators to the dashboard’s use that we have mapped onto the BEhavior and Acceptance fRamework (BEAR) Model, including considerations of Environmental Context/Resources, Contingencies, Implementation Climate, Effort Expectancy, Performance Expectancy, Individuals Involved, and Social Influence.

Speaker(s):
Alvin Jeffery, PhD, RN
U.S. Department of Veterans Affairs

Author(s):
Christine Kimpel, PhD, RN - Vanderbilt University; Amy Guidera, PhD, RN/APN - U.S. Department of Veterans Affairs;

Automatic identification of brain metastasis development in patients with lung cancer using natural language processing

Poster Number: P50
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Large Language Models (LLMs), Natural Language Processing, Real-World Evidence Generation, Population Health
Primary Track: Foundations
Programmatic Theme: Clinical Informatics

Brain metastases (BM) lead to increased mortality. However, lack of data in cancer registries on BM limit real-world studies that could inform management. We developed a BERT-based NLP model to accurately detect BM in MRI reports of cancer patients. Aggregating report-level predictions, the model accurately identified patients with BM in a validation cohort and also revealed additional cases not recorded in the registry, demonstrating its potential in enhancing BM detection and facilitating epidemiological research.

Speaker(s):
Julie Wu, MD PhD
Palo Alto Veterans Affairs Healthcare System

Julie Wu


Author(s):
Aparajita Khan, PhD - Stanford University; Julie Wu, MD PhD - Veterans Affairs Health System Palo Alto; Chloe Su, PhD - Stanford University; June Corrigan, BS - Boston University School of Public Health; Rika Terashima, MD - Stanford University; Megan Chang, BS - Stanford University; Emily Rodriguez, BS - Stanford University; Christopher J. Shin, BS - Stanford University; Akash Shah, BS - Stanford University; Rakshit Kaushik, BS - Stanford University; Allison Kurian, MD - Stanford University; Heather Wakelee, MD - Stanford University; Curtis Langlotz, MD, PhD - Stanford University; Leah Backhus, MD - Stanford University; Michael Kelley, MD - VA Health System, Durham; Nathanael Fillmore, PhD - VA Boston Healthcare System; Summer Han, PhD - Stanford University;

Deidentifying Clinical Text: NLM Scrubber Error Analysis

Poster Number: P51
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Data Sharing, Natural Language Processing, Privacy and Security, Evaluation, Information Extraction
Primary Track: Applications
Programmatic Theme: Clinical Informatics

This poster presents an investigation into error patterns in deidentifying clinical patient notes using the NLM Scrubber. Analyzing 45 patient notes processed by the tool, alongside expert human evaluation, revealed specific limitations in the data scrubbing. Findings illuminate where the tool struggled to ensure complete deidentification, highlighting areas for potential improvement. Insights gained from this study provide groundwork for enhancing the accuracy of deidentification processes, crucial in safeguarding patient privacy in clinical data release.

Speaker(s):
Mohammad Arvan, PhD
University of Illinois at Chicago

Author(s):
Karl Kochendorfer, MD - University of Illinois; Shane Borkowsky, MD - University of Illinois Hospial; Aaron Chaise, MD - University of Illinois at Chicago; Bhrandon Harris, MD - Univeristy of Illinois at Chicago; Natalie Parde, Ph.D. - University of Illinois Chicago;

Evaluating the Reliability of Real-Time Prescription Benefit Tools: A Quality Assessment Study

Poster Number: P52
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Clinical Decision Support, Healthcare Economics/Cost of Care, Informatics Implementation, Internal Medicine or Medical Subspecialty, Usability, Transitions of Care, Education and Training
Primary Track: Applications
Programmatic Theme: Clinical Informatics

Real-Time Prescription Benefit (RTPB) prescriber tools remain underutilized, due to concerns surrounding its reliability. This retrospective cohort study demonstrates that 28% of patients had no pharmacy benefit manager (PBM) identified and/or selected, preventing the tool from functioning. Of the patients with a PBM identified and selected, 25% had partial or no prescription cost information returned. System improvements, provider education and further mandates for prescription plans to support RTPB tools, may increase RTPB tools' reliability.

Speaker(s):
Sharmila Tilak, MD
Brigham & Women's Hospital

Author(s):
Jeffrey Schnipper, MD, MPH - Brigham and Women's Hospital;

External Validation Infrastructures for AI in Healthcare Models

Poster Number: P53
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Evaluation, Deep Learning, Imaging Informatics, Machine Learning
Primary Track: Applications

Multiple commercial AI models have received approval from the US Food and Drug Administration (FDA), though not without concerns about generalizability across new clinical settings and diverse populations. It is imperative to evaluate the impact of these AI models on patient outcomes before their adoption in clinical settings. However, assessing the performance of AI algorithms through external validation studies is stymied by multiple challenges. AI vendors are hesitant to provide their algorithms, and medical institutions cannot share patient data. Moreover, outsourcing this work to enterprises providing AI evaluation services is expensive. To address these challenges, we propose ClExtVal.AI, a framework for developing robust cloud-based infrastructures for the external validation of AI algorithms that equips medical institutions to analyze the generalizability of AI models and their impact on patient safety and health equity. We configure the framework to securely store patient data within the medical institution. AI algorithms, along with their dependencies and license files aimed at controlling access to models, are shared as docker images. We demonstrate the reliability and customizability of our framework by integrating three commercial FDA-approved AI models for breast cancer detection. We are performing large-scale external validation of these models on a diverse dataset of 40,000 mammograms from the Breast Cancer Surveillance Consortium (BCSC). We will comprehensively assess the performance to understand their generalizability by holistically considering various factors, including tumor aggressiveness, demographic diversity, and geographic variations, to investigate hidden biases. Our work will enhance the scope for the clinical adoption of AI technologies.

Speaker(s):
Ojas Ankurbhai Ramwala, BTech
University of Washington, Seattle

Author(s):
Matthew Unrath, BS - Pariveda Solutions; Suzanne Kolb, MPH - University of Washington, Seattle; Christoph Lee, MD, MS, MBA - University of Washington, Seattle;

Leveraging Electronic Health Record Decision Support to Promote Social Risk Factors Screening and Follow-Up among Hospitalized Pediatric Patients

Poster Number: P55
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Health Equity, Clinical Decision Support, Pediatrics, Healthcare Quality, Workflow
Primary Track: Applications
Programmatic Theme: Clinical Informatics

Social risk factors are known to be critical to the health of pediatric patients and yet screening is infrequently done in the inpatient setting. Here, we will describe an improvement initiative to support screening of hospitalized pediatric patients using a vendor-provided EHR tool and development of multiple modes of EHR clinical decision support to foster screening and support resource referrals for at-risk patients. To date, we have observed an improvement in screening rates and resource referrals, but nearly 20% of patients are not screened and 50% of patients at-risk are still not provided with resources. Further analysis of engagement with the individual EHR tools that we developed may help to reveal those that are more efficacious as well as opportunities to optimize the decision support and enhance usability to further support our initiative.

Speaker(s):
Anne Fallon, MD
University of Rochester

Author(s):
Sonia Joshi, MD; Grace Ng; Matthew Present, MD - University of Rochester;

Enhanced Detection of Dementia in the Emergency Department Through Advanced Machine Learning Techniques

Poster Number: P56
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Machine Learning, Chronic Care Management, Clinical Decision Support, Bioinformatics, Healthcare Quality
Primary Track: Applications
Programmatic Theme: Clinical Informatics

We apply three machine learning methods (traditional, positive unlabeled learning, and active learning) using three types of models (Random Forest, XGBoost, and LASSO) to predict dementia among patients 65+ in the emergency department (ED). Our findings indicate that while traditional models show robust performance, incorporating novel learning strategies hold potential for improved identification of dementia cases in the ED, which is crucial for timely intervention.

Speaker(s):
Inessa Cohen, MPH
Yale University

Author(s):
Haipeng Xue, MS - Yale University; Inessa Cohen, MPH - Yale University; Richard Taylor, MD MHS - Yale University; Aidan Gilson - Yale School of Medicine; Isaac Faustino, MS - Yale University; Natalia Festa, MD, MHS - Yale University; James Lai, MD, MHS - New York University; Phillip Magidson, MD, MPH - Johns Hopkins University; Adam Mecca, MD, PhD - Yale University; Debra Tomasino, MA - NYU Grossman School of Medicine; Ula Hwang, MD, MPH - NYU Grossman School of Medicine;

The Future is in Sight–Using Computer Vision for Automated Objective Review of Laparoscopic Cholecystectomies

Poster Number: P57
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Surgery, Machine Learning, Teaching Innovation, Human-computer Interaction, Deep Learning, User-centered Design Methods
Working Group: Surgical and Procedural Informatics Working Group
Primary Track: Applications

Reviewing laparoscopic surgical videos is an underutilized educational opportunity to improve surgical skills. To address this, we developed a novel automated and objective computer vision (CV) model to review laparoscopic cholecystectomies (LC) and detect surgical steps and errors related to technical skills. In our surgical step-segmentation model, we obtained a video-level accuracy of 88%, a video-level F1 of 81%, and a frame-level F1 of 86%. Our CV model demonstrates accurate step-level recognition performance and has the potential to enhance surgical video review with an automated and objective analysis.

Speaker(s):
Joshua Villarreal, MD
Stanford Healthcare

Author(s):
Charlotte Egeland, MD PhD - Copenhagen University Hospital Rigshospitalet, Copenhagen, Denmark; Elaine Sui, MS - Stanford University; Anita Rau, PhD - Stanford University; Shelly Goel, MS - Stanford University; Josiah Aklilu, PhD candidate - Stanford University; Eric Soresnon, MD - Intermountain Health Department of General Surgery; Roger Bohn, PhD - University California San Diego; Teodor Grantcharov, MD PhD - Stanford University Department of General Surgery; Serena Yeung-Levy, PhD - Stanford University; Jeffrey Jopling, MD - Johns Hopkins School of Medicine Department of General Surgery;

Derm2Vec: Image-Text Alignment with Improved Dermatology Diagnosis Classification Robustness and Explainability

Poster Number: P58
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Natural Language Processing, Deep Learning, Diagnostic Systems
Working Group: Natural Language Processing Working Group
Primary Track: Applications
Programmatic Theme: Clinical Informatics

Skin diseases are among the most common reasons for medical consultations. Accurate automatic skin disease classification can help reduce workload on healthcare professionals, and improve patient outcomes. This study aims to develop a scalable method to address the heterogeneity, and improve the interpretability of skin disease classification. Our contributions include (1) a dermatology-centric, static, medical concept embedding (Derm2Vec), which is capable of clustering related dermatological concepts. (2)demonstrating a new method for dermatology image classification that utilizes vector similarities between image and static diagnosis embeddings, with improved +1.3% accuracy through pretraining on a different dataset. (3) showing the potential for better diagnosis interpretability through zero-shot symptom prediction.

Speaker(s):
Yujuan Fu, BSE
University of Washington

Author(s):
Yujuan Fu, BSE - University of Washington; Zhaoyi Sun, Master of Science - University of Washington; Wen-Wai Yim - Augmedix; Meliha Yetisgen, PhD - University of Washington;

Implementation of a Near Real-Time Report to Predict Risk of Carbapenem-Resistant Enterobacterales on Admission to an Acute Care Hospital

Poster Number: P59
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Data Mining, Infectious Diseases and Epidemiology, Informatics Implementation
Primary Track: Applications
Programmatic Theme: Clinical Informatics

Our study aimed to swiftly identify carbapenem-resistant Enterobacterales (CRE) carriers, crucial for effective infection control. Initially, we developed a prediction model using electronic health record (EHR) data, then enhanced it with medical comorbidities and antibiotic therapy data. Leveraging SQL queries, we created a near real-time report integrated into the EHR, providing daily risk scores for new admissions. Clinicians can use this tool to promptly identify CRE risk and implement appropriate measures. Our findings demonstrate the feasibility and utility of this approach in healthcare settings, warranting further exploration.

Speaker(s):
Chad Robichaux, MPH
Emory University

Author(s):
Barney Chan, MS - Emory University; Jessica Howard-Anderson, MD - Emory University; Chris Bower, MPH - Emory University; Radhika Asrani, MPH, MSc - Emory University;

Development of Myelodysplastic Syndrome Clinical Decision Support Tool

Poster Number: P60
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Clinical Decision Support, Usability, Evaluation
Primary Track: Applications
Programmatic Theme: Clinical Informatics

Medical conditions such as Myelodysplastic Syndrome (MDS) are challenging to diagnose due to its complexity and widely variable expertise in disease recognition and management. We developed a prototype to assist clinicians on appropriate MDS evaluation and management and use heuristic evaluation to further develop a MDS clinical decision support tool. We aim to analyze the content of the clinical decision support tool using Delphi method before its implementation.

Speaker(s):
John Muthu, Physician/MD
SUNY Downstate Health Sciences University

Author(s):
Mohammad Faysel, PhD - SUNY Downstate Health Sciences University; David Kaufman, PhD - SUNY Downstate Health Sciences University; Edwin Chiu, MD - SUNY Downstate Health Sciences University;

IASChain: A Blockchain Approach for Individual Access under TEFCA

Poster Number: P61
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Interoperability and Health Information Exchange, Personal Health Informatics, Data Sharing
Primary Track: Applications
Programmatic Theme: Clinical Informatics

TEFCA establishes a nationwide framework for interoperable health information exchange, with individual access as a key component. However, it confronts challenges such as patient mismatching, complex permit control, and difficulties in understanding data use. This study introduces IASChain, a blockchain-based approach that leverages the technology's decentralized, transparent, and immutable nature. It creates an anonymous patient identity, patient-centric permit control, and immutable usage audit, thereby enhancing security and clarity in individual access services within TEFCA’s framework.

Speaker(s):
Zhen Hou, MS
Indiana University

Author(s):
Yan Zhuang, Ph.D. - Indiana University;

Clinical Decision Support System for Automatic Detection of Subjective Pain Intensity from Gaze Data

Poster Number: P62
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Human-computer Interaction, Clinical Decision Support, Biomarkers, Machine Learning, User-centered Design Methods
Working Group: Clinical Decision Support Working Group
Primary Track: Applications
Programmatic Theme: Clinical Informatics

Pain, a significant health challenge, stands to benefit from sensor-based technologies, with eye movements showing promise as objective measures. However, traditional self-reported pain assessments lack objectivity, emphasizing the need for alternative measures. In this study, we aimed to develop a machine-learning engine to detect subjective pain intensity levels from physiological eye movements, potentially serving as a biomarker for pain. We conducted an eye-tracking experiment with 63 adults, 40 pain-free individuals, and 22 chronic pain individuals. machine learning techniques, including Random Forest and Neural Network models, we analyzed eye movement features such as fixations, visits, and saccade metrics. Our results demonstrated the ability of these features to capture nuances in attention related to pain experience, with the Random Forest model achieving an accuracy of 82% and the Neural Network model achieving 78%. These findings underscore the potential of eye movement metrics as objective measures for assessing subjective pain intensity.

Speaker(s):
Doaa Alrefaei, MS
Worcester Polytechnic Institute

Author(s):
Doaa Alrefaei, MS - Worcester Polytechnic Institute; Read Alharbi, PhD - Saudi Electronic University; Soussan Djamasbi, PhD - Worcester Polytechnic Institute; Diane Strong, PhD - Worcester PolytechnicInstitutes;

Enhancing Cervical and Breast Cancer Screening in Individuals with Severe Mental Illness through the Use of Electronic Health Records in a Canadian Mental Health Setting

Poster Number: P63
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Cancer Prevention, Healthcare Quality, Chronic Care Management
Primary Track: Applications
Programmatic Theme: Clinical Informatics

Women with severe mental illness have significantly lower rates of breast and cervical cancer screening, leading to increased poor health outcomes, including death. Over the last year, the Centre for Addiction and Mental Health (CAMH) has looked at implementing informatics tools to better deliver breast and cervical cancer screening to our long-stay inpatients. This poster presentation will focus on the digital supports that are being implemented to support this project and the findings and lessons learned from this work.

Speaker(s):
Tania Tajirian, MD
CAMH

Author(s):
Nasrin Adams, MD, PhD - Centre for Addiction and Mental Health; caroline chessex; Brian Lo, MHI - Centre for Addiction and Mental Health; Jennifer Nicolle, MD CCFP - Centre for Addiction and Mental Health; Cristina de Lasa; Sanjeev Sockalingam; Tania Tajirian, MD - CAMH;

Optimizing Emergency Department Imaging Orders with Machine Learning

Poster Number: P64
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Machine Learning, Large Language Models (LLMs), Imaging Informatics, Workflow, Clinical Decision Support
Primary Track: Applications
Programmatic Theme: Clinical Informatics

Background: Rising ED imaging volumes lead to longer radiology queues and delays in identifying critical findings.

Methods: We developed an LGBM model to predict the likelihood of actionable findings in ED imaging studies using triage data, medical history, orders, lab results, and vital signs from 228,839 imaging studies (2020-2023). We simulated ED imaging queues prioritizing studies with higher predicted probabilities of actionable findings.

Results: The LGBM model predicted actionable findings with an AUROC of 0.72 and AUPRC of 0.65. The LGBM-prioritized queue significantly reduced the mean time from order to imaging completion for actionable cases compared to the control queue (45.1 mins vs 77.5 mins; p<0.05).

Conclusion: Machine learning can estimate the pre-test probability of actionable findings in ED imaging studies, enabling prioritization of studies likely to yield management-altering results. This approach can optimize ED imaging workflows, mitigate adverse effects of overcrowding, and improve patient outcomes.

Speaker(s):
Kevin Tang, BS
Albert Einstein College of Medicine

Author(s):
Kevin Tang, BS - Albert Einstein College of Medicine; Julia Reisler, BS - Stanford University; Christian Rose, MD - Stanford University, School of Medicine; Brian Suffoletto, MD - Stanford University; David Kim, MD PhD - Stanford University;

Leveraging a Provider Builder Program for Note Template Optimization to Reduce Documentation Burden and Note Bloat

Poster Number: P65
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Documentation Burden, Pediatrics, Education and Training
Primary Track: Applications
Programmatic Theme: Clinical Informatics

High documentation burden and “note bloat” contribute to provider burnout and poor note readability. Partnering
with provider builders, we implemented a structured program for outpatient note template optimization. Our early
experience showed consistent decrease in note length and variable effects on documentation time.

Speaker(s):
Hyunjung Shin, MD
Emory University / Children's Healthcare of Atlanta

Author(s):
Kelly Rouster-Stevens, MD - Emory University/Children's Healthcare of Atlanta; Kristina Cossen, MD - Emory University / Children's Healthcare of Atlanta; Wayne Liang, MD MS FAMIA - Children's Healthcare of Atlanta & Emory University; Evan Orenstein, MD - Childrenís Healthcare of Atlanta;

Clinical Decision Support to Improve Appropriate Medication Dosing for Patients with Decreased Kidney Function

Poster Number: P66
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Clinical Decision Support, Clinical Guidelines, Pediatrics
Primary Track: Applications
Programmatic Theme: Clinical Informatics

Renally cleared medications require dose adjustment when patients have decreased kidney function, measured as
estimated creatinine clearance (CrCl). The indication for and the degree of dose adjustment depends on the extent of CrCl reduction. However, reduced CrCl is not easily recognized, especially in pediatric patients, where complex
calculations are required to determine the CrCl. Among hospitalized patients with decreased kidney function, we
found that 35% of orders for amoxicillin, cefepime, meropenem, and piperacillin/tazobactam were too high for their
CrCl.

Speaker(s):
Hyunjung Shin, MD
Emory University / Children's Healthcare of Atlanta

Author(s):
Thomas Dawson, PharmD - Children's Healthcare of Atlanta;

Informatics Challenges in the Study of ADL Compliance

Poster Number: P67
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Information Retrieval, Healthcare Quality, Nursing Informatics, Data Transformation/ETL
Primary Track: Applications
Programmatic Theme: Clinical Informatics

Throughout the 6 years of longitudinal study data collection, we encountered several informatics challenges, including data location changes, data value changes, poor data reliability, and poor data accuracy. Studies relying on electronic health record (EHR) data, especially studies spanning several years, must be aware not only of the challenges in data location changes, value changes, reliability, and accuracy but also of how those issues may increase as the length of the study increases.

Speaker(s):
John Huber, MS
University of Cincinnati Department of Biomedical Informatics

Author(s):
John Huber, MS - University of Cincinnati Department of Biomedical Informatics; Christopher Dandoy, MD, MS - Cincinnati Children’s Hospital Medical Center; Judith Dexheimer, PhD - Cincinnati Children's Hospital;

Using equity X design to conceptualize, develop, and test cancer symptom self-management mobile app for people in rural areas

Poster Number: P68
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Diversity, Equity, Inclusion, Accessibility, and Health Equity, Mobile Health, Self-care/Management/Monitoring
Working Group: Nursing Informatics Working Group
Primary Track: Applications
Programmatic Theme: Clinical Informatics

Human-centered design approaches that intentionally center equity are needed to mitigate the root causes of inequity and create inclusive and sustainable interventions. equity X design is an approach that recognizes existing inequities in healthcare and targets the social determinants of health. For people with cancer living in rural areas, access inequity significantly impacts both cancer outcomes and quality of life. The purpose of this project is to describe our process of conceptualizing, developing, and prototyping a symptom self-management intervention for people with cancer living in rural areas delivered via mobile app called OASIS (i.e., Oncology Associated Symptoms & Individualized Strategies). We engaged stakeholders, articulated our individual identities, operationalized the active elements of OASIS, iteratively tested low-fidelity and high-fidelity prototypes, and envisioned successful process and clinical outcomes. By embracing equity X design, clinicians and scientists can develop interventions promote equitable and just healthcare for individuals, families, and communities.

Speaker(s):
Stephanie Gilbertson-White, PhD, APRN-BC, FAAN
University of Iowa, College of Nursing and Internal Medicine

Author(s):
Heath Davis, MS, MLIS, FAMIA - University of Iowa, Carver College of Medicine;

Natural Language Processing Models to Confirm Monoclonal Gammopathy of Undetermined Significance and Progression Diagnoses in Veterans’ Electronic Health Records

Poster Number: P69
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Natural Language Processing, Cancer Prevention, Machine Learning
Primary Track: Applications
Programmatic Theme: Clinical Informatics

We developed and evaluated 5 natural language processing (NLP) classification models (1 rule-based, 4 machine learning-based) to confirm diagnoses of monoclonal gammopathy of undetermined significance (MGUS) and related progression in Veterans' electronic health record data. The rule-based model achieved the best classification accuracy for MGUS, while the machine learning-based model using support vector machines performed best for progression. In conclusion, the best performing NLP classification model for disease confirmation depends on data characteristics, e.g., classification balance.

Speaker(s):
Mei Wang, MS
Washington University in St. Louis

Author(s):
Yao-Chi Yu, PhD - Washington University in St. Louis; Lawrence Liu, MD - City of Hope; Martin Schoen, MD - Saint Louis Univeristy; Graham Colditz, MD, DrPH - Washington University in St. Louis; Theodore Thomas, MD - Washington University in St. Louis; Su-Hsin Chang, PhD - Washington University in St. Louis;

Gap Analysis: Mapping Emergency Department Data to a Common Data Model for Opioid-Use Disorder Surveillance

Poster Number: P70
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Data Standards, Data Transformation/ETL, Informatics Implementation, Clinical Decision Support, Health Equity
Primary Track: Applications
Programmatic Theme: Clinical Informatics

We performed a gap analysis to assess the feasibility of mapping electronic health record data from the Clinical Emergency Data Registry (CEDR) to the Observational Medical Outcomes Partnership (OMOP) Common Data Model (CDM). While most CEDR fields could be mapped to OMOP, challenges remain in handling protected health information and addressing complex data transformations. This highlights the importance of standardized data models in enhancing opioid surveillance efforts.

Speaker(s):
Inessa Cohen, MPH
Yale University

Author(s):
Inessa Cohen, MPH - Yale University; Richard Taylor, MD MHS - Yale University; Zihan Diao, AB - Yale University; Pawan Goyal, MD, MHA, FAMIA, FHIMSS, PMP - American College of Emergency Physicians; Kathryn Hawk, MD, MHD - Yale University; Bill Malcom, PMP - American College of Emergency Physicians (ACEP); Caitlin Malicki, MPH - Yale University; Brian Sweeney, A&P - American College of Emergency Physicians (ACEP); Dhruv Sharma, MS - American College of Emergency Physicians; Scott Weiner, MD, MPH - Brigham and Women’s Hospital; Arjun Venkatesh, MD, MBA, MHS - Yale University;

ADE Identification in EHR Data: Weight Gain Among Pediatric Patients Diagnosed with ASD and Treated with Atypical Antipsychotic Medications

Poster Number: P71
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Patient Safety, Drug Discoveries, Repurposing, and Side-effect, Pediatrics
Primary Track: Applications
Programmatic Theme: Clinical Informatics

Observational EHR data has yet to be broadly leveraged for pharmacovigilance and prior work has not focused on on continuous outcomes. Using the example of atypical antipsychotics (ATAP)-induced weight gain among pediatric patients, this exploratory study highlights how EHR data can be used to better understand adverse drug events in vulnerable populations less commonly included in drug trials. Such insight could better support patients, parents, and providers in balancing benefits and risks of ATAP therapy.

Speaker(s):
Sharon Davis, PhD
Vanderbilt University Medical Center

Author(s):
Joshua Smith, PhD - Vanderbilt University Medical Center;

Clustering and Summarizing Cancer Clinical Trial Eligibility Criteria

Poster Number: P72
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Natural Language Processing, Cancer Prevention, Large Language Models (LLMs)
Primary Track: Applications
Programmatic Theme: Clinical Informatics

Clinical trials are vital for medical advancement but reviewing their protocols is demanding and requires medical expertise. Our goal is to streamline this by creating a patient-focused chatbot that queries patients on trial criteria to match them with suitable trials. The first step includes identifying common criteria among thousands of breast cancer trial eligibility criteria. Our approach uses SBERT for embeddings, HDBSCAN for clustering, BERTopic, and GPT-4 for text representation. We used our previous annotated data to evaluate the clustering purity and entropy. Our result shows the effectiveness and feasibility of automated cluster clinical trial eligibility criteria.

Speaker(s):
Yumeng Yang, MS
The University of Texas Health Science Center at Houston (UTHealth) School of Biomedical Informatics

Author(s):
Yumeng Yang, MS - The University of Texas Health Science Center at Houston (UTHealth) School of Biomedical Informatics; Kirk Roberts, PhD - University of Texas Health Science Center at Houston;

Agreement of Electronic Health Record and Medicare Claims Data for Amyotrophic Lateral Sclerosis Patients

Poster Number: P73
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Data Standards, Data Transformation/ETL, Real-World Evidence Generation
Primary Track: Foundations
Programmatic Theme: Clinical Informatics

Researchers used different approaches to assess the quality of real-world data (RWD) that detect and address data errors to produce reliable real-world evidence. We used the data source agreement method to assess the physical existence of data points in electronic health records versus claims data for amyotrophic lateral sclerosis patients. Our preliminary results showed disagreement between both databases in multiple data elements, which warranted more analyses to localize the exact underlying reason for the disagreement.

Speaker(s):
Yahia Mohamed
UMKC

Author(s):
Yahia Mohamed - UMKC; Lemuel Waitman, PhD - University of Missouri; Xing Song, PhD - University of Missouri; Tamara McMahon - UMKC School of Medicine; Suman Sahil - University of Missouri-Kansas City School of Medicine;

A support-vector-machines-based natural language processing model to confirm multiple myeloma diagnosis in electronic health records

Poster Number: P74
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Natural Language Processing, Machine Learning, Cancer Prevention
Primary Track: Applications
Programmatic Theme: Clinical Informatics

With advancements in natural language processing (NLP), we developed a support-vector-machine-based NLP model to effectively confirm the diagnoses of multiple myeloma (MM) using Veterans’ electronic health records. The model achieved an AUC of 0.943 in the testing set and was able to detect dates of MM diagnosis with 81.8% matched gold standard within 30 days. The top 5 important features were identified. We demonstrate a viable method to replace manual chart review for future research.

Speaker(s):
Mei Wang, MS
Washington University in St. Louis

Author(s):
Yao-Chi Yu, PhD - Washington University in St. Louis; Lawrence Liu, MD - City of Hope; Martin Schoen, MD - Saint Louis University; Graham Colditz, MD, DrPH - Washington University in St. Louis; Theodore Thomas, MD - Washington University in St. Louis; Su-Hsin Chang, PhD - Washington University in St. Louis;

Enhancing Active Learning for Annotation Sampling over Large-Scale Corpus via Vector-Based Indexing

Poster Number: P75
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Natural Language Processing, Data Mining, Information Retrieval
Working Group: Natural Language Processing Working Group
Primary Track: Applications
Programmatic Theme: Clinical Informatics

Active learning offers a path to efficient NLP model development by strategically selecting informative data for annotation. However, traditional methods struggle with large clinical datasets due to the computational cost of processing the entire corpus in each sampling round. This work proposes a novel vector-based active learning approach for clinical note sampling for annotation. Our preliminary experiments demonstrated that this approach offers a promising solution.

Speaker(s):
Jianlin Shi, MS, MD
The Division of Epidemiology, School of Medicine, University of Utah; VA Salt Lake City Healthcare System

Author(s):

Development of a Generative Artificial Intelligence Data Pipeline to Automate the Capture of Unstructured MRI Data for Prostate Cancer Care

Poster Number: P76
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Data Mining, Natural Language Processing, Surgery
Primary Track: Applications
Programmatic Theme: Clinical Informatics

While many web-based LLM tools are available, they require manual entry of each prompt and are not approved for use with protected health information (PHI). We developed a detailed prompt instructing an internal HIPAA-compliant LLM to extract specified data elements when presented with prostate MRI reports. Accuracy ranged from 53.4%-90.0%. LLMs can be used to flexibly extract text data from imaging reports with high accuracy and low up-front programming requirements.

Speaker(s):
Anobel Odisho, MD, MPH
University of California, San Francisco

Author(s):
Anobel Odisho, MD, MPH - University of California, San Francisco; Andrew Liu, MD - University of California, San Francisco; William Pace, BA - UCSF; Marvin Carlisle, BA - UCSF; Robert Krumm, BA - UCSF; Janet Cowan, MS - UCSF; Peter Carroll, MD - UCSF; Matthew Cooperberg, MD - UCSF;

Phenotyping Depressive Disorders using the All of Us Research Program

Poster Number: P77
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Phenomics and Phenome-wide Association Studies, Reproducibility, Usability
Primary Track: Applications
Programmatic Theme: Clinical Informatics

This study assesses the generalizability of a rule-based phenotyping algorithm for Major Depressive Disorder (MDD) using electronic health records. Cohort analysis reveals demographic consistencies with existing literature. Healthcare utilization and medication distribution patterns align with prior findings, except for the PsychRx phenotype, which warrants further investigation. Preliminary findings indicate the algorithm is generalizable and exhibits promise for enhancing reproducible research and informing clinical decision-making in the realm of mental health.

Speaker(s):
Nick Souligne, M.Eng
University of Arizona

Author(s):
Nick Souligne, M.Eng - University of Arizona; Vignesh Subbian, PhD - University of Arizona;

Lab Labeler: Harmonizing diverse classifications & lessons learned

Poster Number: P78
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Large Language Models (LLMs), Data Transformation/ETL, Usability
Primary Track: Applications

LLMs allow for efficient labeling of extensive datasets, which is a challenge across medicine. Thus, we developed a customizable tool to classify mislabeled labs according to institutional LIS categories. It was tested at UC Irvine with promising results. Through prompt engineering on a sample of 100, we created a custom GTP model that categorized ~5000 labs into corresponding categories with a correlation rate of 96%, reducing the manual labeling time from weeks to hours.

Speaker(s):
Monil Patel, MD
UC Irvine Health

Author(s):
Jacob Franklin, M.D. - VUMC;

Continuous Prediction of Emergency Department Disposition with Multimodal Deep Learning

Poster Number: P79
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Deep Learning, Clinical Decision Support, Information Extraction, Machine Learning, Precision Medicine
Primary Track: Applications
Programmatic Theme: Clinical Informatics

Background: Emergency Department (ED) physicians face challenges in rapidly determining patient disposition, leading to prolonged stays and overcrowding. This study aims to improve disposition accuracy and reduce decision-making time using static and continuous patient information.

Methods: We developed a transformer-based model using a multimodal dataset from 169,244 ED visits (2020-2023), incorporating triage data, orders, continuous vitals, lab results, and radiology results to predict discharge, hospital admission, or ICU admission throughout a patient's ED stay.

Results: Our model achieved an Area Under the Receiver Operating Characteristic (AUROC) curve of 0.80 at triage time and 0.91 at final disposition time.

Conclusion: A multimodal machine learning model for continuous prediction of patient disposition in the ED has the potential to assist providers in real-time management of patient flow, improving efficiency and patient outcomes.

Speaker(s):
Kevin Tang, BS
Albert Einstein College of Medicine

Author(s):
Kevin Tang, BS - Albert Einstein College of Medicine; Julia Reisler, BS - Stanford University; Tom Jin, MS - Stanford University; David Kim, MD PhD - Stanford University;

Fitness of EHR Data for Secondary Use

Poster Number: P80
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Informatics Implementation, Usability, Information Visualization
Primary Track: Applications
Programmatic Theme: Clinical Informatics

Not much is known about the fitness of use of electronic health record data for secondary use at larger integrated health systems. We used the data completeness dimensions defined by Weiskopf et al. to assess the fitness of EHR data for secondary uses at Ochsner health system. This work provides valuable insights into the degree of data completeness over time and has the potential to inform future efforts to develop data completeness guidelines for secondary use cases. This work highlights the extent of data completeness at an integrated health system over a period of 8 years. Increased adoption of EHRs in the last twelve years has increased the documentation and breadth of data in the EHR. However, data density, after temporal and linear adjustments as described by Sperrin et al., has not changed very much over the years and may be the limiting factor when determining the amount of “usable” data for secondary use.

Speaker(s):
Meenakshi Mishra, PhD, MPH, MS
Ochsner Health

Author(s):

Knowledge Augmented Causal Discovery: Application in Chronic Low Back Pain Through Large Language Models and Knowledge Graphs

Poster Number: P81
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Large Language Models (LLMs), Knowledge Representation and Information Modeling, Causal Inference
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics

Causal discovery can help understand causes and health outcomes in complex diseases such as chronic low back pain, but determining constraints derived from knowledge expertise in large datasets is challenging. This study explores using large language models (LLMs) and knowledge graphs (KG) to augment domain knowledge during the causal discovery process. Compared to a causal model by experts, the data-driven algorithm finds fewer associations, suggesting the data lacks robust signals for causative connections considered by the experts. A KG model offers a structured approach to distilling insights from diverse literature sources and could reduce false positives and negatives in future structural learning algorithms.

Speaker(s):
Abel Torres Espin, PhD
University of Waterloo

Author(s):
Abel Torres Espin, PhD - University of Waterloo; Conor O'Neill, MD - University of California San Francisco; Paul Anderson, PhD - California Polytechnic State University;

Elevating the Precision Oncology Data Repository (PODR) to an Enterprise Solution to Support the Cancer Moonshot

Poster Number: P83
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Data Sharing, Imaging Informatics, Omics (genomics, metabolomics, proteomics, transcriptomics, etc.) and Integrative Analyses, Information Retrieval, Information Extraction, Information Visualization
Primary Track: Foundations
Programmatic Theme: Clinical Research Informatics

The Precision Oncology Data Repository (PODR) built within the Department of Veterans Affairs (VA) supports multiple programs, including the Cancer Moonshot program1. The upgraded version of PODR; PODR-DTS (Data Tracking System) can report data aggregation metrics and quickly respond to queries such as: ‘For how many patients with non-small cell lung cancer, stage 3 and up, a KRAS or EGFR mutation do you have CT scans?’

Speaker(s):
Antwan Baker, MS Computer Science
The Department of Veterans Affairs

Author(s):
Allie Skahen, MS - The Department Of Veterans Affairs; Danne Elbers, PhD - VA Boston CSP / MAVERIC;

The Proliferation of Virtual Care Technologies in the Veterans Health Administration: A Portfolio Review of Related Health Services Research

Poster Number: P84
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Telemedicine, Mobile Health, Information Extraction
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics

The quantity and complexity of virtual care modalities offered within the Veterans Health Administration (VHA) over the past two decades has resulted in a diverse portfolio of related research. Portfolio reviews conducted by the VHA’s consortia of research are collecting key information about studies examining virtual care use and/or effectiveness. Portfolio reviews may inform how VHA can better understand knowledge gaps and identify future funding opportunities, such as those focused on virtual care.

Speaker(s):
Nicholas McMahon
Department of Veterans Affairs

Author(s):

Statewide Claims Data: Lessons Learned in Investigating Use of Fertility Preservation Services in Rhode Island

Poster Number: P85
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Healthcare Economics/Cost of Care, Healthcare Quality, Clinical Guidelines
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics

Rhode Island (RI) became the first state to pass a mandate requiring commercial insurance plans to pay for fertility preservation (FP) procedures prior to gonadotoxic treatments. To investigate the impact of this mandate on use of FP services statewide, computational approaches were developed to identify cohorts of incident breast cancer cases in reproductive-aged females from the All-Payers Claims Database in RI. The challenges of inferring clinical practices from claims data underscore its uses and limitations.

Speaker(s):
Farahnaz Maroof, M.S.
Brown University

Author(s):
Farahnaz Maroof, M.S. - Brown University; Karen Crowley, PhD - Brown University - Brown Center for Biomedical Informatics; Elizabeth Chen, PhD - Brown University; May-Tal Sauerbrun-Cutler, M.D. - Department of Obstetrics and Gynecology, Brown University, Providence, RI;

Utility of Named Entity Recognition and Large Language Models in Cancer Research Informatics: Complimentary Application of Prodigy and LLAMA2

Poster Number: P86
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Deep Learning, Large Language Models (LLMs), Natural Language Processing
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics

Named Entity Recognition (NER) and Large Language Models (LLMs) can be used to extract critical cancer-related data from clinical notes. Using Prodigy and LLAMA2 platforms, we achieved high accuracy in identifying tumor biomarkers, cancer subtypes, and environmental exposures across several cancer types. NER and LLM approaches complement each other, when applied creatively to different use cases. Future enhancements include standardizing and integrating data outputs into broader research frameworks.

Speaker(s):
Ferris Hussein, Computer Science
NYU Langone Health

Author(s):
Rimma Belenkaya, MS, MA - NYU Langone; Jessica Sommer, BS - NYU Langone Health;

Validation of Social Determinants of Health Extraction in Oropharyngeal Cancers

Poster Number: P87
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Natural Language Processing, Cancer Prevention, Information Extraction
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics

Social Determinants of Health (SDoH) factors impact cancer outcomes. 70% of Oropharyngeal Cancers (OPC) are due to HPV, which in turn is sexually transmitted. In this work, we apply SODA, an existing BERT-based natural language processing (NLP) pipeline to extract SDoH factors in an OPC cohort and evaluate its extraction of sexual activity status using a bag of words model. Our bag of words model achieves an F1 score of .97 in predicting SODA's labels.

Speaker(s):
Protiva Rahman, PhD
University of Florida

Author(s):
Shama Karanth, PhD - University of Florida; Jiang Bian, PhD - University of Florida;

Global Disparities in Suicide Research Using Natural Language Processing

Poster Number: P88
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Global Health, Natural Language Processing, Health Equity
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics

We examined the gaps in suicide research between Low- and Middle-Income Countries (LMICs) and High-Income Countries (HICs) throughout history by employing NLP as well as latent themes in LMIC articles.

Speaker(s):
Hayoung Donnelly, Ph.D.
University of Pennsylvania

Author(s):
Danielle Mowery, PhD, MS, MS, FAMIA - University of Pennsylvania; Gregory Brown, Ph.D - University of Pennsylvania; Michael Steinberg, M.A. - University of Pennsylvania;

Training Longformer Models to Infer Surgery Outcomes in Inflammatory Bowel Disease Patients

Poster Number: P89
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Natural Language Processing, Deep Learning, Machine Learning
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics

This poster presents a work-in-progress on developing a neural network based model to predict surgery outcomes in patients with inflammatory bowel diseases (IBD), based on information available in just the clinical notes. The poster summarizes the results of two Longformer models trained on all specialty notes and just gastroenterology notes, respectively, and evaluates them as a retrospective cohort study. Results show improved performance on models trained over gastroenterology notes alone.

Speaker(s):
V.G.Vinod Vydiswaran, Ph.D.
University of Michigan

Author(s):
V.G.Vinod Vydiswaran, Ph.D. - University of Michigan; Deahan Yu, MHI; Jiazhao Li, MS - University of Michigan; Ryan Stidham, MD, MS - University of Michigan;

Machine Learning Enabled Dementia Risk Prediction Using Longitudinal Cardiovascular Risk Factors

Poster Number: P90
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Data Mining, Biomarkers, Population Health
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics

Prediction models that incorporate cardiovascular risk factors are well established but demonstrate limited validity. The incorporation of longitudinal cardiovascular risk factors may be an avenue to improve predictive accuracy but has yet been explored. We developed sets of machine learning enabled cross-sectional and longitudinal prediction models on pooled cohort data and found that adding change rates of time-variant risk factors provides statistically significant but limited improved in prediction performance.

Speaker(s):
Jingzhi Yu, BA
Northwestern University Feinberg School of Medicine

Author(s):
John Stephen, MS - Northwestern University; Denise Scholtens, PhD - Northwestern University; Lucia Petito, PhD - Northwestern University; Norrina Allen, PhD, MPH - Northwestern University;

Use of value sets to filter FHIR based data for desired clinical concepts

Poster Number: P91
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Controlled Terminologies, Ontologies, and Vocabularies, Chronic Care Management, Clinical Guidelines, Information Extraction, Internal Medicine or Medical Subspecialty
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics

Care management is challenging for patients with multiple chronic conditions, who frequently see providers in multiple care settings. Progress has been made towards interoperability across care teams, but there has been little research investigating the impacts on patient care. We are using the eCarePlanner application to integrate patient data across multiple care settings and investigate treatment efficacies for hypertension and chronic kidney disease in these patients. A significant challenge in effectively analyzing this large volume of patient records is finding ways to quickly identify the most relevant data points.

Value sets are collections of structured data codes designed to represent a comprehensive grouping of a medical condition or concept; they are available for download through the Value Set Authority Center (VSAC), a large repository of value sets submitted by independent organizations. Value sets can be used to create a catalog or superset of all codes relating to a particular concept.

We have identified about 50 general concepts of interest relating to hypertension treatment and chronic kidney disease, including comorbid conditions, medication reactions, and other complications. We then queried VSAC and reviewed and downloaded the most relevant value sets for each concept, consisting of ICD10, SNOMED, and LOINC codes. The value sets will be implemented as filters in the eCarePlanner application, in order to quickly identify relevant data. The goal is to illuminate external data in the patient’s health record, which may have altered treatment decisions and outcomes, if it was immediately available to all providers.

Speaker(s):
Michelle Bobo, BA, MS
Oregon Health & Science University

Author(s):

Feasibility Pilot of Linking Patient-Reported Outcomes Collected in Community Pharmacies with Electronic Health Record Data

Poster Number: P92
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Interoperability and Health Information Exchange, Information Retrieval, Patient / Person Generated Health Data (Patient Reported Outcomes)
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics

Community pharmacies provide a novel setting for collection of patient-reported outcomes (PROs) but are not included in health information exchange (HIE) networks. This study attempted to link PROs data collected in pharmacies to data from a statewide HIE. Enrollment included 13 patients in 3 pharmacies representing 3 counties. Successful linkage between sources occurred for all patients, demonstrating data can be successfully linked between community pharmacies and HIEs.

Speaker(s):
Heather Jaynes
Purdue University, College of Pharmacy

Author(s):
Margie Snyder, PharmD, MPH, FCCP - Purdue University; Molly Nicols, PharmD - Purdue University; Katelyn Hettinger-Riddell, PharmD - Purdue University; Beverly Musick, MS - Indiana University; Susan Perkins, PhD - Indiana University;

An Evaluation of Large Language Models for the Retrospective Identification of Surgical Site Infections

Poster Number: P93
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Surgery, Machine Learning, Natural Language Processing, Large Language Models (LLMs)
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics

In this study, we propose a framework for utilizing machine learning methods for the retrospective identification of surgical site infections. We trained and tested 3 standard machine learning models and performed 0-shot prompt engineering using 3 LLMs based on a cohort of labeled EHR notes. All models showed potential for SSI identification, but LLMs did not demonstrate improved performance over standard models.

Speaker(s):
Parker Evans, MD
Vanderbilt University Medical Center

Author(s):
Parker Evans, MD - Vanderbilt University Medical Center; Cosmin Bejan, PhD - Vanderbilt University Medical Center;

Creating validated Phenopackets from Tabular Data

Poster Number: P94
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Bioinformatics, Interoperability and Health Information Exchange, Data Standards, Data Mining, Data Sharing
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics

This poster introduces a novel software tool aimed at transforming healthcare data from tabular formats like CSV and Excel into phenopackets. Phenopackets are a structured form of healthcare data representation that enhances the computability and comparability of clinical information by including detailed modules on phenotype, genotype, disease, patient information, and metadata. The tool streamlines the conversion process through a user-friendly workflow that involves preprocessing, parsing values into standardized codes from international ontologies, assembling the mapped information into Phenopackets, and validating and writing the data in JSON format. The resultant phenopackets facilitate precise data analysis, offering compatibility with a range of applications such as LIRICAL, Exomiser, and SvAnna, and ensuring data is unambiguous and reusable. Although proficiency in Python is required, the tool's design minimizes the necessary programming input by abstracting complex functions. Future developments aim to enhance accessibility for clinicians with limited programming expertise, potentially enhancing data analysis in specialized healthcare centers.

Speaker(s):
Jan-Filip Rehburg, Research Fellow
Berlin Institute of Health @ Charite

Author(s):
Jan-Filip Rehburg, Research Fellow - Berlin Institute of Health @ Charite; Daniel Danis, Ph.D. - The Jackson Laboratory for Genomic Medicine; Peter Robinson, Prof., MD - Berlin Institute of Health at Charité; Sylvia Thun, Prof., MD - Berlin Institute of Health at Charité;

Toward Granular Social Determinants of Health (SDoH) Coding: A Semantics AI Framework to Extract and Encode SDoH enabled by Large Language Models

Poster Number: P95
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Large Language Models (LLMs), Natural Language Processing, Information Extraction, Controlled Terminologies, Ontologies, and Vocabularies
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics

We propose a framework that leverages Social Determinants of Health (SDoH) ontological knowledge to guide a large language model (specifically GPT-4) to identify SDoH elements from free-text electronic health record notes and map them to standard granular SDoH concepts. Here, we leverage Intelligent Medical Objects (IMO) coding of granular SDoH content collected through real world patient problem list. First, given a note and broad SDoH categories, we prompt GPT-4 to identify all SDoH elements and classify each element to a broad category along with providing a rationale and the evidence note text. Subsequently, we prompt the model to select final granular concepts given its previous response. Our framework achieved an F1-score of 90.18 in identifying SDoH elements, when evaluated on 30 MIMIC-III notes. Among the correctly identified elements, the accuracies in classifying the broad SDoH category and mapping to granular IMO concepts were 86.63% and 95.45%, respectively.

Speaker(s):
Surabhi Datta, PhD
Intelligent Medical Objects

Author(s):
Surabhi Datta, PhD - Intelligent Medical Objects; Hunki Paek, PhD - Intelligent Medical Objects; Kyeryoung Lee, PhD - Intelligent Medical Objects; Liang-Chin Huang, PhD - Intelligent Medical Objects; Jingqi Wang, PhD - Intelligent Medical Objects; Xiaoyan Wang, PhD - Intelligent Medical Objects;

Safety Evaluation Metrics of LLM-Powered Mental Health Chatbot

Poster Number: P96
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Large Language Models (LLMs), Evaluation, Standards
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics

Our study emphasizes the development of comprehensive evaluation metrics for assessing the clinical safety of Large Language Models (LLMs) in mental health care, addressing the gap in current methodologies that often prioritize technical performance over safety and ethical considerations. We proposed a collaborative framework involving mental health and LLM experts to enhance the reliability of health chatbots.
A two-phase evaluation process was established, beginning with the creation of 100 benchmark questions reflecting real-life clinical scenarios, followed by an expert review using five guideline questions in the following areas: adherence to practice guidelines, health risk management, consistency in critical situations, assessment of resource provision, and empowerment of users in managing their health. This method was applied to assess ChatGPT-3.5's responses across various clinical situations, resulting in an overall average score of 7.2 from mental health experts.
The findings highlight the need for a standardized approach to chatbot evaluation. Our framework lays the groundwork for future work in developing metrics for accuracy, empathy, and privacy, aiming for the responsible integration of chatbots into healthcare and building trust among users and professionals.

Speaker(s):
Jung In Park, PhD, RN, FAMIA
UC Irvine

Author(s):
Jung In Park, PhD, RN, FAMIA - UC Irvine;

UMLS API Functionality Comparison

Poster Number: P97
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Controlled Terminologies, Ontologies, and Vocabularies, Interoperability and Health Information Exchange, Privacy and Security, Terminology Systems, Usability, User-centered Design Methods, Informatics Implementation, Information Retrieval
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics

This poster explores the utilization of the Unified Medical Language System (UMLS) through application programming interfaces (APIs) for biomedical informatics. Three API categories were tested, comparing Postman API samples, HTML directory paths, and older code methodologies. Results showed faster responses with newer API code, while older methods offer increased security and customization. A functionality tool book is proposed to streamline API access and understanding. Further technical evaluations are suggested to determine the most suitable methods for different settings.

Speaker(s):
Tamer Khatib, Undergraduate Student
Department of Department of Biomedical and Health Information Sciences, University of Illinois Chicago, Chicago, Illinois, USA

Andrew Boyd, MD
University of Illinois at Chicago

Author(s):
Andrew Boyd, MD - University of Illinois at Chicago; Tamer Khatib, Undergraduate - Department of Department of Biomedical and Health Information Sciences, University of Illinois Chicago, Chicago, Illinois, USA;

OMOP for an Academic Medical Research Insititution

Poster Number: P98
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Terminology Systems, Usability, User-centered Design Methods, Standards, Teaching Innovation, Interoperability and Health Information Exchange, Educational Collaboration, Data Standards
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics

Clinical informaticists face several challenges when leveraging medical record data, including spurious data resulting from non-standard terminologies and inconsistent documentation. Common data models, such as OMOP, can improve the accuracy of data extraction but introduce other complications such as rigid conventions and additional jargon. To better serve researchers who are unfamiliar with OMOP, we have built tools to take advantage of the strengths of OMOP, while minimizing its challenges for our OMOP-naive researchers.

Speaker(s):
Kelli Hodge, MS, RN
Health Data Compass

Author(s):

Phenome Wide Association Study on Veterans with Lung Cancer Diagnosis Based on Smoking Status

Poster Number: P99
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Bioinformatics, Information Visualization, Disease Models
Primary Track: Foundations
Programmatic Theme: Clinical Research Informatics

Lung Cancer disproportionally effects veterans in the United States. This motivates recent research in survival trends in for Veterans with Lung Cancer. Utilizing Phe-codes, we conducted a Phenotype wide association study (PheWAS) on 80,491 veterans with lung cancer based on smoking status recorded in the electronic health record. The study concluded that patients in the cohort with a recorded smoking status are much more likely to have hypertension, lipid metabolism disorders, and hyperlipidemia.

Speaker(s):
Christopher Guardo, BS in Mathematics
Vanderbilt University Medical Center

Author(s):
Bhavnisha Patel, MSPS - Vanderbilt University Medical Center; Robert Winters, BA - Vanderbilt University Medical Center; Henry Ong, PhD - Vanderbilt University Medical Center; Wei-Qi Wei, MD, PhD - Vanderbilt University; Stephen Deppen, PhD - Vanderbilt University Medical Center;

External Validation of the2024 ASCVD Risk Model in the All of Us Cohort

Poster Number: P100
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Fairness and Elimination of Bias, Racial Disparities, Chronic Care Management
Primary Track: Applications

The 2024 PREVENT equations have recently been developed as a race-free approach to estimate 10-year ASCVD risk. In this study we externally validated these equations in the All of Us Research Cohort and demonstrated that the PREVENT equations are significantly better calibrated in this disease-enriched cohort as compared to previous approaches. These findings indicate the robustness of the PREVENT equations as a risk-assessment method.

Speaker(s):
Nicholas Jackson, PhD Student Biomedical Informatics
Vanderbilt University

Author(s):
Nicholas Jackson, PhD Student Biomedical Informatics - Vanderbilt University; Lina Sulieman, PhD - Vanderbilt University Medical Center;

Computable Phenotype Feature Selection Utilizing Generative Pre-Trained Transformer 4 (GPT-4) through Prompt-based Few-shot Learning

Poster Number: P101
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Large Language Models (LLMs), Data Mining, Terminology Systems, Informatics Implementation
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics

This study presents a novel, high-efficiency approach to selecting features for computable phenotypes (CPs) which are essential for the development of algorithms in healthcare research. Traditional methods of identifying these features from electronic health records (EHRs) have been manually intensive, requiring significant time and expertise in clinical informatics. This study harnesses the capabilities of Generative Pre-trained Transformer 4 (GPT-4), a large language model, utilizing its few-shot learning ability to automate the process of feature selection. By integrating biomedical ontologies, the method effectively maps specified phenotypes to their corresponding relevant EHR features, streamlining the development of algorithms for clinical research, population health studies, and personalized medicine. The GPT-4-based approach provides a scalable and efficient solution, capable of handling a wide range of phenotypes, including those poorly defined or complex. The study demonstrates the robustness of this pipeline through its application to various phenotype inputs and the subsequent use of these features in algorithm development.

Speaker(s):
Victor Castro, MS
Mass General Brigham

Author(s):
Victor Castro, MS - Mass General Brigham; Vivian Gainer, MS - Mass General Brigham; Nich Wattanasin - Mass General Brigham; Michael Mendis - Mass General Brigham; Shawn Murphy, MD, Ph.D. - Massachusetts General Hospital;

A Performance-Based Voting Framework for Structuring Clinical Notes

Poster Number: P102
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Natural Language Processing, Information Retrieval, Controlled Terminologies, Ontologies, and Vocabularies, Data Standards
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics

This study introduces a performance-based voting framework integrating multiple Natural Language Processing models and clinical assertion frameworks for transforming unstructured clinical text into structured information. Employing Document Assembler, Sentence Detector, Tokenizer, PubMed and clinical BioBERT models, and BiLSTM-CNN-Char frameworks, it identifies medical entities, categorizes assertions, and maps to standardized ontologies. Preliminary evaluation on MIMIC-III dataset demonstrates the framework's effectiveness in structuring clinical notes, with detailed performance metrics for Entity, Polarity, and Tense classifications.

Speaker(s):
Behnaz Eslami, PhD Student
Loyola University Chicago

Author(s):
Behnaz Eslami, MS - Loyola University Chicago; Benjamin Strickland, DO - Loyola University Medical Center; Holly Mattix-Kramer, MD MPH; Dmitriy Dligach, Ph.D. - Loyola University Chicago; Mohammad Samie Tootooni, PhD - Loyola University Chicago;

Assessing Impact and Contextualizing the Value of Virtual Care Technologies: An Outcomes Framework for the Veterans Health Administration

Poster Number: P103
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Telemedicine, Mobile Health, Participatory Approach/Science
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics

Veterans, their family members, caregivers, and clinical team members are increasingly using a variety of virtual care (VC) technologies as part of daily clinical practice and interactions with the healthcare system. Recently, Veterans Health Administration’s (VHA) Office of Connected Care in collaboration with VHA researchers, developed a framework for understanding and demonstrating the impacts of virtual care technologies among different stakeholder groups and the VHA healthcare system.

Speaker(s):
Saige Calkins, M.A.
Department of Veteran Affairs

Author(s):

Application of Large Language Models for Systematic Literature Review Tasks: An Epidemiology Use Case

Poster Number: P104
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Large Language Models (LLMs), Infectious Diseases and Epidemiology, Real-World Evidence Generation, Informatics Implementation
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics

Large language models, such as Generative Pre-trained Transformers (GPT), are promising mechanisms to conduct systematic literature review tasks more efficiently. Our objective focuses on abstract screening to validate various GPT strategies against human agreement performance. We developed prompts asking about relevance of abstracts on a 1 (strongly irrelevant) to 5 (strongly relevant) Likert Scale. When setting Likert Score to 5, GPT 4.0 models using prompts with examples performed as good or better than human agreement.

Speaker(s):
James Rogers
Merck

Author(s):
Dong Wang - Merck; Varun Kumar Nomula, Master in Science - Merck & Co., Inc.; Yehua Wang, MSPH - University of Florida; Nazleen Khan, PhD - Merck & Co., Inc.; Katrina Mott; Peter Fiduccia, PhD - Merck; Mehmet Burcu, PhD, MS - Merck & Co., Inc.; Xinyue Liu, PhD - Merck & Co., Inc.;

Synthetic Data Augmentation Enhance Disease Named Entity Recognition

Poster Number: P105
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Information Extraction, Natural Language Processing, Large Language Models (LLMs), Machine Learning
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics

Synthetically generated labelled examples from language models have the potential address coverage gaps and imbalances in training corpora. We use Unified Medical Language System (UMLS) sourced prompts in conjunction with ChatGPT model to generate clinical mentions of diseases and show a small but significant improvement when including synthetic text.

Speaker(s):
John Osborne, PhD
University of Alabama at Birmingham

Author(s):
Kuleen Sasse;

Design of A Clinical Trial Protocol Extraction Workflow using Pre-Trained Large Language Models and Retrieval Augmented Generation

Poster Number: P106
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Natural Language Processing, Large Language Models (LLMs), User-centered Design Methods
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics

Our research introduces an innovative architecture that combines Large Language Models, Retrieval Augmented Generation, and a persistent vector database to enhance clinical trial registration at ClinicalTrials.gov. Employing both qualitative and quantitative evaluations on generated responses and retriever, we aimed for alignment with PRS standards. Preliminary findings revealed rich semantic detail but also highlighted discrepancies with PRS submission structures, which drive our ongoing research in prompt engineering to improve data generation.

Speaker(s):
Ramya Sri Baluguri, MS in Medical Informatics
University of California, Davis

Author(s):

EHR Data Abstraction for Post-Radiation Therapy Mortality Prediction

Poster Number: P107
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Data Mining, Disease Models, Information Extraction, Information Retrieval, Machine Learning
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics

We describe the abstraction of clinical, radiographic, and laboratory electronic health record data for all patients who consulted in a radiation oncology department at an academic hospital system from 2018 – 2024. A radiation oncologist collaborated with medical informaticists for a year to develop SQL queries that captured clinical data elements that may be associated with post-radiation therapy mortality.

Data elements were abstracted for 5,303,296 patient encounters for 38,262 patients, including 1,131,158 patient clinical summaries, performance status scores recorded in 804,706 radiation oncologist clinical notes, structured cancer stages stages for 23,857 patients, 810,847 ICD-10 diagnosis codes, 40,704,088 laboratory results, 1,941,361 radiologist dictations, 172,467 pathologist narratives, 199,952 antineoplastic therapy infusions, 48,540 surgeries, 30,630 courses of radiation therapy, 14,557,521 prescription medication records, and substance use histories tied to 1,380,138 encounters.

Many data elements could only be located with oncologist collaboration searching the EHR interface for clues to understand the data storage structure.

Post-radiation therapy prognostic data were sucessfully extracted for 38,262 patients, but with difficulty. Data storage structures built without data domain understanding may impede subsequent analyses.

Speaker(s):
Kendall Kiser, MD, MS
BJC Healthcare / Washington University in St. Louis

Author(s):
Kendall Kiser, MD, MS - BJC Healthcare / Washington University in St. Louis; Ashish Vaidyanathan, n/a - Washington University in St. Louis; Matthew Schuelke, PhD - Institute for Informatics, Washington University School of Medicine in St. Louis; Joshua Denzer, PhD - BJC Healthcare; Trudy Landreth, MS - Institute for Informatics; Christopher Abraham, MD, MS - Department of Radiation Oncology, Washington University School of Medicine in St. Louis; Adam Wilcox, PhD - Washington University in St. Louis;

Association Between Antidepressant Prescription and Post-COVID Services Utilization: An EHR-Based Event Analysis from the RECOVER Program

Poster Number: P108
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Infectious Diseases and Epidemiology, Healthcare Economics/Cost of Care, Population Health, Disease Models, Real-World Evidence Generation
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics

This study examines the association between antidepressant treatment and reduced healthcare services utilization following COVID-19 infection. Using large scale, routinely collected electronic health records from five health systems across NYC, patients on antidepressant treatment prior to COVID-19 infection had significantly fewer emergency department, inpatient, outpatient, and psychiatric visits post-infection compared to controls. These results suggest a potential off-label use for antidepressants in mitigating long-term COVID-19 symptoms and PASC, with future directions potentially examining dosage effects.

Speaker(s):
Veer Vekaria, BS
Weill Cornell Medicine / NewYork-Presbyterian Hospital

Author(s):
Stephenson Strobel, MD, PhD - Weill Cornell Medicine; Mark Olfson, MD, MPH - Columbia University; Jyotishman Pathak, PhD - Weill Cornell Medical College, Cornell University;

Weight and blood-based markers of cachexia predict disability, hospitalization and worse survival in cancer immunotherapy patients

Poster Number: P109
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Biomarkers, Cancer Prevention, Real-World Evidence Generation
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics

This study examined the association between several routinely collected cancer-associated cachexia markers and key clinical outcomes to ICI therapy – disability-free, hospitalization-free, and overall survival. Consensus weight-based cachexia, WLGS, and NLR measured at the time of ICI initiation are predictive of worse disability-free, hospitalization-free, and overall survival in cancer patients receiving ICI therapy, and may be useful for clinical prognostication and decision-making.

Speaker(s):
Steven Tran
Northwestern University - Feinberg School of Medicine

Author(s):
Steven Tran - Northwestern University - Feinberg School of Medicine; Noah Forrest, BS - Northwestern University - Feinberg School of Medicine; Vijeeth Guggilla, BA - Northwestern University; Geovanni Perottino, MD - Stanford University; Jodi Johnson, PhD - Northwestern University; Jeffrey Sosman, MD - Northwestern University; Ishan Roy, MD, PhD - Northwestern University; Theresa Walunas, PhD - Northwestern University;

The Disability Representation in the All of Us Research Program

Poster Number: P110
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Disability, Accessibility, and Human Function, Information Extraction, Surveys and Needs Analysis, Data Mining
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics

Disability are rarely documented in structured formate in EHR. The All of Us Research program is collecting self-reported disability from participants. We quantified self-reported disability percentages in All of Us and compared it to CDC and Census, and quantified three disability categories in EHR. The All of Us percentages match CDC but higher than census. EHR percentages were lower for cognitive and vision. Using only EHR might exclude participants with incomplete records who are mainly underrepresented populations.

Speaker(s):
Lina Sulieman, PhD
Vanderbilt University Medical Center

Author(s):
Rob Cronin, MD - Ohio State University; Qingxia Chen, PhD - Vanderbilt University Medical Center; Paul Harris, PhD - Vanderbilt University; Huiding Chen, BSc - Vanderbilt Univeristy; Brandy Mapes, MLIS - Vanderbilt University Medical Center;

A Causal Inference Representation of Survival for Patients with Non-Metastatic Pancreatic Ductal Adenocarcinoma

Poster Number: P111
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Causal Inference, Knowledge Representation and Information Modeling, Disease Models, Health Equity
Primary Track: Foundations
Programmatic Theme: Clinical Research Informatics

Pre-specified, transparent research protocol and analysis plan are critical when replicating a target trial using observational data such as from electronic health records. A comprehensive directed acyclic graph (DAG) was developed to model treatment decision for patients with non-metastatic pancreatic adenocarcinoma using triangulation of a literature review, multidisciplinary meeting observations, and interviews with subject-matter experts. This knowledge-driven DAG illuminated multiple confounders such as tumor biology, latent patient health status, and access to health care.

Speaker(s):
Frances Hsu, MS
Oregon Health & Science University

Author(s):
Nicole Weiskopf, PhD - Oregon Health & Science University; Emerson Chen, MD - Oregon Health & Science University; Eric Hall, PhD - OHSU-PSU School of Public Health;

STOP-HCV-HCC Program in South Texas: A privacy-preserving approach to enable remote data access and analytics at Federally Qualified Health Centers

Poster Number: P112
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Cancer Prevention, Data Sharing, Interoperability and Health Information Exchange
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics

The STOP-HCV-HCC clinical program, a collaborative effort aimed at combating hepatocellular carcinoma (HCC) in South Texas, utilized a privacy-preserving data analytics platform by TripleBlind to overcome challenges in privacy-preserving data access across electronic health record systems (EHRs). This innovative approach enabled secure, efficient analysis of decentralized data without the need for data transfer, significantly reducing the time required for quarterly reporting from days to under four minutes. The successful application of secure multi-party computation and federated data techniques demonstrates the platform's potential to enhance program management, ensure patient privacy, and improve data accuracy, offering a scalable solution for healthcare data analytics.

Speaker(s):
Gharib Gharibi, PhD
TripleBlind

Author(s):
Catherine Craven, PhD, MA, MLS, FAMIA - Consultant/MU/UTHSA; Raudel Bobadilla Bobadilla, MPH, CHW-I - University of Texas Health Science Center San Antonio, San Antonio, Texas; Edward Yao, PhD - University of Texas Health Science Center San Antonio, San Antonio, Texas; Caelob Castillo, NA - University of Texas Health Science Center San Antonio, San Antonio, Texas; Julio Cerroblanco, MBA - University of Texas Health Science Center San Antonio, San Antonio, Texas; Bertha E. Flores, PhD, RN, WHNP-BC - University of Texas Health Science Center San Antonio, San Antonio, Texas; Mamta Jain, MD - UT Southwestern Medical Center, Dallas, Texas; Alex Radunksy, PhD, MPH - UT Southwestern Medical Center, Dallas, Texas; Mikulas Pleask, BS - TripleBlind; Craig Gentry, PhD - TripleBlind; Riddhiman Das, MS - TripleBlind; Gharib Gharibi, PhD - TripleBlind;

Ontology-based Database Design for an Injury Research Platform

Poster Number: P113
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Controlled Terminologies, Ontologies, and Vocabularies, Informatics Implementation, Knowledge Representation and Information Modeling, Interoperability and Health Information Exchange
Primary Track: Foundations
Programmatic Theme: Clinical Research Informatics

This poster presents the design of the ontology-based database for an expanding web-based platform aimed at facilitating injury research and patient care of violence survivors. Developed by a collaborative effort between informaticians and clinicians to organize data into a semi-relational, key-value structured database, the ontology prioritizes flexibility and interoperability to accommodate diverse data sources, whose elements are not known ahead of time, as well as evolving research needs. Furthermore, the ontology is built to technically control the behavior of the software platform. More than just a concept mapping, the ontology is integrated with standardized terminologies and resources such as SNOMED CT and HL-7 FHIR and incorporates temporal modeling for longitudinal data. Injury cases and data, with a special focus on bruises, are laid out over more than 500 concepts and their relationships, which are organized into three main hierarchies to describe injuries, patients, and images. Almost 400 concepts are extracted from image metadata, enhancing the richness of captured information. Designed with attributes tailored for software platform integration, the ontology enhances data usability and analysis for both research and clinical applications, underscoring its utility as a specialized tool within the broader landscape of biomedical informatics.

Speaker(s):
Mohammad Qodrati, MD
George Mason University

Author(s):
Mohammad Qodrati, MD - George Mason University; Amin Amin Nayebi, PhD - George Mason University; Michał Markiewicz, Dr.-Ing. - Jagiellonian University, Kraków, Poland; David Lattanzi, PhD - George Mason University; Katherine Scafide, RN, PhD - George Mason University; Janusz Wojtusiak, PhD - George Mason University;

Difference-in-Differences Associations Between SSRI Prescription Strength and Healthcare Services Utilization After SARS-CoV-2 Infection

Poster Number: P114
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Infectious Diseases and Epidemiology, Drug Discoveries, Repurposing, and Side-effect, Chronic Care Management, Healthcare Economics/Cost of Care
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics

This study investigates the potential impact of different doses of commonly prescribed SSRIs, citalopram (Celexa) and escitalopram (Lexapro), on healthcare utilization post-COVID-19 infection. Using large scale, routinely collected electronic health records from five health systems across NYC, the key results show varied effects of SSRIs on inpatient, outpatient, and emergency department visits. These findings contribute to understanding the role of antidepressants in managing post-COVID-19 symptoms and may inform future research on complementary treatment strategies.

Speaker(s):
Veer Vekaria, BS
Weill Cornell Medicine / NewYork-Presbyterian Hospital

Author(s):
Stephenson Strobel, MD, PhD - Weill Cornell Medicine; Mark Olfson, MD, MPH - Columbia University; Jyotishman Pathak, PhD - Weill Cornell Medical College, Cornell University;

Leveraging Spiritual-BERT for Characterizing Spiritual Care Documentation in EHRs of Older Adults with Heart Failure

Poster Number: P116
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Advanced Disease, Real-World Evidence Generation, Healthcare Quality, Nursing Informatics, Natural Language Processing, Aging in Place
Primary Track: Foundations
Programmatic Theme: Clinical Research Informatics

This study employs Transformer-based natural language processing (NLP) to detect and characterize spiritual care documentation in the electronic health records (EHR) of older adults with heart failure. Retraining Bio-Clinical-BERT on EHR data, we developed Spiritual-BERT and validated its robust performance. Extracting data from over 2 million EHR notes using Spiritual-BERT, only 2.4% mentioned spiritual care, highlighting NLP's potential for enhancing patient-centered care. Future research will explore documentation variations across patient demographics to identify specific cohorts in need of spiritual care, thereby facilitating tailored interventions to enhance patient-centered care.

Speaker(s):
Alaa Albashayreh, PhD, MSHI, RN
University of Iowa

Author(s):
Alaa Albashayreh, PhD, MSHI, RN - University of Iowa; Nahid Zeinali, MS - University of Iowa; Yuwen Ji, MSN, RN - University of Iowa; Nanle Joseph Gusen, MSW, BSN, RN - University of Iowa; Stephanie Gilbertson White, PhD, APRN-BC, FAAN - University of Iowa;

Algorithmic Matching of Unique Device Information to Electronic Health Record Data

Poster Number: P117
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Real-World Evidence Generation, Data Transformation/ETL, Rule-based artificial intelligence
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics

There remains a large volume of medical device data in electronic health records that remain unmapped to a unique device identifier (UDI), which is used to monitor devices during after-market release. We propose methods to match the data to the UDI in the Department of Veterans Affairs (VA), one that makes use of a rule-based string-matching algorithm for our semi-structured data.

Speaker(s):
Julie Kim, PharmD
Department of Veterans Affairs

Author(s):
Julie Kim, PharmD - Department of Veterans Affairs; Ashley Spann, MD, MSACI - Vanderbilt University Medical Center; Michael McLemore, BSN - Vanderbilt University Medical Center; Tina French, RN, CPHQ - Vanderbilt University Medical Center; Mohammed Al-Garadi, PhD - VUMC; Michael Matheny, MD, MS, MPH, FACMI, FAMIA - Vanderbilt University Medical Center;

Usage trends for traditional- and neo-pronouns in electronic health record clinical notes

Poster Number: P118
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Diversity, Equity, Inclusion, Accessibility, and Health Equity, Health Equity, Fairness and Elimination of Bias
Primary Track: Applications
Programmatic Theme: Consumer Health Informatics

We examined pronoun usage in electronic health records (EHRs) at four academic medical centers. We found an increasing trend in the use of preferred pronouns such as "she/her/hers" and "they/them/their," while common pronouns "herself" and "himself" usage declined. Neopronoun usage remains infrequent without noticeable upward trends. This preliminary analysis suggests a shift in pronoun documentation, setting the stage for more comprehensive national studies on EHR pronoun use and patient satisfaction.

Speaker(s):
David Hanauer, MD
University of Michigan

Author(s):

Refresh, Refresh, Refresh: Measuring Repeated Access to the Patient Portal While Waiting for Sensitive Test Results

Poster Number: P120
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Patient Engagement and Preferences, Data Sharing, Data Mining
Primary Track: Applications
Programmatic Theme: Consumer Health Informatics

Patients now have immediate access to their test results as soon as they become available. We studied portal audit logs to quantify the extent to which patients refreshed their portal awaiting new results and measured the association with increased messaging. We observed that patients refreshed for 41.5% of results an average of 4 times before the result was available. Patients who refreshed were significantly more likely to send a message to their clinical team.

Speaker(s):
Bryan Steitz, PhD
Vanderbilt University Medical Center

Author(s):
Liz Salmi - Beth Israel Deaconess Medical Center; Robert Turer, MD, MSE - UT Southwestern Medical Center; Uday Suresh, MS - Vanderbilt University Department of Biomedical Informatics; Scott MacDonald, MD - UC Davis Health System; Catherine DesRoches, DrPH - OpenNotes, Beth Israel Deaconess Medical Center; S. Trent Rosenbloom, MD, MPH, FACMI, FAMIA - Vanderbilt University Medical Center Dept of Biomedical Informatics;

Factors to promote usage of wearable technology among Black and Latino adults at risk for atrial fibrillation

Poster Number: P121
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Diversity, Equity, Inclusion, Accessibility, and Health Equity, Tracking and Self-management Systems, Patient Engagement and Preferences, Self-care/Management/Monitoring
Primary Track: Applications
Programmatic Theme: Consumer Health Informatics

Wearable technology can facilitate community surveillance of atrial fibrillation (AF), but adoption among Black and Latino adults–a group at high risk for missed AF and worse outcomes–is low. This study aimed to identify facilitators and barriers to wearable technology among Black and Latino adults through qualitative interviews with diverse stakeholders. Community-based interventions providing AF education and tailored technological support may increase the usage of wearable technology for AF surveillance among Black and Latino adults.

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

Author(s):
Christianna Pepingco, BS - Columbia University; Jacquelyn Taylor, PhD, PNP-BC, RN - Columbia University School of Nursing; Elaine Wan, MD - Division of Cardiology, Department of Medicine, Vagelos College of Physicians and Surgeons, Columbia University;

Quantifying Patient Portal Use Among Patients and Care Partners Managing Alzheimer’s Disease and Related Dementias

Poster Number: P122
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Patient Engagement and Preferences, Chronic Care Management, Data Mining, Aging in Place
Primary Track: Applications
Programmatic Theme: Consumer Health Informatics

Patients with Alzheimer’s Disease and Related Dementias (ADRD) commonly require support from care partners to manage their healthcare. Patient portals provide tools that support care partners to manage and coordinate care, but portal enrollment and use remains low. This study investigates audit logs to understand the specific portal functionality accessed by patients and care partners managing ADRD as a critical step to identifying opportunities and interventions that better support this population.

Speaker(s):
Bryan Steitz, PhD
Vanderbilt University Medical Center

Author(s):
Uday Suresh, MS - Vanderbilt University Department of Biomedical Informatics; S. Trent Rosenbloom, MD, MPH, FACMI, FAMIA - Vanderbilt University Medical Center Dept of Biomedical Informatics; Jessica Ancker, MPH, PhD, FACMI - Vanderbilt University Medical Center;

LLM Validates Cancer Patient’s Pan-Cancer Clinical Data Elements

Poster Number: P123
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Natural Language Processing, Machine Learning, Controlled Terminologies, Ontologies, and Vocabularies, Real-World Evidence Generation, Workflow, Clinical Guidelines
Primary Track: Applications
Programmatic Theme: Consumer Health Informatics

There are 100,000 MSKCC cancer patients with genomic data that is an asset for researchers but, can be enhanced if complemented with clinical data. MSKCC employs data curators to manually obtain clinical data which takes 1-1.5 days for each patient. To tackle this, we collaborate with Realyze Intelligence to harness large language models (LLM) to automate clinical concepts. The advantage is speed and efficiency in extracting data for many patients in real time.

Speaker(s):
Andrew Niederhausern, BS
MSKCC

Author(s):
Nadia Bahadur, Masters of Clinical Research - Memorial Sloan Kettering Cancer Center; Andrew Niederhausern - MSKCC; Gary Wallace, Bachelor of Science - Realyze Intelligence; Carlos Martinez, Master of Science - Memorial Sloan Kettering Cancer Center; Gilan Saadawi, Doctor of Philosophy, Doctor of Medicine - Realyze Intelligence; John Philip, MS - Memorial Sloan Kettering Cancer Center;

Characterizing Laypeople’s Use of Large Language Models vs. Search Engines for Health Queries

Poster Number: P124
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Large Language Models (LLMs), Delivering Health Information and Knowledge to the Public, Human-computer Interaction
Primary Track: Applications
Programmatic Theme: Consumer Health Informatics

We surveyed 2,002 laypeople about whether they use LLMs for health queries, the types of queries posed, and their reactions to LLM versus search engine responses. 1,913 (96%) participants reported using search engines for health and 642 (32%) participants reported using LLMs for health. Use cases were similar, though LLMs were rated slightly less useful than search engines. However, they also elicited slightly less negative feelings and were perceived as more human and less biased.

Speaker(s):
Nina Singh, BS
NYU Grossman School of Medicine

Author(s):
Tamir Mendel, PhD - NYU Tandon School of Engineering; Batia Wiesenfeld, PhD - NYU Stern School of Business; Devin Mann, MD - NYU Grossman School of Medicine; Oded Nov, PhD - NYU Tandon School of Engineering;

Exploring User Engagement and Its Relationship with Weight Loss Outcomes in a Digital Weight Management Intervention by Analyzing App Usage Data

Poster Number: P125
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Mobile Health, Patient Engagement and Preferences, Patient / Person Generated Health Data (Patient Reported Outcomes)
Primary Track: Applications

This study examines the impact of user engagement on digital weight loss interventions using app usage data from the Habit app. Out of 31 participants, 29.03% achieved a 5% weight reduction during the 24-week intervention. The study found significant differences in app and feature usage between successful and unsuccessful weight loss groups, particularly in the features used from week 7 to week 12. Additionally, it revealed significant correlations between user engagement and weight loss outcomes.

Speaker(s):
Lidan Zhang, Master
Worcester Polytechnic Institute

Author(s):
Lidan Zhang, Master - Worcester Polytechnic Institute; Bengisu Tulu, PhD - Worcester Polytechnic Institute; Sherry Pagoto, PhD - University of Connecticut;

New Approaches to Staffing and Data Delivery Services

Poster Number: P126
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Administrative Systems, Data Sharing, Data Standards
Working Group: Clinical Research Informatics Working Group
Primary Track: Foundations
Programmatic Theme: Consumer Health Informatics

Staffing core data delivery services is a major challenge in clinical research data delivery. We have implemented several strategies to improve our data delivery services, including restructuring management, partnering closely with our IRB to streamline compliance approvals, and implementing tiered services. Due to the changes we have made in the past year, we cut our data request queue in half.

Speaker(s):
Rashawnda Lacy, Masters of Health Informatics
Health Data Compass

Author(s):

Evaluation Of Diagnosis And Triage Performance Of Ada, Webmd Symptom Checkers, And ChatGPT On Patients Presenting For Urgent Primary Care

Poster Number: P127
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Evaluation, Diagnostic Systems, Large Language Models (LLMs), Mobile Health, Personal Health Informatics, Governance of Artificial Intelligence
Primary Track: Applications
Programmatic Theme: Consumer Health Informatics

Diagnostic and triage output from symptom checkers (SCs), as well as Large Language Models (LLMs) like ChatGPT, may influence patient care decisions. Little research exists on evaluation for these system using real patient cases. In this study, we examine WebMD, Ada Health, and ChatGPT versions 3.5 and 4.0 diagnostic and triage accuracy on patients visiting a primary care clinic.

Speaker(s):
Daven Crossland, B.A.
Brown University

Author(s):
Hamish Fraser, MBChB, MRCP, MSc - Brown University; Chloe Kim, B.A. - Brown University; Caroline Bailey, BS - Brown University; Kenon Graham, BA - Brown University; Ross Hilliard, MD, FACP - MaineHealth; Ian Bacher - Brown University; Drew Nagle, MD - Brown Medicine; Megan Ranney, MD, MPH - Dept. of Emergency Medicine, Alpert Medical School, Brown Univ.; Tracy Madsen, M.D., Ph.D - Brown University; Daven Crossland, B.A. - Brown University;

Understanding Activity Patterns in Dementia Patients Using Wearables

Poster Number: P128
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Mobile Health, Tracking and Self-management Systems, Ubiquitous Computing and Sensors, Aging in Place
Primary Track: Applications
Programmatic Theme: Consumer Health Informatics

Wearables enable activity tracking for dementia patients, with many employing step-count averages over multiple days to analyze trends. This study examined the impact of different time windows for averaging step counts on preserving activity variability based on the step counts of 7 days collected using Fitbit Inspire 2 with a sliding window approach. Our analysis suggests that shorter windows captured natural day-to-day variations, whereas longer windows over-smoothed fluctuations, obscuring significant changes.

Speaker(s):
Saitejaswi Cherukupalli, Master of Science
Indiana University

Author(s):
Saitejaswi Cherukupalli, Master of Science - Indiana University; Charanjit Kaur, Master's in Health Informatics - Indiana University Purdue University; Jean-Francois Daneault, PhD - Rutgers University; Hee Tae Jung;

Leveraging advanced facial recognition technologies to annotate a large corpus of patient photos

Poster Number: P129
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Evaluation, Healthcare Quality, Real-World Evidence Generation
Primary Track: Foundations
Programmatic Theme: Consumer Health Informatics

We investigated the use of Microsoft's Face AI API to annotate a large corpus of patient photos with the following attributes: size, position, center, angle and occlusions of the face, exposure and blur level of the photo, the complexity of background and number of faces in one photo. We present the detailed steps needed to secure the annotation environment and the evaluation results of a test set of photos leveraging this annotation tool.

Speaker(s):
Chenyang Li, MD
Columbia University in the City of New York

Author(s):
Gregory Hruby, PhD - NewYork-Presbyterian Hospital; Yelstin Fernandes, BA - Center for Patient Safety Science, Columbia University Medical Center; Stanley Adelman, BA - Center for Patient Safety Science, Columbia University Medical Center; Jo Applebaum, MPH - Center for Patient Safety Science, Columbia University Medical Center; I-Fong Lehman, DrPH - Center for Patient Safety Science, Columbia University Medical Center; Neha Thummala, MPH - Center for Patient Safety Science, Columbia University Medical Center; Hojjat Salmasian, MD, MPH, PhD, FAMIA - Children's Hospital of Philadelphia; Jason Adelman, MD, MS - Columbia University Medical Center;

Increasing Health Insurance Literacy Among International Student: preliminary results of user-centered design study

Poster Number: P131
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Personal Health Informatics, User-centered Design Methods, Qualitative Methods, Human-computer Interaction, Education and Training
Primary Track: Applications
Programmatic Theme: Consumer Health Informatics

From interviews and surveys with international students in the first stage of a user-centered design methodology to develop a tool to increase health insurance literacy among international students, four main themes arose: frustrating billing processes and confirmation, confusion if a provider is in-network, difficulty finding information regarding insurance plan or documents, and distrust of information provided by insurance representatives. Potential solutions pointed at a website to aid in organizing, accessing, and centralizing health insurance information.

Speaker(s):
Rachael Kang, BS, MS
University of Maryland Baltimore County

Author(s):
Rachael Kang, BS, MS - University of Maryland Baltimore County; Siddharth Monga, BS - University of Maryland - Baltimore County; Neel Bhesaniya, BS - University of Maryland - Baltimore County; Mohammad Arshad, BS - University of Maryland - Baltimore County; Tera Reynolds, PhD, MPH, MS - University of Maryland Baltimore County;

Enabling Superior Personalised and Precise Cardiac Rehabilitation at Home: From Concept to Commercialization

Poster Number: P132
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Behavioral Change, Delivering Health Information and Knowledge to the Public, User-centered Design Methods, Health Equity, Mobile Health, Diversity, Equity, Inclusion, Accessibility, and Health Equity
Working Group: Climate, Health and Informatics Working Group
Primary Track: Applications
Programmatic Theme: Consumer Health Informatics

This poster presents a web-based digital health platform for at-home cardiac rehabilitation (CR). The platform, developed through a clinician-led codesign cycle, features interfaces for both clinicians and patients. The study also provides a SWOT analysis and a business model with a delivery framework for commercialization. The solution aims to enhance accessibility and engagement in CR, addressing challenges such as cost and geographical constraints as well as being as suitable for vulnerable communities such as CALD (culturally and linguistically diverse) communities . The platform’s development and potential impact transcend healthcare system sand the solution is being designed and developed concurrently in both Australian and US healthcare contexts.

Speaker(s):
Nilmini Wickramasinghe, PhD
La Trobe University

Author(s):
Nalika Ulapane, PhD; Vijay Gehlot, PhD; ELLIOT SLOANE, PhD, CCE, FAIMBE, FHIMSS;

The Evolution of Telehealth Services: The Impact of COVID on Utilization Trends

Poster Number: P133
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Patient Engagement and Preferences, Population Health, Healthcare Quality
Primary Track: Applications
Programmatic Theme: Consumer Health Informatics

Telehealth services have undergone a remarkable transformation, especially in the context of the COVID-19 pandemic, which accelerated their utilization across hospitals in the United States. This study investigates the extent to which telehealth services were utilized before COVID-19, the expansion during the pandemic, and the sustained utilization post-COVID. The findings of this study would better allow policymakers and healthcare organizations strategize to maximize the benefits of telehealth while addressing potential barriers to its widespread implementation.

Speaker(s):
Oghale Asagbra, PhD
East Carolina University

Author(s):

Time-optimized Prediction of Late Cancer Diagnosis

Poster Number: P134
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Machine Learning, Cancer Prevention, Reproducibility, Evaluation, Clinical Decision Support
Primary Track: Applications
Programmatic Theme: Consumer Health Informatics

Our research highlights challenges using machine learning methods in predicting late cancer diagnosis, particularly for prostate and breast cancer. Moreover, the generalization of the prediction modeling can also be another issue. The presented research highlights challenges in late cancer diagnosis, particularly for prostate and breast cancer, emphasizing factors like low awareness and socioeconomic disparities. The work suggests that it is possible to construct models for predicting late diagnosis by using gender-specific machine learning models, optimizing lookback windows, and addressing missing data through imputation techniques. However, prediction accuracies are very low, indicating that socioeconomic variables alone cannot be used as predictors reliably.

Speaker(s):
HUAN-JU SHIH, Doctoral Student
George Mason University

Author(s):
HUAN-JU SHIH, Doctoral Student - George Mason University; Janusz Wojtusiak, PhD - George Mason University; Hua Min, PhD - George Mason University;

“I need someone I trust”: Limitations of information systems for young adult gay and bisexual men seeking mental healthcare

Poster Number: P135
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Diversity, Equity, Inclusion, Accessibility, and Health Equity, Human-computer Interaction, Qualitative Methods
Primary Track: Applications
Programmatic Theme: Consumer Health Informatics

Disparities in finding mental health care providers exist for young adult gay and bisexual men. To understand how information systems contribute to this disparity I conducted semi-structured interviews to understand their search for mental health care. Due to their unique information needs, they have a more difficult time finding a provider resulting in increased work load. This pilot study highlights a need for further understanding to make information more accessible during the search process.

Speaker(s):
Matthew Ackerman
Rutgers University

Author(s):
Matthew Ackerman - Rutgers University;

Proxy Access in Patient Portal and Its Impact on Patient Experience

Poster Number: P136
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Patient Engagement and Preferences, Healthcare Quality, Personal Health Informatics
Primary Track: Applications
Programmatic Theme: Consumer Health Informatics

Patient portals facilitate communication between patients and healthcare providers while offering various functionalities for managing health information. Among these functionalities, proxy access allows designated individuals, such as caregivers or family members, to access a patient's health records and communicate with healthcare professionals on their behalf. The aim of this study is to examine the relationship between hospitals that offer their patients access to the proxy access functionality and their overall patient satisfaction score.

Speaker(s):
Oghale Asagbra, PhD
East Carolina University

Author(s):

Clusters of Social Isolation Trajectories

Poster Number: P137
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Machine Learning, Data Mining, Behavioral Change
Primary Track: Applications

Research has addressed social isolation (SI) and its health outcomes; however, machine learning methods have not been applied to detect SI trajectories. The Health and Retirement Study was utilized to examine SI trajectories. Interpolation and extrapolation were performed to handle unequal length of trajectories. K-means algorithm was employed to cluster SI. Findings show consistent SI scores over time for the majority. In contrast, one cluster showed worsening scores, possibly due to a decline in social participation.

Speaker(s):
Ghaida Alsadah, Health Services Research, PhD
George Mason University

Author(s):
Ghaida Alsadah, Health Services Research, PhD - George Mason University; Janusz Wojtusiak, PhD - George Mason University;

Advancing Clinical Observations through Biomedical Ontologies

Poster Number: P138
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Controlled Terminologies, Ontologies, and Vocabularies, Knowledge Representation and Information Modeling, Interoperability and Health Information Exchange
Primary Track: Applications
Programmatic Theme: Public Health Informatics

Biomedical ontologies are crucial for organizing and standardizing domain-specific knowledge in the biomedical field. However, there is a gap in accurately capturing abnormal states in clinical findings. This research study focuses on leveraging biomedical ontologies to represent abnormal clinical observations, such as elevated blood creatinine levels and prolonged QT intervals. The goal is to enhance existing ontologies to support complex clinical research and healthcare data management by identifying and addressing gaps in representation.

Speaker(s):
Mathias Brochhausen, Ph.D.
University of Arkansas for Medical Sciences

Author(s):
Mitra Rocca, Dipl.-Inform. Med, FAMIA - Food and Drug Administration, University of Arkansas for Medical Sciences; Jonathan Bona, PhD - University of Arkansas for Medical Sciences (UAMS);

Using Large Language Models for the Classification of Social Media Posts Covering Xylazine Wounds

Poster Number: P139
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Large Language Models (LLMs), Health Equity, Natural Language Processing, Data Mining, Drug Discoveries, Repurposing, and Side-effect, Patient / Person Generated Health Data (Patient Reported Outcomes)
Primary Track: Applications
Programmatic Theme: Public Health Informatics

This research seeks to use the GPT3.5 LLM model to automate the multiclass classification of Reddit posts relevant to Xylazine-associated wounds. The LLM was prompted using examples of manually classified posts from a previous study. It is expected that the LLM will require multiple iterations of prompting fine-tuning to produce a successful classification of the posts. NLP methodologies will be used to parse a successful classified output for distribution analysis.

Speaker(s):
JaMor Hairston, MSHI, MS
Emory University SOM Department of Biomedical Informatics

Author(s):

What is health—WHO definition and public perspectives

Poster Number: P140
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Population Health, Surveys and Needs Analysis, Personal Health Informatics
Primary Track: Applications
Programmatic Theme: Public Health Informatics

We compared the public’s perspectives on the definition of health based on three online surveys and the definition established by the World Health Organization. Over 80% of the respondents agreed that health is the state of mental, physical, and social well-being, regardless of disease. Our results indicate that the respondents, a well-educated group, understand health, prevention, self-ownership of health, and the complex nature of health. A more important question for such groups would be how to motivate them to maintain a healthy lifestyle.

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

Author(s):
Yuchun Zhou, PhD - Ohio University; Temiloluwa Sokoya, PhD - Jackson State University; Xia Jing, PhD - Clemson University;

The Prospect of LLMs-Based Communication in Primary Care

Poster Number: P141
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Change Management, Chronic Care Management, Healthcare Quality, User-centered Design Methods, Workflow, Natural Language Processing, Deep Learning, Machine Learning
Primary Track: Applications
Programmatic Theme: Public Health Informatics

This study reviews how LLMs have been applied in primary care and suggests their applicability in the future. Literature was collected from five databases. Using Covidence, 34 articles were selected and analyzed. The results suggest that LLMs could facilitate communication between patients and the primary care team. Furthermore, LLMs can contribute to primary care by allowing patients and the primary care team to interact with each other so that patients can be prioritized.

Speaker(s):
Dogeun Park, B.A. (Bachelor of Arts) & B.E. (Bachelor of Engineering)
Dept. of Artificial Intelligence, Hallym University

Dogeun Park, B.A. (Bachelor of Arts) & B.E. (Bachelor of Engineering)
Hallym University

Author(s):
Dogeun Park, BA, BE - Dept. of Artificial Intelligence Convergence, Hallym University; Kwangsub So, ME - Dept. of Artificial Intelligence Convergence, Hallym University; Dong-Soo Shin, RN, PhD - School of Nursing, Hallym University and Center for Primary Care, Harvard Medical School; Dong-Ok Won, PhD - Dept. of Artificial Intelligence Convergence and Dept. of Neurology, College of Medicine, Hallym University; David Duong, MD, MPH - Center for Primary Care, Harvard Medical School;

The acceptance of wearable technology amongst those with cognitive impairment and their carers

Poster Number: P142
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: User-centered Design Methods, Qualitative Methods, Patient Engagement and Preferences
Primary Track: Policy
Programmatic Theme: Public Health Informatics

We conducted a qualitative study into the acceptability of wearable technologies amongst those with cognitive impairments and their carers. Twenty-one semi-structured interviews were conducted after two weeks of technology usage. Key themes from our analysis included the ease of use, data transparency and usefulness. These results highlighted the need for researchers to consider factors, such as the difficulty level of cognitive testing games, to improve acceptance and enjoyment of wearables.

Speaker(s):
Sarah Wilson, MSc, BSc
Newcastle University

Author(s):
Sarah Wilson, MSc, BSc - Newcastle University; Emily Beswick, PhD - Trinity College Dublin; Rachel Morrell, MSc - University College London; Sharandeep Bhogal, PhD - University College London; Clare Tolley - Newcastle University; Timothy Whitfield, PhD - University College London; Kieran Wing, MSc - University College London; Riona McArdle, PhD - Newcastle University; Zuzana Walker, MD - University College London; Sarah Slight, PhD - Newcastle University;

Impact of Extreme Heat Days on Healthcare Utilization and Expenditures Among Members of a Large National Payor

Poster Number: P143
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Environmental Health and Climate Informatics, Population Health, Real-World Evidence Generation
Primary Track: Applications
Programmatic Theme: Public Health Informatics

An informatics-informed evaluation leveraging a rich foundation of medical claims, third-party weather data, and advanced analytics was used to characterize extreme heat events from 2022 to 2023 and associated increases in healthcare utilization and related-expenditures of ~5 – 10% among members of a large national payor residing across the United States.

Speaker(s):
Amanda Zaleski, PhD
Aetna

Author(s):
Amanda Zaleski, PhD - Aetna; Annie Laurie Hines, PhD - CVS Health; Kelly Craig, PhD - CVS Health; Samrat Kulkarni, MS - CVS Health; Eric Simoni, MBA - CVS Health; Dorothea Verbrugge, MD - CVS Health; Eric Hamilton, MS - CVS Health; Daniel Knecht, M.D., M.B.A. - CVS Health;

Effectiveness of avatar-based technology in patient education for improving chronic disease knowledge and self-care behavior

Poster Number: P144
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Education and Training, Human-computer Interaction, Chronic Care Management, User-centered Design Methods, Self-care/Management/Monitoring
Primary Track: Applications
Programmatic Theme: Public Health Informatics

In general, avatar-based technology was found beneficial for patient education in improving chronic disease knowledge and self-care behavior. The emergence of conversational agents indicates that personalized user experience is becoming increasingly popular for patient education. Future studies that use avatar-based technology in patient education may benefit from a personalized design that considers different patient groups and cultural contexts to enhance patient engagement.

Speaker(s):
Po-Yin Yen, PhD, RN, FAMIA, FAAN
Washington University in St. Louis

Author(s):
Qingfan An, MRes - Umeå University; Po-Yin Yen, PhD, RN, FAMIA, FAAN - Washington University in St. Louis;

A Visualization Tool for Geospatial Phenotype Prevalences in VA Patients

Poster Number: P145
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Information Visualization, Phenomics and Phenome-wide Association Studies, Population Health
Primary Track: Applications
Programmatic Theme: Public Health Informatics

High-throughput phenotyping strategies are capable of classifying large volumes of patients. However, translating this data to real world applications is challenging. We have developed GeoPheno, a tool which displays the geospatial prevalences of EHR-based phenotypes in the Veteran population over time. Our flexible tool can display data from a wide array of phenotypes and is integrated with the CIPHER phenotype library, allowing users to view the definitions of the conditions being visualized.

Speaker(s):
Michael Murray, MS
VA Boston Healthcare System

Author(s):
Michael Murray, MS - VA Boston Healthcare System; Yuk-Lam Ho, MPH; Vidul Ayakulangara Panickan; Jacqueline Honerlaw, RN, MPH; David Heise; Keith Connatser; Sumitra Muralidhar, PhD - Veterans Health Association; Kelly Cho, PhD;

A mixed-method comparison of SDOH documentation between an academic medical center and a community health center

Poster Number: P146
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Participatory Approach/Science, Documentation Burden, Population Health
Primary Track: Foundations

Most biomedical health informatics research are carried out within academic medical centers (AMC), including
natural language process (NLP) for SDoH. In contrast, very little research has been carried out at community health
centers (CHC), which care for a disproportionate share of individuals with high unmet social needs.These differences and similarities between two AMC and two CHC clinics in the documentation of SDoH are explored.

Speaker(s):
Carolin Spice
University of Washington

Author(s):

Identifying Rural and Urban Participants in All of Us

Poster Number: P147
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Data Transformation/ETL, Population Health, Precision Medicine, Racial Disparities
Primary Track: Foundations
Programmatic Theme: Public Health Informatics

Rural residents in the US are at an increased risk of health disparities. We implemented a data transformation model in the NIH’s All of Us program that linked 3-digit zip code data to electronic health record data. We were able to distinguish between rural and urban areas by calculating a novel metric termed the ‘rurality score’. This model will allow targeted population health studies within All of Us and advance public health research.

Speaker(s):
Anas Awan, M.S.
National Human Genome Research Institute

Author(s):
Kate Chaillet, B.A. - National Human Genome Research Institute; Dayo Shittu, M.S., M.P.H. - National Human Genome Research Institute; Tam Tran; Tracey Ferrara, PhD - National Human Genome Research Institute; Joshua Denny, MD, MS - National Institutes of Health; Huan Mo, MD, MS;

Associations between Online Health Information Behaviors and Trust in the US Healthcare System

Poster Number: P148
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Social Media and Connected Health, Delivering Health Information and Knowledge to the Public, Patient Engagement and Preferences
Primary Track: Foundations

Distrust in healthcare can lead to lower healthcare engagement and inferior clinical outcomes. With patients increasingly likely consume online health information, it is important to understand the associations between online health information behaviors and trust in healthcare. Using a publicly available survey dataset, we found links between both health information consumption and perceived misinformation and low trust in the healthcare system. Future work is needed to understand the role of information source in this relationship.

Speaker(s):
Jessica Ray, PhD
University of Florida

Author(s):
Sherrine Fils-Aime, Undergraduate Student - University of Florida; Katie Kloss, BA - University of Florida;

ML Based Capacity Estimation for Advance Care Planning

Poster Number: P149
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Machine Learning, Clinical Decision Support, Population Health, Chronic Care Management, Critical Care, Workflow
Primary Track: Applications
Programmatic Theme: Public Health Informatics

The study addresses the challenge of identifying patient population for advanced care planning by leveraging Electronic Health Record (EHR) data to predict 1-year mortality risk scores using the Laboratory-Based Risk Stratification (LabRS) model. The study outlines the application of the prediction tool and explores how a risk stratified patient list can guide providers for capacity estimation and workflow planning in healthcare systems. Through a structured workflow, patients are identified, reviewed, and referred to appropriate ACP programs.

Speaker(s):
Biplab S Bhattacharya, PhD
Geisinger Health System

Author(s):

Exploring Data Quality Problems and Data Quality Management Practices in Citizen Science: Barriers and Facilitators in the COVID-19 Tracking Project

Poster Number: P150
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Data Standards, Participatory Approach/Science, Delivering Health Information and Knowledge to the Public
Primary Track: Applications
Programmatic Theme: Public Health Informatics

We conducted a data quality assessment of the COVID Tracking Project (CTP)’s publicly available data sets, measuring the integrity, consistency, and completeness of the data produced by this citizen-led participatory science project. We found that the CTP aggregated and published timely public health data with high levels of integrity and consistency. Problems with missing data were traced back to issues with state- and local-level reporting systems and inconsistent data standards across states and territories.

Speaker(s):
Donald McCormick, MSHI
Arkansas Hospital Association

Author(s):
Toni Jaudon, student - University of Arkansas for Medical Sciences; Petronella Mlambo, Bachelor's Degree in Computer Engineering - Arkansas Blue Cross and Blue Shield; Ash Richison, MPH - UAMS; Gunes Koru, PhD, FAMIA - University of Arkansas for Medical Sciences, Northwest Regional Campus;

Real-world Evidence of Emerging Health Disparities in Severe Alcoholic Hepatitis Incidence in the Post-COVID Era

Poster Number: P151
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Real-World Evidence Generation, Health Equity, Population Health
Primary Track: Applications
Programmatic Theme: Public Health Informatics

The COVID-19 pandemic alters social norms of work, life, and interpersonal communications, which triggers new health disparities in the risk of severe alcoholic hepatitis (sAH) and other mental health issues that share similar socioeconomic stressors. This work generated comprehensive real-world evidence of the emerging health disparities from a national and up-to-date medical claims dataset. Our findings indicate a fast-increasing trend of risk drinking and concerning new patterns of health disparities.

Speaker(s):
Jiangqiong Li, Ph.D.
Indiana University

Author(s):
Jing Su, PhD - Indiana University School of Medicine;

The Relationship between Air Pollution and Patient Portal Messaging in an Asthmatic Population

Poster Number: P153
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Patient Engagement and Preferences, Environmental Health and Climate Informatics, Mobile Health
Primary Track: Applications

Asthma is a common chronic illness in children and adults. It is well-known in scientific literature that a common cause of asthma exacerbations is air pollution. There is no literature found on the patient portal activity of patients when exposed to unfavorable environmental circumstances. This study focuses on evaluating the relationship between patient portal activity for an asthmatic population when air pollution is high.

Speaker(s):
Marily Barron
Department of Biomedical Informatics, Vanderbilt University

Author(s):
S. Trent Rosenbloom, MD, MPH, FACMI, FAMIA - Vanderbilt University Medical Center Dept of Biomedical Informatics; Laurie Novak, PhD - Vanderbilt University Medical Center Dept of Biomedical Informatics; Vern Kerchberger, MD - Vanderbilt University Medical Center Dept of Biomedical Informatics; Marily Barron - Department of Biomedical Informatics, Vanderbilt University;

Investigating the Stakeholder Requirements for Publicly Accessible Maternal Health Data: Empirically Guiding the Design and Development of the Arkansas Maternal Health Data Dashboard

Poster Number: P154
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Delivering Health Information and Knowledge to the Public, Population Health, Qualitative Methods, Data Sharing
Working Group: Public Health Informatics Working Group
Primary Track: Applications
Programmatic Theme: Public Health Informatics

This study investigated the requirements of multiple stakeholders for publicly accessible maternal health data to ensure the usefulness of the first Maternal Health Data Dashboard for the State of Arkansas, designed and developed by our team. Following a qualitative research approach and conducting semi-structured interviews, we identified 121 user stories grouped under eight themes. The identified requirements provide empirical guidance for our design and development activities, which can also inform similar initiatives tackled in other localities.

Speaker(s):
Gunes Koru, PhD, FAMIA
University of Arkansas for Medical Sciences, Northwest Regional Campus

Author(s):
Rachel Purvis, PhD - University of Arkansas for Medical Sciences; Jennifer Callaghan-Koru, PhD; Toni Jaudon, student - University of Arkansas for Medical Sciences; Gunes Koru, PhD, FAMIA - University of Arkansas for Medical Sciences, Northwest Regional Campus;

A Multi-data Source Study on Scientific Investigations and Research Investment on African American Health During COVID-19

Poster Number: P155
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Diversity, Equity, Inclusion, Accessibility, and Health Equity, Racial Disparities, Population Health
Primary Track: Policy

This study pioneers a multi-data source analysis of COVID-19 research on African American (AA) health, utilizing publications, grants, and clinical trials. We aim to elucidate the landscape of research output, investment, and focus areas, identifying key researchers, institutions, and funding trends pertinent to the topic. Our findings will highlight research gaps, facilitating targeted future investigations. This analysis not only informs stakeholders but also lays the groundwork for extending research benefits to other underserved populations.

Speaker(s):
Siobahn Grady, Ph.D.
North Carolina Central University

Author(s):
Fei Yu - UNC at Chapel Hill; Emily Jones, MLIS - University of North Carolina at Chapel Hill; Siobahn Grady, PhD - North Carolina Central University;

Healthcare services consumption among Heart Failure patients in Western New York during the COVID-19 pandemic

Poster Number: P156
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Chronic Care Management, Data Standards, Disease Models, Healthcare Quality, Healthcare Quality, Informatics Implementation, Infectious Diseases and Epidemiology
Primary Track: Applications
Programmatic Theme: Public Health Informatics

We conducted a retrospective study based on the Electronic Healthcare Records (EHR) of Kaleida Health in Western New York, from January 2019 until December 2021. Data were mapped to the Observational Medical Outcomes Partnership (OMOP) Common Data Model (CDM) version 5.3.1. The study included 4,138 HF patients and revealed higher rates of healthcare services utilization among HF patients after the outbreak of the COVID-19 pandemic. This trend was observed both in cardiac and noncardiac encounters.

Speaker(s):
Samuel Tiosano, MD, MPH
University at Buffalo

Author(s):
Samuel Tiosano, MD, MPH - University at Buffalo; Skyler Resendez - The University at Buffalo; Crystal Tomlin, PhD - University at Buffalo; Muhammad Piracha, MBBS - University at Buffalo; Nicholas Sass, MD - University at Buffalo; Prahalad Rangan, PhD - University at Buffalo; Peter Elkin, MD, MACP, FACMI, FNYAM, FAMIA, FIAHSI - Department of Biomedical Informatics, Jacobs School of Medicine and Biomedical Sciences, University at Buffalo, State University of New York;

Infodemic Management using Natural Language Processing: A COVID-19 Case Study

Poster Number: P157
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Natural Language Processing, Infectious Diseases and Epidemiology, Data Mining
Working Group: Knowledge Discovery and Data Mining Working Group
Primary Track: Applications
Programmatic Theme: Public Health Informatics

Considering the lingering consequences of Coronavirus disease 2019 (COVID-19), the world needs to
be better equipped with infodemic management technologies in case of a future outbreak. One of the major bottle-
necks in the management of future infodemics is the lack of verified information with balanced class annotations. This is likely to be further amplified by a lack of understanding of the evolution of disease-related topics over time.The current era of global change has been associated with growth in both emerging and re-emerging infectious diseases. Consequently, researchers need to be capable of detecting and disseminating verified information related to such diseases. Resurgence waves of COVID-19 cases have been observed worldwide since 20201. During this unprecedented crisis, people have increasingly relied on online information sources for dealing with the pandemic. A real-time assessment of X (formerly known as Twitter) discussions can be useful for timely addressing of public health emergency responses. In this direction, we sought to use natural language processing (NLP) techniques to classify information available from X related to the surveillance and prevention of COVID-19

Speaker(s):
Reshma Kar, PhD
Saint Peter’s University

Author(s):
Venu Devadi, MS - Saint Peters University; Braja Gopal Patra, PhD - Weill Cornell Medicine; Dept of Population Health Sciences; Div of Health Informatics;

Efficient Detection System of Vaping Discussions on Reddit: Towards Enhanced Public Health Monitoring

Poster Number: P158
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Natural Language Processing, Deep Learning, Social Media and Connected Health
Primary Track: Applications

This study proposed an effective natural language process (NLP) system to identify vaping-related discussions on Reddit, using an ensemble of pre-trained BERT models. Addressing the gap in conventional social media surveillance methods, this methodology enhances public health monitoring by accurately capturing vaping dialogues, paving the way for targeted intervention strategies.

Speaker(s):
Yang Ren, MS
Univeristy of South Carolina

Author(s):
Dezhi Wu, PhD - University of South Carolina; Erin Kasson, MS - university of washington st louis; Patricia Cavazos-Rehg, PHD - university of washington st louis; Ming Huang, PhD - UTHealth;

Utilization of Information Architecture for Querying Electronic Health Record (EHR) Data

Poster Number: P159
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Information Retrieval, Reproducibility, Information Extraction
Primary Track: Foundations
Programmatic Theme: Translational Bioinformatics

The application of information architecture to research projects assists both investigators and programmers when querying electronic health record (EHR) data. Our establishment of a web-based platform that serves as a query repository for major data domains and individual projects has decreased project development costs as well as improved accuracy and efficiency.

Speaker(s):
Aaron Thomas, MSHI
UNC-Chapel Hill

Author(s):
Purvi Khanuja, Pharm D, MS Health Informatics; Bill Ross, BS - UNC-Chapel Hill; Kaitlin Calhoun, PhD - UNC-Chapel Hill;

Multi-Omics Integrative Risk Model for Alzheimer’s Disease in a Large-Scale Biobank

Poster Number: P160
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Omics (genomics, metabolomics, proteomics, transcriptomics, etc.) and Integrative Analyses, Bioinformatics, Computational Biology, Machine Learning
Primary Track: Foundations
Programmatic Theme: Translational Bioinformatics

Alzheimer’s Disease (AD) is a neurodegenerative disease characterized by genetic heterogeneity, which complicates prediction of genetic risk associated with the condition. We constructed a multi-omics integrative risk model for AD, leveraging imputed transcriptomic and proteomic expression information from the Alzheimer’s Disease Sequencing Project (ADSP). Gene-disease and protein-disease associations were identified in ADSP and validated in the UK Biobank. This study highlights the importance of integrating multi-omics data to predict genetic risk for complex diseases.

Speaker(s):
Rasika Venkatesh, B.S.
University of Pennsylvania Perelman School of Medicine

Author(s):
Anni Moore, Graduate Student - University of Pennsylvania; Rachit Kumar, B.S. - University of Pennsylvania; Yuki Bradford, MS - University of Pennsylvania; Marylyn Ritchie, PhD - University of Pennsylvania, Perelman School of Medicine;

A Computable Phenotyping Pipeline for Rapid Learning Health Systems with the Molecular Oncology Almanac

Poster Number: P161
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Precision Medicine, Bioinformatics, Clinical Decision Support, Real-World Evidence Generation
Primary Track: Applications
Programmatic Theme: Translational Bioinformatics

We extend the Molecular Oncology Almanac (MOAlmanac) to clinical panel sequencing results from the Veterans Affairs (VA) National Precision Oncology Program (NPOP) with a computable phenotyping pipeline that enables interpretation of results without preselection of diagnosis and incorporation of environmental exposures. We developed a pipeline to phenotype single-nucleotide variants (SNVs), insertion/deletions (indels), copy number variations (CNVs), and fusions present in VA NPOP data, integrate environmental exposure data and then annotate them using the MOAlmanac.

Speaker(s):
Nhan Do, MD, MS, Clinical Informatics Diplomate
Boston VA HCS

Author(s):
Theodore Feldman, PhD - VA Boston Healthcare System; Brendan Reardon; Paul Marcantonio, Computer Science; Robert Zwolinski, B.A. - VA Boston Healthcare System; Anthony Szema, MD - Zucker School of Medicine; Danne Elbers, PhD - VA Boston CSP / MAVERIC; Nathanael Fillmore, PhD - VA Boston Healthcare System; Eliezer Van Allen, MD - Dana Farber Cancer Insitute; Michael Kelley, MD - Durham VAMC; Mary Brophy, MD; Nhan Do, MD, MS, Clinical Informatics Diplomate - Boston VA HCS;

Reducing technical barriers to reusing computable phenotypes

Poster Number: P162
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Reproducibility, Data Sharing, Knowledge Representation and Information Modeling
Primary Track: Applications

Computable phenotypes play important roles in cohort identification for prospective and observational research, but many barriers make reuse of computable phenotypes difficult. Increasing their potential for reuse has many benefits, including saving time and resources, and enabling more readily comparable clinical studies. We have developed a model for packaging any computable biomedical knowledge artifact as a Knowledge Object (KO), a compound digital object containing executable code, a service to run the code, metadata, and a unique persistent identifier. Here, we demonstrate how the core classification knowledge in a rule-based computable phenotype can be reimplemented and packaged as a KO carrying the computable phenotype, two services to deploy and run it, and relevant metadata. The goal of this work is to reduce technical barriers and facilitate computable phenotype reuse by researchers who have no SQL expertise, or whose input data may not be formatted like the data in a typical proprietary or custom-made relational database.

Speaker(s):
Marisa Conte, MLIS
University of Michigan

Author(s):
Marisa Conte, MLIS - University of Michigan; Anurag Bangera, MS - University of Michigan; Farid Seifi, PhD - University of Michigan Medical School;

Graph Neural Network-Driven Interactive Visualization of Signal Pathways for Personalized Drug Target Discovery

Poster Number: P163
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Bioinformatics, Cancer Prevention, Cancer Genetics, Computational Biology, Pharmacogenomics
Working Group: Genomics and Translational Bioinformatics Working Group
Primary Track: Applications
Programmatic Theme: Translational Bioinformatics

The advent of precision medicine has markedly transformed cancer treatment, shifting towards a model that recognizes and addresses the individual molecular differences among patients. This paradigm hinges on selecting therapeutic drugs aimed at specifically targeting the aberrant signaling pathways that fuel tumor growth and progression. However, the intricate nature of intracellular signaling networks, combined with the diversity of genetic alterations in cancer, significantly complicates the identification of optimal therapeutic targets. The web of interactions among genes and proteins within cancer signaling pathways, characterized by complex feedback loops and cross-talk, plays a crucial role in determining cellular fate. Alterations such as copy number variations (CNVs), insertions and deletions (INDELs), single nucleotide variations (SNVs), and gene expression changes can disrupt these pathways, propelling uncontrolled cell proliferation and survival. Addressing this issue necessitates a comprehensive understanding of these pathways and their interactions within the tumor microenvironment and the immune system.Utilizing Graph Neural Networks (GNNs) with advanced visualization tools offers a powerful approach to overcome challenges in personalized cancer treatment. This combination provides a clear platform for clinicians and researchers to understand and use model predictions for informed therapy choices. Our aim is to improve patient outcomes through targeted therapy by integrating these computational methods to navigate drug selection complexities. For each patient, we analyze DNA, CNV, and RNA expression data, aligning these with 15 cancer signaling pathways visualized to show genes, processes, and interactions clearly. This setup helps clarify pathway dynamics for better decision-making.

Speaker(s):
Hui Li, Phd
University of Texas Health Science Center at Houston

Author(s):
jinlian wang, PhD - UTHealth;

Leveraging Large Language Models to Extract Acupoint Body Location Relations from Textual Knowledge Sources

Poster Number: P164
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Information Extraction, Large Language Models (LLMs), Natural Language Processing
Primary Track: Applications

Extracting relations for acupuncture point locations is crucial for understanding the anatomical and functional aspects of acupuncture. This study investigates the use of four models—long short-term memory (LSTM) network, pretrained GPT-3.5-turbo, finetuned GPT-3.5-turbo, and pretrained GPT-4—for this purpose. Employing a structured approach, we define five relation types and use a specific prompt for extraction. Finetuned GPT-3.5-turbo demonstrates superior performance (Overall macro-average F1: 0.91), highlighting the potential of large language models in biomedical research.

Speaker(s):
Yiming Li, Master
UTHealth Science Center Houston

Author(s):
Yiming Li - UTHealth Science Center Houston; Xueqing Peng, PhD - Yale University; Cui Tao, PhD - Mayo Clinic; Hua Xu, Ph.D - Yale University; Na Hong; Na Hong, PhD - Yale University;

A Data-Driven, Computational Drug Repurposing Strategy Suggests New Therapeutic Options to Treat Diabetic Foot Ulcer

Poster Number: P165
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Bioinformatics, Computational Biology, Advanced Disease, Biomarkers
Primary Track: Applications
Programmatic Theme: Translational Bioinformatics

Diabetic foot ulcers (DFUs) are a serious complication with limited treatment options. This study explored drug repurposing as a potential solution. Using DISGENET, we identified DFU-associated genes, and with ENRICHR, we pinpointed relevant biological pathways. Finally, we identified relevant drugs utilizing the associated genes. This approach revealed drug candidates like benzbromarone, cefotetan, and prazosin. Further research is needed to validate their therapeutic potential for DFUs, but this study highlights the potential of computational drug repurposing to expedite the discovery of DFU treatments.

Speaker(s):
Tahir Abdulrehman, PhD
McMaster University

Author(s):
Tahir Abdulrehman, PhD - McMaster University; Shameer Khader, PhD - Imperial College London; Kamlesh Yadav, PhD - School of Engineering Medicine, Texas A&M University;

PubMed Extracted Biomedical Relations Can Enhance Literature-Based Drug Repurposing

Poster Number: P166
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Natural Language Processing, Drug Discoveries, Repurposing, and Side-effect, Data Mining, Deep Learning, Information Extraction
Primary Track: Applications
Programmatic Theme: Translational Bioinformatics

The extraction of biomedical relations from scientific literature can be a critical component for downstream tasks such as drug repurposing, but the noisiness of this extraction process means that often only curated resources are used. Here we compare PubMed extracted biomedical relations to a comprehensive curated resource (RTX-KG2) and show that at least for drug repurposing, KGs could potentially benefit from the inclusion of NLP sourced relations using transformer-based extraction approaches.

Speaker(s):
John Osborne, PhD
University of Alabama at Birmingham

Author(s):
Kuleen Sasse; Abdullateef Almudaifer; Richard Kennedy, MD PhD - University of Alabama at Birmingham;

Multivariate longitudinal trajectories of response to GLP-1 RA therapy

Poster Number: P167
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Precision Medicine, Real-World Evidence Generation, Deep Learning, Population Health
Primary Track: Applications
Programmatic Theme: Translational Bioinformatics

Treatment response in chronic diseases is a multifaceted and heterogeneous process that evolves over time and differs per patient. In practice, it is often defined as a binary outcome at a single point in time, which discards valuable temporal information. The goal of this proposed research is to apply deep learning-based clustering methods to multivariate temporal real-world data to identify subgroups of patients with distinct GLP-1 agonist drug response trajectories.

Speaker(s):
Bhargav Vemuri, MPH
University of Washington

Author(s):
Peter Tarczy-Hornoch, MD - University of Washington; Jennifer Hadlock, MD - Institute for Systems Biology;

Retrospective observational investigation of postoperative atrial fibrillation following coronary artery bypass graft surgery in nationwide sample from Epic Cosmos database

Poster Number: P18
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Clinical Guidelines, Clinical Decision Support, Internal Medicine or Medical Subspecialty
Primary Track: Applications
Programmatic Theme: Clinical Informatics

New guidelines recommend anticoagulation for postoperative atrial fibrillation (POAF) after a coronary artery bypass graft for at least 60 days if there is low risk of bleeding. Using Epic Cosmos data, this study determines differences in demographics and complications rate between patients anticoagulated for POAF and patients not on anticoagulation. We see mortality rates at 1 year fall in anticoagulated patients and no significant difference between rates of ischemic stroke.

Speaker(s):
Kathryn Tsai, B.S.
Carle Illinois College of Medicine

Author(s):
Daniel Cheah, MS - Carle Illinois College of Medicine; Yousra Khalid, MBBS - Carle Foundation Hospital; Bryan McConomy, MD - Carle Health; Krishna Kurpad, MBBS - Carle Foundation Hospital; Sanjay Mehta, MD - Carle Foundation Hospital, Carle Illinois College of Medicine;

Using Contactless Sensors to Remotely Monitor Falling

Poster Number: P130
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Self-care/Management/Monitoring, Aging in Place, Tracking and Self-management Systems, Evaluation
Primary Track: Applications
Programmatic Theme: Consumer Health Informatics

One in three adults over the age of 65 falls each year making falls the leading cause of injury and death for older adults. The primary goal of this study is to evaluate a contact-free, passive sensor to monitor an elderly population for falls. Preliminary results demonstrate, across a total of 343 fall and non-fall events, the contactless sensor demonstrated a 0% false positive rate during stunt session activities and a 72% sensitivity for falls.

Speaker(s):
Alisha Johnson, PhD, RN
University of Missouri

Author(s):
Chang-Chun (Max) Chen, PhD - University of Arizona, School of Nursing; Shu-Fen Wung, PhD, MS, RN, ACNP-BC, FAAN - University of Arizona; Marjorie Skubic, PhD - University of Missouri;

Evaluating Patient Access to Abortion Healthcare after Dobbs v. Jackson Women’s Health Organization with Cross-Institutional Aggregate EHR Data

Poster Number: P152
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Real-World Evidence Generation, Diversity, Equity, Inclusion, Accessibility, and Health Equity, Data Mining, Data Sharing, Telemedicine, Healthcare Quality, Population Health, Causal Inference
Primary Track: Applications
Programmatic Theme: Public Health Informatics

When Dobbs v. Jackson Women’s Health Organization (“Dobbs”) overturned Roe v. Wade, several states enacted “trigger laws" that immediately banned or severely restricted access to abortion services. We evaluated whether trigger laws significantly changed patient access to abortion services and their modality of such access. Data were collected from a graphical user interface (GUI) of Cosmos, a cross-institution, Epic-facilitated data aggregation tool of a U.S. Census-representative population of over 240 million patients and 4 billion patient encounters [3]. We used a diff-in-diff analysis of monthly cross-sectional data collected pre- and post-Dobbs in the number of encounters for pharmacological abortions (prescribing mifepristone or misoprostol)offered via telehealth or in-person – for patients residing in trigger-law states to those for patients residing in states neighboring the trigger law states without bans or severe abortion restrictions at the time or in states with shield laws. We find a significant decrease in the number of encounters for pharmacological abortions by patients residing in states with trigger laws compared to neighboring states or states with shield laws (-45.9%, 95% CI [-83.6%, -6.78%]). We found a marginally significant increase in telehealth appointments for pharmacological abortions by patients residing in trigger law states. We have preliminary quantitative evidence of how patients in trigger law states are changing their abortion and reproductive care access. This study provides a preliminary application on how cross-institutional efforts to aggregate EHR data can aid in evaluating how health-based policies change patient decision-making and health-based outcomes and variably impact patients, especially for health equity.

Speaker(s):
Ratnalekha Viswanadham, PhD
NYU Grossman School of Medicine

Author(s):
Yuhan (Betty) Cui, MS - NYU Grossman School of Medicine; Chelsea Twan, MS - NYU Grossman School of Medicine; Devin Mann, MD - NYU Grossman School of Medicine;

Helios: A Platform for Early Childhood Amblyopia Detection using Fixation Eye Movements

Poster Number: P168
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Large Language Models (LLMs), Workflow, Deep Learning, Imaging Informatics, Machine Learning
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics

Amblyopia is a neurodevelopmental disorder that is the primary cause of preventable monocular visual impairment among children, with prevalence rates of 3% to 5% globally and 2% to 3% in the United States, that is, 3% to 4% in those aged 1 to 5 years with 677,000 children affected in the US and 99.2 million children globally. Our study offers a novel, machine learning-based framework that analyzes eye movement data with a level of objectivity and precision previously unattainable with standard screening techniques. Utilizing advanced signal processing, we converted raw eye-tracking data from a pediatric cohort into a series of detailed images that reflect the complex dynamics of ocular movement. These images, capturing subtle variations indicative of amblyopia, were interpreted by the fine-tuned Gemini model. A domain expert in ophthalmology was instrumental in crafting the prompts used to train the model, ensuring clinical relevance in the interpretations. The classification results of the Gemini model were promising, showing a notable ability to distinguish between normal and amblyopic subjects. The model was particularly adept at identifying more pronounced cases of amblyopia, highlighting its potential as a scalable and non-invasive diagnostic tool. Our interdisciplinary approach bridges computational innovation with practical medical expertise, demonstrating the utility of AI in enhancing diagnostic methodologies. This research advances medical informatics and underscores the transformative impact of AI in healthcare. It paves the way for further studies to refine the model and explore its integration into clinical practice, with the ultimate goal of improving amblyopia detection and outcomes in pediatric ophthalmology.

Speaker(s):
Dipak Upadhyaya, PhD Student
Case Western Reserve University

Author(s):
Aasef Shaikh, MD, PhD - Department of Neurology, Case Western Reserve University, School of Medicine; Katrina Prantzalos, MS - Case Western Reserve University; Pedram Golnari, MD - Case Western Reserve University; Fatema Ghasia, Associate Professor/ MD - Visual Neuroscience Laboratory, Cole Eye Institute, Cleveland Clinic, Cleveland; Satya Sahoo;

Evaluating the Impact of Lab Test Results on Large Language Models Generated Differential Diagnoses from Clinical Case Vignettes

Poster Number: P169
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Large Language Models (LLMs), Clinical Decision Support, Information Retrieval, Personal Health Informatics, Diagnostic Systems
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics

Background
While existing studies evaluate the accuracy of differential diagnosis derived from case reports by large language models (LLMs), the role of lab test results in these differential diagnosis predictions is unknown. This study aims to evaluate the role of lab test results in improving the accuracy of differential diagnosis using five large language models namely, GPT-4, GPT-3.5, Llama-2, Claude2, and Mixtral using clinical case reports collected from the PMC-Patients dataset.
Methods
Clinical vignettes were manually generated to let LLMs consider the patient's Age, Gender, Symptoms, Lab tests, and the full Case Report and any pertinent details to formulate the response. Using optimized prompts five LLMs were used to generate the top 10, top 5 differential diagnoses, and the top diagnosis for the presented clinical vignette with and without lab test results. The predictions automatically evaluated the response using GPT-4 in terms of ‘Exact Match:1.0’, ‘Relevant:0.5’, and ‘Incorrect:0.0’ by comparing the actual and predicted diagnosis. Based on the score’s accuracy is calculated as the final metric.
Results
All models performed superior when lab test result information is included in the case report. The Mixtral-8x7B LLM performed as the best model with 50-55% accuracy in all three cases including lab test result information. The performance of the best model is reduced by 30-35% when lab test result information is excluded.
Conclusion
This study reports that the accuracy of differential diagnoses from LLMs improves substantially when lab test results are included, underscoring their critical role in accurate medical diagnosis.

Speaker(s):
Zhe He, PhD, FAMIA
Florida State University

Author(s):
Qiao Jin, M.D. - National Institutes of Health; Shubo Tian; Karim Hanna, MD - University of South Florida Health; Zhiyong Lu, PhD - National Library of Medicine, NIH; Zhe He, PhD, FAMIA - Florida State University;

Evaluating Clinical Text Summarization Efficacy by Open-Source Large Language Models through Downstream Predictive Performances

Poster Number: P170
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Large Language Models (LLMs), Natural Language Processing, Deep Learning, Machine Learning
Primary Track: Foundations

LLMs demonstrate potential for various medical applications, with clinical text summarization being an important one, which synthesizes key information from large volumes of clinical notes and could enable clinicians to spend more time on patient care. To this end, understanding the capabilities of LLMs to summarize complex clinical notes is crucial. However, creating ground truth summaries of prolonged clinical notes for evaluation can be difficult or even infeasible. This study aims to explore an alternative approach to assessing whether LLMs can identify valuable information in the clinical notes by leveraging downstream clinical NLP tasks. Our experiments indicate that generated summaries by open-source LLMs could maintain more than 90% of predictive values for ICD coding and readmission prediction, and the downstream modeling presented different patterns than ROUGE in comparing summaries from different LLMs. We also found these generated summaries were beneficial for enhancing downstream modeling through data augmentation.

Speaker(s):
Jinghui Liu, PhD
The Australian e-Health Research Centre, CSIRO

Author(s):
Anthony Nguyen, PhD - The Australian e-Health Research Centre, CSIRO;

Filling the gaps: leveraging large language models for temporal harmonization of clinical text across multiple medical visits for clinical prediction

Poster Number: P173
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Large Language Models (LLMs), Deep Learning, Natural Language Processing, Machine Learning, Health Equity, Clinical Decision Support
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics

Electronic health records offer great promise for early disease detection, treatment evaluation, information discovery, and other important facets of precision health. Clinical notes, in particular, may contain nuanced information about a patient’s condition, treatment plans, and history that structured data may not capture. As a result, and with advancements in natural language processing, clinical notes have been increasingly used in supervised prediction models. To predict long-term outcomes such as chronic disease and mortality, it is often advantageous to leverage data occurring at multiple time points in a patient’s history. However, these data are often collected at irregular time intervals and varying frequencies, thus posing an analytical challenge. Here, we propose the use of large language models (LLMs) for robust temporal harmonization of clinical notes across multiple visits. We compare multiple state-of-the-art LLMs in their ability to generate useful information during time gaps, and evaluate performance in supervised deep learning models.

Speaker(s):
Inyoung Choi, B.S.
University of Pennsylvania

Author(s):

Ambient AI Scribes: A Qualitative Evaluation of Clinician Perspectives

Poster Number: P174
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Large Language Models (LLMs), Documentation Burden, Informatics Implementation
Primary Track: Applications
Programmatic Theme: Clinical Informatics

AI-powered ambient scribe tools can help address clinical documentation burden. Our objective is to use the RE-AIM/PRISM framework to evaluate an ambient AI scribe pilot from the perspective of clinicians, through written surveys and semi-structured interviews. Pre vs post survey analysis includes a comparison of EHR burden, burnout, and provider acceptance. The post survey also assesses clinician satisfaction of the tool and tool usability. We also plan to conduct semi-structured interviews of pilot clinicians with a deductive approach to coding and thematic analysis that incorporates inductive thematic coding to explore emerging themes. Preliminary results of a separate midpoint survey (n=36) found 92% of clinicians thought the tool was easy to learn to use. The majority of pilot clinicians (75%) thought the tool improved documentation efficiency, but only 36% of pilot clinicians thought the tool improved documentation quality. Pilot clinicians believed the tool saved an average of 20 minutes of documentation time per half-day of clinic. Several areas for improvement were identified including 1.) incorporation of interactive voice commands and natural language editing, 2.) utilization of full context from the chart and 3.) better summarization for complex encounters with multiple problems. Lessons learned include reflections on how to identify which clinicians may be most likely to benefit from using the tool. The results from our pilot will also help to inform strategic plans to scale the solution to 1500 users. We are currently in the process of administering post-surveys, conducting pre vs post survey analysis, and launching recruitment for the semi-structured interviews.

Speaker(s):
Shreya Shah, MD
Stanford University

Author(s):
Stephen Ma, MD, PhD - Stanford University School of Medicine; Margaret Smith, MBA - Stanford School of Medicine; Anna Devon-Sand, MPH - Stanford University; Yejin Jeong, BA - Stanford University; Trevor Crowell, BA - Stanford University; Clarissa Delahaie, BAS - Stanford Health Care; Caroline Hsia, MEng - Stanford Health Care; Betsy Yang, MD - Stanford University; April Liang, MD - Stanford University; Steven Lin, MD - Stanford University School of Medicine; Tait Shanafelt, MD - Stanford University; Michael Pfeffer, MD - Stanford University; Christopher Sharp, MD - Stanford University School of Medicine; Patricia Garcia, MD - Stanford University, School of Medicine;

BioCiteGPT: Recurrent Retrieval-Augmented Large Language Models for Faithful Biomedical Citation Recommendation

Poster Number: P175
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Large Language Models (LLMs), Natural Language Processing, Knowledge Representation and Information Modeling
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics

Citation recommendation is the task of providing relevant articles to cite for a given text, and plays a key role in scientific research. Due to the ever-growing literature, finding appropriate citations is becoming more challenging and effective citation recommenders are of greater importance. In this work, we present a LLM-based recurrent retrieval-augmented framework for citation recommendation. In addition, our recommender works on a sentence level, as opposed to other models which are article or topic-level. Generative LLMs such as ChatGPT have shown effectiveness on a variety of tasks, but are prone to hallucinations (creating imaginary information), especially when used for citation recommendation. Our approach combines generative LLMs and retrieval augmentation whereby the two approaches are used to enhance each other. In addition, we train a LLaMA2-based discriminator which further distinguishes whether an article is relevant to a given sentence, further enhancing the reliability of our process. We show that combining these approaches improves the reliability of citations generated by ChatGPT. Our study is supported by a dataset we constructed consisting of sentences from PubMed articles related to Alzheimer's disease, along with information about articles cited by those sentences.

Speaker(s):
Jeffrey Zhang, PhD
Yale University

Author(s):
Qianqian Xie, PhD - Yale University; Jeffrey Zhang, PhD - Yale University; Yan Wang, PhD - Yale University; Fongci Lin, PhD - Yale University; Yi-chung Wang, BS - Yale University; Ziqing Ji, BS - Yale University; Haoting Chen, BS - Yale University; Qingyu Chen, PhD - Yale University; Hua Xu, Ph.D - Yale University;

Assessing Methods to Handle Missing Electronic Health Record Data for Prediction Models

Poster Number: P176
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Machine Learning, Clinical Decision Support, Critical Care, Pediatrics
Primary Track: Foundations
Programmatic Theme: Clinical Informatics

We used electronic health record data to generate a simulated complete dataset and evaluated multiple methods for imputation and their effects on predictive performance. We found that last observation carried forward and native support of missing data by machine learning models, such as gradient boosted trees, offered the least computationally intensive method of handling missing data with good performance.

Speaker(s):
Jean Digitale, MPH, RN
UCSF

Author(s):
Deborah Franzon, MD - UCSF; Mark Pletcher, MD MPH - UCSF; Charles McCulloch, PhD - UCSF; Efstathios Gennatas, MBBS, PhD - UCSF;

Cross-Organization Aggregate EHR Audit Log Data Imputation

Poster Number: P177
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Machine Learning, Data Standards, Data Sharing
Primary Track: Applications

Implementing electronic health records (EHRs) has generated valuable vendor audit log data. However, missing data poses challenges for longitudinal and multi-site analysis. Advanced machine learning (ML) techniques may better capture missing patterns and preserve those observations compared to basic imputation methods. The primary objective of this study is to compare the efficacy of specific imputation strategies across diverse EHR usage metric data fields and consider implementation feasibility for organizations.

Speaker(s):
Huan Li, MS
Yale

Author(s):
Huan Li, MS - Yale; Nate Apathy, PhD - University of Maryland; A J Holmgren, PhD - University of California, San Francisco; Edward Melnick, MD - Yale University, School of Medicine; Robert McDougal, Ph.D. - Yale School of Public Health;

A Study of Challenges In Algorithmic Transportability Between VHA Sites

Poster Number: P178
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Machine Learning, Interoperability and Health Information Exchange, Reproducibility, Controlled Terminologies, Ontologies, and Vocabularies, Data Standards, Data Transformation/ETL
Primary Track: Foundations
Programmatic Theme: Clinical Informatics

There is a rising awareness of the difficulties in transporting predictive medical algorithms outside their health systems of origin. But, the challenges facing algorithm re-use within a single health system are less understood. In a study of Veterans Health Administration care sites, we found models of outpatient visit frequency trained on single-site data performed very differently -- usually worse -- at other sites. Differences in patient populations and data representations -- especially lab labeling -- contributed to performance differences.

Speaker(s):
Kimberley Cox, MS
Vanderbilt University

Author(s):
Kimberley Cox, MS - Vanderbilt University; Carlos Grijalva, MD, MPH - Vanderbilt University Medical Center; Candace McNaughton, MD - University of Toronto; Michael Matheny, MD, MS, MPH, FACMI, FAMIA - Vanderbilt University Medical Center; Thomas Lasko, MD, PhD - Vanderbilt University Medical Center; Amanda Mixon, MD - Vanderbilt University Medical Center;

Relapse Prediction through Convolutional Auto-encoders and Clustering for Patients with Psychotic Disorders using Wearable Data

Poster Number: P180
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Mobile Health, Deep Learning, Machine Learning, Personal Health Informatics
Primary Track: Applications
Programmatic Theme: Consumer Health Informatics

Digital phenotyping is crucial for remotely monitoring psychotic-disorder relapse. We developed a novel relapse-prediction model using neural-network-based anomaly detection and clustering. Analyzing e-Prevention data from 10 patients over six months, we created multivariate time-series profiles, extracted latent features via convolutional autoencoders, and identified relapse clusters. Our model showed promising results (PR-AUC = 0.877, ROC-AUC = 0.548, harmonic-mean = 0.675), highlighting the power of unsupervised learning and the importance of sleep data in relapse detection.

Speaker(s):
Casey Taylor, PhD
Johns Hopkins University

Author(s):
Yujie Yan; Traci Speed, MD, PhD - Johns Hopkins University; Casey Taylor, PhD - Johns Hopkins University;

Outpatient Portal Use and Physiological Outcomes Management during Pregnancy

Poster Number: P181
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Mobile Health, Racial Disparities, Patient Engagement and Preferences, Chronic Care Management
Primary Track: Applications

Our retrospective study investigated the association between hemoglobin A1c (A1c), systolic and diastolic blood pressure (BP) and outpatient portal use during pregnancy. Our results provide evidence to support a relationship between outpatient portal use and physiological outcomes during pregnancy. Our findings emphasize the possible association with improved maternal and infant outcomes. Further research should focus on the implementation and design of patient portals for pregnancy.

Speaker(s):
Naleef Fareed, PhD MBA
The Ohio State University Dept Biomedical Informatics

Author(s):
Athena Stamos, BS - The Ohio State University; Naleef Fareed, PhD MBA - The Ohio State University Dept Biomedical Informatics;

Impact of Clinical Trial Virtualization on Recruitment in Underserved Communities for a Type 2 Diabetes mHealth Intervention

Poster Number: P182
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Mobile Health, Diversity, Equity, Inclusion, Accessibility, and Health Equity, Evaluation, Health Equity
Primary Track: Applications
Programmatic Theme: Consumer Health Informatics

mHealth interventions are increasingly investigated in randomized clinical trials (RCTs), but representation of underserved populations remains a challenge. The move towards virtualization of clinical trials during the COVID-19 pandemic has highlighted the viability of decentralized clinical trials utilizing technology and, if adopted, may reduce barriers to participation in RCTs, even after the pandemic. In this study, we discuss an approach to virtualizing an RCT for a Type 2 Diabetes (T2DM) mHealth intervention for a medically underserved population. We compare demographic, clinical, and technical literacy characteristics of participants recruited in-person pre-pandemic versus virtually after the pandemic’s onset. Gender, education, age, HbA1c, employment status, technical literacy, combined family income, and language spoken at home were not significantly different between participants recruited in-person versus virtually. Race/ethnicity and birth location were significantly different, although this may be due to differences in demographic characteristics between patients receiving care at the two FQHCs pre-pandemic, and the remaining four FQHCs. Overall, our approach did not lead to any delays in recruitment or significant changes in the population recruited into the study. Consequently, these virtualization strategies may be used in future trials testing mHealth and other technological interventions, potentially reaching a broader and more diverse population, without exacerbating the under-representation of medically underserved populations or increasing the demands on busy, FQHCs and other under-resourced practices.

Speaker(s):
Elizabeth Campbell, MS, MSPH, PhD
Columbia University Department of Biomedical Informatics

Author(s):
Elizabeth Campbell, MS, MSPH, PhD - Columbia University Department of Biomedical Informatics; Pooja Desai, BA, MA - Columbia University Irving Medical Center; Arlene Smaldone, PhD - Columbia University; Haomiao Jia, PhD - Columbia University; Andrea Cassells, MPH - Clinical Directors Network; Jacqeline Cortez, MPH - Clinical Directors Network; TJ Lin, MPH - Clinical Directors Network; Jonathan Tobin, PhD - Clinical Directors Network; Lena Mamykina, PhD - Columbia University;

Improving Generalizability of Extracting Social eterminants of Health Using Large Language Models through Prompt-tuning

Poster Number: P183
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Natural Language Processing, Large Language Models (LLMs), Information Extraction
Primary Track: Foundations
Programmatic Theme: Clinical Research Informatics

The progress in natural language processing (NLP) using large language models (LLMs) has greatly improved patient information extraction from clinical narratives. However, most methods based on the fine-tuning strategy have limited transfer learning ability for cross-domain applications. This study proposed a novel approach that employs a soft prompt-based learning architecture, which introduces trainable prompts to guide LLMs toward desired outputs. We examined two types of LLM architectures, including encoder-only GatorTron and decoder-only GatorTronGPT, and evaluated their performance for the extraction of social determinants of health (SDoH) using a cross-institution dataset from the 2022 n2c2 challenge and a cross-disease dataset from the University of Florida (UF) Health. The results show that decoder-only LLMs with prompt tuning achieved better performance in cross-domain applications. GatorTronGPT achieved the best F1 scores for both datasets, outperforming traditional fine-tuned GatorTron by 8.9% and 21.8% in a cross-institution setting, and 5.5% and 14.5% in a cross-disease setting.

Speaker(s):
Cheng Peng, PhD
University of Florida

Author(s):
zehao yu - University of Florida; Kaleb Smith, PhD - NVIDIA; Weihsuan Jenny Lo-Ciganic, PhD - University of Florida; Jiang Bian, PhD - University of Florida; Yonghui Wu, PhD - University of Florida;

Prompt Engineering for Portable Clinical Note Structure Awareness

Poster Number: P184
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Natural Language Processing, Large Language Models (LLMs), Interoperability and Health Information Exchange, Information Retrieval
Primary Track: Applications
Programmatic Theme: Clinical Informatics

Clinical note structure and metadata elicitation are critical for many downstream clinical NLP applications, but portability, robustness, and cost of existing solutions continues to hamper the adaptability and adoption across healthcare systems. Prompt engineering strategies for large language models have demonstrated the potential for generalizable performance at a relatively low cost. Our study exhibits the promise of a dynamic few-shot prompting technique for section classification and proposes future approaches for document structure awareness.

Speaker(s):
Kurt Miller, M.S.
University of Minnesota

Author(s):
Steven Bedrick - Oregon Health & Science University; Hongfang Liu, PhD - University of Texas Health Science Center at Houston;

SmartMedicalCodex - IA-Based Wizard for CIE-10 Coding

Poster Number: P185
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Natural Language Processing, Deep Learning, Large Language Models (LLMs), Controlled Terminologies, Ontologies, and Vocabularies
Primary Track: Applications
Programmatic Theme: Clinical Informatics

This paper presents SmartMedicalCodex (SMC), a deep learning system developed to assist in the ICD-10 coding
process within electronic health records (EHR) in Spain. SMC fills a significant gap in current solutions as most are
designed for English or Chinese contexts and require full diagnostics and procedure coding. Using recurrent neural
networks and transformers, SMC provides a comprehensive interpretation of EHRs in Spanish, emphasizing its
commitment to transparent and inclusive technology development that respects and understands the diversity within
the Spanish-speaking healthcare landscape. The system is built on a transformer-based Mixture of Experts and
rigorously tuned to an augmented dataset. It is also validated using real clinical data. SMC models have obtained
high F1 scores, mainly around 0.9 points, and they were trained using an increased dataset. This is especially
remarkable given the difficulty of predicting total codes across different medical specialties and languages.

Speaker(s):
Carlos Luis Parra-Calderon, MSc
"Virgen del Rocio" University Hospital - Institute of Biomedicine of Seville

Author(s):
Adrian Ramos Cápitas, BEng - Keedio; Juan Carlos Rivera Domínguez, MSc - Keedio;

Representation and Extraction of Drug Information: A Study Focusing on the Needs of Older Adults

Poster Number: P186
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Natural Language Processing, Machine Learning, Information Extraction, Deep Learning
Primary Track: Foundations
Programmatic Theme: Clinical Informatics

As the population ages, older adults (roughly those aged 65+) face the challenge of managing complex medication
regimens, requiring significant information about the drugs they are taking. However, supporting medication information technologies like alerts, apps, and chatbots face hurdles due to the unstructured nature of the authoritative source of drug information–FDA Structured Product Labels, commonly known as “drug labels”–which are almost entirely free-text documents. This paper investigates the effectiveness of using natural language processing (NLP) to extract critical drug information from these documents using a small set of 20 drug labels annotated with drug indications, contraindications, adverse reactions, dosages, and usage instructions. The study employs (and makes publicly available) a BERT model pre-trained on 122 thousand drug labels. We also explored transfer learning using an existing adverse reaction dataset. Despite the limited size of our pilot data, resulting in a promising performance, including an F1 of 94 for adverse reactions, 80 for dosages, 90 for instructions, and 76 for contraindications, though only 52 for indications. With further development, this line of research can lead to enhanced medication management for older adults.

Speaker(s):
Abayomi Adegunlehin, MS
University Of Texas Health Science Center at Houston

Author(s):
Kirk Roberts, PhD - University of Texas Health Science Center at Houston; Francis Ifiora, MS - University of Texas Health Science Center at Houston; Tasneem Kaochar, MS - University of Texas Health Science Center at Houston;

Data-driven automated classification algorithms for acute health conditions: Applying PheNorm to Anaphylaxis

Poster Number: P187
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Natural Language Processing, Real-World Evidence Generation, Machine Learning, Population Health
Primary Track: Foundations
Programmatic Theme: Public Health Informatics

Accurate identification of anaphylaxis using observational data is important for medical product safety surveillance, but difficult to both diagnose clinically and recognize algorithmically. Traditional phenotyping methods rely on expensive gold standard training data and manual feature engineering. Herein we apply an automated approach, PheNorm, to create a computable phenotype for identifying patients with anaphylaxis using NLP, machine learning, and low-cost silver-standard training labels. Performance was comparable to a recently published, higher-cost manual phenotyping effort.

Speaker(s):
Joshua Smith, PhD
Vanderbilt University Medical Center

Author(s):
Joshua Smith, PhD - Vanderbilt University Medical Center; Daniel Park, BS - Vanderbilt University Medical Center; Jill Whitaker, MSN - Vanderbilt University Medical Center; Michael McLemore, BSN - Vanderbilt University Medical Center; Robert Winter, BA - Vanderbilt University Medical Center; Arvind Ramaprasan, MS - Kaiser Permanente Washington Health Research Institute; David Cronkite, MS - Kaiser Permanente Washington Health Research Institute; Saranrat Wittayanukorn, PhD - US Food and Drug Administration; Danijela Stojanovic, PharmD, PhD - US Food and Drug Administration; Yueqin Zhao, PhD - US Food and Drug Administration; Sarah Dutcher, PhD - US Food and Drug Administration; Kevin Johnson, MD, MS - University of Pennsylvania; David Carrell, PhD - Kaiser Permanente Washington Health Research Institute; Brian Williamson, PhD - Kaiser Permanente Washington Health Research Institute;

Real-Time Automated Billing for Tobacco Treatment: A CDS Hook Approach for Simulating Clinician Facing Coding Prompts Within EHRs

Poster Number: P188
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Natural Language Processing, Interoperability and Health Information Exchange, Rule-based artificial intelligence, Workflow
Primary Track: Applications
Programmatic Theme: Clinical Informatics

A machine learning prediction model, CigStopper, demonstrated automation of smoking cessation billing using a CDS Hook in a simulated clinical workflow. The approach showcases the real-time interoperability of an externally hosted API interfacing with EHRs using a public sandbox proxy with SMART on FHIR standards. This proof-of-concept emphasizes the functional potential for scalable, EHR-agnostic CDS tools to improve clinical care by automating billing processes with real-time decision support.

Speaker(s):
Derek Baughman, MD
Vanderbilt University

Author(s):
Layth Qassem, PharmD - VUMC; Eric Morgan, MS - Plasma FHIR; Lina Sulieman, PhD - Vanderbilt University Medical Center; Hilary Tindle, MD, MPH - Vanderbilt University Medical Center; Michael Matheny, MD, MS, MPH, FACMI, FAMIA - Vanderbilt University Medical Center; Daniel Fabbri, PhD - Vanderbilt; Adam Wright, PhD - Vanderbilt University Medical Center;

LEME: Open-Sourced Large Language Models for Vision Research

Poster Number: P189
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Natural Language Processing, Deep Learning, Machine Learning
Primary Track: Applications
Programmatic Theme: Clinical Informatics

Despite the potential of Large Language Models (LLMs) in healthcare, they lack domain specificity, often resulting in inaccurate, outdated, and hallucinated responses. Specifically, there is currently no publicly available LLM in Ophthalmology. This study aims to develop and validate EyeLLaMA – Ophthalmology domain-specific LLMs using over 150K instructions, covering six specific applications. The results demonstrate EyeLLaMA consistently outperformed other open-sourced LLMs by a large margin in all the applications.

Speaker(s):
Qingyu Chen, PhD
Yale University

Author(s):

Health Text Simplification: An Annotated Corpus for Digestive Cancer Education and Novel Strategies for Reinforcement Learning

Poster Number: P190
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Natural Language Processing, Large Language Models (LLMs), Deep Learning
Primary Track: Foundations
Programmatic Theme: Consumer Health Informatics

We introduce the Simplified Digestive Cancer (SimpleDC) corpus, comprising parallel texts of original and simplified cancer education materials. We explore supervised fine-tuning, reinforcement learning (RL), and an innovative reinforcement learning with human feedback (RLHF) applied to Llama 2. We benchmark against GPT-4 in zero-shot and few-shot settings. Our novel RLHF methodology improves simplification performance over SFT alone. All trained Llama 2 models outperform GPT-4 in automatic adequacy metrics and manual assessments of meaning preservation.

Speaker(s):
Md Mushfiqur Rahman, Graduate Research Assistant
George Mason University

Author(s):
Md Mushfiqur Rahman, Graduate Research Assistant - George Mason University; Mohammad Sabik Irbaz, PhD Student - George Mason University; Kai North, PhD Student - George Mason University; Michelle Williams, PhD - George Mason University; Marcos Zampieri, PhD - George Mason University; Kevin Lybarger, PhD - George Mason University;

Behavioral Testing of Implicit Bias: the Downstream Impact of the Targeted Replacement of Pejorative and Laudative Terms in Unstructured Clinical Notes

Poster Number: P191
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Natural Language Processing, Health Equity, Fairness and Elimination of Bias, Machine Learning
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics

Pejorative language in unstructured clinical notes (e.g., "unkempt pt") can be indicative of implicit clinician bias towards a patient. We hypothesize that, just as pejorative terms negatively influence clinicians reading notes, they negatively alter predictions made by models trained on notes. In this study, we use the Behavioral Testing framework with a BERT-based model fine-tuned on predicting diagnosis codes to show that even when clinical details are held constant pejorative terms significantly alter model output.

Speaker(s):
Paul Heider, PhD
Medical University of South Carolina

Author(s):
Dmitry Scherbakov, PhD - Medical University of South Carolina; Jihad Obeid, MD - Medical University of South Carolina;

Feasibility and Initial Impact of Automated Digital Monitoring in Post-Angioplasty Care

Poster Number: P192
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Patient / Person Generated Health Data (Patient Reported Outcomes), Chronic Care Management, Healthcare Quality, Telemedicine, Behavioral Change, Self-care/Management/Monitoring, Patient Engagement and Preferences, Health Equity
Primary Track: Applications
Programmatic Theme: Clinical Informatics

This study assessed an automated program's impact on post-PCI care for 20 patients, revealing successful implementation and 20% requiring medication adjustments due to elevated blood pressure. Patients reported high satisfaction (4.9/5), indicating effective support. This grant-funded initiative suggests automated programs could transform cardiac catheterization labs into health activation centers, potentially revolutionizing immediate post-PCI care by improving adherence and satisfaction and reducing major adverse cardiovascular events.

Speaker(s):
Lina (Yimeng) Du, BS; MD in progress
UC Davis School of Medicine

Author(s):

Review of sexually transmitted infection test results in MyChart among adolescent and young adult patients at a large urban hospital: Implications for future interventions

Poster Number: P193
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Patient Engagement and Preferences, Patient / Person Generated Health Data (Patient Reported Outcomes), Behavioral Change, Infectious Diseases and Epidemiology, Racial Disparities
Primary Track: Applications
Programmatic Theme: Public Health Informatics

Almost half of all reported sexually transmitted infection (STI) cases are concentrated youth aged 15 to 24 years old. Patient portals may support STI prevention via disclosure validated test results to sex partners. We examine use of MyChart to view test results after STI testing among youth at a large urban hospital.

We analyzed Epic EHR and patient portal data among patients aged 15 to 25 years between March 2022 and April 2023. Patients were included in the analysis if they received STI testing within the study period. Chi-square and one-way ANOVA tests were used to compare characteristics across three categories of reviewing test results in MyChart: a) reviewing in 2 weeks or less, b) reviewing in more than two weeks, and c) no review.

A total of 19,193 unique patients were included in the analysis. Median age at first STI test was 21.3 (18.9-23.4) and 72.5% identified as female. 60.8% of patients reviewed their STI test result on the day it was posted. Younger age was associated with not viewing test results in MyChart. Male patients comprised 29.3% of participants who reviewed results in more than two weeks and 37.3% of patients did not review results. Hispanic/Latino patients were overrepresented among patients who did not review test results.

Findings suggest both the promise of novel STI interventions that exploit test result review as well as the need to focus on how to improve timely review among younger adolescents, male patients, and Hispanic/Latino communities.

Speaker(s):
Kevon-Mark Jackman, DrPH, MPH
Johns Hopkins School of Medicine

Author(s):
Kevon-Mark Jackman, DrPH, MPH - Johns Hopkins School of Medicine; Laura Prichett, PhD, MHS - Johns Hopkins University; Yong Zeng, MD, ScM - Johns Hopkins University; Yongyi Lu, BA - Johns Hopkins University; Bareng Nonyane, PhD, MSc - Johns Hopkins University; Kevin Johnson, MD, MS - University of Pennsylvania; Harold Lehmann, MD, PhD - Johns Hopkins University; Maria Trent, MD, MPH - Johns Hopkins University;

Who’s Using the Portal and for What? Spanish-speaking Patients’ Portal Message Use across 3 Health Systems in North Texas

Poster Number: P194
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Patient Engagement and Preferences, Diversity, Equity, Inclusion, Accessibility, and Health Equity, Mobile Health, Personal Health Informatics
Primary Track: Applications
Programmatic Theme: Consumer Health Informatics

Understanding portal use patterns, including who is sending messages and about what, may enhance interventions to increase portal uptake and use among Spanish-speakers, a large and underserved population. Findings from this study of portal message data from three healthcare systems in North Texas suggest that interventions should include education of care partners and instruction about appropriate use of messaging versus structured portal functions.

Speaker(s):
Robin Higashi, PhD
UT Southwestern Medical Center

Author(s):
Emily Repasky, MA - UTSW Medical Center; Luis Pena, BS - UT Southwestern Medical Center; Libertad Gracia, BA CHW - UT Southwestern Medical Center; Bhaskar Thakur, Ph.D./Biostatistician - UT Southwestern Medical Center; Samuel McDonald, MD - UT Southwestern; Christoph Lehmann, MD, FAAP, FACMI, FIAHSI - UT Southwestern; Ann Navar, MD, PhD - UT Southwestern Medical Center; Eric Peterson, MD - UT Southwestern; DuWayne Willett, MD - University of Texas Southwestern Medical Center; Ellen O'Connell, MD - UT Southwestern Medical Center; Brett Moran, MD - Parkland Health; Christopher Clark, MPA - Parkland Health; Ferdinand Velasco, MD - Texas Health Resources; Andrew Masica, M.D., M.S.C.I. - Texas Health Resources; Clark Walker, MPH - Texas Health Resources; Dwayne Hoelscher, DNP - Texas Health Resources; Robert Turer, MD, MSE - UT Southwestern Medical Center;

Paucity of AI implementation in Clinical Pediatrics: A Systematic Review

Poster Number: P195
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Pediatrics, Machine Learning, Informatics Implementation
Primary Track: Applications
Programmatic Theme: Clinical Informatics

There is abundant research on AI, however, much is focused on the adult population with few algorithms actually implemented in clinical practice. In this abstract we review studies that represent AI implementation in pediatric clinical care. We report on study demographics, implementation settings, predictive, care process, outcome, and human performance measures. We reviewed 8004 identified articles for inclusion based on the title and abstract and rejected articles which were not truly AI or not implemented in pediatric clinical settings, resulting in 126 articles for full text review. and 17 met the inclusion/exclusion parameters. AI systems in 14 studies (82%) were used by providers and clinicians, 2 were (12%) used by a patient/parent, while 1 (6%) AI system was used by clinical research coordinators. Six studies (35%) were implemented in the emergency department setting, 6 (35%) in the outpatient setting, 4 (24%) in ICUs, 3 in the inpatient setting, and 2 (12%) in primary care; 5 (30%) were implemented in multiple settings. Post-implementation predictive performance with comparisons to prior validations were not reported in 70% of studies. The majority (12, 70%) reported improvements in care process measures. AI implementation improved patient outcomes in 4 studies while impact of AI on outcomes was not reported in 7 (41%). Only One included measurement on human performance. The number of AI implementations in pediatrics is minimal with no standardized reporting on outcome metrics. More comprehensive evaluations will help understand mechanisms of impact.

Speaker(s):
Swaminathan Kandaswamy, PhD
Emory University School of Medicine

Author(s):
Lindsey Knake, MD, MS - The University of Iowa Carver College of Medicine; Adam Dziorny, MD, PhD - University of Rochester, School of Medicine and Dentistry; Sean Hernandez, MD - Wake Forest Baptist Medical Center; Allison McCoy, PhD, ACHIP, FACMI, FAMIA - Vanderbilt University Medical Center; Philip Hagedorn, MD, MBI - Cincinnati Children's Hospital and Medical Center; Evan Orenstein, MD - Childrenís Healthcare of Atlanta; Eric Kirkendall, MD, MBI - Wake Forest Baptist School of Medicine/Advocate Health; Matthew Malloy, MD - UC Department of Pediatrics; Naveen Muthu, MD - Children's Healthcare of Atlanta; Avinash Murugan - Yale New Haven Hospital; Jonathan Beus, MD, MSCR - Children's Healthcare of Atlanta; Mark Mai, MD, MHS - Children's Healthcare of Atlanta; Brooke Luo, MD - Children's Hospital of Philadelphia; Juan Chaparro, MD, MS - Nationwide Children's Hospital;

Typologies of Patient Portal Use Among Patients with HIV: A Longitudinal Multichannel Sequence Analysis

Poster Number: P196
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Personal Health Informatics, Mobile Health, Chronic Care Management, Patient Engagement and Preferences
Primary Track: Applications
Programmatic Theme: Consumer Health Informatics

Among patients with HIV (n=10,742), we employed multichannel sequence analysis to identify longitudinal typologies of patient portal use. Among five identified typologies, the largest groups represented patients with low use of features or those who primarily use the portal to refill medications. Patient characteristics varied across typologies by race, housing status, substance use disorder, and alcohol use disorder, highlighting opportunities to address barriers for patients who have access to portals but little engagement.

Speaker(s):
Ashley Griffin, PhD, MSPH
Veterans Affairs Palo Alto Health Care System & Stanford University

Author(s):
Ashley Griffin, PhD, MSPH - Veterans Affairs Palo Alto Health Care System & Stanford University; Lara Troszak, MA - VA; Tigran Avoundjian, PhD - VA; Stephanie Shimada, PhD - Department of Veterans Affairs; Shayna Cave, MS - VA; Keith McInnes, ScD - Bedford VA Medical Center; Amanda Midboe, PhD - VA Palo Alto Health Care System;

Are Patients and Care Partners Speaking the Same Language? Language Characteristics of Spanish Speakers’ Patient Portal Messages

Poster Number: P197
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Personal Health Informatics, Diversity, Equity, Inclusion, Accessibility, and Health Equity, Health Equity, Patient Engagement and Preferences, Mobile Health
Primary Track: Applications
Programmatic Theme: Consumer Health Informatics

Knowledge is limited on the language characteristics of patient portal messaging among Spanish-speaking patients. This retrospective study evaluated the rates of Spanish vs. English messaging, the frequency of patient vs. care partner messaging, and the rate of language concordance between patient and healthcare team members among a random sample of 297 Spanish-speaking patient portal users from three health systems in Dallas, TX using mixed methods.

Speaker(s):
Robert Turer, MD, MSE
UT Southwestern Medical Center

Author(s):
Emily Repasky, MA - UTSW Medical Center; Luis Peña, B.S. - UT Southwestern Medical Center; Libertad Gracia, BA, CHW - UT Southwestern Medical Center; Bhaskar Thakur, Ph.D./Biostatistician - UT Southwestern Medical Center; Samuel McDonald, MD - UT Southwestern; Christoph Lehmann, MD, FAAP, FACMI, FIAHSI - UT Southwestern; Ann Navar, MD, PhD - UT Southwestern Medical Center; Eric Peterson, MD - UT Southwestern; DuWayne Willett, MD - University of Texas Southwestern Medical Center; Ellen O'Connell, MD - UT Southwestern Medical Center; Brett Moran, MD - Parkland Health; Christopher Clark, MPA - Parkland Health; Ferdinand Velasco, MD - Texas Health Resources; Andrew Masica, M.D., M.S.C.I. - Texas Health Resources; Clark Walker, MPH - Texas Health Resources; Dwayne Hoelscher, DNP - Texas Health Resources; Robin Higashi, PhD - UT Southwestern Medical Center;

Tobacco Use in the Socioeconomically Disadvantaged: Insights from Aggregate Electronic Health Record Data

Poster Number: P198
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Population Health, Health Equity, Racial disparities, Causal Inference, Data Mining, Data Transformation/ETL, Real-World Evidence Generation, Qualitative Methods
Primary Track: Foundations
Programmatic Theme: Clinical Research Informatics

Introduction
Tobacco use is significantly higher among socioeconomically disadvantaged populations. We sought to determine the association of smoking status and social needs using geocoded social vulnerability indexes (SVI) and patient reported social determinants of health (SDOH) in a nationally representative sample using aggregate EHR data.

Methods
This cross-sectional study included all adult (age≥18 years), base patients (≥2 face-to-face encounters in any 2-year interval) with documented smoking status, age, legal sex, and ≥1 social need measure in the Epic Cosmos Database. Multivariable logistic regression was used to examine associations.

Results
Of >240 million unique patients in Cosmos, 182.2 million (75.8%) were base patients. Compared to the overall current smoking prevalence (14.3%), smoking was highest among patients in the highest (75% or more) SVI Theme Household Composition Percentile (20.8%) and highest SDOH Food Scarcity categories (34.7%). Within each social need measure, current smoking prevalence increased sequentially in a dose-response association with more disadvantageous categories. The highest odds of current smoking were for patients in the most disadvantageous SVI Theme Socioeconomic Percentile (AOR: 2.553, 95% CI: 2.549-2.556, p<.0001) and SDOH Financial Resource Strain (AOR: 4.506, 95% CI: 4.454-4.560, p<.0001) categories.

Discussion
This nationwide analysis is the largest study to date to show the associations of geocoded SVI and patient reported SDOH measures with smoking status. There was a strong, dose-response relationship with less advantageous social need categories and higher odds of smoking. These findings support smoking cessation interventions that address the social needs beyond tobacco use.

Speaker(s):
Craig Jarrett, MD/MBA
University Hospitals Cleveland

Author(s):

Differences in Medication Fill Metrics in the Electronic Health Records and Linked Claims Data Among Population Covered by Different Types of Insurance

Poster Number: P199
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Population Health, Real-World Evidence Generation, Clinical Decision Support
Primary Track: Foundations
Programmatic Theme: Clinical Informatics

This study stresses the importance of evaluating the accuracy of medication fill data from pharmacy/EHR linkage with claims data for reliable point-of-care interventions. We present retrospective findings from Johns Hopkins Health Plans patients, comparing medication metrics from EHR and claims data across payers. Our results underscore the utility of linked EHR-pharmacy data for measuring medication adherence, complexity, and high-caution. However, insurance-specific variations raise questions about adequacy of EHR-pharmacy or claims data for population-based medication metrics.

Speaker(s):
Chintan Pandya, PhD, MPH, MBBS
Johns Hopkins University

Author(s):
Chintan Pandya, PhD, MPH, MBBS - Johns Hopkins University; Bayan Hijazi, PhD - Johns Hopkins University; Christopher Kitchen, MS; Thomas Richards, MS - Johns Hopkins University; Klaus Lemke, PhD - Johns Hopkins University; Jonathan Weiner, DrPH - Johns Hopkins University; Hadi Kharrazi, MD, PhD, FAMIA, FACMI - Johns Hopkins University;

Imputing Missing Data via Deep Learning in Federated Settings

Poster Number: P200
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Privacy and Security, Deep Learning, Data Mining
Primary Track: Foundations

Longitudinal health data are vital in supporting clinical decision, but the fragmented nature of the healthcare system and the data collection process may lead to incomplete and missing data values that may diminish data usability and lead to biased analyses. In this on-going study, we aim at investigating the applicability of emerging deep learning models for imputing clinical data using distributed datasets, enabling the training of powerful data imputation models while protecting privacy.

Speaker(s):
Luca Bonomi, PhD
Vanderbilt University Department of Biomedical Informatics

Author(s):
Luca Bonomi, PhD - Vanderbilt University Department of Biomedical Informatics; Liyue Fan, PhD - University of North Carolina at Charlotte;

Understanding clinical reasoning with physiological-model-estimated phenotypes in an intensive care unit: a qualitative study

Poster Number: P201
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Qualitative Methods, Critical Care, Human-computer Interaction, Biomarkers, Machine Learning, Data Mining, Clinical Decision Support
Primary Track: Foundations
Programmatic Theme: Clinical Research Informatics

We perform a qualitative study with concept mapping and thematic analysis to understand how health care professionals (HCPs) reason with model-estimated phenotypes of unmeasurable endocrine physiological processes in tube-fed ICU patients. By identifying salient themes from comparing captured concepts maps between HCP mindsets and computer-electronic health record-based interpretations, we have a clearer understanding of the cognitive mapping between clinical language and physiological mechanics, advancing the goal of incorporating newly-estimated high-fidelity physiological phenotypes for HCPs.

Speaker(s):
Yanran Wang, PhD Candidate
BIOS, DBMI, University of Colorado Anschutz Medical Campus

Author(s):
Yanran Wang, PhD Candidate - BIOS, DBMI, University of Colorado Anschutz Medical Campus; Lena Mamykina, PhD - Columbia University; Cecilia C. Low Wang, MD - Division of Endocrinology, Metabolism and Diabetes, Department of Medicine, University of Colorado School of Medicine; Palak Choksi, MD - Division of Endocrinology, Metabolism and Diabetes, Department of Medicine, University of Colorado School of Medicine; James Mitchell - University of Colorado; George Hripcsak, MD - Columbia University Irving Medical Center; David Albers, PhD - University of Colorado, Department of Biomedical Informatics;

Use of Patient Portal Messages to Support an Age-Friendly Health System

Poster Number: P202
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Rule-Based Artificial Intelligence, Workflow, Patient / Person Generated Health Data (Patient Reported Outcomes), Personal Health Informatics
Primary Track: Applications
Programmatic Theme: Consumer Health Informatics

Background: Patient portal secure messaging can support age-friendly dementia care, yet little is known about care partner use of the portal and how message concerns relate to age-friendly issues.
Methods: We first assessed the feasibility of automating care partner identification from patient portal messages by developing and testing a natural language processing (NLP) rule-based classification system from portal messages of 1,973 unique patients 65 and older. Second, two independent reviewers manually coded a randomly selected sample of portal messages for 987 persons with dementia to identify the frequency of expressed needs from the 4M domains of an Age Friendly Health System (medications, mentation, mobility, and what matters).
Results: A total of 267 (13.53%) of 1,973 messages sent from older adults’ portal accounts were identified through manual coding as sent by a nonpatient author. The NLP model performance to identify nonpatient authors demonstrated an AUC of 0.90. Most messages sent from the accounts of persons with dementia contained content relevant to the 4Ms (60%, 601/987), with the breakdown as follows: medications – 36% (357/987), mobility – 10% (101/987), mentation – 16% (153/987), and what matters (aligning care with specific health goals and care preferences) – 21%, 207/987.
Conclusions: Patient portal messaging offers an avenue to identify care partners and meet the informational needs of persons with dementia and their care partners.

Speaker(s):
Kelly Gleason, PhD, RN
Johns Hopkins University

Author(s):
Danielle Powell, PhD, AuD - University of Maryland; Athena DeGennaro, BS - Johns Hopkins University; Mingche MJ Wu, MS - Johns Hopkins University; Talan Zhang, MS - Johns Hopkins University; Hillary Lum, MD, PhD - University of Colorado Health; Jennifer Portz, PhD - University of Colorado Health; Jennifer Wolff, PhD - Johns Hopkins University;

Implementation and Adoption of an Order-Based Surgical Case Request Tool Across Diverse Subspecialty Clinics

Poster Number: P203
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Surgery, Workflow, Change Management
Primary Track: Applications
Programmatic Theme: Clinical Informatics

A flexible order-based case request tool was implemented across diverse surgical subspecialty clinics at a safety net hospital with the objective to reduce repetitive data entry and provide a comprehensive electronic health record (EHR) audit trail. The novel process increased surgical case dwell time within the EHR. Clinics with higher adoption significantly reduced unnecessary or “in error” cases. Limitations in the tool’s ability to handle complex procedure coding may limit adoption.

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

Author(s):
Alyssa Low, BS - UT Southwestern Medical Center; Robert Turer, MD, MSE - UT Southwestern Medical Center; Derek Ngai, MD - UT Southwestern; Christoph Lehmann, MD, FAAP, FACMI, FIAHSI - UT Southwestern; Hongzhao Ji; Hongzhao Ji, MD - UT Southwestern Medical Center;

Lack of Exchanged Data Fidelity Limits the Clinician’s Decision Making

Poster Number: P205
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Usability, Interoperability and Health Information Exchange, Governance of Artificial Intelligence, Natural Language Processing, Clinical Decision Support, Data Sharing, Machine Learning, Patient Safety
Primary Track: Foundations
Programmatic Theme: Clinical Informatics

The exchange of electronic health records has dramatically increased from 28% of hospital adoption in 2011 to 96% hospital adoption by 2021.1 Clinical usability of the data being exchanged must be prioritized over the volume.
In this study, 68.1% of analyzed CCDAs had low fidelity of data not allowing for clinical decision-making in determining follow-up related procedures and timeliness. Based on these findings and the need for high-quality data in clinical settings and for training Gen AI models, we recommend that AI readiness checklists must require high-quality clinical data for the training models to prevent hallucinations.

Speaker(s):
Jeffery E Anderson, MD
VA

Author(s):
Aziz Sulayman, MD, M.Ed - J P Systems, INC; Gay Stahr, BS - J P Systems; Todd Turner, BA, MBA - VA;

Development and Evaluation of an Online Contraceptive Decision Aid for Transgender and Gender-Nonconforming Assigned Female at Birth Persons

Poster Number: P206
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: User-centered Design Methods, Diversity, Equity, Inclusion, Accessibility, and Health Equity, Patient Engagement and Preferences, Delivering Health Information and Knowledge to the Public
Primary Track: Applications
Programmatic Theme: Consumer Health Informatics

We performed a study to develop and evaluate an online contraceptive decision aid, MyChoiceForAll, for patient education and decision assistance for the transgender and gender-nonconforming assigned female at birth population. We reviewed the existing resources, conducted a focus group interview, and consulted with a medical professional for development of MyChoiceForAll, which included 15 contraceptive methods and could provide customized recommendations based on a user’s background, preferences, and medical conditions. Evaluations showed that MyChoiceForAll could provide appropriate recommendations for 94% of the 105 test cases. User survey with 18 statements on system usefulness, usability, and general impression indicated >80% positive responses. A follow-up interview confirmed most findings from the survey. The results suggest that MyChoiceForAll can provide useful system functions and user-friendly features to address the unique needs of the target audience. Further research is required to fully implement the tool and to assess its health impacts in natural settings.

Speaker(s):
Vi-Anh Hoang, B.S.
Washington University School of Medicine in St. Louis

Author(s):
Vi-Anh Hoang, B.S. - Washington University School of Medicine in St. Louis; Molly Redman, MS - Arizona State University; Joyce Wang, MD - University of Washington School of Medicine; Jenny Brian, PhD - Arizona State University; Dongwen Wang, PhD - Arizona State University;

Using informatics methods and design thinking to improve genetic diagnosis in nephrology

Poster Number: P207
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: User-centered Design Methods, Clinical Decision Support, Precision Medicine, Qualitative Methods, Diagnostic Systems
Primary Track: Foundations
Programmatic Theme: Clinical Informatics

Following the design thinking process pioneered by IDEO, we organized a workshop to inform the creation of a prototype real-time genetic diagnosis tool for nephrologists as a use case. The output of the workshop were design requirements for the prototype.

Speaker(s):
Katrina Romagnoli, PhD, MS, MLIS
Geisinger

Author(s):
Katrina Romagnoli, PhD, MS, MLIS - Geisinger; Zachary Salvati, MS, CGC - Geisinger Clinic; Darren Johnson - Geisinger Medical Center; Heather Ramey, MS - Geisinger; Alexander Chang, MD - Geisinger; Marc Williams, MD - Marc S. Williams;