- Home
- 2025 Annual Symposium Gallery
- Towards Safer AI in Healthcare: Using Storytelling to Educate About Unintended Harms of AI-Powered Tools
Custom CSS
double-click to edit, do not edit in source
11/18/2025 |
5:00 PM – 6:30 PM |
International Ballroom (Posters)
Poster Session 3
Presentation Type: Posters
Description
Join us for an engaging Poster Session where ideas come to life through one-on-one conversations with presenters. Explore a diverse range of topics, learn directly from the researchers behind the work, and dive deeper into the studies that spark your interest. This is your opportunity to connect with others who share your passions, exchange perspectives, and build new professional relationships. Whether you’re looking to gain insights, ask questions, or network with peers, the Poster Session offers a dynamic, interactive environment to expand your knowledge and your professional circle.
The Informatics of Value-Base Care: Missing Opportunities
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Workforce Development, Teaching Innovation, Curriculum Development
Primary Track: Foundations
Programmatic Theme: Academic Informatics / LIEAF
Value-based care (VBC) is transforming the healthcare landscape by shifting the focus from the volume of services provided to the quality of care delivered. In this model, providers are rewarded based on patient health outcomes, with the aim of encouraging a more patient-centered approach. Achieving VBC, however, requires robust health information technology (IT) systems to collect and analyze data that can both define and assess the success of these value-based efforts. The purpose of this poster is to thoroughly explore the various components of VBC using a modern informatics framework, specifically through a structured Informatics Stack. By examining each layer of the stack, we will identify opportunities where informatics research can move VBC research forward.
Speaker:
Chen Dun, MHS
Johns Hopkins University
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Workforce Development, Teaching Innovation, Curriculum Development
Primary Track: Foundations
Programmatic Theme: Academic Informatics / LIEAF
Value-based care (VBC) is transforming the healthcare landscape by shifting the focus from the volume of services provided to the quality of care delivered. In this model, providers are rewarded based on patient health outcomes, with the aim of encouraging a more patient-centered approach. Achieving VBC, however, requires robust health information technology (IT) systems to collect and analyze data that can both define and assess the success of these value-based efforts. The purpose of this poster is to thoroughly explore the various components of VBC using a modern informatics framework, specifically through a structured Informatics Stack. By examining each layer of the stack, we will identify opportunities where informatics research can move VBC research forward.
Speaker:
Chen Dun, MHS
Johns Hopkins University
Chen
Dun,
MHS - Johns Hopkins University
Development of Billing for Point-of-Care Ultrasound (POCUS) in the Pediatric Emergency Department
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Workflow, Imaging Informatics, Pediatrics
Primary Track: Foundations
Programmatic Theme: Clinical Informatics
Point-of-care ultrasound (POCUS) is an essential tool for pediatric emergency medicine (PEM) physicians. We conducted walkthroughs of POCUS users workflow from completion of POCUS study to documentation which informed our development of structured forms for documentation. We then analyzed the rate of images with standardized documentation per month using a statistical process control chart. We also collected the number of charges for each different POCUS Current Procedural Terminology (CPT) codes after starting this program.
Speaker:
Carrie Ng, MD
Emory University School of Medicine, Children's Healthcare of Atlanta
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Workflow, Imaging Informatics, Pediatrics
Primary Track: Foundations
Programmatic Theme: Clinical Informatics
Point-of-care ultrasound (POCUS) is an essential tool for pediatric emergency medicine (PEM) physicians. We conducted walkthroughs of POCUS users workflow from completion of POCUS study to documentation which informed our development of structured forms for documentation. We then analyzed the rate of images with standardized documentation per month using a statistical process control chart. We also collected the number of charges for each different POCUS Current Procedural Terminology (CPT) codes after starting this program.
Speaker:
Carrie Ng, MD
Emory University School of Medicine, Children's Healthcare of Atlanta
Carrie
Ng,
MD - Emory University School of Medicine, Children's Healthcare of Atlanta
Integrating a Breast Cancer Risk Model into a Clinical Workflow
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: User-centered Design Methods, Workflow, Machine Learning, Personal Health Informatics, Cancer Prevention
Primary Track: Applications
Programmatic Theme: Clinical Informatics
Breast cancer is the most commonly diagnosed cancer in women. A Breast Cancer Risk Estimator (BCRE) machine learning model was developed to enable outreach to high-risk patients that need mammogram screening. In this poster, we present a clinical workflow with a user-centered design that enables immediate patient outreach, minimizes unnecessary touchpoints and integrates with the EHR as well as current breast cancer protocols. Program evaluation includes monitoring patient outreach status and outcomes.
Speaker:
Rebecca Maff, MS
Geisinger
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: User-centered Design Methods, Workflow, Machine Learning, Personal Health Informatics, Cancer Prevention
Primary Track: Applications
Programmatic Theme: Clinical Informatics
Breast cancer is the most commonly diagnosed cancer in women. A Breast Cancer Risk Estimator (BCRE) machine learning model was developed to enable outreach to high-risk patients that need mammogram screening. In this poster, we present a clinical workflow with a user-centered design that enables immediate patient outreach, minimizes unnecessary touchpoints and integrates with the EHR as well as current breast cancer protocols. Program evaluation includes monitoring patient outreach status and outcomes.
Speaker:
Rebecca Maff, MS
Geisinger
Rebecca
Maff,
MS - Geisinger
Automated Assessments of Clinical Encounters: Provider Perspectives
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Usability, User-centered Design Methods, Artificial Intelligence, Information Visualization, Health Equity, Diversity, Equity, Inclusion, and Accessibility, Human-computer Interaction
Primary Track: Applications
Programmatic Theme: Clinical Informatics
Care quality and outcomes are influenced by the quality of patient-provider communication; ConverSense was developed to automatically assess and provide clinicians with feedback on interpersonal communication in patient-provider interactions, including implicit bias. Through interviews with 24 primary care providers who received feedback from ConverSense from simulated clinical encounters, we identified 6 themes about the usefulness of ConverSense. Findings provide guidance for improved usability and transparency of automated assessment and feedback on patient-provider communication.
Speaker:
Claire Lai, B.S.
Univeristy of Washington
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Usability, User-centered Design Methods, Artificial Intelligence, Information Visualization, Health Equity, Diversity, Equity, Inclusion, and Accessibility, Human-computer Interaction
Primary Track: Applications
Programmatic Theme: Clinical Informatics
Care quality and outcomes are influenced by the quality of patient-provider communication; ConverSense was developed to automatically assess and provide clinicians with feedback on interpersonal communication in patient-provider interactions, including implicit bias. Through interviews with 24 primary care providers who received feedback from ConverSense from simulated clinical encounters, we identified 6 themes about the usefulness of ConverSense. Findings provide guidance for improved usability and transparency of automated assessment and feedback on patient-provider communication.
Speaker:
Claire Lai, B.S.
Univeristy of Washington
Claire
Lai,
B.S. - Univeristy of Washington
Motivational Messaging in a 1-week Trial of a Digital Health Program for Older Family Care Partners of Persons with Heart Failure
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Usability, Patient Engagement and Preferences, Qualitative Methods, Nursing Informatics
Primary Track: Applications
We assessed the usability and acceptability of motivational messaging in a digital health physical activity program for older family caregivers of persons with heart failure, through a 1-week, two-phase field test. Phase 1 participants found automated messages insufficiently motivating, leading to refinements in Phase 2, where a health coach manually delivered messages. Participants preferred personalized, encouraging messages with a human touch. Findings highlight the need for tailored, human-centered digital health interventions to enhance older caregiver engagement.
Speaker:
Dawon Baik, PhD
University of Colorado
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Usability, Patient Engagement and Preferences, Qualitative Methods, Nursing Informatics
Primary Track: Applications
We assessed the usability and acceptability of motivational messaging in a digital health physical activity program for older family caregivers of persons with heart failure, through a 1-week, two-phase field test. Phase 1 participants found automated messages insufficiently motivating, leading to refinements in Phase 2, where a health coach manually delivered messages. Participants preferred personalized, encouraging messages with a human touch. Findings highlight the need for tailored, human-centered digital health interventions to enhance older caregiver engagement.
Speaker:
Dawon Baik, PhD
University of Colorado
Dawon
Baik,
PhD - University of Colorado
Development and Validation of a Machine Learning Model to Predict Postoperative Morbidity and Mortality of Patients with COVID-19 who Underwent Major Surgery
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Surgery, Machine Learning, Clinical Decision Support, Patient Safety, Healthcare Quality
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
Active coronavirus disease (COVID-19) increases the risk of morbidity and mortality after surgery. This investigation utilized machine learning algorithms to predict postoperative outcomes based on preoperative features.
Multiple supervised machine learning algorithms were trained on 153 features for 10,613 patients with COVID-19 who underwent major surgery in 2022, and validated on 5,269 patients from 2021. Patients with COVID-19 in 2021 and 2022 had a significantly higher 30-day composite mortality or major morbidity rate of 17.0% and 11.8%, respectively, compared to 3.2% and 3.0% of patients without (p<0.001). LightGBM was the best performing algorithm, with an area under the receiver operating characteristic curve and F1 score of 0.865 and 0.512 when training on 2022 data, and 0.898 and 0.617 when validating on 2021 data, respectively. This approach accurately identified patients with COVID-19 who experienced postoperative morbidity and mortality, and may provide a more objective approach to risk stratification.
Speaker:
Sean McDermott, MD
University of Pittsburgh
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Surgery, Machine Learning, Clinical Decision Support, Patient Safety, Healthcare Quality
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
Active coronavirus disease (COVID-19) increases the risk of morbidity and mortality after surgery. This investigation utilized machine learning algorithms to predict postoperative outcomes based on preoperative features.
Multiple supervised machine learning algorithms were trained on 153 features for 10,613 patients with COVID-19 who underwent major surgery in 2022, and validated on 5,269 patients from 2021. Patients with COVID-19 in 2021 and 2022 had a significantly higher 30-day composite mortality or major morbidity rate of 17.0% and 11.8%, respectively, compared to 3.2% and 3.0% of patients without (p<0.001). LightGBM was the best performing algorithm, with an area under the receiver operating characteristic curve and F1 score of 0.865 and 0.512 when training on 2022 data, and 0.898 and 0.617 when validating on 2021 data, respectively. This approach accurately identified patients with COVID-19 who experienced postoperative morbidity and mortality, and may provide a more objective approach to risk stratification.
Speaker:
Sean McDermott, MD
University of Pittsburgh
Sean
McDermott,
MD - University of Pittsburgh
Insights on Social Isolation from a National Health Information Survey
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Social Media and Connected Health, Population Health, Surveys and Needs Analysis
Primary Track: Applications
Programmatic Theme: Public Health Informatics
US Surgeon General’s Advisory in 2023 highlighted the loneliness epidemic and importance of social connections on health. The Health Information National Trends Survey data from 2022 was analyzed and PROMIS isolation scores (>=60 high isolation) were used to characterize association with 3 variables. 20% with low social isolation scores had never used social media. There was a significant association between perceived quality of care and trust in the healthcare system with PROMIS social isolation scores.
Speaker:
Madhur Thakur, MS Health Informatics
University of Minnesota
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Social Media and Connected Health, Population Health, Surveys and Needs Analysis
Primary Track: Applications
Programmatic Theme: Public Health Informatics
US Surgeon General’s Advisory in 2023 highlighted the loneliness epidemic and importance of social connections on health. The Health Information National Trends Survey data from 2022 was analyzed and PROMIS isolation scores (>=60 high isolation) were used to characterize association with 3 variables. 20% with low social isolation scores had never used social media. There was a significant association between perceived quality of care and trust in the healthcare system with PROMIS social isolation scores.
Speaker:
Madhur Thakur, MS Health Informatics
University of Minnesota
Madhur
Thakur,
MS Health Informatics - University of Minnesota
Mapping the Use of Social Media Analytics in Firearm Injury Exposure Research: A Scoping Review
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Social Media and Connected Health, Natural Language Processing, Public Health, Data Mining, Evaluation, Quantitative Methods
Primary Track: Applications
Programmatic Theme: Public Health Informatics
This scoping review examined how social media analytics have been used to study firearm injury exposure. Sixteen English-language studies published between 2014–2023, primarily studying Twitter/X, used applied methods such as natural language processing and sentiment analysis. Findings highlight limited platform diversity, methodological inconsistencies, and suggest that social media analytics offer untapped potential to strengthen public health surveillance of firearm injury exposure.
Speaker:
Michele Flynch, PHD, RN
Columbia University School of Nursing
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Social Media and Connected Health, Natural Language Processing, Public Health, Data Mining, Evaluation, Quantitative Methods
Primary Track: Applications
Programmatic Theme: Public Health Informatics
This scoping review examined how social media analytics have been used to study firearm injury exposure. Sixteen English-language studies published between 2014–2023, primarily studying Twitter/X, used applied methods such as natural language processing and sentiment analysis. Findings highlight limited platform diversity, methodological inconsistencies, and suggest that social media analytics offer untapped potential to strengthen public health surveillance of firearm injury exposure.
Speaker:
Michele Flynch, PHD, RN
Columbia University School of Nursing
Michele
Flynch,
PHD, RN - Columbia University School of Nursing
Phenotypes of stigma expressed by people who use drugs on Reddit
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Social Media and Connected Health, Natural Language Processing, Large Language Models (LLMs), Public Health, Health Equity
Primary Track: Foundations
Programmatic Theme: Public Health Informatics
In this study, we analyzed 1.03 million Reddit posts from six drug-related subreddits using LLM-based classification to identify 56,446 stigma expressions. K-means clustering of 17 validated dimensions revealed three distinct phenotypes: Internalized Stigma (34.5%) characterized by self-directed stigma and avoidant coping; Public Stigma (38.9%) featuring healthcare barriers and power dynamics; and Righteous Indignation (26.6%) showing structural critique and analytical language. These phenotypes inform targeted anti-stigma interventions for people who use drugs.
Speaker:
Layla Bouzoubaa, MSPH
Drexel University
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Social Media and Connected Health, Natural Language Processing, Large Language Models (LLMs), Public Health, Health Equity
Primary Track: Foundations
Programmatic Theme: Public Health Informatics
In this study, we analyzed 1.03 million Reddit posts from six drug-related subreddits using LLM-based classification to identify 56,446 stigma expressions. K-means clustering of 17 validated dimensions revealed three distinct phenotypes: Internalized Stigma (34.5%) characterized by self-directed stigma and avoidant coping; Public Stigma (38.9%) featuring healthcare barriers and power dynamics; and Righteous Indignation (26.6%) showing structural critique and analytical language. These phenotypes inform targeted anti-stigma interventions for people who use drugs.
Speaker:
Layla Bouzoubaa, MSPH
Drexel University
Layla
Bouzoubaa,
MSPH - Drexel University
Correcting Informative Censoring in Treatment Effect Estimation with Large Scale, Real-World Health Data
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Real-World Evidence Generation, Quantitative Methods, Causal Inference
Primary Track: Foundations
Programmatic Theme: Clinical Research Informatics
Observational studies play a crucial role in evaluating the effectiveness and safety of medical treatments, but are prone to various biases which can distort causal effect estimates. While Inverse Probability of Treatment Weighting (IPTW) adjusts for confounding, Inverse Probability of Censoring Weighting (IPCW) accounts for bias introduced by informative censoring. In this study, we systematically evaluate the performance of IPTW, IPCW, and their combination using both simulation experiments and real-world health data.
Speaker:
Hsin Yi Chen, B.S.
Columbia University
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Real-World Evidence Generation, Quantitative Methods, Causal Inference
Primary Track: Foundations
Programmatic Theme: Clinical Research Informatics
Observational studies play a crucial role in evaluating the effectiveness and safety of medical treatments, but are prone to various biases which can distort causal effect estimates. While Inverse Probability of Treatment Weighting (IPTW) adjusts for confounding, Inverse Probability of Censoring Weighting (IPCW) accounts for bias introduced by informative censoring. In this study, we systematically evaluate the performance of IPTW, IPCW, and their combination using both simulation experiments and real-world health data.
Speaker:
Hsin Yi Chen, B.S.
Columbia University
Hsin Yi
Chen,
B.S. - Columbia University
Evaluating Federated Learning’s Application in Epidemiology: Identifying Risk Factors for Treatment-Resistant Depression
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Real-World Evidence Generation, Privacy and Security, Causal Inference, Fairness and elimination of bias, Machine Learning
Primary Track: Applications
Programmatic Theme: Public Health Informatics
Federated learning (FL) approach enables analysis across data sources while preserving privacy. Despite its increasing adoption in machine learning tasks, FL’s application in statistical inferences remain limited. We piloted federated algorithms for statistical modeling to identify risk factors for treatment-resistant depression and compare the federated results from those obtained on centralized data. Our study piloted and evaluated a federated analysis framework for common healthcare statistical models, presenting a repeatable process for federated epidemiological analyses. Our results showed high concordance between the federated and centralized analyses in different scenarios, demonstrating the reliability of federated approaches.
Speaker:
Echo Wang, DrPH
Merck
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Real-World Evidence Generation, Privacy and Security, Causal Inference, Fairness and elimination of bias, Machine Learning
Primary Track: Applications
Programmatic Theme: Public Health Informatics
Federated learning (FL) approach enables analysis across data sources while preserving privacy. Despite its increasing adoption in machine learning tasks, FL’s application in statistical inferences remain limited. We piloted federated algorithms for statistical modeling to identify risk factors for treatment-resistant depression and compare the federated results from those obtained on centralized data. Our study piloted and evaluated a federated analysis framework for common healthcare statistical models, presenting a repeatable process for federated epidemiological analyses. Our results showed high concordance between the federated and centralized analyses in different scenarios, demonstrating the reliability of federated approaches.
Speaker:
Echo Wang, DrPH
Merck
Echo
Wang,
DrPH - Merck
A Multilevel Source-of-Bias Model in Real-World Healthcare Evidence: A Scoping Review
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Real-World Evidence Generation, Clinical Decision Support, Informatics Implementation
Primary Track: Foundations
Programmatic Theme: Clinical Research Informatics
To organize knowledge about biases in the development of evidence based on Real World Data (EHR in particular), we developed a multi-level framework spanning healthcare, data management, and research, analyzing 153 bias sources. Biases at patient, provider, IT, informatics, and analytics levels were found to interconnect, transfer, and amplify across healthcare systems. This framework can provide a foundation for developing level-specific debiasing models to enhance reliability real-world evidence derived from individual centers and collaborative networks.
Speaker:
Haeun Lee, MS
Johns Hopkins University
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Real-World Evidence Generation, Clinical Decision Support, Informatics Implementation
Primary Track: Foundations
Programmatic Theme: Clinical Research Informatics
To organize knowledge about biases in the development of evidence based on Real World Data (EHR in particular), we developed a multi-level framework spanning healthcare, data management, and research, analyzing 153 bias sources. Biases at patient, provider, IT, informatics, and analytics levels were found to interconnect, transfer, and amplify across healthcare systems. This framework can provide a foundation for developing level-specific debiasing models to enhance reliability real-world evidence derived from individual centers and collaborative networks.
Speaker:
Haeun Lee, MS
Johns Hopkins University
Haeun
Lee,
MS - Johns Hopkins University
Uncovering Bias in Real-World Data: Challenges in Inpatient Mortality
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Real-World Evidence Generation, Causal Inference, Data Standards
Primary Track: Foundations
This study highlights the challenges of using electronic health record (EHR) data for real-world evidence generation, focusing on potential discrepancies of inpatient mortality. By emulating a clinical trial on dexamethasone, researchers showed that the inconsistencies in mortality documentation, specifically when coinciding with discharge, can influence outcome interpretations. Despite modest overall effects, the findings stress the need for standardized data practices to ensure accurate, bias-free analyses in observational research.
Speaker:
Wesley Anderson, PhD.
Critical Path Institute
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Real-World Evidence Generation, Causal Inference, Data Standards
Primary Track: Foundations
This study highlights the challenges of using electronic health record (EHR) data for real-world evidence generation, focusing on potential discrepancies of inpatient mortality. By emulating a clinical trial on dexamethasone, researchers showed that the inconsistencies in mortality documentation, specifically when coinciding with discharge, can influence outcome interpretations. Despite modest overall effects, the findings stress the need for standardized data practices to ensure accurate, bias-free analyses in observational research.
Speaker:
Wesley Anderson, PhD.
Critical Path Institute
Wesley
Anderson,
PhD. - Critical Path Institute
ECG-Enabled Clustering and Dynamic Time Warping for Cardiovascular and Cardiorenal Patient Stratification
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Quantitative Methods, Precision Medicine, Machine Learning
Primary Track: Applications
Dynamic Time Warping (DTW) was used to cluster ECG features from FinnGen data, identifying HFpEF (heart failure with preserved ejection fraction) phenogroups and CKD (chronic kidney disease) risk. DTW outperformed clustering methods using all or key extracted ECG features (PR interval, QRS duration), yielding more enriched clusters. These findings highlight DTW's potential for improving ECG-based clustering, patient stratification, and risk assessment in cardiovascular and kidney health.
Speaker:
Sally Zhao, MS
Pfizer
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Quantitative Methods, Precision Medicine, Machine Learning
Primary Track: Applications
Dynamic Time Warping (DTW) was used to cluster ECG features from FinnGen data, identifying HFpEF (heart failure with preserved ejection fraction) phenogroups and CKD (chronic kidney disease) risk. DTW outperformed clustering methods using all or key extracted ECG features (PR interval, QRS duration), yielding more enriched clusters. These findings highlight DTW's potential for improving ECG-based clustering, patient stratification, and risk assessment in cardiovascular and kidney health.
Speaker:
Sally Zhao, MS
Pfizer
Sally
Zhao,
MS - Pfizer
SOFA Produces Uncertain Mortality Predictions: Reconsidering the Use of SOFA in Triage
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Quantitative Methods, Machine Learning, Critical Care
Primary Track: Applications
Programmatic Theme: Clinical Informatics
In the event of a disaster, many US institutions have guidelines that triage patients for scarce resources. Many of these guidelines rank-order patients based on thresholds in models that predict mortality, such as the Sequential Organ Failure Assessment (SOFA) score. Using conformal prediction, we show that the SOFA score is a highly uncertain predictor of mortality at the patient level, calling into question the use of threshold-based models such as SOFA in crisis triage.
Speaker:
Katarina Pejcinovic, M.S.
Oregon Health & Science University
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Quantitative Methods, Machine Learning, Critical Care
Primary Track: Applications
Programmatic Theme: Clinical Informatics
In the event of a disaster, many US institutions have guidelines that triage patients for scarce resources. Many of these guidelines rank-order patients based on thresholds in models that predict mortality, such as the Sequential Organ Failure Assessment (SOFA) score. Using conformal prediction, we show that the SOFA score is a highly uncertain predictor of mortality at the patient level, calling into question the use of threshold-based models such as SOFA in crisis triage.
Speaker:
Katarina Pejcinovic, M.S.
Oregon Health & Science University
Katarina
Pejcinovic,
M.S. - Oregon Health & Science University
Patient Perspectives on Clinical Artificial Intelligence
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Qualitative Methods, Workflow, Patient Engagement and Preferences
Primary Track: Applications
Programmatic Theme: Clinical Informatics
In health care, artificial intelligence (AI) resources are often developed and implemented without patient or community input. Community engagement approaches need to adapt to the complex topic of AI. We urgently need to engage patients in the processes of developing and implementing AI tools in health care. Using semi-structured interviews and co-design workshops, this study aims to understand participants’ information needs and communication preferences related to AI in general and AI use in their care.
Speaker:
Joyce Harris, MA
Vanderbilt University Medical Center
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Qualitative Methods, Workflow, Patient Engagement and Preferences
Primary Track: Applications
Programmatic Theme: Clinical Informatics
In health care, artificial intelligence (AI) resources are often developed and implemented without patient or community input. Community engagement approaches need to adapt to the complex topic of AI. We urgently need to engage patients in the processes of developing and implementing AI tools in health care. Using semi-structured interviews and co-design workshops, this study aims to understand participants’ information needs and communication preferences related to AI in general and AI use in their care.
Speaker:
Joyce Harris, MA
Vanderbilt University Medical Center
Joyce
Harris,
MA - Vanderbilt University Medical Center
Methodological reporting of mixed-methods studies of health informatics interventions: A systematic review
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Qualitative Methods, Quantitative Methods, Real-World Evidence Generation, Education and Training
Primary Track: Applications
This systematic review aims to characterize the types of informatics studies that have utilized a mixed-methods study for evaluating clinician-facing informatics interventions. We found that 24% of studies that reported using mixed-methods design did not show integration of their quantitative and qualitative data. Some studies also incorrectly described or omitted to report their core mixed-methods design. These findings suggest a need for educating the informatics research community about implementing best practices for mixed methods research.
Speaker:
Oliver Nguyen, MSHI
University of Wisconsin at Madison
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Qualitative Methods, Quantitative Methods, Real-World Evidence Generation, Education and Training
Primary Track: Applications
This systematic review aims to characterize the types of informatics studies that have utilized a mixed-methods study for evaluating clinician-facing informatics interventions. We found that 24% of studies that reported using mixed-methods design did not show integration of their quantitative and qualitative data. Some studies also incorrectly described or omitted to report their core mixed-methods design. These findings suggest a need for educating the informatics research community about implementing best practices for mixed methods research.
Speaker:
Oliver Nguyen, MSHI
University of Wisconsin at Madison
Oliver
Nguyen,
MSHI - University of Wisconsin at Madison
Claims-Based Identification of Primary Care Providers in Georgia
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Public Health, Real-World Evidence Generation, Delivering Health Information and Knowledge to the Public
Primary Track: Foundations
Programmatic Theme: Public Health Informatics
Claims data is a valuable data source to understand provider availability. Claims and eligibility information from the Georgia APCD were analyzed to determine annual primary care provider (PCP) counts and ratios by provider type, day-of-week, and county for 2018-2023. PCP-to-members ratios were lower for urban counties; both urban and rural PCP ratios declined from 2018-2022 and increased in 2023. Additionally, PCP availability was lower on weekends and nurse practitioners expanded in the PCP workforce.
Speaker:
Caleb Hightower, MSPH
Georgia Tech Research Institute
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Public Health, Real-World Evidence Generation, Delivering Health Information and Knowledge to the Public
Primary Track: Foundations
Programmatic Theme: Public Health Informatics
Claims data is a valuable data source to understand provider availability. Claims and eligibility information from the Georgia APCD were analyzed to determine annual primary care provider (PCP) counts and ratios by provider type, day-of-week, and county for 2018-2023. PCP-to-members ratios were lower for urban counties; both urban and rural PCP ratios declined from 2018-2022 and increased in 2023. Additionally, PCP availability was lower on weekends and nurse practitioners expanded in the PCP workforce.
Speaker:
Caleb Hightower, MSPH
Georgia Tech Research Institute
Caleb
Hightower,
MSPH - Georgia Tech Research Institute
Clustering County-level Overdose Mortality Trajectories in Ohio: A Latent Profile Analysis Approach
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Public Health, Population Health, Quantitative Methods
Primary Track: Policy
Programmatic Theme: Public Health Informatics
To analyze county-level differences in opioid mortality, a Latent Profile Analysis was used to cluster opioid overdose trajectories in Ohio counties from 2010-2020, finding four classes. After justifying clusters via opioid seizure data, significant social determinants of health variables were analyzed and only “Economic Context” variables were found to be statistically overrepresented. Therefore, policy makers should consider targeting underlying economic disparities in addition to other common public health interventions.
Speaker:
Jaeho Lee, N/A
Stanford Stats for Social Good
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Public Health, Population Health, Quantitative Methods
Primary Track: Policy
Programmatic Theme: Public Health Informatics
To analyze county-level differences in opioid mortality, a Latent Profile Analysis was used to cluster opioid overdose trajectories in Ohio counties from 2010-2020, finding four classes. After justifying clusters via opioid seizure data, significant social determinants of health variables were analyzed and only “Economic Context” variables were found to be statistically overrepresented. Therefore, policy makers should consider targeting underlying economic disparities in addition to other common public health interventions.
Speaker:
Jaeho Lee, N/A
Stanford Stats for Social Good
Jaeho
Lee,
N/A - Stanford Stats for Social Good
3D Modeling of Objects Responsible for Foreign Body Injuries in Children: A Novel Web-Based Approach for Targeted Public Health Prevention and Pediatric Safety
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Precision Medicine, Imaging Informatics, Public Health, Healthcare Quality, Information Extraction, Patient Safety, Pediatrics, Information Visualization
Primary Track: Applications
Programmatic Theme: Public Health Informatics
Injuries caused by foreign bodies in children represent a significant public health concern. This project develops a novel web-based platform using 3D scanning of foreign bodies from Susy Safe registry, enabling risk profiling by linking object morphology, biomechanical dynamics of aspiration or ingestion, child demographics, socio-familial context, clinical outcomes. This approach supports the identification of object-specific risk profiles and fosters targeted prevention strategies to improve pediatric safety and public awareness.
Speaker:
Cinzia Anna Maria Papappicco, I have 2 Degrees: 1) Science in Nursing; 2) Computer Science
Unit of Biostatistics, Epidemiology and Public Health, Department of Cardiac, Thoracic and Vascular Sciences, University of Padua, Padua, Italy
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Precision Medicine, Imaging Informatics, Public Health, Healthcare Quality, Information Extraction, Patient Safety, Pediatrics, Information Visualization
Primary Track: Applications
Programmatic Theme: Public Health Informatics
Injuries caused by foreign bodies in children represent a significant public health concern. This project develops a novel web-based platform using 3D scanning of foreign bodies from Susy Safe registry, enabling risk profiling by linking object morphology, biomechanical dynamics of aspiration or ingestion, child demographics, socio-familial context, clinical outcomes. This approach supports the identification of object-specific risk profiles and fosters targeted prevention strategies to improve pediatric safety and public awareness.
Speaker:
Cinzia Anna Maria Papappicco, I have 2 Degrees: 1) Science in Nursing; 2) Computer Science
Unit of Biostatistics, Epidemiology and Public Health, Department of Cardiac, Thoracic and Vascular Sciences, University of Padua, Padua, Italy
Cinzia Anna Maria
Papappicco,
I have 2 Degrees: 1) Science in Nursing; 2) Computer Science - Unit of Biostatistics, Epidemiology and Public Health, Department of Cardiac, Thoracic and Vascular Sciences, University of Padua, Padua, Italy
Age- and Sex-Specific Cutoffs for Obesity-Related Indices to Improve Early Detection of Metabolic Syndrome Risk in Taiwanese Adults
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Population Health, Public Health, Precision Medicine
Primary Track: Applications
Programmatic Theme: Public Health Informatics
This study identified optimal age- and sex-specific cutoffs for BMI, TyG index, and CVAI in predicting metabolic syndrome (MetS) in Taiwanese adults, based on data from 30,382 subjects. Most identified cutoffs for these obesity-related indices demonstrated AUCs and sensitivity ranging from acceptable to outstanding (>0.70–0.90). Integrating these cutoffs into screening protocols can enhance early detection, prevention, and management of MetS risk.
Speaker:
Chien-Yeh Hsu, PhD
National Taipei University of Nursing and Health Sciences
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Population Health, Public Health, Precision Medicine
Primary Track: Applications
Programmatic Theme: Public Health Informatics
This study identified optimal age- and sex-specific cutoffs for BMI, TyG index, and CVAI in predicting metabolic syndrome (MetS) in Taiwanese adults, based on data from 30,382 subjects. Most identified cutoffs for these obesity-related indices demonstrated AUCs and sensitivity ranging from acceptable to outstanding (>0.70–0.90). Integrating these cutoffs into screening protocols can enhance early detection, prevention, and management of MetS risk.
Speaker:
Chien-Yeh Hsu, PhD
National Taipei University of Nursing and Health Sciences
Chien-Yeh
Hsu,
PhD - National Taipei University of Nursing and Health Sciences
Clustering Health Indicators: A Framework for Evaluating Population Health
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Population Health, Public Health, Quantitative Methods
Primary Track: Applications
Programmatic Theme: Public Health Informatics
In this study, we developed a practical health indicator index, or HII, as a means of assessing population health, by assigning clusters to 29 health indicators. The data we used are from a previous survey study on public perception of health. These indicators are clustered by hierarchical clustering and K-means (K=7). Cluster designations were determined by identifying patterns in preventative health, metabolic health, behavioral factors, and social determinants. The results highlight a need to include both clinical and nonclinical factors for comprehensive health assessments. In future work, we will seek to refine the HII using additional clustering methods and adding the validation of real-world data.
Speaker:
Yujia sun, MS
Clemson University
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Population Health, Public Health, Quantitative Methods
Primary Track: Applications
Programmatic Theme: Public Health Informatics
In this study, we developed a practical health indicator index, or HII, as a means of assessing population health, by assigning clusters to 29 health indicators. The data we used are from a previous survey study on public perception of health. These indicators are clustered by hierarchical clustering and K-means (K=7). Cluster designations were determined by identifying patterns in preventative health, metabolic health, behavioral factors, and social determinants. The results highlight a need to include both clinical and nonclinical factors for comprehensive health assessments. In future work, we will seek to refine the HII using additional clustering methods and adding the validation of real-world data.
Speaker:
Yujia sun, MS
Clemson University
Yujia
sun,
MS - Clemson University
Reducing Order Friction in Pediatric EHR Systems: A Quality Improvement Initiative
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Pediatrics, Workflow, Qualitative Methods
Primary Track: Applications
Programmatic Theme: Clinical Informatics
Efficient electronic health record (EHR) systems are critical for optimizing workflow and ensuring patient safety. However, many EHR systems are plagued with inefficiencies and excessive alerts. Order friction (OF) is an EHR vendor-derived metric describing how difficult it is for clinicians to enter orders, and includes the sum of all interruptive alerts, procedure duplicate checks, medication warnings, and changed fields within the order composer. Preliminary data shows high OF correlates with increased clinician stress, burnout, and delays in patient care. We devised a framework to prioritize order optimization. We focused on total order changes, or how many times a user changes a default, to prioritize orders with most OF and most volume. Oral solutions comprised most of the top medications, as pediatric care relies on weight-based dosing. We streamlined the order process for acetaminophen and ibuprofen oral solutions, reducing the previous requirement of four clicks for every-six-hour dosing as needed to zero clicks or CPO. Our data also reveals that the friction of preference lists is greater than the friction for order sets. User configured preference lists and order sets further reduced OF. In specific department preference lists, order design can be improved by setting default values for commonly used settings, thereby reducing the number of clicks and adjustments required to complete an order. The next steps are to focus on preference list orders and departments with the highest OF scores. We aim for an impactful reduction in clinician burden and improvement in patient care efficiency.
Speaker:
Bayley Bennett, MD
Emory University
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Pediatrics, Workflow, Qualitative Methods
Primary Track: Applications
Programmatic Theme: Clinical Informatics
Efficient electronic health record (EHR) systems are critical for optimizing workflow and ensuring patient safety. However, many EHR systems are plagued with inefficiencies and excessive alerts. Order friction (OF) is an EHR vendor-derived metric describing how difficult it is for clinicians to enter orders, and includes the sum of all interruptive alerts, procedure duplicate checks, medication warnings, and changed fields within the order composer. Preliminary data shows high OF correlates with increased clinician stress, burnout, and delays in patient care. We devised a framework to prioritize order optimization. We focused on total order changes, or how many times a user changes a default, to prioritize orders with most OF and most volume. Oral solutions comprised most of the top medications, as pediatric care relies on weight-based dosing. We streamlined the order process for acetaminophen and ibuprofen oral solutions, reducing the previous requirement of four clicks for every-six-hour dosing as needed to zero clicks or CPO. Our data also reveals that the friction of preference lists is greater than the friction for order sets. User configured preference lists and order sets further reduced OF. In specific department preference lists, order design can be improved by setting default values for commonly used settings, thereby reducing the number of clicks and adjustments required to complete an order. The next steps are to focus on preference list orders and departments with the highest OF scores. We aim for an impactful reduction in clinician burden and improvement in patient care efficiency.
Speaker:
Bayley Bennett, MD
Emory University
Bayley
Bennett,
MD - Emory University
HELPeR: A Novel Digital-Health Librarian Supporting Ovarian Cancer Patients and Caregivers Through Tailored Health Information
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Patient Engagement and Preferences, Nursing Informatics, Information Retrieval, Delivering Health Information and Knowledge to the Public
Primary Track: Applications
Programmatic Theme: Consumer Health Informatics
Background: Ovarian cancer (OvCa) presents significant informational challenges for patients and caregivers. HELPeR (Health E-Librarian with Personalized Recommendations) is a digital platform designed to provide tailored health information based on disease trajectory, user needs, and preferences. This study examined user characteristics, online information-seeking behaviors, and perceptions of the recommended content.
Methods: A 12-week study recruited OvCa patients and caregivers in person and online. Participants completed surveys at baseline, week 6, and week 12. They received a guided demonstration of HELPeR and were encouraged to use the system independently. Engagement metrics, including accessed articles, bookmarks, and feedback submissions, were analyzed.
Results: Among 20 participants, most relied on search engines (patients: 71.5%, caregivers: 69.3%) but used online recommendation systems infrequently. Patients accessed an average of 29 articles, bookmarked 11, and submitted 19 feedback responses, whereas caregivers accessed 13 articles, bookmarked 2, and submitted 19 responses. Both groups rated recommended articles highly for relevance and clarity, with symptom/treatment-related content being the most accessed.
Conclusion: Findings suggest that OvCa patients and caregivers actively seek online health information and benefit from tailored recommendations. However, the sample was skewed toward users with higher education and digital literacy. Patients engaged more than caregivers, potentially reflecting differing informational needs or caregiving burdens. Future research should explore barriers to caregiver engagement and assess HELPeR’s impact on long-term health outcomes. Digital platforms like HELPeR have the potential to bridge informational gaps and enhance the care experience.
Speaker:
Youjia Wang, BSN, RN
University of Pittsburgh School of Nursing
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Patient Engagement and Preferences, Nursing Informatics, Information Retrieval, Delivering Health Information and Knowledge to the Public
Primary Track: Applications
Programmatic Theme: Consumer Health Informatics
Background: Ovarian cancer (OvCa) presents significant informational challenges for patients and caregivers. HELPeR (Health E-Librarian with Personalized Recommendations) is a digital platform designed to provide tailored health information based on disease trajectory, user needs, and preferences. This study examined user characteristics, online information-seeking behaviors, and perceptions of the recommended content.
Methods: A 12-week study recruited OvCa patients and caregivers in person and online. Participants completed surveys at baseline, week 6, and week 12. They received a guided demonstration of HELPeR and were encouraged to use the system independently. Engagement metrics, including accessed articles, bookmarks, and feedback submissions, were analyzed.
Results: Among 20 participants, most relied on search engines (patients: 71.5%, caregivers: 69.3%) but used online recommendation systems infrequently. Patients accessed an average of 29 articles, bookmarked 11, and submitted 19 feedback responses, whereas caregivers accessed 13 articles, bookmarked 2, and submitted 19 responses. Both groups rated recommended articles highly for relevance and clarity, with symptom/treatment-related content being the most accessed.
Conclusion: Findings suggest that OvCa patients and caregivers actively seek online health information and benefit from tailored recommendations. However, the sample was skewed toward users with higher education and digital literacy. Patients engaged more than caregivers, potentially reflecting differing informational needs or caregiving burdens. Future research should explore barriers to caregiver engagement and assess HELPeR’s impact on long-term health outcomes. Digital platforms like HELPeR have the potential to bridge informational gaps and enhance the care experience.
Speaker:
Youjia Wang, BSN, RN
University of Pittsburgh School of Nursing
Youjia
Wang,
BSN, RN - University of Pittsburgh School of Nursing
Effectiveness of Participant Support in a Digital Health Trial: Insights from the DIAMANTE Study
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Patient Engagement and Preferences, Mobile Health, User-centered Design Methods, Health Equity
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
Digital health interventions can improve chronic disease management, particularly for those from historically underrepresented populations. A secondary analysis of the DIAMANTE RCT quantified and described the impact of automated text messages and human support outreach on participant retention. Our findings suggest that built-in digital and human support greatly improved retention (with over half of the participants re-engaging in the trial through this outreach), underscoring the need to tailor intervention engagement in the future.
Speaker:
Courtney Lyles, PhD
UC Davis Center for Healthcare Policy and Research
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Patient Engagement and Preferences, Mobile Health, User-centered Design Methods, Health Equity
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
Digital health interventions can improve chronic disease management, particularly for those from historically underrepresented populations. A secondary analysis of the DIAMANTE RCT quantified and described the impact of automated text messages and human support outreach on participant retention. Our findings suggest that built-in digital and human support greatly improved retention (with over half of the participants re-engaging in the trial through this outreach), underscoring the need to tailor intervention engagement in the future.
Speaker:
Courtney Lyles, PhD
UC Davis Center for Healthcare Policy and Research
Courtney
Lyles,
PhD - UC Davis Center for Healthcare Policy and Research
Quantifying Information Needs of Patients and Caregivers Managing Dementia through Classification of Patient-Initiated Messages
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Patient Engagement and Preferences, Large Language Models (LLMs), Chronic Care Management
Primary Track: Applications
Programmatic Theme: Consumer Health Informatics
This study examines how patients with dementia and care partners use patient portal messaging to manage care. Using an expert-derived taxonomy of information needs, we classified the content of all messages sent after a dementia diagnosis. Among over 35000 messages, the majority discussed medications and managing clinical concerns, including changes to treatment plans. Understanding information needs is a critical step to providing resources and proactive support that improve the work of caregiving, coordination, and self-management.
Speaker:
Bryan Steitz, PhD
Vanderbilt University Medical Center
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Patient Engagement and Preferences, Large Language Models (LLMs), Chronic Care Management
Primary Track: Applications
Programmatic Theme: Consumer Health Informatics
This study examines how patients with dementia and care partners use patient portal messaging to manage care. Using an expert-derived taxonomy of information needs, we classified the content of all messages sent after a dementia diagnosis. Among over 35000 messages, the majority discussed medications and managing clinical concerns, including changes to treatment plans. Understanding information needs is a critical step to providing resources and proactive support that improve the work of caregiving, coordination, and self-management.
Speaker:
Bryan Steitz, PhD
Vanderbilt University Medical Center
Bryan
Steitz,
PhD - Vanderbilt University Medical Center
Uncovering Pharmacogenomic Variability in Prostate cancer Using Big Data
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Omics (genomics, metabolomics, proteomics, transcriptomics, etc.) and Integrative Analyses, Real-World Evidence Generation, Precision Medicine, Clinical Decision Support
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
Pharmacogenomics improves clinical outcomes by enabling personalized medication, particularly for cancer patients. We analyzed 16 key pharmacogenes, including CYP2D6, in ~3,500 prostate cancer patients from the All of Us Program using WGS and EHR data. All patients had at least one actionable PGx phenotype (median is 5). 61.1% had both an actionable phenotype and exposure to an affected medication. These findings support integrating PGx into clinical care to optimize treatment outcomes for prostate cancer patients.
Speaker:
Huiyi Yang, PhD Student
University of Utah
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Omics (genomics, metabolomics, proteomics, transcriptomics, etc.) and Integrative Analyses, Real-World Evidence Generation, Precision Medicine, Clinical Decision Support
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
Pharmacogenomics improves clinical outcomes by enabling personalized medication, particularly for cancer patients. We analyzed 16 key pharmacogenes, including CYP2D6, in ~3,500 prostate cancer patients from the All of Us Program using WGS and EHR data. All patients had at least one actionable PGx phenotype (median is 5). 61.1% had both an actionable phenotype and exposure to an affected medication. These findings support integrating PGx into clinical care to optimize treatment outcomes for prostate cancer patients.
Speaker:
Huiyi Yang, PhD Student
University of Utah
Huiyi
Yang,
PhD Student - University of Utah
Clinical Nurses’ Readiness to Interpret AI-Powered Predictive Models: A Quantitative Assessment
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Nursing Informatics, Surveys and Needs Analysis, Workforce Development, Artificial Intelligence, Education and Training
Primary Track: Foundations
Programmatic Theme: Clinical Informatics
This study examined the statistical knowledge and health numeracy related to probability and likelihood, which are essential for understanding these AI-based models and systems. The findings revealed specific areas of weakness that need to be addressed.
Speaker:
INSOOK CHO, PhD
Inha University
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Nursing Informatics, Surveys and Needs Analysis, Workforce Development, Artificial Intelligence, Education and Training
Primary Track: Foundations
Programmatic Theme: Clinical Informatics
This study examined the statistical knowledge and health numeracy related to probability and likelihood, which are essential for understanding these AI-based models and systems. The findings revealed specific areas of weakness that need to be addressed.
Speaker:
INSOOK CHO, PhD
Inha University
INSOOK
CHO,
PhD - Inha University
The effectiveness of AI chatbots in alleviating mental distress and promoting health behaviors among Adolescents and Young Adults (AYAs): a systematic review and meta-analysis.
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Nursing Informatics, Population Health, Delivering Health Information and Knowledge to the Public, Personal Health Informatics, Patient Engagement and Preferences, Public Health, Artificial Intelligence, Large Language Models (LLMs)
Primary Track: Applications
Programmatic Theme: Public Health Informatics
This study evaluates chatbot effectiveness in alleviating mental distress and promoting health behaviors among adolescents and young adults. A meta-analysis revealed small-to-moderate improvements in depressive, anxiety, stress, psychosomatic symptoms; negative affect, life satisfaction, self-ambivalence, and health behaviors. Dialogue system methods, deployment formats, and the use of reminders were identified as key moderators of success. While generative AI chatbots show promise, further research is required to validate their long-term impacts and establish robust safety protocols.
Speaker:
Xinyu Feng, Master
The Hong Kong Polytechnic University
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Nursing Informatics, Population Health, Delivering Health Information and Knowledge to the Public, Personal Health Informatics, Patient Engagement and Preferences, Public Health, Artificial Intelligence, Large Language Models (LLMs)
Primary Track: Applications
Programmatic Theme: Public Health Informatics
This study evaluates chatbot effectiveness in alleviating mental distress and promoting health behaviors among adolescents and young adults. A meta-analysis revealed small-to-moderate improvements in depressive, anxiety, stress, psychosomatic symptoms; negative affect, life satisfaction, self-ambivalence, and health behaviors. Dialogue system methods, deployment formats, and the use of reminders were identified as key moderators of success. While generative AI chatbots show promise, further research is required to validate their long-term impacts and establish robust safety protocols.
Speaker:
Xinyu Feng, Master
The Hong Kong Polytechnic University
Xinyu
Feng,
Master - The Hong Kong Polytechnic University
An NLP Pipeline for Structuring Tremor Severity Ratings from Semi-Structured Clinical Notes
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
Tremor assessment relies on subjective clinician observations, limiting large-scale machine learning applications. We evaluated LLaMA3 (70B) for automated tremor classification, converting clinician notes into binary labels across seven body parts. Benchmarking against structured annotations (2,980 labels), the model achieved a macro F1 score of 0.893. Performance varied by body part, with strong agreement for hand tremors but lower accuracy for face and trunk. Future work will refine misclassifications and assess generalizability to less structured documentation.
Speaker:
Jeanne Powell, PhD
Emory University
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
Tremor assessment relies on subjective clinician observations, limiting large-scale machine learning applications. We evaluated LLaMA3 (70B) for automated tremor classification, converting clinician notes into binary labels across seven body parts. Benchmarking against structured annotations (2,980 labels), the model achieved a macro F1 score of 0.893. Performance varied by body part, with strong agreement for hand tremors but lower accuracy for face and trunk. Future work will refine misclassifications and assess generalizability to less structured documentation.
Speaker:
Jeanne Powell, PhD
Emory University
Jeanne
Powell,
PhD - Emory University
Identifying palliative care needs and cardiovascular symptoms in Dutch clinical notes: An evaluation of the text mining application NimbleMiner
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Natural Language Processing, Information Extraction, Nursing Informatics
Primary Track: Foundations
Programmatic Theme: Clinical Informatics
The early identification of palliative care needs in cardiovascular patients is hindered by an administrative burden of patient identification tools. Natural language processing can be leveraged to extract information indicating palliative care needs from clinical notes in the electronic health record. This study evaluated the clinical text mining application NimbleMiner for the identification of symptoms indicating palliative care needs in Dutch clinical notes. We achieved excellent performance, demonstrating potential for future applications.
Speaker:
Chloé Desmedt, Master of Science in Nursing and Midwifery
KU Leuven
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Natural Language Processing, Information Extraction, Nursing Informatics
Primary Track: Foundations
Programmatic Theme: Clinical Informatics
The early identification of palliative care needs in cardiovascular patients is hindered by an administrative burden of patient identification tools. Natural language processing can be leveraged to extract information indicating palliative care needs from clinical notes in the electronic health record. This study evaluated the clinical text mining application NimbleMiner for the identification of symptoms indicating palliative care needs in Dutch clinical notes. We achieved excellent performance, demonstrating potential for future applications.
Speaker:
Chloé Desmedt, Master of Science in Nursing and Midwifery
KU Leuven
Chloé
Desmedt,
Master of Science in Nursing and Midwifery - KU Leuven
An NLP Method to Identify Macro Guideline Sentences in Radiology Reports
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Natural Language Processing, Information Extraction, Deep Learning
Primary Track: Applications
Programmatic Theme: Academic Informatics / LIEAF
Radiology reports contain system-generated macro guidelines alongside patient-specific findings, which can interfere with information extraction (IE) models. This study presents a ClinicalBERT-based approach to identify and filter macro guideline sentences, improving patient-specific data extraction. Using an annotated dataset of 8,860 sentences, our model achieved superior performance over a TF-IDF + SVM baseline, with an F1-score increase from 0.9544 to 0.9705. Filtering macros enhances dataset quality for downstream NLP tasks, though further refinements are needed to address misclassification challenges.
Speaker:
Zhaoyi Sun, Master of Science
University of Washington
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Natural Language Processing, Information Extraction, Deep Learning
Primary Track: Applications
Programmatic Theme: Academic Informatics / LIEAF
Radiology reports contain system-generated macro guidelines alongside patient-specific findings, which can interfere with information extraction (IE) models. This study presents a ClinicalBERT-based approach to identify and filter macro guideline sentences, improving patient-specific data extraction. Using an annotated dataset of 8,860 sentences, our model achieved superior performance over a TF-IDF + SVM baseline, with an F1-score increase from 0.9544 to 0.9705. Filtering macros enhances dataset quality for downstream NLP tasks, though further refinements are needed to address misclassification challenges.
Speaker:
Zhaoyi Sun, Master of Science
University of Washington
Zhaoyi
Sun,
Master of Science - University of Washington
Streamlining Temporal Information Extraction: Integrating Rule-Based Methods into MedspaCy for Clinical Application
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Natural Language Processing, Information Extraction, Data Mining
Primary Track: Applications
Programmatic Theme: Clinical Informatics
Electronic health records (EHRs) contain detailed records of a patient’s health over time, making it crucial to accurately identify the timing of a clinical event. In this work, we introduce a rule-based temporal entity extraction package for medspaCy, which can adapt to any spaCy-styled tokenizers, as well as handle additional types of implicit expressions.
Speaker:
Mengke Hu, PHD
University of Utah
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Natural Language Processing, Information Extraction, Data Mining
Primary Track: Applications
Programmatic Theme: Clinical Informatics
Electronic health records (EHRs) contain detailed records of a patient’s health over time, making it crucial to accurately identify the timing of a clinical event. In this work, we introduce a rule-based temporal entity extraction package for medspaCy, which can adapt to any spaCy-styled tokenizers, as well as handle additional types of implicit expressions.
Speaker:
Mengke Hu, PHD
University of Utah
Mengke
Hu,
PHD - University of Utah
Autocomplete Using LLMs for Simplifying Health-related Texts
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Natural Language Processing, Delivering Health Information and Knowledge to the Public, Large Language Models (LLMs)
Primary Track: Applications
Programmatic Theme: Consumer Health Informatics
Health-literacy is critical for improving health outcomes. Unfortunately, it can be challenging to create educational materials that are accessible and few tools help content creators to write materials. In this paper, we explore autocompletion as a tool to help create content. We examine GPT3.5 Davinci for autocompletion on a corpus of simplified medical texts, which achieves ap- proximately 20% accuracy. We conclude with a number of potential options for improving this baseline performance.
Speaker:
David Kauchak, PhD
Pomona College
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Natural Language Processing, Delivering Health Information and Knowledge to the Public, Large Language Models (LLMs)
Primary Track: Applications
Programmatic Theme: Consumer Health Informatics
Health-literacy is critical for improving health outcomes. Unfortunately, it can be challenging to create educational materials that are accessible and few tools help content creators to write materials. In this paper, we explore autocompletion as a tool to help create content. We examine GPT3.5 Davinci for autocompletion on a corpus of simplified medical texts, which achieves ap- proximately 20% accuracy. We conclude with a number of potential options for improving this baseline performance.
Speaker:
David Kauchak, PhD
Pomona College
David
Kauchak,
PhD - Pomona College
Identifying Documented Goals of Care Conversations Using Definition-informed Chain-of-Thought Prompting (DiCoT)
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Natural Language Processing, Bioinformatics, Deep Learning
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
Goals-of-care discussions (GoCD) are essential for aligning care with patient priorities, yet their documentation varies, challenging automated detection using Natural Language Processing (NLP). While fine-tuned BERT models perform well in identifying GoCD in clinical notes, their reliance on large annotated datasets and limited context length constrains their utility. Large Language Models (LLMs) offer a promising alternative, requiring minimal labeled data while handling longer text sequences.
In this study, we present the first evaluation of an LLM for GoCD detection, comparing its performance against a fine-tuned BERT model. We assessed both zero-shot performance and the utility of a novel GoCD-focused version of Chain-of-Thought prompting, Definition-informed CoT (DiCOT), on the task of extracting GoCDs. Our findings suggest that LLMs prompted using DiCoT can match end-to-end fine-tuning models in identifying GoCD without requiring annotated training data. This offers a practical alternative as it reduces the need for resource-intensive annotation requirements while maintaining performance.
Speaker:
Kevin Li, PhD Candidate
University of Washington
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Natural Language Processing, Bioinformatics, Deep Learning
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
Goals-of-care discussions (GoCD) are essential for aligning care with patient priorities, yet their documentation varies, challenging automated detection using Natural Language Processing (NLP). While fine-tuned BERT models perform well in identifying GoCD in clinical notes, their reliance on large annotated datasets and limited context length constrains their utility. Large Language Models (LLMs) offer a promising alternative, requiring minimal labeled data while handling longer text sequences.
In this study, we present the first evaluation of an LLM for GoCD detection, comparing its performance against a fine-tuned BERT model. We assessed both zero-shot performance and the utility of a novel GoCD-focused version of Chain-of-Thought prompting, Definition-informed CoT (DiCOT), on the task of extracting GoCDs. Our findings suggest that LLMs prompted using DiCoT can match end-to-end fine-tuning models in identifying GoCD without requiring annotated training data. This offers a practical alternative as it reduces the need for resource-intensive annotation requirements while maintaining performance.
Speaker:
Kevin Li, PhD Candidate
University of Washington
Kevin
Li,
PhD Candidate - University of Washington
Bridging Gaps in HIV Care: Usability Evaluation of a mHealth App for Identifying and Retaining Individuals with Non-Viral Suppression
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Mobile Health, User-centered Design Methods, Usability, Public Health
Primary Track: Applications
Programmatic Theme: Public Health Informatics
Despite improved outcomes, geographic disparities persist, with higher rates of people with HIV who have not achieved viral suppression living in the Deep South. In response, we developed the Drive to Zero app, targeting individuals who have not achieved viral suppression or lacked evidence of care within the past 12 months, aligning with 2022-2025 National HIV/AIDS Strategy Goals. Our usability evaluation revealed high user satisfaction, demonstrating that our user-centered design supported a usable app.
Speaker:
Fabiana Dos Santos, PhD, MSN, RN
Columbia University School of Nursing
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Mobile Health, User-centered Design Methods, Usability, Public Health
Primary Track: Applications
Programmatic Theme: Public Health Informatics
Despite improved outcomes, geographic disparities persist, with higher rates of people with HIV who have not achieved viral suppression living in the Deep South. In response, we developed the Drive to Zero app, targeting individuals who have not achieved viral suppression or lacked evidence of care within the past 12 months, aligning with 2022-2025 National HIV/AIDS Strategy Goals. Our usability evaluation revealed high user satisfaction, demonstrating that our user-centered design supported a usable app.
Speaker:
Fabiana Dos Santos, PhD, MSN, RN
Columbia University School of Nursing
Fabiana
Dos Santos,
PhD, MSN, RN - Columbia University School of Nursing
Developing a Cancer Patient Navigation System: Usability Results
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Mobile Health, Usability, User-centered Design Methods, Personal Health Informatics, Health Equity
Primary Track: Applications
Programmatic Theme: Consumer Health Informatics
We co-designed a patient navigation system, consisting of a smartphone app for cancer patients and a web-based dashboard for their navigators. The app collects patient-reported barriers to care and suggests relevant services. Navigators receive alerts and review this data via the dashboard. Usability testing with five cancer patients and six navigators showed both tools were easy to use and demonstrated good usability.
Speaker:
Ming-Yuan Chih, PhD
University of Kentucky
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Mobile Health, Usability, User-centered Design Methods, Personal Health Informatics, Health Equity
Primary Track: Applications
Programmatic Theme: Consumer Health Informatics
We co-designed a patient navigation system, consisting of a smartphone app for cancer patients and a web-based dashboard for their navigators. The app collects patient-reported barriers to care and suggests relevant services. Navigators receive alerts and review this data via the dashboard. Usability testing with five cancer patients and six navigators showed both tools were easy to use and demonstrated good usability.
Speaker:
Ming-Yuan Chih, PhD
University of Kentucky
Ming-Yuan
Chih,
PhD - University of Kentucky
Assessing the Quality and Usability of Mobile Dental Health Apps Using the MARS Framework
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Mobile Health, Usability, Education and Training, User-centered Design Methods
Primary Track: Applications
Programmatic Theme: Consumer Health Informatics
This study evaluates the quality and usability of mobile dental health apps using the Mobile App Rating Scale (MARS). After screening over 600 apps across iOS and Android platforms, 27 met the inclusion criteria. Apps were assessed across engagement, functionality, aesthetics, and information quality. Findings highlight top-performing apps and key design gaps, offering actionable insights for developers and healthcare professionals to enhance digital oral health interventions and user experience.
Speaker:
Prathibha Bondili, Masters
University of Pittsburgh
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Mobile Health, Usability, Education and Training, User-centered Design Methods
Primary Track: Applications
Programmatic Theme: Consumer Health Informatics
This study evaluates the quality and usability of mobile dental health apps using the Mobile App Rating Scale (MARS). After screening over 600 apps across iOS and Android platforms, 27 met the inclusion criteria. Apps were assessed across engagement, functionality, aesthetics, and information quality. Findings highlight top-performing apps and key design gaps, offering actionable insights for developers and healthcare professionals to enhance digital oral health interventions and user experience.
Speaker:
Prathibha Bondili, Masters
University of Pittsburgh
Prathibha
Bondili,
Masters - University of Pittsburgh
Machine Learning of Remote Video Interviews for Quantification of Cognitive Impairment and Psychological Well-Being in Older Adults
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Machine Learning, Telemedicine, Clinical Decision Support
Primary Track: Applications
Population aging is increasing the MCI healthcare burden, worsening mental health issues amid a workforce shortage. This study demonstrates the feasibility of monitoring 1) psychological well-being (anxiety, depression, loneliness, and social engagement; 80% F1 score overall) in people with MCI through facial videos and 2) cognitive decline over 12 months (76% F1 score) through speech analysis. Our future work will conduct field studies using our models to develop clinical decision support for dementia care clinics.
Speaker:
Merna Bibars, MSc.
Georgia Institute of Technology
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Machine Learning, Telemedicine, Clinical Decision Support
Primary Track: Applications
Population aging is increasing the MCI healthcare burden, worsening mental health issues amid a workforce shortage. This study demonstrates the feasibility of monitoring 1) psychological well-being (anxiety, depression, loneliness, and social engagement; 80% F1 score overall) in people with MCI through facial videos and 2) cognitive decline over 12 months (76% F1 score) through speech analysis. Our future work will conduct field studies using our models to develop clinical decision support for dementia care clinics.
Speaker:
Merna Bibars, MSc.
Georgia Institute of Technology
Merna
Bibars,
MSc. - Georgia Institute of Technology
Machine Learning-Based Prediction of Cancer-Related Cardiac Dysfunction in Cardio-Oncology Patients Using the All-of-Us Research Program
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Machine Learning, Quantitative Methods, Cancer Prevention, Chronic Care Management, Delivering Health Information and Knowledge to the Public, Nursing Informatics, Patient / Person Generated Health Data (Patient Reported Outcomes), Precision Medicine
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
Background: Cardio-Oncology (CO) patients face overlapping symptoms and complex treatment challenges, which may increase their risk of developing cancer-related cardiac dysfunction (CRCD), including heart failure and cardiomyopathy. Identifying key factors that predict the occurrence of CRCD is crucial for improving prognosis. This study aims to identify the predictors of CRCD in CO patients and assess their impact on clinical outcomes.
Methods: This study used All of Us data, incorporating EHR and survey responses. Final sample included 5,729 participants, with 19% with CRCD and 81% without CRCD. To address class imbalance, class weighting was applied. We implemented three machine learning (ML) models: Logistic Regression, Random Forest, and XGBoost. For preprocesing, the dataset was split into Train - Test (80/20) subsets, and robust scaling and one-hot encoding were applied. Each model was trained using the train dataset and evaluated based on predictions made on the test dataset to assess performance.
Results: XGBoost achieved the most balanced performance with the highest Accuracy (0.8148) and F1 score (0.8962). Feature importance analysis identified number of CO symptoms as the strongest predictor, followed by age, income, and gender, while SHAP analysis highlighted discrimination, neighborhood disorder, and number of CO symptoms as key contributors to CRCD risk. An AUC of 0.75 indicates fair to good predictive accuracy.
Conclusion: These findings underscore the significance of Cardio-Oncology symptoms, sociodemographic factors, and social determinants of health in predicting CRCD risk. Integrating these factors into early screening and risk stratification may enhance clinical decision-making and patient outcomes.
Speaker:
Cheongin "Rachel" Im, MSN, RN
The University of Texas at Austin School of Nursing
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Machine Learning, Quantitative Methods, Cancer Prevention, Chronic Care Management, Delivering Health Information and Knowledge to the Public, Nursing Informatics, Patient / Person Generated Health Data (Patient Reported Outcomes), Precision Medicine
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
Background: Cardio-Oncology (CO) patients face overlapping symptoms and complex treatment challenges, which may increase their risk of developing cancer-related cardiac dysfunction (CRCD), including heart failure and cardiomyopathy. Identifying key factors that predict the occurrence of CRCD is crucial for improving prognosis. This study aims to identify the predictors of CRCD in CO patients and assess their impact on clinical outcomes.
Methods: This study used All of Us data, incorporating EHR and survey responses. Final sample included 5,729 participants, with 19% with CRCD and 81% without CRCD. To address class imbalance, class weighting was applied. We implemented three machine learning (ML) models: Logistic Regression, Random Forest, and XGBoost. For preprocesing, the dataset was split into Train - Test (80/20) subsets, and robust scaling and one-hot encoding were applied. Each model was trained using the train dataset and evaluated based on predictions made on the test dataset to assess performance.
Results: XGBoost achieved the most balanced performance with the highest Accuracy (0.8148) and F1 score (0.8962). Feature importance analysis identified number of CO symptoms as the strongest predictor, followed by age, income, and gender, while SHAP analysis highlighted discrimination, neighborhood disorder, and number of CO symptoms as key contributors to CRCD risk. An AUC of 0.75 indicates fair to good predictive accuracy.
Conclusion: These findings underscore the significance of Cardio-Oncology symptoms, sociodemographic factors, and social determinants of health in predicting CRCD risk. Integrating these factors into early screening and risk stratification may enhance clinical decision-making and patient outcomes.
Speaker:
Cheongin "Rachel" Im, MSN, RN
The University of Texas at Austin School of Nursing
Cheongin "Rachel"
Im,
MSN, RN - The University of Texas at Austin School of Nursing
Impact of Socioeconomic Status on Maternal Mental Health among Pregnant Women Using NIH All of Us Data
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Machine Learning, Quantitative Methods, Natural Language Processing, Informatics Implementation, Large Language Models (LLMs)
Primary Track: Applications
Programmatic Theme: Clinical Informatics
This study examines the impact of socioeconomic factors such as education, marital status, health insurance, employment, and income on maternal mental health (MMH) among pregnant women in the NIH All of Us Research Program. We developed predictive models to identify MMH risks early, with the Random Forest model achieving the highest AUROC (0.90). Our findings highlight the significant impact of socioeconomic status on MMH. These insights can inform targeted policies and interventions to improve maternal well-being and reduce adverse pregnancy outcomes.
Speaker:
Uma Sarder, Ms in Data Science
Meharry Medical College, Nashville, TN
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Machine Learning, Quantitative Methods, Natural Language Processing, Informatics Implementation, Large Language Models (LLMs)
Primary Track: Applications
Programmatic Theme: Clinical Informatics
This study examines the impact of socioeconomic factors such as education, marital status, health insurance, employment, and income on maternal mental health (MMH) among pregnant women in the NIH All of Us Research Program. We developed predictive models to identify MMH risks early, with the Random Forest model achieving the highest AUROC (0.90). Our findings highlight the significant impact of socioeconomic status on MMH. These insights can inform targeted policies and interventions to improve maternal well-being and reduce adverse pregnancy outcomes.
Speaker:
Uma Sarder, Ms in Data Science
Meharry Medical College, Nashville, TN
Uma
Sarder,
Ms in Data Science - Meharry Medical College, Nashville, TN
Leveraging Health Data and Machine Learning to Predict No-Show Propensity Factors in Glaucoma Patients: A Health Informatics Approach to Improve Clinic Efficiency and Patient Care
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Machine Learning, Personal Health Informatics, Clinical Decision Support
Primary Track: Applications
Programmatic Theme: Clinical Informatics
This study combines predictive modeling and inference using electronic health records to identify factors influencing no-show risk for glaucoma patients at their first visit. Data from 8,504 patients generated a no-show propensity factor (NSPF), with XGBoost achieving 89% sensitivity. Higher NSPF was linked to younger age, Black race, lower socioeconomic status, and worse clinical conditions. These findings identify at-risk patients and suggest tailored interventions, such as social support, to improve appointment adherence and care continuity.
Speaker:
Hai-Wei Liang, PhD
Department of Ophthalmology, University of Pittsburgh School of Medicine
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Machine Learning, Personal Health Informatics, Clinical Decision Support
Primary Track: Applications
Programmatic Theme: Clinical Informatics
This study combines predictive modeling and inference using electronic health records to identify factors influencing no-show risk for glaucoma patients at their first visit. Data from 8,504 patients generated a no-show propensity factor (NSPF), with XGBoost achieving 89% sensitivity. Higher NSPF was linked to younger age, Black race, lower socioeconomic status, and worse clinical conditions. These findings identify at-risk patients and suggest tailored interventions, such as social support, to improve appointment adherence and care continuity.
Speaker:
Hai-Wei Liang, PhD
Department of Ophthalmology, University of Pittsburgh School of Medicine
Hai-Wei
Liang,
PhD - Department of Ophthalmology, University of Pittsburgh School of Medicine
Comparison of time series clustering approaches for identifying subphenotypes of acute cardiology patients
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Machine Learning, Information Extraction, Personal Health Informatics, Nursing Informatics
Primary Track: Foundations
Programmatic Theme: Clinical Research Informatics
Highly dimensional inpatient time series data are routinely collected but rarely used in clinical decision-making due to computational complexity in modeling. Although various time series clustering approaches aim to reduce dimensions while minimizing information loss, it is unknown whether they identify distinct or similar illness severity trajectories among acute cardiology inpatients. Thus, the aim of this study is to compare clustering agreement and performance across approaches as well as characterize identified subphenotypes.
Speaker:
Reanna Panagides, MS, RN
University of Virginia
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Machine Learning, Information Extraction, Personal Health Informatics, Nursing Informatics
Primary Track: Foundations
Programmatic Theme: Clinical Research Informatics
Highly dimensional inpatient time series data are routinely collected but rarely used in clinical decision-making due to computational complexity in modeling. Although various time series clustering approaches aim to reduce dimensions while minimizing information loss, it is unknown whether they identify distinct or similar illness severity trajectories among acute cardiology inpatients. Thus, the aim of this study is to compare clustering agreement and performance across approaches as well as characterize identified subphenotypes.
Speaker:
Reanna Panagides, MS, RN
University of Virginia
Reanna
Panagides,
MS, RN - University of Virginia
Machine Learning Enhances Diagnosis of Necrotizing Soft Tissue Infections with Superior Accuracy
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Machine Learning, Infectious Diseases and Epidemiology, Diagnostic Systems
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
Machine learning enhances necrotizing soft tissue infection (NSTI) diagnosis by leveraging standard clinical data to produce an interpretable risk prediction tool. This approach surpasses the Laboratory Risk Indicator for Necrotizing Fasciitis (LRINEC)—a commonly used risk stratification score that underperforms in external validation. The new model shows strong potential for seamless clinical integration, improving accuracy and supporting timely decision-making to help reduce mortality from NSTI.
Speaker:
Anita Subbarao, MD
University of Washington
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Machine Learning, Infectious Diseases and Epidemiology, Diagnostic Systems
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
Machine learning enhances necrotizing soft tissue infection (NSTI) diagnosis by leveraging standard clinical data to produce an interpretable risk prediction tool. This approach surpasses the Laboratory Risk Indicator for Necrotizing Fasciitis (LRINEC)—a commonly used risk stratification score that underperforms in external validation. The new model shows strong potential for seamless clinical integration, improving accuracy and supporting timely decision-making to help reduce mortality from NSTI.
Speaker:
Anita Subbarao, MD
University of Washington
Anita
Subbarao,
MD - University of Washington
Enhanced Atrial Fibrillation Detection in ICU: Leveraging Novel ECG-Derived Features with Machine Learning
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Machine Learning, Evaluation, Quantitative Methods, Artificial Intelligence, Diagnostic Systems, Nursing Informatics, Informatics Implementation, Data transformation/ETL
Primary Track: Applications
Programmatic Theme: Clinical Informatics
Atrial fibrillation (AF) monitoring is crucial yet hindered by false alarms in ICUs. We developed machine learning models combining ECG features (engineered and foundation model-based) and demographics to distinguish true vs. false AF. Clinicians annotated 723 alarms, labeling 161 (22.3%) as false. A logistic regression model achieved 87% AUROC, reduced false alarms reduced to 14.3%, and maintaining 90% sensitivity. This approach may enhance AF alarm accuracy and reduce alarm fatigue in critical care.
Speaker:
Andrew Lu, MSc, RN
Center for Data Science, Nell Hodgson Woodruff School of Nursing, Emory University
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Machine Learning, Evaluation, Quantitative Methods, Artificial Intelligence, Diagnostic Systems, Nursing Informatics, Informatics Implementation, Data transformation/ETL
Primary Track: Applications
Programmatic Theme: Clinical Informatics
Atrial fibrillation (AF) monitoring is crucial yet hindered by false alarms in ICUs. We developed machine learning models combining ECG features (engineered and foundation model-based) and demographics to distinguish true vs. false AF. Clinicians annotated 723 alarms, labeling 161 (22.3%) as false. A logistic regression model achieved 87% AUROC, reduced false alarms reduced to 14.3%, and maintaining 90% sensitivity. This approach may enhance AF alarm accuracy and reduce alarm fatigue in critical care.
Speaker:
Andrew Lu, MSc, RN
Center for Data Science, Nell Hodgson Woodruff School of Nursing, Emory University
Andrew
Lu,
MSc, RN - Center for Data Science, Nell Hodgson Woodruff School of Nursing, Emory University
Balanced Cluster Canonical Correlation Analysis for Multimodal Data Integration in Alzheimer's Disease
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Machine Learning, Diagnostic Systems, Bioinformatics
Primary Track: Foundations
This study addresses the challenge of dataset class imbalance in Canonical Correlation Analysis (CCA), a fundamental statistical method for elucidating relationships between two sets of variables. We propose Balanced Cluster CCA, a novel approach that mitigates imbalance through re-weighting, ensuring unbiased class representation. Our method enhances the balance of the representations of the majority and minority projections, thereby improving the robustness and accuracy in the context of Alzheimer's Disease (AD) detection studies. The efficacy of Balanced Cluster CCA in addressing imbalance is demonstrated through rigorous experimental evaluation, comprising subsequent classification tasks on two sets of AD datasets. Results indicate that our approach significantly improves the performance of standard CCA in imbalanced scenarios, with potential implications for broader applications in biomedical research. Beyond AD research, Balanced Cluster CCA offers promising applications across biomedical domains where data imbalance commonly undermines analytical precision and clinical utility.
Speaker:
Boning Tong, MSE
University of Pennsylvania
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Machine Learning, Diagnostic Systems, Bioinformatics
Primary Track: Foundations
This study addresses the challenge of dataset class imbalance in Canonical Correlation Analysis (CCA), a fundamental statistical method for elucidating relationships between two sets of variables. We propose Balanced Cluster CCA, a novel approach that mitigates imbalance through re-weighting, ensuring unbiased class representation. Our method enhances the balance of the representations of the majority and minority projections, thereby improving the robustness and accuracy in the context of Alzheimer's Disease (AD) detection studies. The efficacy of Balanced Cluster CCA in addressing imbalance is demonstrated through rigorous experimental evaluation, comprising subsequent classification tasks on two sets of AD datasets. Results indicate that our approach significantly improves the performance of standard CCA in imbalanced scenarios, with potential implications for broader applications in biomedical research. Beyond AD research, Balanced Cluster CCA offers promising applications across biomedical domains where data imbalance commonly undermines analytical precision and clinical utility.
Speaker:
Boning Tong, MSE
University of Pennsylvania
Boning
Tong,
MSE - University of Pennsylvania
LLM-Agent AI for Cancer Fact-Checking on Online Platforms
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Social Media and Connected Health, Artificial Intelligence
Primary Track: Applications
We developed a multi-agent LLM system to detect, verify, and respond to cancer-related misinformation (CRM) on YouTube. Using automated transcription, zero-shot classification, and evidence retrieval via PubMed, the system achieved strong accuracy and precision. This approach enables scalable, real-time CRM mitigation to support evidence-based cancer communication on social media.
Speaker:
Mohammed Al-Garadi, PhD
VUMC
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Social Media and Connected Health, Artificial Intelligence
Primary Track: Applications
We developed a multi-agent LLM system to detect, verify, and respond to cancer-related misinformation (CRM) on YouTube. Using automated transcription, zero-shot classification, and evidence retrieval via PubMed, the system achieved strong accuracy and precision. This approach enables scalable, real-time CRM mitigation to support evidence-based cancer communication on social media.
Speaker:
Mohammed Al-Garadi, PhD
VUMC
Mohammed
Al-Garadi,
PhD - VUMC
Brief Translation Quality Measure for Patient-reported Outcome Measures Using Machine Translations
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Patient / Person Generated Health Data (Patient Reported Outcomes), Evaluation
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
We describe the development and psychometric properties of the Translation Quality Measure (TQM) for the assessment of the quality and readability of translations for patient-reported outcome measures, with a specific focus on language models.
Speaker:
Sheng-Chieh Lu, PhD
The University of Texas MD Anderson Cancer Center
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Patient / Person Generated Health Data (Patient Reported Outcomes), Evaluation
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
We describe the development and psychometric properties of the Translation Quality Measure (TQM) for the assessment of the quality and readability of translations for patient-reported outcome measures, with a specific focus on language models.
Speaker:
Sheng-Chieh Lu, PhD
The University of Texas MD Anderson Cancer Center
Sheng-Chieh
Lu,
PhD - The University of Texas MD Anderson Cancer Center
Examining Nurses’ Acceptance of LLM-RR Based Nursing Manual Search Engine Using a Combined TAM and UTAUT Model
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Nursing Informatics, Surveys and Needs Analysis
Primary Track: Applications
This study evaluated nurses' acceptance of LLM-RR-based nursing manual search engine with 111 nurses. PLS-SEM analysis showed effort expectancy, social influence, and performance expectancy significantly influenced attitude, which strongly predicted intention to use (R²=0.39-0.47). Facilitating conditions had no significant effect. Results indicate successful implementation requires user-friendly interfaces, organizational support, and improved technical resources for adoption in nursing practice. The system addresses current limitations including dispersed documentation and inefficient searches that hinder information access.
Speaker:
Ye Eun Park, Bachelor
Seoul National University Bundang Hospital
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Nursing Informatics, Surveys and Needs Analysis
Primary Track: Applications
This study evaluated nurses' acceptance of LLM-RR-based nursing manual search engine with 111 nurses. PLS-SEM analysis showed effort expectancy, social influence, and performance expectancy significantly influenced attitude, which strongly predicted intention to use (R²=0.39-0.47). Facilitating conditions had no significant effect. Results indicate successful implementation requires user-friendly interfaces, organizational support, and improved technical resources for adoption in nursing practice. The system addresses current limitations including dispersed documentation and inefficient searches that hinder information access.
Speaker:
Ye Eun Park, Bachelor
Seoul National University Bundang Hospital
Ye Eun
Park,
Bachelor - Seoul National University Bundang Hospital
Enhancing Transparency in Large Language Model-Based Research with PromptLog: A Tool for Prompt History Tracking and Reporting
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Natural Language Processing, Human-computer Interaction
Primary Track: Foundations
Programmatic Theme: Academic Informatics / LIEAF
LLMs assist researchers with various tasks, but success depends on prompt quality, which requires iterative refinements. However, the frequent lack of transparency in these iterations hinders research replicability. To address this, we propose PromptLog, a tool for tracking and reporting prompt history. By documenting prompt iterations, PromptLog contributes to establishing best practices in LLM-based research at AMIA.
Speaker:
Lu Wang, Master
Stevens Institute of Technology
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Natural Language Processing, Human-computer Interaction
Primary Track: Foundations
Programmatic Theme: Academic Informatics / LIEAF
LLMs assist researchers with various tasks, but success depends on prompt quality, which requires iterative refinements. However, the frequent lack of transparency in these iterations hinders research replicability. To address this, we propose PromptLog, a tool for tracking and reporting prompt history. By documenting prompt iterations, PromptLog contributes to establishing best practices in LLM-based research at AMIA.
Speaker:
Lu Wang, Master
Stevens Institute of Technology
Lu
Wang,
Master - Stevens Institute of Technology
Large Language Models for Generative Mental Health Tasks: A Scoping Review
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Natural Language Processing, User-centered Design Methods, Diversity, Equity, Inclusion, and Accessibility, Legal, Ethical, Social and Regulatory Issues, Fairness and elimination of bias, Real-World Evidence Generation
Primary Track: Applications
Programmatic Theme: Clinical Informatics
Large language models (LLMs) demonstrate potential in generative mental health tasks, such as therapy and counseling, yet face challenges in evaluation and clinical validation. This scoping review of 16 studies (2020–2024) highlights applications including Cognitive Behavioral Therapy (CBT) interventions, clinical assistance, and emotional support, predominantly using GPT models. While showing short-term efficacy in anxiety and loneliness, studies relied on heterogeneous, non-standardized evaluations and underassessed safety and fairness. Key priorities include adopting validated frameworks, open-source development, and positioning LLMs as adjuncts to clinicians. Multimodal interfaces and ethical collaboration are essential to ensure equitable, safe integration before clinical deployment.
Speaker:
Yining Hua, MSc
Harvard T.H. Chan School of Public Health
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Natural Language Processing, User-centered Design Methods, Diversity, Equity, Inclusion, and Accessibility, Legal, Ethical, Social and Regulatory Issues, Fairness and elimination of bias, Real-World Evidence Generation
Primary Track: Applications
Programmatic Theme: Clinical Informatics
Large language models (LLMs) demonstrate potential in generative mental health tasks, such as therapy and counseling, yet face challenges in evaluation and clinical validation. This scoping review of 16 studies (2020–2024) highlights applications including Cognitive Behavioral Therapy (CBT) interventions, clinical assistance, and emotional support, predominantly using GPT models. While showing short-term efficacy in anxiety and loneliness, studies relied on heterogeneous, non-standardized evaluations and underassessed safety and fairness. Key priorities include adopting validated frameworks, open-source development, and positioning LLMs as adjuncts to clinicians. Multimodal interfaces and ethical collaboration are essential to ensure equitable, safe integration before clinical deployment.
Speaker:
Yining Hua, MSc
Harvard T.H. Chan School of Public Health
Yining
Hua,
MSc - Harvard T.H. Chan School of Public Health
NutriRAG: Unleashing the Power of Large Language Models for Food Identification and Classification through Retrieval Methods
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Natural Language Processing, Precision Medicine
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
Accurate classification of unstructured food entries from diet-tracking apps remains challenging due to natural language variability and limited database coverage. This study developed NutriRAG, a retrieval-augmented LLM framework, to automate food classification and analyze dietary patterns from free-text entries. RAG-enhanced GPT-4 achieved superior accuracy (Micro F1: 82.24) over fine-tuned BERT and non-RAG LLMs. Applied to an RCT (NCT04259632), NutriRAG revealed metabolic links between meal timing (p<0.05), demonstrating its potential for precision nutrition research.
Speaker:
Huixue Zhou, PhD
University of Minnesota
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Natural Language Processing, Precision Medicine
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
Accurate classification of unstructured food entries from diet-tracking apps remains challenging due to natural language variability and limited database coverage. This study developed NutriRAG, a retrieval-augmented LLM framework, to automate food classification and analyze dietary patterns from free-text entries. RAG-enhanced GPT-4 achieved superior accuracy (Micro F1: 82.24) over fine-tuned BERT and non-RAG LLMs. Applied to an RCT (NCT04259632), NutriRAG revealed metabolic links between meal timing (p<0.05), demonstrating its potential for precision nutrition research.
Speaker:
Huixue Zhou, PhD
University of Minnesota
Huixue
Zhou,
PhD - University of Minnesota
Understanding Large Language Models’ Rating Behaviors on Novel Psychoactive Substances using Reddit
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Natural Language Processing, Social Media and Connected Health, Public Health, Information Retrieval
Primary Track: Applications
Programmatic Theme: Public Health Informatics
We used Reddit discussions on six novel psychoactive substances to generate answers to expert questions and evaluated these responses using four large language models (LLMs). Larger LLMs were more decisive in their ratings, particularly when objective evaluation metrics were present. Self-judging models showed stronger agreement with their own answers under objective criteria but were less consistent under subjective metrics. These results highlight model size and metric objectivity as key factors in LLM judgment patterns.
Speaker:
Swati Rajwal, PhD
Emory University
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Natural Language Processing, Social Media and Connected Health, Public Health, Information Retrieval
Primary Track: Applications
Programmatic Theme: Public Health Informatics
We used Reddit discussions on six novel psychoactive substances to generate answers to expert questions and evaluated these responses using four large language models (LLMs). Larger LLMs were more decisive in their ratings, particularly when objective evaluation metrics were present. Self-judging models showed stronger agreement with their own answers under objective criteria but were less consistent under subjective metrics. These results highlight model size and metric objectivity as key factors in LLM judgment patterns.
Speaker:
Swati Rajwal, PhD
Emory University
Swati
Rajwal,
PhD - Emory University
Evaluating Large Language Models for Summarizing Long Clinical Texts and Longitudinal Patient Trajectories
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Natural Language Processing, Artificial Intelligence, Clinical Decision Support
Primary Track: Applications
Programmatic Theme: Clinical Informatics
This study evaluates the capabilities of large language models (LLMs) in summarizing multi-modal longitudinal Electronic Health Records (EHR). We investigate the impact of input modalities, timestamps, and long-context settings on LLM performance using three tasks: discharge summarization, daily progress note assessment and plan generation, and binary diagnosis prediction. Results show that while Retrieval-Augmented-Generation (RAG) improved performance, models still struggled with temporal reasoning and long-context clinical reasoning remains a challenge.
Speaker:
Maya Kruse, MS
University of Colorado
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Natural Language Processing, Artificial Intelligence, Clinical Decision Support
Primary Track: Applications
Programmatic Theme: Clinical Informatics
This study evaluates the capabilities of large language models (LLMs) in summarizing multi-modal longitudinal Electronic Health Records (EHR). We investigate the impact of input modalities, timestamps, and long-context settings on LLM performance using three tasks: discharge summarization, daily progress note assessment and plan generation, and binary diagnosis prediction. Results show that while Retrieval-Augmented-Generation (RAG) improved performance, models still struggled with temporal reasoning and long-context clinical reasoning remains a challenge.
Speaker:
Maya Kruse, MS
University of Colorado
Maya
Kruse,
MS - University of Colorado
Identifying Genitourinary Symptoms in Medical Notes with Open-Source Large Language Models
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Natural Language Processing, Artificial Intelligence
Primary Track: Foundations
Early detection of diseases relied on the accurate identification of clinical signs and symptoms. However, most of the symptom-related data in electronic health records (EHRs) was in unstructured clinical notes, posing challenges for automated extraction. This study explored the feasibility of using the Llama 3.3-70B large language model (LLM) to extract genitourinary symptoms from clinical text and map them to corresponding ICD-10 codes. Signs and symptoms from ICD-10 code R30-R39 were included. Using clinical notes from the MTSamples database, we evaluated three prompt-engineering strategies with an assumption-free constraints approach achieving the best performance. This method achieved an F1-score of 0.92 for symptom and sign extraction and 0.89 for ICD-10 code mapping. These findings showed the potential of using locally deployed LLMs to improve clinical text processing while maintaining data privacy.
Speaker:
Yunbing Bai, MS
University of Utah
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Natural Language Processing, Artificial Intelligence
Primary Track: Foundations
Early detection of diseases relied on the accurate identification of clinical signs and symptoms. However, most of the symptom-related data in electronic health records (EHRs) was in unstructured clinical notes, posing challenges for automated extraction. This study explored the feasibility of using the Llama 3.3-70B large language model (LLM) to extract genitourinary symptoms from clinical text and map them to corresponding ICD-10 codes. Signs and symptoms from ICD-10 code R30-R39 were included. Using clinical notes from the MTSamples database, we evaluated three prompt-engineering strategies with an assumption-free constraints approach achieving the best performance. This method achieved an F1-score of 0.92 for symptom and sign extraction and 0.89 for ICD-10 code mapping. These findings showed the potential of using locally deployed LLMs to improve clinical text processing while maintaining data privacy.
Speaker:
Yunbing Bai, MS
University of Utah
Yunbing
Bai,
MS - University of Utah
Detecting PTSD in Clinical Interviews: A Comparative Analysis of NLP Methods and Large Language Models
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Natural Language Processing, Artificial Intelligence, Deep Learning, Diagnostic Systems
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
Post-Traumatic Stress Disorder (PTSD) remains underdiagnosed in clinical settings, presenting opportunities for automated detection to identify patients. This study evaluates natural language processing approaches for detecting PTSD from clinical interview transcripts. We compared general and mental health-specific transformer models (BERT/RoBERTa), embedding-based methods (SentenceBERT/LLaMA), and large language model prompting strategies (zero-shot/few-shot/chain-of-thought) using the DAIC-WOZ dataset. Domain-specific models significantly outperformed general models (Mental-RoBERTa F1=0.643 vs. RoBERTa-base 0.485). LLaMA embeddings with neural networks achieved the highest performance (F1=0.700). Zero-shot prompting using DSM-5 criteria yielded competitive results without training data (F1=0.657). Performance varied significantly across symptom severity and comorbidity status, with higher accuracy for severe PTSD cases and patients with comorbid depression. Our findings highlight the potential of domain-adapted embeddings and LLMs for scalable screening while underscoring the need for improved detection of nuanced presentations and offering insights for developing clinically viable AI tools for PTSD assessment.
Speaker:
Trevor Cohen, MBChB, PhD
Biomedical Informatics and Medical Education, University of Washington
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Natural Language Processing, Artificial Intelligence, Deep Learning, Diagnostic Systems
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
Post-Traumatic Stress Disorder (PTSD) remains underdiagnosed in clinical settings, presenting opportunities for automated detection to identify patients. This study evaluates natural language processing approaches for detecting PTSD from clinical interview transcripts. We compared general and mental health-specific transformer models (BERT/RoBERTa), embedding-based methods (SentenceBERT/LLaMA), and large language model prompting strategies (zero-shot/few-shot/chain-of-thought) using the DAIC-WOZ dataset. Domain-specific models significantly outperformed general models (Mental-RoBERTa F1=0.643 vs. RoBERTa-base 0.485). LLaMA embeddings with neural networks achieved the highest performance (F1=0.700). Zero-shot prompting using DSM-5 criteria yielded competitive results without training data (F1=0.657). Performance varied significantly across symptom severity and comorbidity status, with higher accuracy for severe PTSD cases and patients with comorbid depression. Our findings highlight the potential of domain-adapted embeddings and LLMs for scalable screening while underscoring the need for improved detection of nuanced presentations and offering insights for developing clinically viable AI tools for PTSD assessment.
Speaker:
Trevor Cohen, MBChB, PhD
Biomedical Informatics and Medical Education, University of Washington
Trevor
Cohen,
MBChB, PhD - Biomedical Informatics and Medical Education, University of Washington
Using Large Language Models for Thematic Analysis of Cognitive Concerns in Subjective Cognitive Decline: An EHR-Based Study
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Natural Language Processing, Data Standards, Artificial Intelligence
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
Subjective cognitive decline (SCD) often represents the earliest identifiable stage of cognitive impairment. We conducted a retrospective study using electronic health records (EHRs) and large language models (LLMs) to categorize cognitive concerns in patients aged ≥65 with SCD. Our findings highlight the prevalence of specific types of concerns, particularly memory issues, and demonstrate the potential of LLMs for scalable, nuanced phenotyping of early cognitive symptoms to support targeted screening and intervention.
Speaker:
Liqin Wang, PhD
Brigham and Women's Hospital
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Natural Language Processing, Data Standards, Artificial Intelligence
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
Subjective cognitive decline (SCD) often represents the earliest identifiable stage of cognitive impairment. We conducted a retrospective study using electronic health records (EHRs) and large language models (LLMs) to categorize cognitive concerns in patients aged ≥65 with SCD. Our findings highlight the prevalence of specific types of concerns, particularly memory issues, and demonstrate the potential of LLMs for scalable, nuanced phenotyping of early cognitive symptoms to support targeted screening and intervention.
Speaker:
Liqin Wang, PhD
Brigham and Women's Hospital
Liqin
Wang,
PhD - Brigham and Women's Hospital
LLM-Match: An Open-Sourced Patient Matching Model Based on Large Language Models and Retrieval-Augmented Generation
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Natural Language Processing, Artificial Intelligence
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
Patient matching is the process of linking patients to appropriate clinical trials by accurately identifying and matching their medical records with trial eligibility criteria. We propose LLM-Match, a novel framework for patient matching leveraging fine-tuned open-source large language models. Our approach consists of four key components. First, a retrieval-augmented generation (RAG) module extracts relevant patient context from a vast pool of electronic health records (EHRs). Second, a prompt generation module constructs input prompts by integrating trial eligibility criteria (both inclusion and exclusion criteria), patient context, and system instructions. Third, a fine-tuning module with a classification head optimizes the model parameters using structured prompts and ground-truth labels. Fourth, an evaluation module assesses the fine-tuned model's performance on the testing datasets. We evaluated LLM-Match on four open datasets—n2c2, SIGIR, TREC 2021, and TREC 2022—using open-source models, comparing it against TrialGPT, Zero-Shot, and GPT-4-based closed models. LLM-Match outperformed all baselines.
Speaker:
Xiaodi Li, Ph.D.
Mayo Clinic
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Natural Language Processing, Artificial Intelligence
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
Patient matching is the process of linking patients to appropriate clinical trials by accurately identifying and matching their medical records with trial eligibility criteria. We propose LLM-Match, a novel framework for patient matching leveraging fine-tuned open-source large language models. Our approach consists of four key components. First, a retrieval-augmented generation (RAG) module extracts relevant patient context from a vast pool of electronic health records (EHRs). Second, a prompt generation module constructs input prompts by integrating trial eligibility criteria (both inclusion and exclusion criteria), patient context, and system instructions. Third, a fine-tuning module with a classification head optimizes the model parameters using structured prompts and ground-truth labels. Fourth, an evaluation module assesses the fine-tuned model's performance on the testing datasets. We evaluated LLM-Match on four open datasets—n2c2, SIGIR, TREC 2021, and TREC 2022—using open-source models, comparing it against TrialGPT, Zero-Shot, and GPT-4-based closed models. LLM-Match outperformed all baselines.
Speaker:
Xiaodi Li, Ph.D.
Mayo Clinic
Xiaodi
Li,
Ph.D. - Mayo Clinic
Enhancing Clinical Prediction Models through LLM-Agent
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Natural Language Processing, Artificial Intelligence
Primary Track: Applications
This study presents an LLM-agent framework to enhance clinical prediction models for in-hospital mortality in heart failure and COPD patients. By integrating GPT-4 into traditional ML pipelines, the system improves hyperparameter tuning and post-hoc prediction refinement. Using MIMIC-IV data, AUC performance improved across four ML models, most notably with XGBoost. Results demonstrate that LLM-agents can boost model accuracy, interpretability, and scalability, offering promising support for clinical risk stratification and decision-making.
Speaker:
Mohammed Al-Garadi, PhD
VUMC
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Natural Language Processing, Artificial Intelligence
Primary Track: Applications
This study presents an LLM-agent framework to enhance clinical prediction models for in-hospital mortality in heart failure and COPD patients. By integrating GPT-4 into traditional ML pipelines, the system improves hyperparameter tuning and post-hoc prediction refinement. Using MIMIC-IV data, AUC performance improved across four ML models, most notably with XGBoost. Results demonstrate that LLM-agents can boost model accuracy, interpretability, and scalability, offering promising support for clinical risk stratification and decision-making.
Speaker:
Mohammed Al-Garadi, PhD
VUMC
Mohammed
Al-Garadi,
PhD - VUMC
Developing an LLM-based conversational agent for Primary-Care Pre-visit Planning (PCP-Bot)
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Natural Language Processing, Documentation Burden, Workflow, Clinical Decision Support, Healthcare Quality, Patient Engagement and Preferences, User-centered Design Methods
Primary Track: Applications
Programmatic Theme: Clinical Informatics
Pre-visit planning is critical for optimizing clinical workflows, yet traditional over-structured questionnaires burden patients and providers. We developed PCP-Bot, an LLM-based conversational assistant using GPT-4o to collect and summarize patient histories into concise, structured reports. PCP-Bot demonstrated high coherence, readability, and ease of use with minimal hallucinations through simulated clinical scenarios and physician evaluations. Results indicate strong physician acceptance, suggesting the potential to enhance efficiency and patient care in primary care settings overall.
Speaker:
Amogh Ananda Rao, MBBS, MS
University of Pennsylvania
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Natural Language Processing, Documentation Burden, Workflow, Clinical Decision Support, Healthcare Quality, Patient Engagement and Preferences, User-centered Design Methods
Primary Track: Applications
Programmatic Theme: Clinical Informatics
Pre-visit planning is critical for optimizing clinical workflows, yet traditional over-structured questionnaires burden patients and providers. We developed PCP-Bot, an LLM-based conversational assistant using GPT-4o to collect and summarize patient histories into concise, structured reports. PCP-Bot demonstrated high coherence, readability, and ease of use with minimal hallucinations through simulated clinical scenarios and physician evaluations. Results indicate strong physician acceptance, suggesting the potential to enhance efficiency and patient care in primary care settings overall.
Speaker:
Amogh Ananda Rao, MBBS, MS
University of Pennsylvania
Amogh
Ananda Rao,
MBBS, MS - University of Pennsylvania
Evaluating LLM-based reranking method for medication term normalization
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Information Retrieval, Evaluation
Primary Track: Applications
Programmatic Theme: Clinical Informatics
This study evaluates the use of large language models (LLMs) for selecting accurate RXCUI codes from IMO Precision Normalize search results via retrieval-augmented generation. By testing three configurations of Normalize1 and reranking top candidates using GPT-4o and GPT-4o-mini, we assessed performance on 436 medication terms. GPT-4o improved RXCUI selection accuracy by 12.4% over IMO Precision Normalize alone. These findings highlight the potential of LLMs to enhance medication normalization and reduce manual review in clinical workflows.
Speaker:
Tavleen Singh, PhD
IMO Health
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Information Retrieval, Evaluation
Primary Track: Applications
Programmatic Theme: Clinical Informatics
This study evaluates the use of large language models (LLMs) for selecting accurate RXCUI codes from IMO Precision Normalize search results via retrieval-augmented generation. By testing three configurations of Normalize1 and reranking top candidates using GPT-4o and GPT-4o-mini, we assessed performance on 436 medication terms. GPT-4o improved RXCUI selection accuracy by 12.4% over IMO Precision Normalize alone. These findings highlight the potential of LLMs to enhance medication normalization and reduce manual review in clinical workflows.
Speaker:
Tavleen Singh, PhD
IMO Health
Tavleen
Singh,
PhD - IMO Health
Multi-Agent-Based Automated Clinical Data Extraction and Validation from Breast Cancer Pathology Reports Using Llama 3 Language Models
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Information Extraction, Artificial Intelligence
Primary Track: Applications
This study assesses Llama 3.1-70B and Llama 3.3-70B models for automated extraction of clinical data from 1,200 breast cancer pathology reports. Using a multi-agent system involving Extraction and Validation agents with iterative cycles and majority voting, the models achieved ~98% accuracy in extracting tumor size and site under zero-shot conditions. Ongoing analyses aim to validate extraction of metastasis, lymphatic invasion, and pTNM stage.
Speaker:
Sunghyeon Park, MA
The Catholic University of Korea
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Large Language Models (LLMs), Information Extraction, Artificial Intelligence
Primary Track: Applications
This study assesses Llama 3.1-70B and Llama 3.3-70B models for automated extraction of clinical data from 1,200 breast cancer pathology reports. Using a multi-agent system involving Extraction and Validation agents with iterative cycles and majority voting, the models achieved ~98% accuracy in extracting tumor size and site under zero-shot conditions. Ongoing analyses aim to validate extraction of metastasis, lymphatic invasion, and pTNM stage.
Speaker:
Sunghyeon Park, MA
The Catholic University of Korea
Sunghyeon
Park,
MA - The Catholic University of Korea
Exploring Variables Through CIPHER Data Dictionaries
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Knowledge Representation and Information Modeling, Phenomics and Phenome-wide Association Studies, Data Standards
Primary Track: Foundations
Programmatic Theme: Clinical Research Informatics
Phenotype libraries expedite research using electronic health records data by allowing reuse of curated definitions.
Variables from publicly available datasets and surveys are also used to define phenotypes, however these variables may be difficult to find and cite. The Centralized Interactive Phenomics Resource (CIPHER) phenotype library
developed a data dictionary interface to enable browsing of variable descriptions and referencing of variables in phenotype definitions using the Million Veteran Program biobank as a test case.
Speaker:
Michael Murray, MS
VA Boston Healthcare System
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Knowledge Representation and Information Modeling, Phenomics and Phenome-wide Association Studies, Data Standards
Primary Track: Foundations
Programmatic Theme: Clinical Research Informatics
Phenotype libraries expedite research using electronic health records data by allowing reuse of curated definitions.
Variables from publicly available datasets and surveys are also used to define phenotypes, however these variables may be difficult to find and cite. The Centralized Interactive Phenomics Resource (CIPHER) phenotype library
developed a data dictionary interface to enable browsing of variable descriptions and referencing of variables in phenotype definitions using the Million Veteran Program biobank as a test case.
Speaker:
Michael Murray, MS
VA Boston Healthcare System
Michael
Murray,
MS - VA Boston Healthcare System
A novel approach to combine structured and narrative data annotation to improve EHR navigation: a demonstration study using OMOP.
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Knowledge Representation and Information Modeling, Informatics Implementation, Data Standards
Primary Track: Applications
Programmatic Theme: Clinical Informatics
Although electronic health records (EHRs) are capable of linking structured and unstructured data at the
encounter or patient level, a fully structured EHR that can significantly improve data synthesis is absent. We have
previously designed a semantically oriented EHR navigation dashboard but encountered challenges to conduct a
usability evaluation in real-world settings due to laborious manual annotations of structured and narrative data. We therefore developed a user-focused tool to facilitate these annotations using data stored in an Observational Medical Outcomes Partnership (OMOP) database. Time to annotate the concepts of an inpatient encounter went from 14 hours with a manual approach using a custom form to 3 hours and 42 minutes using our novel editor. We describe our editor and demonstrate its use with a clinical use case using real patient data to populate our navigation dashboard as a preparation for a future usability evaluation in real clinical settings.
Speaker:
Nicholas Timkovich, MSHI
UAB Health System
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Knowledge Representation and Information Modeling, Informatics Implementation, Data Standards
Primary Track: Applications
Programmatic Theme: Clinical Informatics
Although electronic health records (EHRs) are capable of linking structured and unstructured data at the
encounter or patient level, a fully structured EHR that can significantly improve data synthesis is absent. We have
previously designed a semantically oriented EHR navigation dashboard but encountered challenges to conduct a
usability evaluation in real-world settings due to laborious manual annotations of structured and narrative data. We therefore developed a user-focused tool to facilitate these annotations using data stored in an Observational Medical Outcomes Partnership (OMOP) database. Time to annotate the concepts of an inpatient encounter went from 14 hours with a manual approach using a custom form to 3 hours and 42 minutes using our novel editor. We describe our editor and demonstrate its use with a clinical use case using real patient data to populate our navigation dashboard as a preparation for a future usability evaluation in real clinical settings.
Speaker:
Nicholas Timkovich, MSHI
UAB Health System
Nicholas
Timkovich,
MSHI - UAB Health System
Evaluation of Semantic Models for Representing Biospecimen and Data Sharing Permissions in Biobank Consent Forms
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Knowledge Representation and Information Modeling, Data Sharing, Large Language Models (LLMs)
Primary Track: Foundations
Programmatic Theme: Clinical Informatics
Concern for potential misuse of biospecimens and genomic data contributes to donation hesitancy. While consent forms delimit biospecimen and data use agreements, missingness of permissions metadata complicates large-scale stewardship. Therefore, we will evaluate two semantic models, HL7 FHIR and W3C ODRL, using 20 biobank consent forms to assess their effectiveness in making biospecimen sharing permissions machine-readable towards engendering trust between consenting contributors and investigators seeking permission.
Speaker:
Taylor Harrison, MS, MBS
Mayo Clinic
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Knowledge Representation and Information Modeling, Data Sharing, Large Language Models (LLMs)
Primary Track: Foundations
Programmatic Theme: Clinical Informatics
Concern for potential misuse of biospecimens and genomic data contributes to donation hesitancy. While consent forms delimit biospecimen and data use agreements, missingness of permissions metadata complicates large-scale stewardship. Therefore, we will evaluate two semantic models, HL7 FHIR and W3C ODRL, using 20 biobank consent forms to assess their effectiveness in making biospecimen sharing permissions machine-readable towards engendering trust between consenting contributors and investigators seeking permission.
Speaker:
Taylor Harrison, MS, MBS
Mayo Clinic
Taylor
Harrison,
MS, MBS - Mayo Clinic
Knowledge Graph for Propositional Reasoning: A Multi-Case Study of Clinical Classifications Software Refined (CCSR) for ICD-10-PCS
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Knowledge Representation and Information Modeling, Controlled Terminologies, Ontologies, and Vocabularies, Interoperability and Health Information Exchange, Standards, Information Retrieval
Primary Track: Applications
Programmatic Theme: Clinical Informatics
Knowledge graph excels in handling sophisticated relationships which are integral to forming and validating propositions. We used a knowledge graph to represent the intricacy between ICD-10-PCS concepts with their compositional character codes, which enabled reasoning on the inclusion criteria for CCSR clinical categories to identifying qualified concepts. The discovered pitfalls of applying the inclusion criteria and the new rule exploration demonstrates the critical role of knowledge graph in propositional reasoning.
Speaker:
Zheng Milgrom, M.D., M.P.H.
Semedy Inc.
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Knowledge Representation and Information Modeling, Controlled Terminologies, Ontologies, and Vocabularies, Interoperability and Health Information Exchange, Standards, Information Retrieval
Primary Track: Applications
Programmatic Theme: Clinical Informatics
Knowledge graph excels in handling sophisticated relationships which are integral to forming and validating propositions. We used a knowledge graph to represent the intricacy between ICD-10-PCS concepts with their compositional character codes, which enabled reasoning on the inclusion criteria for CCSR clinical categories to identifying qualified concepts. The discovered pitfalls of applying the inclusion criteria and the new rule exploration demonstrates the critical role of knowledge graph in propositional reasoning.
Speaker:
Zheng Milgrom, M.D., M.P.H.
Semedy Inc.
Zheng
Milgrom,
M.D., M.P.H. - Semedy Inc.
Detecting Manuscripts Related to Computable Phenotypes Using a Transformer-based Language Model
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Knowledge Representation and Information Modeling, Artificial Intelligence, Phenomics and Phenome-wide Association Studies, Large Language Models (LLMs)
Primary Track: Applications
Programmatic Theme: Public Health Informatics
Identifying relevant manuscripts for phenomics knowledgebases is a complex and time-consuming task. We developed a Transformer-based language model using a fine-tuned BioBERT model to detect manuscripts related to computable phenotypes. To address BioBERT’s 512-token limit, we introduced a sliding-window method, segmenting documents into multiple segments and aggregating classification scores. Our model significantly outperformed the default approach (AUC: 0.99 vs. 0.83, Accuracy: 0.95 vs. 0.72). This method enhances automated phenotyping literature identification, improving knowledgebase development efficiency.
Speaker:
Junghoon Chae, PhD
Oak Ridge National Laboratory
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Knowledge Representation and Information Modeling, Artificial Intelligence, Phenomics and Phenome-wide Association Studies, Large Language Models (LLMs)
Primary Track: Applications
Programmatic Theme: Public Health Informatics
Identifying relevant manuscripts for phenomics knowledgebases is a complex and time-consuming task. We developed a Transformer-based language model using a fine-tuned BioBERT model to detect manuscripts related to computable phenotypes. To address BioBERT’s 512-token limit, we introduced a sliding-window method, segmenting documents into multiple segments and aggregating classification scores. Our model significantly outperformed the default approach (AUC: 0.99 vs. 0.83, Accuracy: 0.95 vs. 0.72). This method enhances automated phenotyping literature identification, improving knowledgebase development efficiency.
Speaker:
Junghoon Chae, PhD
Oak Ridge National Laboratory
Junghoon
Chae,
PhD - Oak Ridge National Laboratory
North Carolina Health Data Utility (HDU) Initiative
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Interoperability and Health Information Exchange, Population Health, Data Sharing
Primary Track: Applications
Programmatic Theme: Public Health Informatics
North Carolina is developing a health data utility (HDU) to enhance data sharing across health care, public health, and social services, while upholding strict data privacy and security standards. Expanding on NC HealthConnex, the statewide health information exchange (HIE), an HDU integrates health-related, non-clinical data to improve care coordination and address community health needs. Role-based access controls ensure necessary, secure, and standardized data exchange to support whole-person care and population health efforts.
Speaker:
Jessica Tenenbaum, PhD
Duke University
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Interoperability and Health Information Exchange, Population Health, Data Sharing
Primary Track: Applications
Programmatic Theme: Public Health Informatics
North Carolina is developing a health data utility (HDU) to enhance data sharing across health care, public health, and social services, while upholding strict data privacy and security standards. Expanding on NC HealthConnex, the statewide health information exchange (HIE), an HDU integrates health-related, non-clinical data to improve care coordination and address community health needs. Role-based access controls ensure necessary, secure, and standardized data exchange to support whole-person care and population health efforts.
Speaker:
Jessica Tenenbaum, PhD
Duke University
Jessica
Tenenbaum,
PhD - Duke University
Enhancing Dietary Data Interoperability: LLM-Assisted Ontology Expansion for Dietary Lifestyle Information
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Interoperability and Health Information Exchange, Large Language Models (LLMs), Patient / Person Generated Health Data (Patient Reported Outcomes)
Primary Track: Applications
Programmatic Theme: Consumer Health Informatics
This study explores the potential of Large Language Models (LLMs) in expanding the Dietary Lifestyle Ontology (DILON) to improve dietary data interoperability. Using in-context two-shot learning for semantic parsing and ReACT-based prompting for ontology curation, Claude 3.7 Sonnet achieved 93% accuracy in integrating new dietary concepts. The findings demonstrate the effectiveness of LLM-human collaboration in semi-automating lifelog ontology expansion, facilitating the interoperability of dietary lifelog data.
Speaker:
Hyeoneui Kim, PhD
Seoul National University
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Interoperability and Health Information Exchange, Large Language Models (LLMs), Patient / Person Generated Health Data (Patient Reported Outcomes)
Primary Track: Applications
Programmatic Theme: Consumer Health Informatics
This study explores the potential of Large Language Models (LLMs) in expanding the Dietary Lifestyle Ontology (DILON) to improve dietary data interoperability. Using in-context two-shot learning for semantic parsing and ReACT-based prompting for ontology curation, Claude 3.7 Sonnet achieved 93% accuracy in integrating new dietary concepts. The findings demonstrate the effectiveness of LLM-human collaboration in semi-automating lifelog ontology expansion, facilitating the interoperability of dietary lifelog data.
Speaker:
Hyeoneui Kim, PhD
Seoul National University
Hyeoneui
Kim,
PhD - Seoul National University
Scaling a Patient Portal Integrated Diabetes Application Using FHIR: A Multisite Experience
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Interoperability and Health Information Exchange, Data Sharing, Informatics Implementation
Primary Track: Applications
Programmatic Theme: Clinical Informatics
Introduction: We built My Diabetes Care, a patient-facing app using SMART on FHIR for cross-EHR interoperability. Methods: We used the Consolidated Framework for Implementation Research to study facilitators and barriers to implementation across two organizations. Results: Key themes include differences in user authentication (intervention), focus on 21st Century Cures Act (outer setting), competing priorities (inner setting), and limited experience with FHIR (individuals, process). Conclusions: Addressing barriers can optimize digital solutions, improving outcomes and reducing disparities.
Speaker:
Nicolás Prada-Rey, MA
Brigham and Women's Hospital
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Interoperability and Health Information Exchange, Data Sharing, Informatics Implementation
Primary Track: Applications
Programmatic Theme: Clinical Informatics
Introduction: We built My Diabetes Care, a patient-facing app using SMART on FHIR for cross-EHR interoperability. Methods: We used the Consolidated Framework for Implementation Research to study facilitators and barriers to implementation across two organizations. Results: Key themes include differences in user authentication (intervention), focus on 21st Century Cures Act (outer setting), competing priorities (inner setting), and limited experience with FHIR (individuals, process). Conclusions: Addressing barriers can optimize digital solutions, improving outcomes and reducing disparities.
Speaker:
Nicolás Prada-Rey, MA
Brigham and Women's Hospital
Nicolás
Prada-Rey,
MA - Brigham and Women's Hospital
Do EHRs help or hinder coordination: Examining facilitators and barriers of EHR to CRC Screening Process
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Interoperability and Health Information Exchange, Cancer Prevention, Workflow
Primary Track: Applications
Programmatic Theme: Clinical Informatics
Electronic health records (EHR) and patient portals play a crucial role in improving colorectal cancer screening workflows. This study explores EHR-related facilitators and barriers by conducting and analyzing semi-structured interviews with clinicians and patients. Findings highlight key EHR functionalities that support screening coordination, including Epic Chat and pending orders, while digital health literacy and interoperability issues hinder effectiveness. Addressing these barriers can enhance screening effectiveness and lead to better patient outcomes.
Speaker:
Miad Alfaqih, Phd
University of Florida
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Interoperability and Health Information Exchange, Cancer Prevention, Workflow
Primary Track: Applications
Programmatic Theme: Clinical Informatics
Electronic health records (EHR) and patient portals play a crucial role in improving colorectal cancer screening workflows. This study explores EHR-related facilitators and barriers by conducting and analyzing semi-structured interviews with clinicians and patients. Findings highlight key EHR functionalities that support screening coordination, including Epic Chat and pending orders, while digital health literacy and interoperability issues hinder effectiveness. Addressing these barriers can enhance screening effectiveness and lead to better patient outcomes.
Speaker:
Miad Alfaqih, Phd
University of Florida
Miad
Alfaqih,
Phd - University of Florida
Time-aware Dimension Reduction for Exploring Trends in Literature
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Information Visualization, Natural Language Processing, Machine Learning
Primary Track: Foundations
With a million scientific papers yearly, methods for organizing and exploring research are more important than ever. Though semantic embeddings combined with manifold learning algorithms like t-SNE produce useful maps, these ignore publication date, which is crucial for understanding research progression. As a remedy, we develop a time-aware dimension reduction algorithm that groups similar papers while directly encoding date using radius in polar plots, revealing trends that would not be visible with standard t-SNE-based maps.
Speaker:
Brian Ondov, PhD
Yale School of Medicine
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Information Visualization, Natural Language Processing, Machine Learning
Primary Track: Foundations
With a million scientific papers yearly, methods for organizing and exploring research are more important than ever. Though semantic embeddings combined with manifold learning algorithms like t-SNE produce useful maps, these ignore publication date, which is crucial for understanding research progression. As a remedy, we develop a time-aware dimension reduction algorithm that groups similar papers while directly encoding date using radius in polar plots, revealing trends that would not be visible with standard t-SNE-based maps.
Speaker:
Brian Ondov, PhD
Yale School of Medicine
Brian
Ondov,
PhD - Yale School of Medicine
Longitudinal Patient Journey Mapping to Inform the Design of Technology to Support Breast Cancer Patients
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Human-computer Interaction, Workflow, Information Visualization, Usability
Primary Track: Foundations
Programmatic Theme: Consumer Health Informatics
This study applies a SEIPS framework-based patient journey mapping technique to analyze the interactions between breast cancer patients and the healthcare system, focusing on technology use. Researchers observed 12 patients, 3 longitudinally, coding for macrocognitive processes, barriers, and facilitators. These maps revealed how technology use intersected with clinical processes across stakeholders. This novel mapping approach offers insights into patient experiences and digital processes, informing future system improvements to address information gaps in cancer care delivery.
Speaker:
Uday Suresh, MS
Vanderbilt University Department of Biomedical Informatics
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Human-computer Interaction, Workflow, Information Visualization, Usability
Primary Track: Foundations
Programmatic Theme: Consumer Health Informatics
This study applies a SEIPS framework-based patient journey mapping technique to analyze the interactions between breast cancer patients and the healthcare system, focusing on technology use. Researchers observed 12 patients, 3 longitudinally, coding for macrocognitive processes, barriers, and facilitators. These maps revealed how technology use intersected with clinical processes across stakeholders. This novel mapping approach offers insights into patient experiences and digital processes, informing future system improvements to address information gaps in cancer care delivery.
Speaker:
Uday Suresh, MS
Vanderbilt University Department of Biomedical Informatics
Uday
Suresh,
MS - Vanderbilt University Department of Biomedical Informatics
Eye Tracking to Measure Cognitive Workload During EHR Use
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Human-computer Interaction, Usability, Clinical Decision Support, Evaluation
Primary Track: Applications
Programmatic Theme: Clinical Informatics
We used eye tracking to measure cognitive workload while using EHR. Participants (N=17) underwent 3 simulated tasks related to prescribing antibiotics. Eye-tracking data indicated 85 area of interests (AOI). The top 20% of AOIs held 80% of the attention. We also identified differences across scenarios and participants. Eye tracking is a feasible way to measure cognitive workload. AI applications could personalize EHR, reduce cognitive load associated with common tasks and provide suggestions on automated entry.
Speaker:
Mustafa Ozkaynak, PhD
University of Colorado-Denver | Anschutz Medical Campus
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Human-computer Interaction, Usability, Clinical Decision Support, Evaluation
Primary Track: Applications
Programmatic Theme: Clinical Informatics
We used eye tracking to measure cognitive workload while using EHR. Participants (N=17) underwent 3 simulated tasks related to prescribing antibiotics. Eye-tracking data indicated 85 area of interests (AOI). The top 20% of AOIs held 80% of the attention. We also identified differences across scenarios and participants. Eye tracking is a feasible way to measure cognitive workload. AI applications could personalize EHR, reduce cognitive load associated with common tasks and provide suggestions on automated entry.
Speaker:
Mustafa Ozkaynak, PhD
University of Colorado-Denver | Anschutz Medical Campus
Mustafa
Ozkaynak,
PhD - University of Colorado-Denver | Anschutz Medical Campus
Second Signs and Wrong Patient Orders–a Natural Experiment
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Healthcare Quality, Quantitative Methods, Clinical Decision Support
Primary Track: Applications
We analyzed data from our institution to determine whether the implementation of a “re-enter signature” workflow for opioid orders led to a reduction in wrong patient orders as measured by retract and resubmission rates. In an interrupted time series analysis, we found that implementation of the 2nd order verification had no significant impact on retraction and resubmission rates for opioids orders, either on their own or in comparison to the control of non-opioid orders.
Speaker:
Sophia Hsu, High School Diploma
University of California, San Francisco
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Healthcare Quality, Quantitative Methods, Clinical Decision Support
Primary Track: Applications
We analyzed data from our institution to determine whether the implementation of a “re-enter signature” workflow for opioid orders led to a reduction in wrong patient orders as measured by retract and resubmission rates. In an interrupted time series analysis, we found that implementation of the 2nd order verification had no significant impact on retraction and resubmission rates for opioids orders, either on their own or in comparison to the control of non-opioid orders.
Speaker:
Sophia Hsu, High School Diploma
University of California, San Francisco
Sophia
Hsu,
High School Diploma - University of California, San Francisco
Addressing healthcare disparities for Arab immigrants and refugees in the U.S.
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Healthcare Quality, Qualitative Methods, Patient Engagement and Preferences
Primary Track: Foundations
Programmatic Theme: Clinical Informatics
Every year, countless Arab immigrants and refugees arrive in the United States, seeking safety, stability, and a better future. From navigating complex insurance policies to overcoming cultural and linguistic barriers, they often struggle to access the care they need. The Galileo Project aims to design the perfect health system, by identifying gaps in current healthcare policies, enhancing cultural competency among providers, and leveraging technology to bridge communication barriers.
Speaker:
Carine Yehya, MMCi
Duke University School of Medicine
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Healthcare Quality, Qualitative Methods, Patient Engagement and Preferences
Primary Track: Foundations
Programmatic Theme: Clinical Informatics
Every year, countless Arab immigrants and refugees arrive in the United States, seeking safety, stability, and a better future. From navigating complex insurance policies to overcoming cultural and linguistic barriers, they often struggle to access the care they need. The Galileo Project aims to design the perfect health system, by identifying gaps in current healthcare policies, enhancing cultural competency among providers, and leveraging technology to bridge communication barriers.
Speaker:
Carine Yehya, MMCi
Duke University School of Medicine
Carine
Yehya,
MMCi - Duke University School of Medicine
Building a Healthcare Data Science Platform in the Cloud
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Healthcare Quality, Privacy and Security, Machine Learning
Primary Track: Applications
Programmatic Theme: Clinical Informatics
This case study details the Geisinger AI Lab's cloud-based data science platform, designed for portability, scalability, and security. By using open-source tools and serverless architectures, the platform ensures flexibility, efficient resource use, and HIPAA compliance. We offer strategies and best practices for leveraging cloud computing in healthcare data science.
Speaker:
Grant DeLong, BA
Geisinger
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Healthcare Quality, Privacy and Security, Machine Learning
Primary Track: Applications
Programmatic Theme: Clinical Informatics
This case study details the Geisinger AI Lab's cloud-based data science platform, designed for portability, scalability, and security. By using open-source tools and serverless architectures, the platform ensures flexibility, efficient resource use, and HIPAA compliance. We offer strategies and best practices for leveraging cloud computing in healthcare data science.
Speaker:
Grant DeLong, BA
Geisinger
Grant
DeLong,
BA - Geisinger
Examining the Impact of Healthcare Team Effectiveness on Mortality Outcomes of Heart Failure Patients Using EHR Note Social Network Analysis
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Healthcare Quality, Patient Safety, Quantitative Methods
Primary Track: Foundations
Programmatic Theme: Clinical Informatics
High-quality teamwork has been established as a major contributor to patient safety, but resource-intensive measurement approaches present a barrier to evaluating teamwork interventions, particularly in cross-institutional settings. To measure teamwork, we adapted an algorithm to unobtrusively measure aspects of teamwork using electronic health record note metadata. Analyses indicated that increased shared team experience is associated with lower ICU mortality in patients with heart failure, after controlling for clinical and demographic characteristics.
Speaker:
Claire Layton, MS
University of Florida
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Healthcare Quality, Patient Safety, Quantitative Methods
Primary Track: Foundations
Programmatic Theme: Clinical Informatics
High-quality teamwork has been established as a major contributor to patient safety, but resource-intensive measurement approaches present a barrier to evaluating teamwork interventions, particularly in cross-institutional settings. To measure teamwork, we adapted an algorithm to unobtrusively measure aspects of teamwork using electronic health record note metadata. Analyses indicated that increased shared team experience is associated with lower ICU mortality in patients with heart failure, after controlling for clinical and demographic characteristics.
Speaker:
Claire Layton, MS
University of Florida
Claire
Layton,
MS - University of Florida
Adaptation of IHI Triple Aim Framework to Reflect Need for Improved Documentation in Adverse Contrast Reactions
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Healthcare Quality, Patient Engagement and Preferences, Patient Safety
Primary Track: Foundations
Programmatic Theme: Clinical Informatics
Documentation of adverse contrast reactions (ACR’s) in radiology departments is non-standardized and sometimes subpar or incomplete. ACR’s have been a rising concern with the ever-increasing number of medical imaging especially in the setting of cancer diagnosis and treatment. Quality documentation of these events has been shown to improve management programs for patients undergoing additional imaging. We created a framework that promotes consistent better-quality adverse event documentation.
Speaker:
Evan Ratkus, Bachelor of Science
University of South Florida
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Healthcare Quality, Patient Engagement and Preferences, Patient Safety
Primary Track: Foundations
Programmatic Theme: Clinical Informatics
Documentation of adverse contrast reactions (ACR’s) in radiology departments is non-standardized and sometimes subpar or incomplete. ACR’s have been a rising concern with the ever-increasing number of medical imaging especially in the setting of cancer diagnosis and treatment. Quality documentation of these events has been shown to improve management programs for patients undergoing additional imaging. We created a framework that promotes consistent better-quality adverse event documentation.
Speaker:
Evan Ratkus, Bachelor of Science
University of South Florida
Evan
Ratkus,
Bachelor of Science - University of South Florida
Digital Voices: A Comprehensive Analysis of Google Reviews in Rural High- and Low-Health Ranked Counties
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Healthcare Quality, Natural Language Processing, Social Media and Connected Health, Data Mining
Primary Track: Applications
Programmatic Theme: Public Health Informatics
This study investigates the correlation between county health rankings and Google review scores of rural Eastern North Carolina healthcare facilities. Using county health status and extracted Google review data, the research compares average review scores between high- and low-ranked counties. Preliminary results suggest potential disparities in perceived healthcare quality. Further analysis will employ natural language processing to understand patient sentiment and inform strategies for improving healthcare experiences in underserved areas.
Speaker:
Carlos Perez-Aldana, PhD
ECU
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Healthcare Quality, Natural Language Processing, Social Media and Connected Health, Data Mining
Primary Track: Applications
Programmatic Theme: Public Health Informatics
This study investigates the correlation between county health rankings and Google review scores of rural Eastern North Carolina healthcare facilities. Using county health status and extracted Google review data, the research compares average review scores between high- and low-ranked counties. Preliminary results suggest potential disparities in perceived healthcare quality. Further analysis will employ natural language processing to understand patient sentiment and inform strategies for improving healthcare experiences in underserved areas.
Speaker:
Carlos Perez-Aldana, PhD
ECU
Carlos
Perez-Aldana,
PhD - ECU
Assessing the Impact of Race, Income, and Air Quality on Hospitalization Risk in Pediatric Asthma Patients: A Retrospective Observational Cohort Study
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Health Equity and Social Determinants of Health (SDoH), Health Equity, Information Retrieval, Population Health, Pediatrics
Primary Track: Applications
Programmatic Theme: Clinical Informatics
This study provides insights into the interplay between race, income, air quality, and clinical factors in pediatric asthma hospitalizations. The principal finding of this study is that patients with previous inpatient admissions, Black patients, younger patients, patients residing in areas with lower air quality and lower income have a higher risk of follow-up inpatient admission for asthma within one year. Our results indicate potential gaps and the need for better interventions among vulnerable demographic groups.
Speaker:
Qingrui Wang, BS
JHU
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Health Equity and Social Determinants of Health (SDoH), Health Equity, Information Retrieval, Population Health, Pediatrics
Primary Track: Applications
Programmatic Theme: Clinical Informatics
This study provides insights into the interplay between race, income, air quality, and clinical factors in pediatric asthma hospitalizations. The principal finding of this study is that patients with previous inpatient admissions, Black patients, younger patients, patients residing in areas with lower air quality and lower income have a higher risk of follow-up inpatient admission for asthma within one year. Our results indicate potential gaps and the need for better interventions among vulnerable demographic groups.
Speaker:
Qingrui Wang, BS
JHU
Qingrui
Wang,
BS - JHU
Quantifying the Effects of Social Determinants of Health on Adverse Birth Outcomes in Louisiana using Bayesian Linear Mixed-Effects Models
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Health Equity and Social Determinants of Health (SDoH), Geospatial (GIS) Data/Analysis, Quantitative Methods, Public Health, Racial disparities
Primary Track: Applications
Programmatic Theme: Public Health Informatics
Louisiana is the lowest ranking states in the United States of America in terms of maternal health outcomes. Previous works have highlighted the impact of social determinants of health on the incidence of adverse birth outcomes. That said, we are, to the best of our knowledge, the first to applied Bayesian linear mixed-effects models to this problem. Such models allow us to make more certain conclusions about the effects of social determinants of health on maternal health outcomes. Our models had excellent convergence and performance metrics. From these models we identified several significant effects on the incidence of adverse birth outcomes and explained a large amount of the variance in the rates of adverse birth outcomes. Some of these include Black and/or African American Population and days above 105°F. We also identified several clusters of autocorrelated parishes which seem to affect each other’s incidence of adverse birth outcomes.
Speaker:
José Irizarry Ayala, M.S.
Tulane University School of Medicine
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Health Equity and Social Determinants of Health (SDoH), Geospatial (GIS) Data/Analysis, Quantitative Methods, Public Health, Racial disparities
Primary Track: Applications
Programmatic Theme: Public Health Informatics
Louisiana is the lowest ranking states in the United States of America in terms of maternal health outcomes. Previous works have highlighted the impact of social determinants of health on the incidence of adverse birth outcomes. That said, we are, to the best of our knowledge, the first to applied Bayesian linear mixed-effects models to this problem. Such models allow us to make more certain conclusions about the effects of social determinants of health on maternal health outcomes. Our models had excellent convergence and performance metrics. From these models we identified several significant effects on the incidence of adverse birth outcomes and explained a large amount of the variance in the rates of adverse birth outcomes. Some of these include Black and/or African American Population and days above 105°F. We also identified several clusters of autocorrelated parishes which seem to affect each other’s incidence of adverse birth outcomes.
Speaker:
José Irizarry Ayala, M.S.
Tulane University School of Medicine
José
Irizarry Ayala,
M.S. - Tulane University School of Medicine
Tracking Human Mobility and Behavior During Pandemics: A GPS-Based Analysis of Movement Patterns and Social Distancing
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Geospatial (GIS) Data/Analysis, Public Health, Infectious Diseases and Epidemiology, Data Mining
Primary Track: Applications
Programmatic Theme: Public Health Informatics
This study examines mobility patterns during COVID-19 using GPS data from 196 participants, generating over 232 million data points. Data preprocessing included filtering invalid locations, noise reduction, and matching with SafeGraph POIs. A 250-meter radius was optimal for visit detection. Findings highlight key locations visited, aiding public health efforts. The study demonstrates GPS-based mobility tracking’s potential for understanding disease transmission patterns and adherence to policy.
Speaker:
Sri Surya Krishna Rama Taraka Naren Durbha, Master's in Health Informatics
George Mason University
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Geospatial (GIS) Data/Analysis, Public Health, Infectious Diseases and Epidemiology, Data Mining
Primary Track: Applications
Programmatic Theme: Public Health Informatics
This study examines mobility patterns during COVID-19 using GPS data from 196 participants, generating over 232 million data points. Data preprocessing included filtering invalid locations, noise reduction, and matching with SafeGraph POIs. A 250-meter radius was optimal for visit detection. Findings highlight key locations visited, aiding public health efforts. The study demonstrates GPS-based mobility tracking’s potential for understanding disease transmission patterns and adherence to policy.
Speaker:
Sri Surya Krishna Rama Taraka Naren Durbha, Master's in Health Informatics
George Mason University
Sri Surya Krishna Rama Taraka Naren
Durbha,
Master's in Health Informatics - George Mason University
Using Reweighting to Reduce Bias in Adverse Event Prediction Models
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Fairness and Elimination of Bias, Machine Learning, Health Equity, Artificial Intelligence, Quantitative Methods, Patient Safety
Primary Track: Applications
Programmatic Theme: Clinical Informatics
Algorithmic bias significantly impacts scientific research, particularly in biomedicine, where it may lead to harmful treatments and outcomes for historically disadvantaged groups. One approach to mitigating bias in machine learning pipelines is reweighting, which assigns different weights to training data points to enhance the representation of underrepresented individuals. In this study, we introduce a novel model that employs reweighting to reduce racial disparities in predicting adverse events. We compare its performance against a baseline model and a proprietary model currently used in hospitals. Our findings suggest that a diverse training dataset can achieve high model accuracy without requiring additional fairness interventions.
Speaker:
Amalia Ionescu, PhD
Cedars-Sinai
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Fairness and Elimination of Bias, Machine Learning, Health Equity, Artificial Intelligence, Quantitative Methods, Patient Safety
Primary Track: Applications
Programmatic Theme: Clinical Informatics
Algorithmic bias significantly impacts scientific research, particularly in biomedicine, where it may lead to harmful treatments and outcomes for historically disadvantaged groups. One approach to mitigating bias in machine learning pipelines is reweighting, which assigns different weights to training data points to enhance the representation of underrepresented individuals. In this study, we introduce a novel model that employs reweighting to reduce racial disparities in predicting adverse events. We compare its performance against a baseline model and a proprietary model currently used in hospitals. Our findings suggest that a diverse training dataset can achieve high model accuracy without requiring additional fairness interventions.
Speaker:
Amalia Ionescu, PhD
Cedars-Sinai
Amalia
Ionescu,
PhD - Cedars-Sinai
Informative Missingness in EHR Labs: A Retrospective Study of Biomarker Trajectories Following SARS-CoV-2 Infection
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Fairness and Elimination of Bias, Laboratory Systems and Reporting, Machine Learning, Real-World Evidence Generation, Infectious Diseases and Epidemiology
Primary Track: Applications
Programmatic Theme: Clinical Informatics
This study analyzes electronic health records (EHR) to track biomarker changes following SARS-CoV-2 infection, focusing on informed missingness to enhance clinical prediction models for Long COVID. Using the electronic health records data from National COVID Cohort Collaborative, it examines lymphocyte levels as predictors of Long COVID through multivariate regression. The findings suggest that including missing data deliberately can improve accuracy and clinical applicability.
Speaker:
Saaya Patel, B.S.
Emory University
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Fairness and Elimination of Bias, Laboratory Systems and Reporting, Machine Learning, Real-World Evidence Generation, Infectious Diseases and Epidemiology
Primary Track: Applications
Programmatic Theme: Clinical Informatics
This study analyzes electronic health records (EHR) to track biomarker changes following SARS-CoV-2 infection, focusing on informed missingness to enhance clinical prediction models for Long COVID. Using the electronic health records data from National COVID Cohort Collaborative, it examines lymphocyte levels as predictors of Long COVID through multivariate regression. The findings suggest that including missing data deliberately can improve accuracy and clinical applicability.
Speaker:
Saaya Patel, B.S.
Emory University
Saaya
Patel,
B.S. - Emory University
Exposure and Control matching in Electronic Health Record Studies using Grouped Marching
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Fairness and Elimination of Bias, Causal Inference, Quantitative Methods
Primary Track: Applications
Programmatic Theme: Clinical Informatics
Control selection for COVID cases is confounded by requiring a negative test. Instead, we propose using a routine care enrollment event for all patients; we match a control to each case in increasing time from enrollment to exposure, blind to future events. Negative controls, specifically pap smears and non-pathological fractures, have a reduced or non-significant difference using grouped marching to find controls.
Speaker:
Margaret Hall, MS
Emory University
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Fairness and Elimination of Bias, Causal Inference, Quantitative Methods
Primary Track: Applications
Programmatic Theme: Clinical Informatics
Control selection for COVID cases is confounded by requiring a negative test. Instead, we propose using a routine care enrollment event for all patients; we match a control to each case in increasing time from enrollment to exposure, blind to future events. Negative controls, specifically pap smears and non-pathological fractures, have a reduced or non-significant difference using grouped marching to find controls.
Speaker:
Margaret Hall, MS
Emory University
Margaret
Hall,
MS - Emory University
Predicting breast cancer survival with fairness-aware and interpretable machine learning
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Fairness and Elimination of Bias, Artificial Intelligence, Clinical Decision Support
Primary Track: Foundations
Programmatic Theme: Clinical Informatics
Ensuring fairness in survival analysis models is critical, particularly in breast cancer prognosis, where disparities exist regarding race, marital status, and socioeconomics. We propose FAIM-Surv, a fairness-aware and interpretable framework that mitigates biases in survival models while preserving predictive performance. Using the SEER database, FAIM-Surv improves both performance-disparity-based and ranking-based fairness metrics by 31.5%–48.4%. By balancing performance and fairness, this approach advances the development of fair survival models in oncology.
Speaker:
Mingxuan Liu, PhD student
Duke-NUS Medical School
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Fairness and Elimination of Bias, Artificial Intelligence, Clinical Decision Support
Primary Track: Foundations
Programmatic Theme: Clinical Informatics
Ensuring fairness in survival analysis models is critical, particularly in breast cancer prognosis, where disparities exist regarding race, marital status, and socioeconomics. We propose FAIM-Surv, a fairness-aware and interpretable framework that mitigates biases in survival models while preserving predictive performance. Using the SEER database, FAIM-Surv improves both performance-disparity-based and ranking-based fairness metrics by 31.5%–48.4%. By balancing performance and fairness, this approach advances the development of fair survival models in oncology.
Speaker:
Mingxuan Liu, PhD student
Duke-NUS Medical School
Mingxuan
Liu,
PhD student - Duke-NUS Medical School
A Framework for Evaluating and Improving Metadata Quality in the RADx Data Hub Using CEDAR Templates
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Evaluation, Workflow, Fairness and elimination of bias, Data Standards, Controlled Terminologies, Ontologies, and Vocabularies, Interoperability and Health Information Exchange, Knowledge Representation and Information Modeling, Data Sharing
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
Metadata quality is essential for FAIR data, but evaluation is time-consuming. We present a semi-automated framework that uses CEDAR templates, encoding expected metadata structure and values, to assess metadata validity, completeness, consistency, accuracy, accessibility, uniqueness, and linguistic quality. Applied in the RADx Data Hub, it identified 586 issues from 181 studies, improving metadata quality and data FAIRness. Our framework supports domain adaptability through custom CEDAR metadata templates alongside pluggable evaluation criteria.
Speaker:
Yan Cao, M.S.
Stanford Center for Biomedical Informatics Research
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Evaluation, Workflow, Fairness and elimination of bias, Data Standards, Controlled Terminologies, Ontologies, and Vocabularies, Interoperability and Health Information Exchange, Knowledge Representation and Information Modeling, Data Sharing
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
Metadata quality is essential for FAIR data, but evaluation is time-consuming. We present a semi-automated framework that uses CEDAR templates, encoding expected metadata structure and values, to assess metadata validity, completeness, consistency, accuracy, accessibility, uniqueness, and linguistic quality. Applied in the RADx Data Hub, it identified 586 issues from 181 studies, improving metadata quality and data FAIRness. Our framework supports domain adaptability through custom CEDAR metadata templates alongside pluggable evaluation criteria.
Speaker:
Yan Cao, M.S.
Stanford Center for Biomedical Informatics Research
Yan
Cao,
M.S. - Stanford Center for Biomedical Informatics Research
Interactive Clinical Utility Decision Analytic (iCUDA) Dashboard – Sensitivity Analysis of Predictive Model Performance and Clinical Utility
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Evaluation, Information Visualization, Machine Learning, Diagnostic Systems, Clinical Decision Support
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
The Interactive Clinical Utility Decision Analytic (iCUDA) dashboard is an open-source and hosted tool (URL) designed to assess a predictive model's value for local use. It enables users to quantify expected treatment harms and benefits under local disease prevalence scenarios and tradeoffs. Through interactive sliders, users can explore model cut-off points (empiric and ideal), expected utility, partial areas under the ROC curve, and prevalence and tradeoff ranges making the model useful.
Speaker:
Star Liu, M.S. in Biomedical Informatics and Data Science
Johns Hopkins School of Medicine - Biomedical Informatics and Data Science
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Evaluation, Information Visualization, Machine Learning, Diagnostic Systems, Clinical Decision Support
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
The Interactive Clinical Utility Decision Analytic (iCUDA) dashboard is an open-source and hosted tool (URL) designed to assess a predictive model's value for local use. It enables users to quantify expected treatment harms and benefits under local disease prevalence scenarios and tradeoffs. Through interactive sliders, users can explore model cut-off points (empiric and ideal), expected utility, partial areas under the ROC curve, and prevalence and tradeoff ranges making the model useful.
Speaker:
Star Liu, M.S. in Biomedical Informatics and Data Science
Johns Hopkins School of Medicine - Biomedical Informatics and Data Science
Star
Liu,
M.S. in Biomedical Informatics and Data Science - Johns Hopkins School of Medicine - Biomedical Informatics and Data Science
EHR Use and Academic Performance Among General Surgery Residents: An Exploratory Analysis Using Clustering and Audit Logs
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Education and Training, Workflow, Machine Learning
Primary Track: Applications
Programmatic Theme: Academic Informatics / LIEAF
Using a machine learning approach, electronic health record (EHR) audit logs of general surgery residents were clustered to identify workflow patterns and evaluate associations with resident academic performance. While no significant associations were found, this study serves as an introductory approach to evaluating EHR workflow efficiency and performance for surgical trainees. Future work will involve supervised machine learning approaches and expand on current methods with additional log patterns and continued refinement of resident performance scoring.
Speaker:
Catherine Pratt, MD
University of Cincinnati College of Medicine
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Education and Training, Workflow, Machine Learning
Primary Track: Applications
Programmatic Theme: Academic Informatics / LIEAF
Using a machine learning approach, electronic health record (EHR) audit logs of general surgery residents were clustered to identify workflow patterns and evaluate associations with resident academic performance. While no significant associations were found, this study serves as an introductory approach to evaluating EHR workflow efficiency and performance for surgical trainees. Future work will involve supervised machine learning approaches and expand on current methods with additional log patterns and continued refinement of resident performance scoring.
Speaker:
Catherine Pratt, MD
University of Cincinnati College of Medicine
Catherine
Pratt,
MD - University of Cincinnati College of Medicine
Leveraging Large Language Models to Develop Automatic AI System for Real-time Feedback in Healthcare Education
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Education and Training, Teaching Innovation, Large Language Models (LLMs), Usability
Primary Track: Applications
Programmatic Theme: Academic Informatics / LIEAF
Effective communication is critical for nurse-led innovation, yet nursing curricula often lack such training. This study developed an AI system using Large Language Models to provide real-time feedback on nursing students' elevator pitches. We analyzed 410 pitch scripts with GPT-3.5 and GPT-4o-mini, comparing fine-tuned and few-shot approaches. GPT-4o-mini achieved the lowest RMSE (2.81), with students rating feedback accuracy at 84–97%. Findings show LLMs can deliver objective, scalable feedback, enhancing nursing education without extensive fine-tuning.
Speaker:
Shaowei GUAN, Bachelor
The Hong Kong Polytechnic University
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Education and Training, Teaching Innovation, Large Language Models (LLMs), Usability
Primary Track: Applications
Programmatic Theme: Academic Informatics / LIEAF
Effective communication is critical for nurse-led innovation, yet nursing curricula often lack such training. This study developed an AI system using Large Language Models to provide real-time feedback on nursing students' elevator pitches. We analyzed 410 pitch scripts with GPT-3.5 and GPT-4o-mini, comparing fine-tuned and few-shot approaches. GPT-4o-mini achieved the lowest RMSE (2.81), with students rating feedback accuracy at 84–97%. Findings show LLMs can deliver objective, scalable feedback, enhancing nursing education without extensive fine-tuning.
Speaker:
Shaowei GUAN, Bachelor
The Hong Kong Polytechnic University
Shaowei
GUAN,
Bachelor - The Hong Kong Polytechnic University
Using Big Data to Assess Social and Individual Factors Associated with Pediatric Essential Hypertension: A Cosmos Database Study
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Diversity, Equity, Inclusion, and Accessibility (DEIA), Racial disparities, Health Equity, Pediatrics, Public Health
Primary Track: Policy
Programmatic Theme: Public Health Informatics
This study aims to leverage large-scale data analytics to elucidate the relationship between individual, social, and economic factors in pediatric HTN, utilizing the largest database study to date. We identified 148,581 patients with HTN and 16,973,958 control patients. Patients with HTN were more likely to be obese, live in rural areas, and be black or multiracial. Our findings identify specific risk factors for hypertension and underscore the potential for targeted mitigation efforts.
Speaker:
Zachary West, MD
Children's Healthcare of Atlanta
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Diversity, Equity, Inclusion, and Accessibility (DEIA), Racial disparities, Health Equity, Pediatrics, Public Health
Primary Track: Policy
Programmatic Theme: Public Health Informatics
This study aims to leverage large-scale data analytics to elucidate the relationship between individual, social, and economic factors in pediatric HTN, utilizing the largest database study to date. We identified 148,581 patients with HTN and 16,973,958 control patients. Patients with HTN were more likely to be obese, live in rural areas, and be black or multiracial. Our findings identify specific risk factors for hypertension and underscore the potential for targeted mitigation efforts.
Speaker:
Zachary West, MD
Children's Healthcare of Atlanta
Zachary
West,
MD - Children's Healthcare of Atlanta
Transfer Learning Enhances Outcome Prediction for Out-of-Hospital Cardiac Arrest: Validation Across Diverse Geographic Contexts
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Diversity, Equity, Inclusion, and Accessibility (DEIA), Fairness and elimination of bias, Privacy and Security, Health Equity, Artificial Intelligence, Machine Learning
Primary Track: Applications
Programmatic Theme: Clinical Informatics
Accurate risk stratification is vital in clinical care but challenging in low-resource settings due to limited data. This study applies transfer learning (TL) to adapt a neurological outcome prediction model for out-of-hospital cardiac arrest (OHCA) from a Japanese cohort to Vietnam (243 patients) and Singapore (15,916 patients). TL significantly improved performance in Vietnam (AUROC: 0.807 vs. 0.467), demonstrating its potential as a scalable solution for enhancing clinical models in data-limited settings.
Speaker:
Siqi Li, Bachelor of Science
Duke-NUS Medical School
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Diversity, Equity, Inclusion, and Accessibility (DEIA), Fairness and elimination of bias, Privacy and Security, Health Equity, Artificial Intelligence, Machine Learning
Primary Track: Applications
Programmatic Theme: Clinical Informatics
Accurate risk stratification is vital in clinical care but challenging in low-resource settings due to limited data. This study applies transfer learning (TL) to adapt a neurological outcome prediction model for out-of-hospital cardiac arrest (OHCA) from a Japanese cohort to Vietnam (243 patients) and Singapore (15,916 patients). TL significantly improved performance in Vietnam (AUROC: 0.807 vs. 0.467), demonstrating its potential as a scalable solution for enhancing clinical models in data-limited settings.
Speaker:
Siqi Li, Bachelor of Science
Duke-NUS Medical School
Siqi
Li,
Bachelor of Science - Duke-NUS Medical School
Identifying Pulmonary Embolism from Radiology Reports Using Large Language Models
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Deep Learning, Natural Language Processing, Artificial Intelligence, Large Language Models (LLMs), Data Mining, Critical Care, Data Standards, Clinical Decision Support
Primary Track: Applications
Programmatic Theme: Clinical Informatics
Pulmonary embolism (PE) detection in radiology reports is critical for timely care. We evaluate Llama 3.1 8B and GPT-4o large language models using zero-shot, few-shot, and fine-tuned strategies on the MIMIC-IV-Ext-PE dataset. GPT-4o achieved 99.3% accuracy and 95.75% F1-score, outperforming all Llama variants. Our results highlight the potential of LLMs to automate PE identification, reducing reliance on manual chart review.
Speaker:
Farzad Ahmed, Ph.D. Student in Computer Science
George Mason University
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Deep Learning, Natural Language Processing, Artificial Intelligence, Large Language Models (LLMs), Data Mining, Critical Care, Data Standards, Clinical Decision Support
Primary Track: Applications
Programmatic Theme: Clinical Informatics
Pulmonary embolism (PE) detection in radiology reports is critical for timely care. We evaluate Llama 3.1 8B and GPT-4o large language models using zero-shot, few-shot, and fine-tuned strategies on the MIMIC-IV-Ext-PE dataset. GPT-4o achieved 99.3% accuracy and 95.75% F1-score, outperforming all Llama variants. Our results highlight the potential of LLMs to automate PE identification, reducing reliance on manual chart review.
Speaker:
Farzad Ahmed, Ph.D. Student in Computer Science
George Mason University
Farzad
Ahmed,
Ph.D. Student in Computer Science - George Mason University
PhenoGnet: A Graph-Based Contrastive Learning Framework for Disease Similarity Prediction
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Deep Learning, Machine Learning, Controlled Terminologies, Ontologies, and Vocabularies, Bioinformatics
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
Understanding disease similarity is beneficial in biomedical research, aiding in diagnosis, drug discovery, and treatment planning. We propose PhenoGnet, a novel framework integrating gene interaction networks and the Human Phenotype Ontology (HPO) using graph neural networks. PhenoGnet employs contrastive learning to align gene and phenotype representations based on associations, enhancing disease similarity prediction. Evaluated on expert-curated datasets, PhenoGnet achieves high predictive performance, demonstrating its potential for knowledge discovery, particularly in rare disease research.
Speaker:
Ranga Baminiwatte, PhD Candidate
Clemson University
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Deep Learning, Machine Learning, Controlled Terminologies, Ontologies, and Vocabularies, Bioinformatics
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
Understanding disease similarity is beneficial in biomedical research, aiding in diagnosis, drug discovery, and treatment planning. We propose PhenoGnet, a novel framework integrating gene interaction networks and the Human Phenotype Ontology (HPO) using graph neural networks. PhenoGnet employs contrastive learning to align gene and phenotype representations based on associations, enhancing disease similarity prediction. Evaluated on expert-curated datasets, PhenoGnet achieves high predictive performance, demonstrating its potential for knowledge discovery, particularly in rare disease research.
Speaker:
Ranga Baminiwatte, PhD Candidate
Clemson University
Ranga
Baminiwatte,
PhD Candidate - Clemson University
Variable importance cloud for medical image analysis
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Deep Learning, Artificial Intelligence, Clinical Decision Support
Primary Track: Foundations
Programmatic Theme: Clinical Informatics
Explainable artificial intelligence (XAI) is critical for developing predictive models in healthcare. Traditional methods focus on a single optimized model, overlooking model uncertainty. We propose VIC-image, a robust ensemble-based XAI method tailored to medical image analysis that leverages nearly-optimal models each with Grad-CAM++ applied. Evaluated on the HAM10000 dataset, VIC-Image outperformed Grad-CAM++ with 6.40% higher IoU and 5.29% higher DSC, better aligning with physician annotations. VIC-image enhances robustness and trustworthiness in medical AI applications.
Speaker:
Mingxuan Liu, PhD student
Duke-NUS Medical School
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Deep Learning, Artificial Intelligence, Clinical Decision Support
Primary Track: Foundations
Programmatic Theme: Clinical Informatics
Explainable artificial intelligence (XAI) is critical for developing predictive models in healthcare. Traditional methods focus on a single optimized model, overlooking model uncertainty. We propose VIC-image, a robust ensemble-based XAI method tailored to medical image analysis that leverages nearly-optimal models each with Grad-CAM++ applied. Evaluated on the HAM10000 dataset, VIC-Image outperformed Grad-CAM++ with 6.40% higher IoU and 5.29% higher DSC, better aligning with physician annotations. VIC-image enhances robustness and trustworthiness in medical AI applications.
Speaker:
Mingxuan Liu, PhD student
Duke-NUS Medical School
Mingxuan
Liu,
PhD student - Duke-NUS Medical School
CDEMapper 2.0: Expediting collaborative review and facilitating consensus building during Common Data Elements mapping
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Data Standards, Interoperability and Health Information Exchange, Controlled Terminologies, Ontologies, and Vocabularies
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
In this study, we developed and evaluated CDEMapper 2.0, a collaborative tool designed to streamline the mapping of study variables to NIH Common Data Elements (CDEs). It supports semantic search, large language model (LLM)-based recommendations, multi-role and multi-user management, multiple experts review, iterative refinement, and consensus building during CDE mapping. Usability evaluation demonstrated its potential to expedite collaborative review, improve consistency, and address key barriers to broader CDE adoption.
Speaker:
Vincent Zhang, MS
Yale University
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Data Standards, Interoperability and Health Information Exchange, Controlled Terminologies, Ontologies, and Vocabularies
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
In this study, we developed and evaluated CDEMapper 2.0, a collaborative tool designed to streamline the mapping of study variables to NIH Common Data Elements (CDEs). It supports semantic search, large language model (LLM)-based recommendations, multi-role and multi-user management, multiple experts review, iterative refinement, and consensus building during CDE mapping. Usability evaluation demonstrated its potential to expedite collaborative review, improve consistency, and address key barriers to broader CDE adoption.
Speaker:
Vincent Zhang, MS
Yale University
Vincent
Zhang,
MS - Yale University
Cross-validation of Machine Learning for All-Cause Mortality and Cancer-Specific Mortality Prediction in Prostate and Breast Cancer
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Data Standards, Evaluation, Machine Learning
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
Prostate and breast cancer pose significant health challenges, especially for older and underserved populations. This study utilizes SEER-MHOS patient-reported outcomes to predict one-year and five-year all-cause and cancer-specific mortality, emphasizing standardized machine learning (ML) evaluation. Five of Wojtusiak’s ten criteria guide the analysis to enhance reproducibility and transparency. A two-stage approach was applied to patients aged 65+, incorporating missing data imputation and 30-fold cross-validation for lookback window optimization, followed by predictive modeling with seven ML algorithms. Optimal lookback windows were 1,160 days for prostate cancer, 410 for breast, and 430 for the combined cohort. Random Forest consistently outperformed others (AUC: 0.82 prostate, 0.78 breast). Combining datasets improved sample size and predictive stability (AUC LR: 0.80). Findings highlight the importance of tailored modeling and robust data integration in improving mortality prediction accuracy and mitigating disparities in cancer outcomes.
Speaker:
HUAN-JU (Coco) SHIH, Doctoral Candidate
George Mason University
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Data Standards, Evaluation, Machine Learning
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
Prostate and breast cancer pose significant health challenges, especially for older and underserved populations. This study utilizes SEER-MHOS patient-reported outcomes to predict one-year and five-year all-cause and cancer-specific mortality, emphasizing standardized machine learning (ML) evaluation. Five of Wojtusiak’s ten criteria guide the analysis to enhance reproducibility and transparency. A two-stage approach was applied to patients aged 65+, incorporating missing data imputation and 30-fold cross-validation for lookback window optimization, followed by predictive modeling with seven ML algorithms. Optimal lookback windows were 1,160 days for prostate cancer, 410 for breast, and 430 for the combined cohort. Random Forest consistently outperformed others (AUC: 0.82 prostate, 0.78 breast). Combining datasets improved sample size and predictive stability (AUC LR: 0.80). Findings highlight the importance of tailored modeling and robust data integration in improving mortality prediction accuracy and mitigating disparities in cancer outcomes.
Speaker:
HUAN-JU (Coco) SHIH, Doctoral Candidate
George Mason University
HUAN-JU (Coco)
SHIH,
Doctoral Candidate - George Mason University
Creation of a Novel EHR Provider Utilization Dashboard to Drive Process Improvement and Enhance Provider Well Being
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Data Sharing, Usability, Information Visualization, User-centered Design Methods
Primary Track: Applications
Several electronic health record (EHR) metrics are correlated to provider burnout and job dissatisfaction rates. Enterprise-wide end user metric data review is important to understand EHR burdens and benefits. Cerner Advance (CA) and Lights On Network (LON) metrics provide end user data analytics with limitations on how data can be filtered, disseminated, and viewed. Using HealtheIntent and Tableau analytics a dynamic EHR utilization dashboard was created to present this data in a user friendly format.
Speaker:
Stacey Stokes, MD, MPH
Children's National Hospital
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Data Sharing, Usability, Information Visualization, User-centered Design Methods
Primary Track: Applications
Several electronic health record (EHR) metrics are correlated to provider burnout and job dissatisfaction rates. Enterprise-wide end user metric data review is important to understand EHR burdens and benefits. Cerner Advance (CA) and Lights On Network (LON) metrics provide end user data analytics with limitations on how data can be filtered, disseminated, and viewed. Using HealtheIntent and Tableau analytics a dynamic EHR utilization dashboard was created to present this data in a user friendly format.
Speaker:
Stacey Stokes, MD, MPH
Children's National Hospital
Stacey
Stokes,
MD, MPH - Children's National Hospital
Privacy-Preserving Dysphonia Detection based on Distributed Deep Learning
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Data Sharing, Privacy and Security, Machine Learning
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
This study presents a privacy-preserving dysphonia detection system using distributed deep learning. By integrating speech data from different healthcare systems, we employ federated and split learning to train models without sharing raw data. Split learning, using the SplitFedV1 framework, outperforms federated learning in terms of convergence speed and accuracy. We also adapted the Audio Spectrogram Transformer to the split learning framework, demonstrating its effectiveness across various neural networks.
Speaker:
Jiarui Xu, N/A
ShanghaiTech University
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Data Sharing, Privacy and Security, Machine Learning
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
This study presents a privacy-preserving dysphonia detection system using distributed deep learning. By integrating speech data from different healthcare systems, we employ federated and split learning to train models without sharing raw data. Split learning, using the SplitFedV1 framework, outperforms federated learning in terms of convergence speed and accuracy. We also adapted the Audio Spectrogram Transformer to the split learning framework, demonstrating its effectiveness across various neural networks.
Speaker:
Jiarui Xu, N/A
ShanghaiTech University
Jiarui
Xu,
N/A - ShanghaiTech University
Exploring Data Scraping on ClinicalTrials.gov to Identify Key Variables to Include in an EHR-based Recruitment Tool for Diabetes Clinical Trials
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Data Mining, Population Health, Workflow
Primary Track: Foundations
Programmatic Theme: Clinical Research Informatics
Failure to meet recruitment goals leads to trial termination and delays. We developed an EHR-based recruitment tool by scraping clinicaltrials.gov to identify criteria from 203 diabetes trials. Of 115 terms, 91 were clinically relevant. There was a large overlap in criteria from data scraping and consulting trialists with 3 criteria unique to trialists and 11 criteria uniquely identified through scraping. Automating variable selection via clinicaltrials.gov extraction streamlines recruitment while maintaining clinical relevance.
Speaker:
Sydney Lash, B.S.
University of North Carolina at Chapel Hill
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Data Mining, Population Health, Workflow
Primary Track: Foundations
Programmatic Theme: Clinical Research Informatics
Failure to meet recruitment goals leads to trial termination and delays. We developed an EHR-based recruitment tool by scraping clinicaltrials.gov to identify criteria from 203 diabetes trials. Of 115 terms, 91 were clinically relevant. There was a large overlap in criteria from data scraping and consulting trialists with 3 criteria unique to trialists and 11 criteria uniquely identified through scraping. Automating variable selection via clinicaltrials.gov extraction streamlines recruitment while maintaining clinical relevance.
Speaker:
Sydney Lash, B.S.
University of North Carolina at Chapel Hill
Sydney
Lash,
B.S. - University of North Carolina at Chapel Hill
Finding Rare Disease Experts for the Undiagnosed Disease Network
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Data Mining, Informatics Implementation, Information Retrieval, Information Visualization
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
The NIH-funded Undiagnosed Disease Network (UDN) brings together experts from healthcare organizations across the United States to provide answers for patients and families affected by mysterious conditions where clinicians have been unable to discover the cause. We used “Profiles”, an open-source system for generating searchable online profiles of investigators, to create a website to help the UDN find these experts. Profiles imports and links data from PubMed, ClinicalTrials.gov, GeneReports, OMIM, and more.
Speaker:
Griffin Weber, MD, PhD
Harvard Medical School
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Data Mining, Informatics Implementation, Information Retrieval, Information Visualization
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
The NIH-funded Undiagnosed Disease Network (UDN) brings together experts from healthcare organizations across the United States to provide answers for patients and families affected by mysterious conditions where clinicians have been unable to discover the cause. We used “Profiles”, an open-source system for generating searchable online profiles of investigators, to create a website to help the UDN find these experts. Profiles imports and links data from PubMed, ClinicalTrials.gov, GeneReports, OMIM, and more.
Speaker:
Griffin Weber, MD, PhD
Harvard Medical School
Griffin
Weber,
MD, PhD - Harvard Medical School
A Reciprocal Approach to Understanding ICU Stays in Alcohol-associated Liver Disease: Bridging Disease-specific Modeling and Clinical Knowledge
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Critical Care, Real-World Evidence Generation, Machine Learning, Healthcare Quality, Clinical Decision Support
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
Decompensated alcohol-associated liver disease presents with severe systemic manifestations that often necessitate critical care. ICU length of stay influences patient outcomes, caregiver communication, healthcare costs, and resource allocation. This study compares various machine learning algorithms to determine the best-performing model for predicting ICU length of stay among patients with alcohol-associated liver disease, emphasizing the importance of disease-specific features in model optimization and the bidirectional relationship between machine learning model interpretation and clinical knowledge.
Speaker:
Sue Hyon Kim, MSN, RN
University of Pennsylvania
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Critical Care, Real-World Evidence Generation, Machine Learning, Healthcare Quality, Clinical Decision Support
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
Decompensated alcohol-associated liver disease presents with severe systemic manifestations that often necessitate critical care. ICU length of stay influences patient outcomes, caregiver communication, healthcare costs, and resource allocation. This study compares various machine learning algorithms to determine the best-performing model for predicting ICU length of stay among patients with alcohol-associated liver disease, emphasizing the importance of disease-specific features in model optimization and the bidirectional relationship between machine learning model interpretation and clinical knowledge.
Speaker:
Sue Hyon Kim, MSN, RN
University of Pennsylvania
Sue Hyon
Kim,
MSN, RN - University of Pennsylvania
Mapping Nursing Flowsheet Data to Common Data Models: A Pilot Study
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Controlled Terminologies, Ontologies, and Vocabularies, Nursing Informatics, Data Standards, Interoperability and Health Information Exchange
Primary Track: Foundations
Programmatic Theme: Clinical Informatics
Nurses, the largest healthcare workforce, generate most data through flowsheets, yet national data initiatives generally focus on physician-centered data. This study examines nurse-sensitive data representation in a common data model (CDM). Similar to previous studies, 65.5% of flowsheet rows were mapped to SNOMED CT and LOINC, highlighting a gap in nurse data representation. Further work is needed to identify key flowsheet rows influencing nursing outcomes and develop a more representative CDM.
Speaker:
Malin Britt Lalich, BSN, RN, PHN, PHIT
University of Minnesota
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Controlled Terminologies, Ontologies, and Vocabularies, Nursing Informatics, Data Standards, Interoperability and Health Information Exchange
Primary Track: Foundations
Programmatic Theme: Clinical Informatics
Nurses, the largest healthcare workforce, generate most data through flowsheets, yet national data initiatives generally focus on physician-centered data. This study examines nurse-sensitive data representation in a common data model (CDM). Similar to previous studies, 65.5% of flowsheet rows were mapped to SNOMED CT and LOINC, highlighting a gap in nurse data representation. Further work is needed to identify key flowsheet rows influencing nursing outcomes and develop a more representative CDM.
Speaker:
Malin Britt Lalich, BSN, RN, PHN, PHIT
University of Minnesota
Malin
Britt Lalich,
BSN, RN, PHN, PHIT - University of Minnesota
Developing a Comprehensive Vocabulary for Cannabis Use Documentation
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Controlled Terminologies, Ontologies, and Vocabularies, Knowledge Representation and Information Modeling, Information Extraction, Natural Language Processing
Primary Track: Foundations
Programmatic Theme: Academic Informatics / LIEAF
Cannabis use has steadily increased with medical and recreational legalization in U.S. states, but characterizing use within electronic health records remains challenging. We developed a vocabulary of 135 terms, including names, product forms, routes, reasons for use, and drug-drug interactions through an iterative process based on prior studies, expert recommendations, and analysis of 156,504 clinical notes. This resource aims to fill gaps in existing vocabularies to support cohort identification and decision making in medical contexts.
Speaker:
Wei Wei, PhD
Geisinger
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Controlled Terminologies, Ontologies, and Vocabularies, Knowledge Representation and Information Modeling, Information Extraction, Natural Language Processing
Primary Track: Foundations
Programmatic Theme: Academic Informatics / LIEAF
Cannabis use has steadily increased with medical and recreational legalization in U.S. states, but characterizing use within electronic health records remains challenging. We developed a vocabulary of 135 terms, including names, product forms, routes, reasons for use, and drug-drug interactions through an iterative process based on prior studies, expert recommendations, and analysis of 156,504 clinical notes. This resource aims to fill gaps in existing vocabularies to support cohort identification and decision making in medical contexts.
Speaker:
Wei Wei, PhD
Geisinger
Wei
Wei,
PhD - Geisinger
Utilizing Ontologies to Map the Intervention Landscape of Depression
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Controlled Terminologies, Ontologies, and Vocabularies, Knowledge Representation and Information Modeling, Natural Language Processing, Data transformation/ETL
Primary Track: Foundations
Depression is a complex condition that is treated by both pharmacological and non-pharmacological interventions. In this study, we analyzed electronic health records (EHRs) to explore pharmacological and non-pharmacological interventions for depression using an ontology-driven approach. By creating a pharmacological ontology for antidepressants and applying the DREAMDNPTO ontology and QuickUMLS for non-pharmacological treatments, we can find patterns in depression-related treatment use from MIMIC-IV data. This analysis can provide insights into intervention trends.
Speaker:
Catherine Kim, B.S.
University of Washington
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Controlled Terminologies, Ontologies, and Vocabularies, Knowledge Representation and Information Modeling, Natural Language Processing, Data transformation/ETL
Primary Track: Foundations
Depression is a complex condition that is treated by both pharmacological and non-pharmacological interventions. In this study, we analyzed electronic health records (EHRs) to explore pharmacological and non-pharmacological interventions for depression using an ontology-driven approach. By creating a pharmacological ontology for antidepressants and applying the DREAMDNPTO ontology and QuickUMLS for non-pharmacological treatments, we can find patterns in depression-related treatment use from MIMIC-IV data. This analysis can provide insights into intervention trends.
Speaker:
Catherine Kim, B.S.
University of Washington
Catherine
Kim,
B.S. - University of Washington
Impact of Radiology Order on ED Orthopedic Consult Turnaround Time
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support, Workflow, Healthcare Quality, Data Mining
Primary Track: Applications
Programmatic Theme: Clinical Informatics
This study analyzes one year of Emergency Department (ED) orthopedic consult data (N = 1568) to assess the effects of radiology order timing. Post-consult radiology orders increased turnaround time by 30% (5.42 vs. 3.74 hours, p < 0.001). This pattern was consistent across patient dispositions, suggesting earlier radiology or shifting post-consult radiology to inpatient or observation settings may help reduce delays and improve ED throughput.
Speaker:
Christina Gomez, MIS
Jackson Health Systems
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support, Workflow, Healthcare Quality, Data Mining
Primary Track: Applications
Programmatic Theme: Clinical Informatics
This study analyzes one year of Emergency Department (ED) orthopedic consult data (N = 1568) to assess the effects of radiology order timing. Post-consult radiology orders increased turnaround time by 30% (5.42 vs. 3.74 hours, p < 0.001). This pattern was consistent across patient dispositions, suggesting earlier radiology or shifting post-consult radiology to inpatient or observation settings may help reduce delays and improve ED throughput.
Speaker:
Christina Gomez, MIS
Jackson Health Systems
Christina
Gomez,
MIS - Jackson Health Systems
Impact Of Close Encounter Checkpoint Notifications on G2211 Billing Accuracy in Primary Care Practices
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support, User-centered Design Methods, Workflow, Chronic Care Management, Healthcare Economics/Cost of Care, Informatics Implementation
Primary Track: Applications
Programmatic Theme: Clinical Informatics
Effective January 1, 2024, Healthcare Common Procedure Coding System (HCPCS) billing code G2211 marks a significant advancement in recognizing the complexity and continuity of outpatient visits in primary care. Accurate billing is essential for appropriate reimbursement, benefiting both clinicians and patients. We implemented an alert system to notify clinicians of eligible encounters, significantly improving G2211 utilization and accuracy. These findings highlight the role of automated notifications in optimizing reimbursement and enhancing documentation precision in primary care.
Speaker:
Stefan Mathews, MD, MSHI, FAAFP
Virtua
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support, User-centered Design Methods, Workflow, Chronic Care Management, Healthcare Economics/Cost of Care, Informatics Implementation
Primary Track: Applications
Programmatic Theme: Clinical Informatics
Effective January 1, 2024, Healthcare Common Procedure Coding System (HCPCS) billing code G2211 marks a significant advancement in recognizing the complexity and continuity of outpatient visits in primary care. Accurate billing is essential for appropriate reimbursement, benefiting both clinicians and patients. We implemented an alert system to notify clinicians of eligible encounters, significantly improving G2211 utilization and accuracy. These findings highlight the role of automated notifications in optimizing reimbursement and enhancing documentation precision in primary care.
Speaker:
Stefan Mathews, MD, MSHI, FAAFP
Virtua
Stefan
Mathews,
MD, MSHI, FAAFP - Virtua
Use of Advanced Design Methods for End-of-Life Care
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support, Transitions of Care, Machine Learning, Patient Engagement and Preferences
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
With breast cancer and the progression to metastatic breast cancer affecting millions of women annually, artificial intelligence (AI) and machine learning methodologies have been utilized to pursue the understanding of cellular or biological disease components, promote therapeutic advancements, improve the analysis of radiology procedures and images, and advance prognostication accuracies. However, in the context of end-of-life (EOL) care and supportive care interventions, AI has been used in a limited capacity.
Speaker:
Rachel Brazee, PhD
University of Pittsburgh
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support, Transitions of Care, Machine Learning, Patient Engagement and Preferences
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
With breast cancer and the progression to metastatic breast cancer affecting millions of women annually, artificial intelligence (AI) and machine learning methodologies have been utilized to pursue the understanding of cellular or biological disease components, promote therapeutic advancements, improve the analysis of radiology procedures and images, and advance prognostication accuracies. However, in the context of end-of-life (EOL) care and supportive care interventions, AI has been used in a limited capacity.
Speaker:
Rachel Brazee, PhD
University of Pittsburgh
Rachel
Brazee,
PhD - University of Pittsburgh
Assessment of Inclusion in USCDI v5 of Query Data Elements Required to Execute a Clinical Decision Support Knowledge Base Encoded in Arden Syntax
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support, Standards, Interoperability and Health Information Exchange
Primary Track: Applications
Programmatic Theme: Clinical Informatics
Context: Arden Syntax encodes computable clinical knowledge as Medical Logic Modules (MLMs). USCDI delineates data classes and elements for health data exchange. Objective: Assess the extent to which data shared pursuant to USCDI is adequate for execution of an Arden Syntax knowledge base (KB). Methods: 325 MLMs containing 3268 queries were examined. Result: USCDI includes all queried data elements. Conclusion: Data exchanged pursuant to USCDI would be adequate to execute a KB of MLMs.
Speaker:
Robert Jenders, MD, MS, FACP, FACMI, FHL7, FAMIA
Charles Drew University/UCLA
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support, Standards, Interoperability and Health Information Exchange
Primary Track: Applications
Programmatic Theme: Clinical Informatics
Context: Arden Syntax encodes computable clinical knowledge as Medical Logic Modules (MLMs). USCDI delineates data classes and elements for health data exchange. Objective: Assess the extent to which data shared pursuant to USCDI is adequate for execution of an Arden Syntax knowledge base (KB). Methods: 325 MLMs containing 3268 queries were examined. Result: USCDI includes all queried data elements. Conclusion: Data exchanged pursuant to USCDI would be adequate to execute a KB of MLMs.
Speaker:
Robert Jenders, MD, MS, FACP, FACMI, FHL7, FAMIA
Charles Drew University/UCLA
Robert
Jenders,
MD, MS, FACP, FACMI, FHL7, FAMIA - Charles Drew University/UCLA
Clinical Decision Support Tool to Increase the Diagnosis of Pediatric Acute Respiratory Distress Syndrome to Improve Adherence with Lung Protective Ventilation
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support, Pediatrics, Critical Care, Clinical Guidelines
Primary Track: Applications
Programmatic Theme: Clinical Informatics
This feasibility study evaluated a silently implemented CDS system for identifying PARDS in PICU patients per PALICC-2 guidelines using OI/OSI criteria. Over 47 days in 2024, 415 patients were screened; the CDS flagged 9.88% (41 patients), with 46% (19) meeting PARDS criteria (4.5% of PICU patients). Only 1 patient (5.2%) received lung-protective ventilation per guidelines. Findings suggest under-recognition of PARDS and limited adherence to recommended ventilation strategies.
Speaker:
Ryan Winter, Physician Assistant
Childrens Healthcare of Atlanta
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support, Pediatrics, Critical Care, Clinical Guidelines
Primary Track: Applications
Programmatic Theme: Clinical Informatics
This feasibility study evaluated a silently implemented CDS system for identifying PARDS in PICU patients per PALICC-2 guidelines using OI/OSI criteria. Over 47 days in 2024, 415 patients were screened; the CDS flagged 9.88% (41 patients), with 46% (19) meeting PARDS criteria (4.5% of PICU patients). Only 1 patient (5.2%) received lung-protective ventilation per guidelines. Findings suggest under-recognition of PARDS and limited adherence to recommended ventilation strategies.
Speaker:
Ryan Winter, Physician Assistant
Childrens Healthcare of Atlanta
Ryan
Winter,
Physician Assistant - Childrens Healthcare of Atlanta
Association between no-shows to scheduled appointments and 30-day Risk of Overdose in Patients Prescribed Methadone for Opioid Use Disorder
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support, Patient Safety, Healthcare Quality
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
In the context of the ongoing opioid crisis, this study examined the association between no-shows in the past 30 days and 30-day overdose risk among patients prescribed methadone for opioid use disorder. Analysis of 1,071 patients showed Z-scored no-shows were associated with 30-day overdose risk when adjusting for temporal variables and patient characteristics (OR=1.23, 95%CI:1.19-1.27, p<0.001). These findings suggest attendance tracking could contribute to risk stratification in opioid treatment programs.
Speaker:
Henry Philofsky, MD
University of Rochester Medical Center
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support, Patient Safety, Healthcare Quality
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
In the context of the ongoing opioid crisis, this study examined the association between no-shows in the past 30 days and 30-day overdose risk among patients prescribed methadone for opioid use disorder. Analysis of 1,071 patients showed Z-scored no-shows were associated with 30-day overdose risk when adjusting for temporal variables and patient characteristics (OR=1.23, 95%CI:1.19-1.27, p<0.001). These findings suggest attendance tracking could contribute to risk stratification in opioid treatment programs.
Speaker:
Henry Philofsky, MD
University of Rochester Medical Center
Henry
Philofsky,
MD - University of Rochester Medical Center
Seize the Moment: Clinical Decision Support for Timely Seizure Rescue Medication Orders
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support, Patient Safety, Human-computer Interaction, Pediatrics
Primary Track: Applications
Programmatic Theme: Clinical Informatics
Effective management of in-hospital seizures require timely recognition and intervention, presenting an opportunity for implementation of Clinical Decision Support (CDS) tools to facilitate necessary orders. Despite providers’ recognition of the potential benefit of CDS to prevent potential care delays, post implementation monitoring reveals a persistent discrepancy between alert activation and acceptance. This highlights the need for ongoing research and refinement to improve CDS efficacy in addressing patient safety concerns.
Speaker:
Sarah Thompson, MSHIMI, BSN, RN
Children's Healthcare of Atlanta
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support, Patient Safety, Human-computer Interaction, Pediatrics
Primary Track: Applications
Programmatic Theme: Clinical Informatics
Effective management of in-hospital seizures require timely recognition and intervention, presenting an opportunity for implementation of Clinical Decision Support (CDS) tools to facilitate necessary orders. Despite providers’ recognition of the potential benefit of CDS to prevent potential care delays, post implementation monitoring reveals a persistent discrepancy between alert activation and acceptance. This highlights the need for ongoing research and refinement to improve CDS efficacy in addressing patient safety concerns.
Speaker:
Sarah Thompson, MSHIMI, BSN, RN
Children's Healthcare of Atlanta
Sarah
Thompson,
MSHIMI, BSN, RN - Children's Healthcare of Atlanta
Electronic Strategies for Tailored Exercise to Prevent Falls (eSTEPS): Implementation in a Tertiary Academic Health System
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support, Patient Safety, Healthcare Quality
Primary Track: Applications
Falls are the leading cause of injury and injury-related death in older adults. The Electronic Strategies for Tailored Exercise to Prevent Falls (eSTEPS) study evaluates an EHR-based clinical decision support (CDS) tool that promotes evidence-based exercise referrals to reduce fall risk. Launched in April 2023 within a tertiary academic health system, the CDS triggered in 92% of eligible visits between July 2023 to June 2024, with referrals placed 26% of the time. This report examines reach, adoption, and implementation in year one, and explores patient and clinician factors influencing referral decisions.
Speaker:
Jenna Reisler, MD
University of Texas Medical Branch
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support, Patient Safety, Healthcare Quality
Primary Track: Applications
Falls are the leading cause of injury and injury-related death in older adults. The Electronic Strategies for Tailored Exercise to Prevent Falls (eSTEPS) study evaluates an EHR-based clinical decision support (CDS) tool that promotes evidence-based exercise referrals to reduce fall risk. Launched in April 2023 within a tertiary academic health system, the CDS triggered in 92% of eligible visits between July 2023 to June 2024, with referrals placed 26% of the time. This report examines reach, adoption, and implementation in year one, and explores patient and clinician factors influencing referral decisions.
Speaker:
Jenna Reisler, MD
University of Texas Medical Branch
Jenna
Reisler,
MD - University of Texas Medical Branch
Revising BPA triggers and inclusion criteria helps reduce nurses’ fatigue
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support, Nursing Informatics, Human-computer Interaction
Primary Track: Applications
Programmatic Theme: Clinical Informatics
Best Practice Advisories (BPAs) are rarely designed with nurses in mind. Yet nursing BPAs are 40% of all BPAs at NYC Health + Hospitals. Here, the BPA work group routinely revises alerts to reduce staff fatigue. Using data analysis and interviews, our study provides insight on the impact of the BPA work group, as well as design recommendations to further alleviate nurses’ alert fatigue.
Speaker:
Federica Bologna, MS
Cornell University
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support, Nursing Informatics, Human-computer Interaction
Primary Track: Applications
Programmatic Theme: Clinical Informatics
Best Practice Advisories (BPAs) are rarely designed with nurses in mind. Yet nursing BPAs are 40% of all BPAs at NYC Health + Hospitals. Here, the BPA work group routinely revises alerts to reduce staff fatigue. Using data analysis and interviews, our study provides insight on the impact of the BPA work group, as well as design recommendations to further alleviate nurses’ alert fatigue.
Speaker:
Federica Bologna, MS
Cornell University
Federica
Bologna,
MS - Cornell University
Towards a Structured Classification of Immune-Related Adverse Events to Inform a Clinical Decision Support System for Early Recognition in Cancer Patients on Immune Checkpoint Inhibitors
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support, Mobile Health, Telemedicine, Healthcare Quality, User-centered Design Methods, Controlled Terminologies, Ontologies, and Vocabularies, Chronic Care Management, Informatics Implementation
Primary Track: Applications
Programmatic Theme: Clinical Informatics
Immune checkpoint inhibitors (ICIs) have transformed cancer treatment yet come with significant risks of immune-related adverse events (irAEs). Use of tools to predict irAE development are limited in clinical care. We sought to develop a clinical decision support (CDS) system to facilitate early recognition of irAEs. The CDS system will entail a patient-facing report system wherein cancer patients on ICIs can regularly report symptoms, with data pushed to the provider electronic health record's dashboard.
Speaker:
Onyekachi Ike-Okpe, MD
University Of Florida
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support, Mobile Health, Telemedicine, Healthcare Quality, User-centered Design Methods, Controlled Terminologies, Ontologies, and Vocabularies, Chronic Care Management, Informatics Implementation
Primary Track: Applications
Programmatic Theme: Clinical Informatics
Immune checkpoint inhibitors (ICIs) have transformed cancer treatment yet come with significant risks of immune-related adverse events (irAEs). Use of tools to predict irAE development are limited in clinical care. We sought to develop a clinical decision support (CDS) system to facilitate early recognition of irAEs. The CDS system will entail a patient-facing report system wherein cancer patients on ICIs can regularly report symptoms, with data pushed to the provider electronic health record's dashboard.
Speaker:
Onyekachi Ike-Okpe, MD
University Of Florida
Onyekachi
Ike-Okpe,
MD - University Of Florida
A Machine Learning Approach to Predict Endometriosis Using EHR
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support, Machine Learning, Data Mining
Primary Track: Applications
Programmatic Theme: Clinical Informatics
Endometriosis is a chronic gynecological condition affecting 5-10% of women of reproductive age. Patients with
endometriosis often face delays in diagnosis due to the non-specific nature of symptoms. This study proposes a
predictive machine learning model to detect endometriosis patients using electronic health record data from
gynecology clinic visits at the University of Cincinnati Medical Center. Preliminary results indicated that among
seven machine learning algorithms, naïve bayes performed the best with an AUC of 0.79.
Speaker:
Parand Shams, Ph.D. Student
Cincinnati Children's Hospital Medical Center
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support, Machine Learning, Data Mining
Primary Track: Applications
Programmatic Theme: Clinical Informatics
Endometriosis is a chronic gynecological condition affecting 5-10% of women of reproductive age. Patients with
endometriosis often face delays in diagnosis due to the non-specific nature of symptoms. This study proposes a
predictive machine learning model to detect endometriosis patients using electronic health record data from
gynecology clinic visits at the University of Cincinnati Medical Center. Preliminary results indicated that among
seven machine learning algorithms, naïve bayes performed the best with an AUC of 0.79.
Speaker:
Parand Shams, Ph.D. Student
Cincinnati Children's Hospital Medical Center
Parand
Shams,
Ph.D. Student - Cincinnati Children's Hospital Medical Center
A Clinically Intuitive Approach to Evaluate Performance of Early Warning Scores
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support, Machine Learning, Critical Care, Evaluation, Fairness and elimination of bias
Primary Track: Applications
Programmatic Theme: Clinical Informatics
Early recognition of clinical deterioration improves patient outcomes. We retrospectively assessed two manual EWS—MEWS and NEWS—and three machine learning models: logistic regression, eXtreme Gradient Boosting (XGB), and long short-term memory, using a calendar-day evaluation framework. The calendar-day approach provided positive net benefits and higher utility compared to standard methods for all models, with XGB performing best. This framework offers a more clinically relevant evaluation of EWS performance.
Speaker:
Peng Wu, MS
University of Wisconsin-Madison
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support, Machine Learning, Critical Care, Evaluation, Fairness and elimination of bias
Primary Track: Applications
Programmatic Theme: Clinical Informatics
Early recognition of clinical deterioration improves patient outcomes. We retrospectively assessed two manual EWS—MEWS and NEWS—and three machine learning models: logistic regression, eXtreme Gradient Boosting (XGB), and long short-term memory, using a calendar-day evaluation framework. The calendar-day approach provided positive net benefits and higher utility compared to standard methods for all models, with XGB performing best. This framework offers a more clinically relevant evaluation of EWS performance.
Speaker:
Peng Wu, MS
University of Wisconsin-Madison
Peng
Wu,
MS - University of Wisconsin-Madison
Developing an Interpretable Breast Cancer Prediction Model Incorporating Demographic, Drug, and Disease Information
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support, Human-computer Interaction, Machine Learning, Cancer Prevention
Primary Track: Applications
Breast cancer affects millions of women every year and is a leading cause of cancer related mortality. As such, early detection is critical for improving patient outcomes. In this work, we develop an explainable machine learning based predictive model for breast cancer risk. We report the construction of our cancer cohort utilizing demographic, drug, and disease information. We also report the model's classification accuracy, individual level classification uncertainty, and risk factors identified by the model.
Speaker:
Matthew Murrow, PhD
Vanderbilt University Medical Center
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Clinical Decision Support, Human-computer Interaction, Machine Learning, Cancer Prevention
Primary Track: Applications
Breast cancer affects millions of women every year and is a leading cause of cancer related mortality. As such, early detection is critical for improving patient outcomes. In this work, we develop an explainable machine learning based predictive model for breast cancer risk. We report the construction of our cancer cohort utilizing demographic, drug, and disease information. We also report the model's classification accuracy, individual level classification uncertainty, and risk factors identified by the model.
Speaker:
Matthew Murrow, PhD
Vanderbilt University Medical Center
Matthew
Murrow,
PhD - Vanderbilt University Medical Center
Evaluating Data Quality in Initial and Repeat Blood Pressure Measurements
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Chronic Care Management, Workflow, Information Visualization, Healthcare Quality, Real-World Evidence Generation
Primary Track: Foundations
Programmatic Theme: Clinical Informatics
While evaluating strategies for improving control of hypertension for our patients, we investigated the accuracy of blood pressure measurements within our system. In almost 500,000 measurements taken during primary care visits over 12 months, we noted that only 22% of elevated measurements were repeated and that the distribution of measurements showed terminal-digit bias, including a bias to record values just under the threshold for adjusting treatment, which became more prominent in repeat measurements.
Speaker:
Seneca Harberger, MD
Geisinger
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Chronic Care Management, Workflow, Information Visualization, Healthcare Quality, Real-World Evidence Generation
Primary Track: Foundations
Programmatic Theme: Clinical Informatics
While evaluating strategies for improving control of hypertension for our patients, we investigated the accuracy of blood pressure measurements within our system. In almost 500,000 measurements taken during primary care visits over 12 months, we noted that only 22% of elevated measurements were repeated and that the distribution of measurements showed terminal-digit bias, including a bias to record values just under the threshold for adjusting treatment, which became more prominent in repeat measurements.
Speaker:
Seneca Harberger, MD
Geisinger
Seneca
Harberger,
MD - Geisinger
Assessing the Impact of a Digital Health Intervention on Self-Efficacy in Heart Failure Patients
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Chronic Care Management, Personal Health Informatics, Mobile Health, Human-computer Interaction, Patient / Person Generated Health Data (Patient Reported Outcomes), Delivering Health Information and Knowledge to the Public, Quantitative Methods
Primary Track: Applications
Programmatic Theme: Consumer Health Informatics
Self-efficacy, or one’s confidence in their ability to carry out tasks, has been shown to have a powerful impact on the self-care behaviors of individuals with chronic conditions, including heart failure. This work explores whether using digital health tools for self-care impacts the self-efficacy of participants in a 6-month intervention. We find significant improvements in self-efficacy, and our results identify sociodemographic and psychosocial factors associated with changes in self-efficacy over the course the intervention.
Speaker:
Rachel Tunis, PhD Student
University of Texas at Austin
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Chronic Care Management, Personal Health Informatics, Mobile Health, Human-computer Interaction, Patient / Person Generated Health Data (Patient Reported Outcomes), Delivering Health Information and Knowledge to the Public, Quantitative Methods
Primary Track: Applications
Programmatic Theme: Consumer Health Informatics
Self-efficacy, or one’s confidence in their ability to carry out tasks, has been shown to have a powerful impact on the self-care behaviors of individuals with chronic conditions, including heart failure. This work explores whether using digital health tools for self-care impacts the self-efficacy of participants in a 6-month intervention. We find significant improvements in self-efficacy, and our results identify sociodemographic and psychosocial factors associated with changes in self-efficacy over the course the intervention.
Speaker:
Rachel Tunis, PhD Student
University of Texas at Austin
Rachel
Tunis,
PhD Student - University of Texas at Austin
Physical Activity Phenotypes as Predictors of Pulmonary Rehabilitation Adherence
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Chronic Care Management, Patient Engagement and Preferences, Telemedicine
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
At-home pulmonary rehabilitation (PR) reduces barriers of traditional programs, but patient adherence and dropout remain challenges. In a 12-week at-home PR program for 375 COPD patients, adherence and dropout factors were examined, focusing on physical activity phenotypes. K-means clustering identified three phenotypes, but no significant associations with adherence or dropout were observed. Instead, higher dyspnea severity (mMRC) predicted lower adherence and increased dropout risk, highlighting the need to consider dyspnea severity for improving PR engagement.
Speaker:
Louis Faust, PhD
Mayo Clinic
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Chronic Care Management, Patient Engagement and Preferences, Telemedicine
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
At-home pulmonary rehabilitation (PR) reduces barriers of traditional programs, but patient adherence and dropout remain challenges. In a 12-week at-home PR program for 375 COPD patients, adherence and dropout factors were examined, focusing on physical activity phenotypes. K-means clustering identified three phenotypes, but no significant associations with adherence or dropout were observed. Instead, higher dyspnea severity (mMRC) predicted lower adherence and increased dropout risk, highlighting the need to consider dyspnea severity for improving PR engagement.
Speaker:
Louis Faust, PhD
Mayo Clinic
Louis
Faust,
PhD - Mayo Clinic
Dyadic Digital Health Interventions for Persons with Dementia and Their Family Caregivers: A Systematic Scoping Review
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Chronic Care Management, Mobile Health, User-centered Design Methods, Usability, Patient Engagement and Preferences, Social Media and Connected Health
Primary Track: Applications
Programmatic Theme: Translational Bioinformatics
Innovative user-facing digital health solutions have the potential to meet the unique health needs of the growing number of adults with cognitive impairment, Alzheimer’s disease, or other dementias, along with their family or informal caregivers, but must be accessible, acceptable, and scalable. This scoping review will identify and describe dyadic digital health interventions for adults with cognitive impairment or dementia and their family caregivers. Results will inform future digital health development and implementation.
Speaker:
Cristina de Rosa, PhD, RN
University of Pittsburgh
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Chronic Care Management, Mobile Health, User-centered Design Methods, Usability, Patient Engagement and Preferences, Social Media and Connected Health
Primary Track: Applications
Programmatic Theme: Translational Bioinformatics
Innovative user-facing digital health solutions have the potential to meet the unique health needs of the growing number of adults with cognitive impairment, Alzheimer’s disease, or other dementias, along with their family or informal caregivers, but must be accessible, acceptable, and scalable. This scoping review will identify and describe dyadic digital health interventions for adults with cognitive impairment or dementia and their family caregivers. Results will inform future digital health development and implementation.
Speaker:
Cristina de Rosa, PhD, RN
University of Pittsburgh
Cristina
de Rosa,
PhD, RN - University of Pittsburgh
Automated Quantitative Assessment of Oral Cavity Dysplasia in Mouse Models of Oral Cavity Squamous Cell Carcinoma
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Cancer Prevention, Imaging Informatics, Machine Learning, Precision Medicine
Primary Track: Applications
Dysplasia is a key but poorly understood transition in oral cavity
cancer development. Accurate, quantitative assessment is critical for
animal studies investigating dysplasia-driven tumorigenesis. However,
current evaluations are limited by subjective variability among
pathologists. We hypothesize that inter-rater variability in
pathologist-labeled dysplasia will be high and that AI-based image
analysis can enhance the consistency, robustness, and reproducibility
of dysplasia assessment in mouse models of oral cancer.
Speaker:
Sheel Tanna, B.S.
University of Chicago
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Cancer Prevention, Imaging Informatics, Machine Learning, Precision Medicine
Primary Track: Applications
Dysplasia is a key but poorly understood transition in oral cavity
cancer development. Accurate, quantitative assessment is critical for
animal studies investigating dysplasia-driven tumorigenesis. However,
current evaluations are limited by subjective variability among
pathologists. We hypothesize that inter-rater variability in
pathologist-labeled dysplasia will be high and that AI-based image
analysis can enhance the consistency, robustness, and reproducibility
of dysplasia assessment in mouse models of oral cancer.
Speaker:
Sheel Tanna, B.S.
University of Chicago
Sheel
Tanna,
B.S. - University of Chicago
ChatGPT-4o proves effective in accurately diagnosing dermoscopic images of benign and malignant skin conditions
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Cancer Prevention, Clinical Decision Support, Imaging Informatics, Informatics Implementation, Large Language Models (LLMs), Precision Medicine, Public Health, Telemedicine
Primary Track: Applications
Artificial intelligence (AI), such as ChatGPT-4o, hold promise in dermatologic diagnostics. This study assessed the accuracy of ChatGPT-4o in diagnosing ten common skin lesions using 150 dermoscopic images. Overall, first-attempt accuracy was 78% (117/150), with melanoma diagnosis achieving 100% sensitivity. Accuracy improved significantly to 96% within the top five differentials. Findings suggest ChatGPT-4o is a valuable diagnostic and educational tool, warranting further validation.
Speaker:
Mascha Korsch, PhD
University of Miami
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Cancer Prevention, Clinical Decision Support, Imaging Informatics, Informatics Implementation, Large Language Models (LLMs), Precision Medicine, Public Health, Telemedicine
Primary Track: Applications
Artificial intelligence (AI), such as ChatGPT-4o, hold promise in dermatologic diagnostics. This study assessed the accuracy of ChatGPT-4o in diagnosing ten common skin lesions using 150 dermoscopic images. Overall, first-attempt accuracy was 78% (117/150), with melanoma diagnosis achieving 100% sensitivity. Accuracy improved significantly to 96% within the top five differentials. Findings suggest ChatGPT-4o is a valuable diagnostic and educational tool, warranting further validation.
Speaker:
Mascha Korsch, PhD
University of Miami
Mascha
Korsch,
PhD - University of Miami
Evaluating the Effectiveness of the Family First Prevention Services Act Using NLP and Staggered Difference-in-Differences Model
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Bioinformatics, Public Health, Natural Language Processing, Evaluation
Primary Track: Policy
Programmatic Theme: Public Health Informatics
This study evaluates the Family First Prevention Services Act's (FFPSA) effectiveness in reducing child maltreatment rates across U.S. states using NLP to analyze FFPSA plans and a staggered Difference-in-Differences model. Results indicate FFPSA implementation significantly reduced maltreatment, particularly through prevention strategies and Qualified Residential Treatment Programs. However, kinship care promotion was associated with increased maltreatment. This study highlights the potential of FFPSA in improving outcomes but also identifies areas for further exploration to optimize its implementation.
Speaker:
Zhidi Luo, MS
Northwestern University
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Bioinformatics, Public Health, Natural Language Processing, Evaluation
Primary Track: Policy
Programmatic Theme: Public Health Informatics
This study evaluates the Family First Prevention Services Act's (FFPSA) effectiveness in reducing child maltreatment rates across U.S. states using NLP to analyze FFPSA plans and a staggered Difference-in-Differences model. Results indicate FFPSA implementation significantly reduced maltreatment, particularly through prevention strategies and Qualified Residential Treatment Programs. However, kinship care promotion was associated with increased maltreatment. This study highlights the potential of FFPSA in improving outcomes but also identifies areas for further exploration to optimize its implementation.
Speaker:
Zhidi Luo, MS
Northwestern University
Zhidi
Luo,
MS - Northwestern University
Decoding Visual Narrative Processing: A Multimodal sEEG-Visual Signal Framework for Mapping Neural Dynamics
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Bioinformatics, Computational Biology, Data Mining, Disability, Accessibility, and Human Function
Primary Track: Foundations
Programmatic Theme: Clinical Research Informatics
sEEG amplitude fluctuations corresponded to narrative elements, we observed that it peaked during emotional climaxes and noted consistent directional interactions between the occipital and temporal cortices. These findings underscore how visual features shape brain connectivity, highlighting cortical synchronization in response to salient plot events. This work provides insight into the neural basis of narrative engagement and visual processing. Using a pipeline to analyze trial-based sEEG data from 10 epilepsy subjects watching annotated Hollywood movies.
Speaker:
Yu Sun, Master of Science
Yale
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Bioinformatics, Computational Biology, Data Mining, Disability, Accessibility, and Human Function
Primary Track: Foundations
Programmatic Theme: Clinical Research Informatics
sEEG amplitude fluctuations corresponded to narrative elements, we observed that it peaked during emotional climaxes and noted consistent directional interactions between the occipital and temporal cortices. These findings underscore how visual features shape brain connectivity, highlighting cortical synchronization in response to salient plot events. This work provides insight into the neural basis of narrative engagement and visual processing. Using a pipeline to analyze trial-based sEEG data from 10 epilepsy subjects watching annotated Hollywood movies.
Speaker:
Yu Sun, Master of Science
Yale
Yu
Sun,
Master of Science - Yale
Methodological reporting of studies evaluating the performance of conversational agents on unstructured data analysis in healthcare: Preliminary findings of a rapid review
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Quantitative Methods, Evaluation
Primary Track: Applications
This rapid review aimed to analyze and evaluate the methodological content of research articles written on the use of conversational agents in healthcare settings. Overall, we found highly varied reporting of key methodological components ranging from 26.3% (how users interacted with outputs) to 83.2% (version number of agents used tied with description of prompts used). Future work will focus on developing a reporting guideline for future studies evaluating conversational agents.
Speaker:
Oliver Nguyen, MSHI
University of Wisconsin at Madison
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Quantitative Methods, Evaluation
Primary Track: Applications
This rapid review aimed to analyze and evaluate the methodological content of research articles written on the use of conversational agents in healthcare settings. Overall, we found highly varied reporting of key methodological components ranging from 26.3% (how users interacted with outputs) to 83.2% (version number of agents used tied with description of prompts used). Future work will focus on developing a reporting guideline for future studies evaluating conversational agents.
Speaker:
Oliver Nguyen, MSHI
University of Wisconsin at Madison
Oliver
Nguyen,
MSHI - University of Wisconsin at Madison
Mapping AI’s Impact in Medical Informatics: A Text Mining Approach
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Machine Learning, Natural Language Processing
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
This poster presents preliminary findings from an ongoing study exploring the impact of artificial intelligence in medical informatics through text mining. Based on an analysis of 1,018 research articles, keyword network analysis and bigram topic modeling were used to identify key themes such as EHRs, decision support, diagnostics, and AI-driven communication. The findings highlight AI’s expanding role and emerging trends across medical informatics, including machine learning, deep learning, and large language models.
Speaker:
Yong Hyun Park, MD, PhD
The Catholic University of Korea
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Machine Learning, Natural Language Processing
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
This poster presents preliminary findings from an ongoing study exploring the impact of artificial intelligence in medical informatics through text mining. Based on an analysis of 1,018 research articles, keyword network analysis and bigram topic modeling were used to identify key themes such as EHRs, decision support, diagnostics, and AI-driven communication. The findings highlight AI’s expanding role and emerging trends across medical informatics, including machine learning, deep learning, and large language models.
Speaker:
Yong Hyun Park, MD, PhD
The Catholic University of Korea
Yong Hyun
Park,
MD, PhD - The Catholic University of Korea
Evaluating AI Voice Agents for Patient Education and Follow-up in Ophthalmology
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Large Language Models (LLMs), Patient Engagement and Preferences
Primary Track: Applications
Programmatic Theme: Clinical Informatics
This pilot study evaluates AI voice agents for patient education and follow-up in Diabetic Macular Edema (DME), Age-related Macular Degeneration (AMD), and Vabysmo treatment. Results showed high accuracy (95-98%), no hallucinations across 55 simulated interactions, and positive subjective feedback (Likert scores: 4.5-4.8). AI voice agents demonstrate promise for enhancing patient communication, warranting further clinical evaluation.
Speaker:
Jaison Jain, BS
Stony Brook University School of Medicine
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Large Language Models (LLMs), Patient Engagement and Preferences
Primary Track: Applications
Programmatic Theme: Clinical Informatics
This pilot study evaluates AI voice agents for patient education and follow-up in Diabetic Macular Edema (DME), Age-related Macular Degeneration (AMD), and Vabysmo treatment. Results showed high accuracy (95-98%), no hallucinations across 55 simulated interactions, and positive subjective feedback (Likert scores: 4.5-4.8). AI voice agents demonstrate promise for enhancing patient communication, warranting further clinical evaluation.
Speaker:
Jaison Jain, BS
Stony Brook University School of Medicine
Jaison
Jain,
BS - Stony Brook University School of Medicine
AgentMD: Empowering Language Agents for Risk Prediction with Large-Scale Clinical Tool Learning
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Large Language Models (LLMs), Deep Learning
Primary Track: Foundations
Programmatic Theme: Clinical Informatics
Clinical calculators are essential for diagnosis, prognosis, and treatment planning but often remain underutilized due to usability and dissemination challenges. We propose AgentMD, an LLM-based autonomous agent that curates 2,164 clinical calculators. AgentMD achieved over 85% accuracy in quality checks and a 90%+ pass rate for unit tests. On the RiskQA benchmark, AgentMD significantly outperforms GPT-4 with chain-of-thought prompting (87.7% vs. 40.9%). These results highlight its potential to enhance clinical decision support and workflow efficiency.
Speaker:
Qiao Jin, M.D.
National Institutes of Health
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Large Language Models (LLMs), Deep Learning
Primary Track: Foundations
Programmatic Theme: Clinical Informatics
Clinical calculators are essential for diagnosis, prognosis, and treatment planning but often remain underutilized due to usability and dissemination challenges. We propose AgentMD, an LLM-based autonomous agent that curates 2,164 clinical calculators. AgentMD achieved over 85% accuracy in quality checks and a 90%+ pass rate for unit tests. On the RiskQA benchmark, AgentMD significantly outperforms GPT-4 with chain-of-thought prompting (87.7% vs. 40.9%). These results highlight its potential to enhance clinical decision support and workflow efficiency.
Speaker:
Qiao Jin, M.D.
National Institutes of Health
Qiao
Jin,
M.D. - National Institutes of Health
Human vs. AI: Performance of AI in Literature Review Screening
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Large Language Models (LLMs), Population Health
Primary Track: Foundations
Programmatic Theme: Clinical Research Informatics
This study evaluates the performance of ChatGPT-4o in title and abstract screening for a literature review focused on studies involving researchers as subjects. From a 10% sample of 19,140 citations, AI screening showed lower sensitivity (76.0%) but higher specificity (92.4%) compared to humans (89.5% sensitivity, 77.1% specificity). While general-purpose AI lacks human-level sensitivity, it may augment review processes by improving specificity and identifying overlooked references, potentially reducing screening burden.
Speaker:
Amelia Shunk, MMCi
NYU Grossman School of Medicine
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Large Language Models (LLMs), Population Health
Primary Track: Foundations
Programmatic Theme: Clinical Research Informatics
This study evaluates the performance of ChatGPT-4o in title and abstract screening for a literature review focused on studies involving researchers as subjects. From a 10% sample of 19,140 citations, AI screening showed lower sensitivity (76.0%) but higher specificity (92.4%) compared to humans (89.5% sensitivity, 77.1% specificity). While general-purpose AI lacks human-level sensitivity, it may augment review processes by improving specificity and identifying overlooked references, potentially reducing screening burden.
Speaker:
Amelia Shunk, MMCi
NYU Grossman School of Medicine
Amelia
Shunk,
MMCi - NYU Grossman School of Medicine
Generative AI-Powered Dataset Retrieval Tool
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Information Retrieval, Machine Learning, Large Language Models (LLMs)
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
Retrieving datasets from clinical data warehouses is a time-consuming and requires SQL expertise. We developed the Generative AI-Powered Dataset Retrieval Tool (GAPDART), which leverages Azure OpenAI to transform natural language input into executable SQL queries, enabling users to access the necessary data for predictive modeling without manual query build. We used GAPDART to retrieve data for an inpatient hypoglycemia prediction model. We recorded the average time to generate the AI-created SQL, token usage per variable, and data retrieval time from the clinical data warehouse. GAPDART successfully generated valid SQL queries and retrieved data for the six prediction variables and one outcome variable. On average, each query took 14.95 seconds to generate, utilized approximately 2,689 tokens, and required up to 8 seconds to retrieve data from the clinical data warehouse for each variable. GAPDART demonstrates the application of generative AI to simplify querying clinical data warehouses.
Speaker:
Aileen Wright, MD
Vanderbilt
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Information Retrieval, Machine Learning, Large Language Models (LLMs)
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
Retrieving datasets from clinical data warehouses is a time-consuming and requires SQL expertise. We developed the Generative AI-Powered Dataset Retrieval Tool (GAPDART), which leverages Azure OpenAI to transform natural language input into executable SQL queries, enabling users to access the necessary data for predictive modeling without manual query build. We used GAPDART to retrieve data for an inpatient hypoglycemia prediction model. We recorded the average time to generate the AI-created SQL, token usage per variable, and data retrieval time from the clinical data warehouse. GAPDART successfully generated valid SQL queries and retrieved data for the six prediction variables and one outcome variable. On average, each query took 14.95 seconds to generate, utilized approximately 2,689 tokens, and required up to 8 seconds to retrieve data from the clinical data warehouse for each variable. GAPDART demonstrates the application of generative AI to simplify querying clinical data warehouses.
Speaker:
Aileen Wright, MD
Vanderbilt
Aileen
Wright,
MD - Vanderbilt
Leveraging Large Language Models to Enhance Machine Learning Interpretability and Predictive Performance: A Case Study on Emergency Department Returns for Mental Health Patients
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Information Extraction, Large Language Models (LLMs), Machine Learning
Primary Track: Applications
Programmatic Theme: Clinical Informatics
Emergency department (ED) returns for mental health patients are frequent and often preventable. This study explores the integration of large language models (LLMs) with traditional machine learning to improve prediction and interpretability. LLM-derived features enhanced performance and generated accurate, clinically meaningful explanations. The framework offers a promising step toward actionable, explainable AI in emergency psychiatry.
Speaker:
Abdulaziz Ahmed, PhD
University of Alabama at Birmingham
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Information Extraction, Large Language Models (LLMs), Machine Learning
Primary Track: Applications
Programmatic Theme: Clinical Informatics
Emergency department (ED) returns for mental health patients are frequent and often preventable. This study explores the integration of large language models (LLMs) with traditional machine learning to improve prediction and interpretability. LLM-derived features enhanced performance and generated accurate, clinically meaningful explanations. The framework offers a promising step toward actionable, explainable AI in emergency psychiatry.
Speaker:
Abdulaziz Ahmed, PhD
University of Alabama at Birmingham
Abdulaziz
Ahmed,
PhD - University of Alabama at Birmingham
Efficient Prediction of COPD Outcomes from Clinical Notes Using Filtered Input and Language Model Embeddings
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Information Extraction, Large Language Models (LLMs)
Primary Track: Applications
We present a scalable framework for predicting COPD outcomes using unstructured clinical notes. By combining a lightweight note-filtering classifier with transformer-based embeddings (Clinical-Longformer), we reduced input volume by over 42% while maintaining comparable AUC performance across six outcomes. Our filtered pipeline lowers computational costs with minimal loss in predictive power, supporting real-time risk stratification in resource-constrained settings. This approach enables efficient, large-scale deployment of language model–based prediction in clinical practice.
Speaker:
Mohammed Al-Garadi, PhD
VUMC
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Information Extraction, Large Language Models (LLMs)
Primary Track: Applications
We present a scalable framework for predicting COPD outcomes using unstructured clinical notes. By combining a lightweight note-filtering classifier with transformer-based embeddings (Clinical-Longformer), we reduced input volume by over 42% while maintaining comparable AUC performance across six outcomes. Our filtered pipeline lowers computational costs with minimal loss in predictive power, supporting real-time risk stratification in resource-constrained settings. This approach enables efficient, large-scale deployment of language model–based prediction in clinical practice.
Speaker:
Mohammed Al-Garadi, PhD
VUMC
Mohammed
Al-Garadi,
PhD - VUMC
Team Leadership Structure and Health AI Implementation Outcomes
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Informatics Implementation, Workflow
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
We investigated how team composition and leadership structure influence AI implementation outcomes in healthcare, adjusting for other AI- and clinically relevant confounders. Using results from 114 clinical studies and assuming that senior authorship reflects team leadership, we hypothesized that clinician (vs. technical)-led AI implementations are more likely to demonstrate significant impact. Logistic regression revealed that leadership structure significantly predicted AI impact, with clinical leadership associated with a higher likelihood of impact (OR = 4.55, p = 0.009).
Speaker:
Peilin Li, MS
Cornell University
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Informatics Implementation, Workflow
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
We investigated how team composition and leadership structure influence AI implementation outcomes in healthcare, adjusting for other AI- and clinically relevant confounders. Using results from 114 clinical studies and assuming that senior authorship reflects team leadership, we hypothesized that clinician (vs. technical)-led AI implementations are more likely to demonstrate significant impact. Logistic regression revealed that leadership structure significantly predicted AI impact, with clinical leadership associated with a higher likelihood of impact (OR = 4.55, p = 0.009).
Speaker:
Peilin Li, MS
Cornell University
Peilin
Li,
MS - Cornell University
A Resource-Conscious Approach Using Small LLMs for Structured Infection Indicator Extraction from Home Healthcare Clinical Notes
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Infectious Diseases and Epidemiology, Natural Language Processing, Nursing Informatics, Deep Learning, Large Language Models (LLMs), Clinical Decision Support, Information Extraction
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
Infections account for 80% of the top ten reasons for unplanned hospitalizations in home healthcare. Delivering structured, precise, and context-rich infection indicators from clinical notes is essential for prompt clinical interventions. However, limited resources and rapid patient dynamics present ongoing challenges in HHC settings. This study demonstrates that fine-tuned, resource-conscious small LLMs (<10B parameters) can effectively extract these indicators under realistic constraints, bridging the gap between high-resource LLMs and decentralized resources available in HHC settings.
Speaker:
Zidu Xu, Mphil,MMed
Columbia University
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Infectious Diseases and Epidemiology, Natural Language Processing, Nursing Informatics, Deep Learning, Large Language Models (LLMs), Clinical Decision Support, Information Extraction
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
Infections account for 80% of the top ten reasons for unplanned hospitalizations in home healthcare. Delivering structured, precise, and context-rich infection indicators from clinical notes is essential for prompt clinical interventions. However, limited resources and rapid patient dynamics present ongoing challenges in HHC settings. This study demonstrates that fine-tuned, resource-conscious small LLMs (<10B parameters) can effectively extract these indicators under realistic constraints, bridging the gap between high-resource LLMs and decentralized resources available in HHC settings.
Speaker:
Zidu Xu, Mphil,MMed
Columbia University
Zidu
Xu,
Mphil,MMed - Columbia University
Leveraging Vision-Language Models for Cephalometric Landmark Detection
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Imaging Informatics, Personal Health Informatics, Clinical Decision Support, Informatics Implementation
Primary Track: Applications
Programmatic Theme: Clinical Informatics
Cephalometric analysis plays a critical role in orthodontic diagnosis and treatment planning.This study introduces an innovative approach utilizing pretrained Vision-Language Models (VLMs), specifically CLIP-based LLaVA, to automate cephalometric landmark detection in orthodontics. Using advanced zero-shot and few-shot prompting methods on the Aariz Cephalometric Dataset (1,000 images, 29 landmarks), our approach significantly improved landmark accuracy and workflow efficiency. Challenges remain in precise spatial localization, suggesting further domain-specific fine-tuning and multicentric validation for enhanced clinical utility.
Speaker:
Harsha Kamineni, Health Informatics- Masters
IUPUI
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Imaging Informatics, Personal Health Informatics, Clinical Decision Support, Informatics Implementation
Primary Track: Applications
Programmatic Theme: Clinical Informatics
Cephalometric analysis plays a critical role in orthodontic diagnosis and treatment planning.This study introduces an innovative approach utilizing pretrained Vision-Language Models (VLMs), specifically CLIP-based LLaVA, to automate cephalometric landmark detection in orthodontics. Using advanced zero-shot and few-shot prompting methods on the Aariz Cephalometric Dataset (1,000 images, 29 landmarks), our approach significantly improved landmark accuracy and workflow efficiency. Challenges remain in precise spatial localization, suggesting further domain-specific fine-tuning and multicentric validation for enhanced clinical utility.
Speaker:
Harsha Kamineni, Health Informatics- Masters
IUPUI
Harsha
Kamineni,
Health Informatics- Masters - IUPUI
Enhancing Equity and Efficiency: AI-Powered Eligibility Assessment for Adult Complex Care
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Health Equity, Population Health, Large Language Models (LLMs), Evaluation, Workflow
Primary Track: Applications
Programmatic Theme: Clinical Informatics
The Office of Population Health at our academic medical center provides Adult Complex Care services for high-utilizing patients, but manual eligibility determination is time-consuming. Using a HIPAA-compliant large language model (LLM) platform, we deployed a structured prompt to streamline chart review. In evaluating 27 charts, the model achieved 100% sensitivity and 77% specificity, with over-inclusion due to missing medical/social complexity criteria. Future refinements will enhance specificity, demonstrating LLMs’ potential to improve efficiency, consistency, and equity in eligibility assessment.
Speaker:
Sara Faghihi Kashani, MD, MPH
UCSF
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Health Equity, Population Health, Large Language Models (LLMs), Evaluation, Workflow
Primary Track: Applications
Programmatic Theme: Clinical Informatics
The Office of Population Health at our academic medical center provides Adult Complex Care services for high-utilizing patients, but manual eligibility determination is time-consuming. Using a HIPAA-compliant large language model (LLM) platform, we deployed a structured prompt to streamline chart review. In evaluating 27 charts, the model achieved 100% sensitivity and 77% specificity, with over-inclusion due to missing medical/social complexity criteria. Future refinements will enhance specificity, demonstrating LLMs’ potential to improve efficiency, consistency, and equity in eligibility assessment.
Speaker:
Sara Faghihi Kashani, MD, MPH
UCSF
Sara
Faghihi Kashani,
MD, MPH - UCSF
Evaluating Approaches to Integrating Social Determinants in Synthetic Data Generation
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Geospatial (GIS) Data/Analysis, Evaluation
Primary Track: Applications
Programmatic Theme: Clinical Informatics
Synthetic medical data offers distinct opportunities to facilitate secure data sharing and medical research, particularly when combined with Social Determinants of Health (SDoH), which are essential for promoting equitable healthcare. However, methodologies for effectively integrating SDoH with synthetic clinical data remain underexplored. This study evaluates post-joined and pre-joined approaches for integrating SDoH data with synthetic clinical data generated via a LLM-based generative AI model. The post-joined approach synthesizes clinical data separately and subsequently integrates it with real SDoH data using synthetic geographic identifiers, while the pre-joined approach simultaneously synthesizes clinical and SDoH variables. Evaluations using a real dataset of 8,813 cancer patients demonstrated that synthetic datasets maintain strong clinical-SDoH correlations, though biases emerge near zero-correlation points. Geographic analyses showed similar county-level value distributions but differences in patient geographic distributions between synthetic and real datasets. Future research should enhance generative models to address these biases and refine synthetic data generation.
Speaker:
Yili Zhang, PhD
Georgetown University
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Geospatial (GIS) Data/Analysis, Evaluation
Primary Track: Applications
Programmatic Theme: Clinical Informatics
Synthetic medical data offers distinct opportunities to facilitate secure data sharing and medical research, particularly when combined with Social Determinants of Health (SDoH), which are essential for promoting equitable healthcare. However, methodologies for effectively integrating SDoH with synthetic clinical data remain underexplored. This study evaluates post-joined and pre-joined approaches for integrating SDoH data with synthetic clinical data generated via a LLM-based generative AI model. The post-joined approach synthesizes clinical data separately and subsequently integrates it with real SDoH data using synthetic geographic identifiers, while the pre-joined approach simultaneously synthesizes clinical and SDoH variables. Evaluations using a real dataset of 8,813 cancer patients demonstrated that synthetic datasets maintain strong clinical-SDoH correlations, though biases emerge near zero-correlation points. Geographic analyses showed similar county-level value distributions but differences in patient geographic distributions between synthetic and real datasets. Future research should enhance generative models to address these biases and refine synthetic data generation.
Speaker:
Yili Zhang, PhD
Georgetown University
Yili
Zhang,
PhD - Georgetown University
On the Bias, Fairness and Bias Mitigation for a Wearable-based Freezing of Gait Detection in Parkinson’s Disease
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Fairness and elimination of bias, Machine Learning, Health Equity, Clinical Decision Support, Deep Learning
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
Consistent performance across diverse demographics is essential for wearable-based freezing of gait (FOG) detection in Parkinson's disease (PD). Our study revealed biases in existing wearable-based FOG models across demographic attributes in multiple datasets. Our proposed transfer learning from multi-site datasets and generic activity representations reduced bias (demographic parity ratio +0.027 & +0.039) and increased performance ( F1-score of +0.026 & +0.018). We hope this work encourages the development of unbiased artificial intelligence for PD populations.
Speaker:
Timothy Odonga, Masters
Emory University
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Fairness and elimination of bias, Machine Learning, Health Equity, Clinical Decision Support, Deep Learning
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics
Consistent performance across diverse demographics is essential for wearable-based freezing of gait (FOG) detection in Parkinson's disease (PD). Our study revealed biases in existing wearable-based FOG models across demographic attributes in multiple datasets. Our proposed transfer learning from multi-site datasets and generic activity representations reduced bias (demographic parity ratio +0.027 & +0.039) and increased performance ( F1-score of +0.026 & +0.018). We hope this work encourages the development of unbiased artificial intelligence for PD populations.
Speaker:
Timothy Odonga, Masters
Emory University
Timothy
Odonga,
Masters - Emory University
Fairness in AI-based Bruise Detection: Averaging vs. Separating Predictions
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Fairness and elimination of bias, Imaging Informatics, Diversity, Equity, Inclusion, and Accessibility, Deep Learning
Primary Track: Foundations
Programmatic Theme: Clinical Research Informatics
Artificial intelligence-based bruise detection models may underperform on darker skin tones, raising fairness concerns. We compare two prediction handling approaches, averaging detections per image versus analyzing detections individually, to examine their impact on fairness using Demographic Parity and Equal Opportunity. Using a YOLOv5 model tested on 1,766 images, averaging yielded stable but less sensitive fairness estimates, while separation revealed greater disparities. Results suggest that prediction handling choices can influence fairness outcomes in bruise detection models.
Speaker:
Dharmi Desai, PhD
George Mason University
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Fairness and elimination of bias, Imaging Informatics, Diversity, Equity, Inclusion, and Accessibility, Deep Learning
Primary Track: Foundations
Programmatic Theme: Clinical Research Informatics
Artificial intelligence-based bruise detection models may underperform on darker skin tones, raising fairness concerns. We compare two prediction handling approaches, averaging detections per image versus analyzing detections individually, to examine their impact on fairness using Demographic Parity and Equal Opportunity. Using a YOLOv5 model tested on 1,766 images, averaging yielded stable but less sensitive fairness estimates, while separation revealed greater disparities. Results suggest that prediction handling choices can influence fairness outcomes in bruise detection models.
Speaker:
Dharmi Desai, PhD
George Mason University
Dharmi
Desai,
PhD - George Mason University
Towards Safer AI in Healthcare: Using Storytelling to Educate About Unintended Harms of AI-Powered Tools
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Fairness and elimination of bias, Large Language Models (LLMs), Natural Language Processing
Primary Track: Foundations
Programmatic Theme: Academic Informatics / LIEAF
We introduce a Human-AI Storytelling Framework that uses multi-agent large language models to simulate clinical narratives and expose potential harms from AI-powered healthcare tools. Our method identifies risks early by generating diverse, realistic stories that highlight failure modes and disparities. Evaluation using GPT-4o shows our method outperforms a template-based baseline in storytelling quality (79%) and risk realism (67.5%), helping stakeholders better anticipate and mitigate unintended consequences in healthcare AI systems before the system is developed.
Speaker:
Anthony Rios, PhD
University of Texas at San Antonio
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Fairness and elimination of bias, Large Language Models (LLMs), Natural Language Processing
Primary Track: Foundations
Programmatic Theme: Academic Informatics / LIEAF
We introduce a Human-AI Storytelling Framework that uses multi-agent large language models to simulate clinical narratives and expose potential harms from AI-powered healthcare tools. Our method identifies risks early by generating diverse, realistic stories that highlight failure modes and disparities. Evaluation using GPT-4o shows our method outperforms a template-based baseline in storytelling quality (79%) and risk realism (67.5%), helping stakeholders better anticipate and mitigate unintended consequences in healthcare AI systems before the system is developed.
Speaker:
Anthony Rios, PhD
University of Texas at San Antonio
Anthony
Rios,
PhD - University of Texas at San Antonio
The Use of Computer Vision for Keystroke Detection for an Adaptive Assessment of Individual with Disability Capacity in Using a Smartphone
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Deep Learning, Evaluation
Primary Track: Foundations
Programmatic Theme: Consumer Health Informatics
Existing keystroke assessment methods require environmental modifications that create barriers for users with disabilities. We aim to bridge this gap by developing a computer vision tool that captures keystrokes during natural smartphone interaction. This approach allows us to observe user behavior by capturing keystrokes without the environmental modifications that may influence user behavior. Our tool demonstrates promising results for keystroke detection with an F1 score of 99%.
Speaker:
Firdaus Indradhirmaya, PhD Student
University of Pittsburgh
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Deep Learning, Evaluation
Primary Track: Foundations
Programmatic Theme: Consumer Health Informatics
Existing keystroke assessment methods require environmental modifications that create barriers for users with disabilities. We aim to bridge this gap by developing a computer vision tool that captures keystrokes during natural smartphone interaction. This approach allows us to observe user behavior by capturing keystrokes without the environmental modifications that may influence user behavior. Our tool demonstrates promising results for keystroke detection with an F1 score of 99%.
Speaker:
Firdaus Indradhirmaya, PhD Student
University of Pittsburgh
Firdaus
Indradhirmaya,
PhD Student - University of Pittsburgh
Implementing an AI Clinical Reasoning Model as a Second Site: Enhancements and Lessons Learned
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Data Sharing, Large Language Models (LLMs), Education and Training
Primary Track: Applications
Programmatic Theme: Clinical Informatics
This podium details the implementation of an AI-based Clinical Reasoning (CR) assessment model built at New York University (NYU) at the University of Cincinnati (UC) as a second site. Using previous Named Entity Recognition (NER) and Large Language Model (LLM) approaches, we fine-tuned and adapted previous models to achieve performance at our site. We share key lessons learned about the impact of documentation and technological differences, annotation pipelines, and fine-tuning strategies.
Speaker:
Scott Vennemeyer, BS
University of Cincinnati, College of Medicine
Presentation Time: 05:00 PM - 06:30 PM
Abstract Keywords: Artificial Intelligence, Data Sharing, Large Language Models (LLMs), Education and Training
Primary Track: Applications
Programmatic Theme: Clinical Informatics
This podium details the implementation of an AI-based Clinical Reasoning (CR) assessment model built at New York University (NYU) at the University of Cincinnati (UC) as a second site. Using previous Named Entity Recognition (NER) and Large Language Model (LLM) approaches, we fine-tuned and adapted previous models to achieve performance at our site. We share key lessons learned about the impact of documentation and technological differences, annotation pipelines, and fine-tuning strategies.
Speaker:
Scott Vennemeyer, BS
University of Cincinnati, College of Medicine
Scott
Vennemeyer,
BS - University of Cincinnati, College of Medicine
Towards Safer AI in Healthcare: Using Storytelling to Educate About Unintended Harms of AI-Powered Tools
Category
Poster - Regular
Description
Custom CSS
double-click to edit, do not edit in source
Date: Tuesday (11/18)
Time: 5:00 PM to 6:30 PM
Room: International Ballroom (Posters)
Time: 5:00 PM to 6:30 PM
Room: International Ballroom (Posters)
11/18/2025 06:30 PM (Eastern Time (US & Canada))