Toward Integrating Machine Learning-powered Polysocial Risk Scores into Electronic Health Record Workflows

Presentation Time: 09:45 AM - 10:00 AM

Abstract Keywords: Health Equity and Social Determinants of Health (SDoH), Clinical Decision Support, Artificial Intelligence, Informatics Implementation, User-centered Design Methods
Primary Track: Applications
Programmatic Theme: Clinical Informatics

Social determinants of health (SDoH) account for 80% of modifiable factors driving health disparities. Health systems play a critical role in addressing patients' unmet social needs essential to health outcomes. To integrate social risk management into patient health care, we developed an electronic health record (EHR)-based machine learning-powered pipeline to identify and address unmet social needs associated with hospitalization risk. By quantifying social risk via a polysocial risk score, this tool enables healthcare providers to identify patients at high social risk and prioritize targeted SDoH interventions. However, gaps exist regarding integrating our polysocial risk score tool into clinical flow. Therefore, in this study, through participatory design sessions with healthcare providers and social workers following user-centered design (UCD) principles, we initiated the integration of this predictive model into EHR workflows. This preliminary work lays the foundation for a comprehensive formal user-centered design process to enhance social risk assessment and intervention implementation.

Speaker:
Xing He, Ph.D.
Indiana University

Authors:
Xing He, Ph.D. - Indiana University; Yu Huang, Ph.D. - Indiana University; Yu Hu, MS - University of Florida; Michael Pappa, MS - University of Florida; Natacha Miller, MBA - University of Florida; Megan Gregory, Ph.D. - University of Florida; Serena Jingchuan Guo, MD, PhD - University of Florida; Jiang Bian, PhD - Indiana University;

---

Modeling Competing Clinical Tradeoffs with Concurrent Risk Assessment

Presentation Time: 10:00 AM - 10:15 AM

Abstract Keywords: Clinical Decision Support, Quantitative Methods, Patient Safety, Clinical Guidelines, Evaluation
Primary Track: Applications
Programmatic Theme: Clinical Research Informatics

Clinical decision support (CDS) tools are typically evaluated in isolation, whereas in practice they could be integrated to maximize the benefits for medical decision-making. We modeled prevention of venous thromboembolism (VTE) using validated prediction models and compared treatment rates and clinical outcomes between ideal practice and observed physician behavior. We found that physicians treated patients highly inefficiently, but applying state-of-the-art prediction models concurrently optimized the tradeoff between treatment rates and competing clinical outcomes.

Speaker:
Benjamin Mittman, BA
CWRU School of Medicine

Authors:
Nicholas Casacchia, PharmD, MS - Cleveland Clinic; Bo Hu, PhD - Duke University; Michael Rothberg, MD, MPH - Cleveland Clinic;

---

10-Year Risk Prediction of Higher-Grade AV Block in Patients with First-Degree AV Block

Presentation Time: 10:15 AM - 10:30 AM

Abstract Keywords: Artificial Intelligence, Machine Learning, Clinical Decision Support
Primary Track: Applications
Programmatic Theme: Clinical Informatics

Background: First-degree atrioventricular (AV) block has traditionally been considered benign, but emerging evidence suggests it may indicate a risk of progression to higher-degree AV block. This study developed and externally validated a machine learning model to predict AV block progression using ECG-derived parameters.
Methods: A retrospective cohort study was conducted using 12-lead ECG data from Severance Hospital (development) and Yongin Severance Hospital (external validation). The model was trained with six ECG-derived parameters (RR interval, P duration, PR segment, PR interval, QRS duration, QT interval), along with age and sex, using a Random Forest algorithm.
Results: It achieved an AUROC of 0.823 (AUPRC 0.719) in internal validation and AUROC 0.808 (AUPRC 0.894) in external validation. SHAP analysis identified PR segment, QRS duration, and age as key predictors.
Conclusion: This model enables early risk stratification for AV block progression using widely available ECG parameters, supporting clinical decision-making.

Speaker:
Dong Won Kim, M.S.
Yonsei University

Authors:
HeeYeon Kwon, B.S. - Yonsei University; Je-Wook Park, MD - Yongin Severance Hospital; Hui-Nam Park, M.D. Ph.D. - Yonsei University Health System; Oh-Seok Kwon, Ph.D. - Yonsei University Health System; Changho Han, M.D. Ph.D. - Postdoctoral researcher; Yujeong Kim, Ph.D. - Yonsei University; Dukyong Yoon;

---

Multimodal Data Integration Improves Disease Risk Prediction in the UK Biobank

Presentation Time: 10:30 AM - 10:45 AM

Abstract Keywords: Informatics Implementation, Machine Learning, Knowledge Representation and Information Modeling, Clinical Decision Support, Personal Health Informatics, Real-World Evidence Generation, Data Mining, Clinical Decision Support
Primary Track: Foundations
Programmatic Theme: Translational Bioinformatics

Family health history is an important component to assess risk for common chronic diseases. The integration of electronic health records and genetic data offers great potential to improve disease risk prediction by capturing both clinical and genetic risk factors. We present ALIGATEHR-Gen, a graph attention network that integrates multimodal patient data including diagnosis codes, demographics, and genetic information, along with external medical ontology knowledge. ALIGATEHR-Gen constructs unified patient representations by incorporating genetically inferred first-degree relationships and disease ontology embeddings. We evaluate the predictive performance of ALIGATEHR-Gen across 118 diseases in the UK Biobank and demonstrate that it outperforms state-of-the-art baseline models by an average of at least 6%. A case study on five primary fibrotic and closely related diseases reveals that ALIGATEHR-Gen effectively distinguishes patient subgroups based on clinical and genetic features. These findings illustrate the potential of ALIGATEHR-Gen to advance predictive and interpretable modeling in healthcare.

Speaker:
Xiayuan Huang, PhD.
Yale University

Authors:
Xiayuan Huang, PhD. - Yale University; Hang Zhou, PhD - Yale University; Yitao Hong, Master - Yale University; Xin Zhou, PhD - Yale University; Zuoheng Wang, PhD - Yale University;

---

Quantifying the Impact of Proactive ML Risk Stratification Interventions: A Regression Discontinuity Approach

Presentation Time: 10:45 AM - 11:00 AM

Abstract Keywords: Population Health, Real-World Evidence Generation, Machine Learning, Cancer Prevention, Real-World Evidence Generation
Primary Track: Applications
Programmatic Theme: Clinical Informatics

Geisinger implemented a program that uses machine learning to identify patients who are high-risk for colorectal cancer (CRC) and overdue for screening. Patients with a risk score above the high-risk threshold receive targeted outreach encouraging CRC screening. We applied regression discontinuity (RD) to evaluate the impact of the program and found a 117% increase in CRC screening and a 43% decrease in 2-year mortality. The implications of using RD to evaluate similar programs are discussed.

Speaker:
Elliot Mitchell, PhD
Geisinger

Authors:
Elliot Mitchell, PhD - Geisinger; Minje Park, PhD - The University of Hong Kong; Rebecca Maff, MS - Geisinger; Abdul Tariq, PhD - Children's Hospital of Philadelphia; Keith Boell, MD - Geisinger; David Vawdrey, PhD - Geisinger; Carri Chan, PhD;

---