Uterine Leiomyoma Prediction using Geometric Deep Learning and Freely Available Electronic Health Record Data

Presentation Time: 03:30 PM - 03:45 PM

Abstract Keywords: Machine Learning, Generative AI, and Predictive Modeling, Secondary Use of EHR Data, Informatics Research/Biomedical Informatics Research Methods, Data Mining and Knowledge Discovery
Primary Track: Data Science/Artificial Intelligence
Programmatic Theme: Translational Bioinformatics Using Multi-Modal Patient Data and AI

Healthcare data are fragmented and complex, posing challenges for traditional machine learning methods. Graph Neural Networks (GNNs) provide a solution by modeling relationships within multimodal data. This study develops a GNN-based approach to predict uterine fibroids using knowledge graphs derived from the MIMIC-IV dataset. Our GNN model significantly outperformed conventional ML and is considerably more interpretable. These results demonstrate the potential of GNNs for early disease prediction in clinical settings.
Speaker(s):
Tram Anh Nguyen, PhD student
University of Pennsylvania

Author(s):
Tram Anh Nguyen, PhD student - University of Pennsylvania; Joseph Romano, PhD - University of Pennsylvania;

Predictive Models of Hematoma Expansion after Traumatic Brain Injury

Presentation Time: 03:45 PM - 04:00 PM

Abstract Keywords: Machine Learning, Generative AI, and Predictive Modeling, Medical Imaging, Secondary Use of EHR Data, Biomarker Discovery and Development, Natural Language Processing
Primary Track: Clinical Research Informatics
Programmatic Theme: Translational Bioinformatics Using Multi-Modal Patient Data and AI

After a traumatic brain injury, patients are at risk for hematoma expansion, a potentially fatal complication wherein there is continued and progressive intracranial bleeding. Hematoma expansion is associated with prolonged hospital stay, delayed neurosurgery, and mortality. Currently, there are no predictive models of hematoma expansion that are used in the clinical setting. We trained machine learning models to predict hematoma expansion and applied interpretable methods to help ascertain sensitive risk factors of expansion.
Speaker(s):
Meghan Hutch, PhD
University of Chicago

Author(s):
Meghan Hutch, PhD - University of Chicago; Yuan Luo, PhD - Northwestern University; Andrew Naidech, MD MSPH - Northwestern University Feinberg School of Medicine;

Predicting survival time for critically ill patients with heart failure using conformalized survival analysis

Presentation Time: 04:00 PM - 04:15 PM

Abstract Keywords: Data-Driven Research and Discovery, Advanced Data Visualization Tools and Techniques, Learning Healthcare System
Primary Track: Clinical Research Informatics
Programmatic Theme: Real-World Evidence in Informatics: Bridging the Gap between Research and Practice

Heart failure (HF) is a serious public health issue, particularly for critically ill patients in intensive care units (ICUs). Predicting survival outcomes of critically ill patients with calibrated uncertainty calibration is a difficult yet crucially important task for timely treatment. This study applies a novel approach, conformalized survival analysis (CSA), to predicting the survival time to critically ill HF patients. CSA quantifies the uncertainty of point prediction by accompanying each predicted value with a lower bound guaranteed to cover the true survival time. Utilizing the MIMICIV dataset, we demonstrate that CSA delivers calibrated uncertainty quantification for the predicted survival time, while the methods based on parametric models (e.g., Cox model or the Accelerated Failure Time model) fail to do so. By applying CSA to a large, real-world dataset, the study highlights its potential to improve decision making in critical care, offering a more nuanced and accurate tool for prognostication in a setting where precise predictions and calibrated uncertainty quantification can significantly influence patient outcomes.
Speaker(s):
Jiancheng Ye, PhD
Weill Cornell Medicine

Author(s):

Machine Learning for Building a Lung Cancer Risk Prediction Model Using Electronic Health Record Data

Presentation Time: 04:15 PM - 04:30 PM

Abstract Keywords: Machine Learning, Generative AI, and Predictive Modeling, Outcomes Research, Clinical Epidemiology, Population Health, Real-World Evidence and Policy Making, Secondary Use of EHR Data
Primary Track: Data Science/Artificial Intelligence
Programmatic Theme: Health Data Science and Artificial Intelligence Innovation: From Single-Center to Multi-Site

We aimed to build machine learning-based models to predict lung cancer using the Mayo Clinic Lung Cancer Screening Registry. Five machine learning models—XGBoost, RF, SVM, KNN, and LR—were developed and validated to predict a future diagnosis of lung cancer. According to the area under the curve (AUC) differences, XGBoost (0.89) significantly outperformed the others. This study provides valuable insights into the applicability of machine learning models for lung cancer prediction using EHR data.
Speaker(s):
Nan Huo, M.D, Ph.D
Mayo Clinic

Author(s):

Comparison of Machine Learning Models in Predicting Mental Health Sequelae Following Concussion in Youth

Presentation Time: 04:30 PM - 04:45 PM

Abstract Keywords: Machine Learning, Generative AI, and Predictive Modeling, Clinical Decision Support for Translational/Data Science Interventions, Social Determinants of Health, Secondary Use of EHR Data, Outcomes Research, Clinical Epidemiology, Population Health
Primary Track: Data Science/Artificial Intelligence
Programmatic Theme: Health Data Science and Artificial Intelligence Innovation: From Single-Center to Multi-Site

Youth who experience concussions may be at greater risk for subsequent mental health challenges, making early detection crucial for timely intervention. This study utilized Long Short-Term Memory (LSTM) networks to predict mental health outcomes following concussion in youth and compared its performance to traditional models. We also examined whether incorporating social determinants of health (SDoH) improved predictive power, given the disproportionate impact of concussions and mental health issues on disadvantaged populations. We evaluated the models using accuracy, area under the curve (AUC) of the receiver operating characteristic (ROC), and other performance metrics. Our LSTM model with SDoH data achieved the highest accuracy (0.883) and AUC-ROC score (0.892). Unlike traditional models, our approach provided real-time predictions at each visit within 12 months of the index concussion, aiding clinicians in making timely, visit-specific referrals for further treatment and interventions.
Speaker(s):
Jin Peng, MD, MS, PhD
Nationwide Children's Hospital Research Institute

Author(s):
Jiayuan Chen, BS - Computer Science and Engineering, The Ohio State University, Columbus, Ohio, USA; Changchang Yin, MS - Computer Science and Engineering, The Ohio State University, Columbus, Ohio, USA; Ping Zhang, PhD, FAMIA - The Ohio State University; Jingzhen Yang, PhD, MPH - Center for Injury Research and Policy, The Abigail Wexner Research Institute, Nationwide Children’s Hospital, Columbus, Ohio, USA;