Calvin Choy
Research Mentor: Mohammed Saeed
Mentor Department: Department of Internal Medicine, Medicine
Author(s): Calvin Choy, Mohammed Saeed
Session: Session 1 (9:00 AM – 9:50 AM)
Presentation Type: Poster 131
Abstract
Atrial fibrillation (AFib) is a prevalent cardiac arrhythmia and a major contributor to recurrent stroke, heart failure, and long-term morbidity. Early identification of AFib risk is particularly important in post-cryptogenic-stroke patients, who are routinely monitored yet may experience delayed or undetected AFib onset. This study explores the use of machine learning methods to predict the future development of atrial fibrillation in patients undergoing cardiac monitoring following a cryptogenic stroke. Early prediction of AFib may help identify patients who would benefit from proactive initiation of anticoagulation to prevent recurrent stroke. We utilize a dataset consisting of electrocardiogram (ECG) monitor reports combined with additional patient information, including demographic characteristics, clinical history, and relevant physiological measurements. Several supervised learning models such as logistic regression, support vector machines, random forests, and gradient-boosted decision trees are trained to estimate an individual’s prospective risk of developing AFib. Model performance is evaluated using standard classification metrics, including accuracy, precision, recall, and area under the receiver operating characteristic curve (AUROC). The results demonstrate that machine learning–based models can effectively leverage ECG-derived features alongside clinical data to improve AFib risk prediction compared to traditional rule-based or risk score approaches. Feature importance analysis highlights clinically meaningful predictors and supports interpretability of the models’ decisions. These findings suggest that machine learning approaches have strong potential to enhance AFib surveillance in post-stroke populations, enabling earlier detection and intervention.
