Machine Learning Modeling to Predict Atrial Fibrillation Detection in Embolic Stroke of Undetermined Source Patients

Area of Science:

• Cardiology
• Neurology
• Artificial Intelligence

Background:

• Occult atrial fibrillation (AF) is a significant cause of cardioembolism in embolic stroke of undetermined source (ESUS).
• Implantable cardiac loop recorders (ILRs) detect AF but are underutilized due to cost and inconvenience.
• Undetected AF increases the risk of recurrent ischemic stroke in ESUS patients.

Purpose of the Study:

• To develop and evaluate machine learning (ML) models for predicting AF in ESUS patients.
• To integrate clinical and echocardiography data for enhanced AF prediction.
• To identify cost-effective methods for AF detection in ESUS.

Main Methods:

• A single-center cohort study of 157 consecutive ESUS patients with ILR evaluation.
• Development and hyperparameter tuning of four ML models (Support Vector Machine, Multilayer Perceptron, XGBoost, Random Forest).
• Prediction of AF detection on ILR using clinical and echocardiography variables.

Main Results:

• 20.4% of ESUS patients had occult AF detected by ILR.
• Support Vector Machine achieved the highest AUC (0.736-0.737) for AF prediction.
• Key predictors included age, HDL-C, heart rate, peak mitral A-wave velocity, and left atrial volume.

Conclusions:

• ML models incorporating clinical and echocardiographic data can predict AF in ESUS patients with moderate accuracy.
• This approach may aid in identifying high-risk patients for targeted anticoagulation.
• Further research can refine ML models for improved AF detection in ESUS.