Machine Learning Model-Based Simple Clinical Information to Predict Decreased Left Atrial Appendage Flow Velocity

Area of Science:

• Cardiology
• Medical Informatics
• Artificial Intelligence in Medicine

Background:

• Transesophageal echocardiography (TEE) is standard for assessing left atrial appendage flow velocity (LAAV) but carries risks.
• A simpler, non-invasive method is needed to screen patients with decreased LAAV.
• Machine learning (ML) offers potential for predicting LAAV using accessible clinical data.

Purpose of the Study:

• To investigate the feasibility and accuracy of an ML model in predicting LAAV.
• To develop a simple, applicable method for screening patients with decreased LAAV.
• To assess the diagnostic performance of ML algorithms for LAAV prediction.

Main Methods:

• Analysis of 1039 patients with atrial fibrillation who underwent TEE.
• Development and comparison of three ML models: Support Vector Machine (SVM), Random Forest (RF), and K-Nearest Neighbors (KNN).
• Evaluation of diagnostic accuracy using the area under the receiver operating characteristic curve (AUC).

Main Results:

• 12% of patients had decreased LAAV (< 25 cm/s).
• Decreased LAAV was associated with obesity, persistent AF, heart failure, hypertension, diabetes, stroke, larger left atrium diameter, and higher NT-proBNP levels.
• The RF model demonstrated the best performance with an AUC of 0.89, identifying NT-proBNP as the most impactful predictor.

Conclusions:

• A simple ML model, specifically the Random Forest model, effectively predicts LAAV using clinical information.
• This ML-based tool can serve as a valuable screening method for decreased LAAV.
• The developed tool may assist in risk stratification for patients susceptible to left atrial appendage thrombosis.