Machine learning model for predicting left atrial thrombus or spontaneous echo contrast in non-valvular atrial fibrillation patients based on multimodal echocardiographic parameters

This study aimed to develop a machine learning (ML) model utilizing clinical data and transthoracic echocardiography (TTE) features to predict thrombosis risk in non-valvular atrial fibrillation (NVAF), in contrast to commonly used clinical models. Between January 2020 and December 2023, 402 NVAF patients undergoing catheter ablation or left atrial appendage closure at the First Affiliated Hospital of Guangxi Medical University were prospectively enrolled. Among them, 142 patients (35.3%) exhibited left atrial thrombus or spontaneous echo contrast (LAT/SEC), including 37 cases (9.2%) of LAT, while 260 patients showed no evidence of LAT/SEC. The patients were split into training and validation sets at a ratio of 7:3, with clinical data, biochemical markers, and echocardiographic parameters collected to build the model. Model performance was assessed using accuracy, precision, recall, F1 score, and area under the curve (AUC). The Shapley additive explanations (SHAP) analysis was employed to investigate the contributions of variables. The eXtreme Gradient Boosting (XGBoost) model slightly outperformed the traditional logistic regression model in predicting thrombosis risk in NVAF patients, demonstrating better predictive capability than other ML algorithms, with an AUC of 0.959 (95% CI, 0.925–0.993, P < 0.05). Additionally, XGBoost offered greater clinical net benefit within a threshold probability range of 0.1 to 1.0. SHAP analysis highlighted decreased peak atrial longitudinal strain (PALS), hemodynamic abnormalities, and left atrium spherical remodeling as key factors associated with higher thrombosis risk in NVAF. Utilizing ML techniques to combine clinical risk factors and multimodal echocardiographic parameters can enhance thrombosis risk prediction in NVAF patients, with the XGBoost model showing superior accuracy among various ML algorithms. However, its clinical reliability still needs to be validated in larger, multicenter, prospective cohorts. Not applicable.