Abstract:Background Atrial fibrillation and flutter (AF/AFl) increase the risk of thromboembolic events by promoting clot formation in the left atrium (LA), which can be visualised using transoesophageal echocardiography (TEE). Current guidelines recommend initiation of oral anticoagulation (OAC) in patients with AF/AFl based solely on CHA2DS2VASc score (congestive heart failure, hypertension, age ≥75 years, diabetes mellitus, history of stroke or thromboembolism, vascular disease, age 65 to 74 years, female sex). Yet, some patients develop left LA thrombus (LAT) and experience thromboembolic events despite OAC. Purpose We sought to develop and externally validate a machine learning model for prediction of presence of LAT based on clinical, laboratory and transthoracic echocardiography (TTE) features. Methods We analyzed data from the multicenter, prospective LATTEE registry (Left Atrial Thrombus on Transesophageal Echocardiography) that included patients from 13 sites who underwent TEE before cardioversion or ablation between November 2018 and March 2021. We used XgBoost model to predict presence of LAT in TEE based on 29 clinical features, 10 biomarkers and 5 TTE measurements. We trained and tested the model internally using 10-fold hold-out cross validation and data from 12 sites (N=2489). We then tested the final model externally using data from the 13th site (that had recruited most patients, N=400). We compared the predictive performance with that of CHA2DS2VASc score using areas under receiver operating curve (AUC) and DeLong test. Results In the training and internal testing cohort the median age was 67 (Inter Quartile Range [IQR] 59, 74), 63% were male, 85% received OAC and LAT was found in 8.4%. Ablation was the indication for TEE in 43%, cardioversion in 57%. In internal, 10-fold hold-out cross validation, the model achieved AUC of 0.755 (95% confidence interval [CI]: 0.722, 0.788) while CHA2DS2VASc performed significantly worse with AUC of 0.638 (95% CI: 0.604, 0.673), P<0.0001 (Figure 1). Left ventricular ejection fraction, rhythm at the time of study (AF/AFl or sinus rhythm) and age received the highest feature importance ranking (Figure 2). In the external testing cohort the median age was 67 (IQR 59, 74), 66% were male, 88% received OAC, ablation was the indication for TEE in 49% of cases and LAT was found in 6.8%. In this external cohort, our model achieved AUC of 0.815 (95% CI: 0.741, 0.889) while CHA2DS2VASc performed significantly worse with AUC of 0.684 (95% CI: 0.583, 0.785), P=0.028. Conclusion Machine learning based on readily available clinical data allows accurate prediction of the presence of LAT in patients with AF/AFl irrespective of OAC treatment. Such score could be used to identify patients who should undergo TEE before ablation or cardioversion. Subsequent studies to clinically evaluate such application of our model as well as how the model can predict future thromboembolic events are warranted. Funding Acknowledgement Type of funding sources: None.

542 Non-invasive risk assessment of the left atrial appendage thrombosis using deep learning methods

Determinants of Left Atrial Thrombus or Spontaneous Echo Contrast in Nonvalvular Atrial Fibrillation

Comparison of Left Atrial and Atrial Appendage Mechanics to Stratify the Risk of Left Atrial/atrial Appendage Stasis in Patients with Non-Valvular Atrial Fibrillation

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

P1360Association between left atrial appendage dysfunction and thromboembolic risk in nonvalvular atrial fibrillation

Predicting the presence of left atrial appendage thrombus with clinical features and transthoracic measurements using machine learning

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

Deep learning evaluation of echocardiograms to identify occult atrial fibrillation

An artificial intelligence‐enabled electrocardiogram algorithm for the prediction of left atrial low‐voltage areas in persistent atrial fibrillation

Using Machine Learning to Enhance Prediction of Atrial Fibrillation Recurrence after Catheter Ablation

Novel Models for the Prediction of Left Atrial Appendage Thrombus in Patients with Chronic Nonvalvular Atrial Fibrillation

Predictors of thrombosis of left atrium and its appendage before catheter ablation or cardioversion in patients with non-valvular atrial fibrillation or atrial flutter

Association Betweenleft Atrial Appendage Function Assessed by Speckle-Tracking Echocardiography and the Presence Ofleft Atrial Appendage Stasis in Patients with Atrial Fibrillation

Radiomics prediction models of left atrial appendage hypercoagulability based on machine learning algorithms: an exploration about cardiac computed tomography angiography imaging

Predictors of left atrial appendage emptying velocity: Derivation and validation of CHIRP3(M-1) score

A nomogram for predicting left atrial thrombus or spontaneous echo contrast in non-valvular atrial fibrillation patients using hemodynamic parameters from transthoracic echocardiography

Prediction of atrial fibrillation in patients with embolic stroke with undetermined source using electrocardiogram deep learning algorithm and clinical risk factors

Machine Learning of Cardiac Anatomy and the Risk of New-Onset Atrial Fibrillation After TAVR

Influencing factors and predictive model for left atrial appendage emptying velocity in nonvalvular AF patients

Evaluating the novel parameters for assessing the LAA function and thrombus formation with nonvalvular atrial fibrillation