Artificial intelligence in detecting left atrial appendage thrombus by transthoracic echocardiography and clinical features: the Left Atrial Thrombus on Transoesophageal Echocardiography (LATTEE) registry
Key result
LAT-AI predicts left atrial appendage thrombus with 0.85 AUC, outperforming LVEF and CHA2DS2-VASc.
- AUC 0.85
- 95% CI 0.82-0.89
- P<.0001
- n=4,111
Why the study?
TOE is routinely performed to rule out left atrial appendage thrombus before catheter ablation or cardioversion in patients on chronic OAC, despite causing discomfort.
Does a machine learning model (LAT-AI) based on clinical and TTE features improve the prediction of left atrial appendage thrombus compared to LVEF and CHA2DS2-VASc score in patients on chronic oral anticoagulation?
Population
Patients undergoing TOE before cardioversion or catheter ablation (n = 2827 training, n = 1284 external testing)
Comparison
LAT-AI model vs LVEF and CHA2DS2-VASc score
Design
13-site prospective registry-based model development and external validation study
Authors
Discussion
Loading...
Member takes
Implication
May support better LAA thrombus prediction than LVEF or CHA2DS2-VASc in anticoagulated patients; leaves open whether it safely reduces TEE use.
Study Design
Type
Cohort (n=4,111)
Multicenter
Yes
Structured PICO
Does a machine learning model (LAT-AI) based on clinical and TTE features improve the prediction of left atrial appendage thrombus compared to LVEF and CHA2DS2-VASc score in patients on chronic oral anticoagulation?
P
Population
4,111 patients (2,827 in development cohort, 1,284 in external testing cohort) on chronic oral anticoagulation who underwent transoesophageal echocardiography (TOE) before cardioversion or catheter ablation from a 13-site prospective registry.
I
Intervention
Machine learning model (LAT-AI) using clinical and transthoracic echocardiography (TTE) features to predict left atrial appendage thrombus.
C
Comparator
Left ventricular ejection fraction (LVEF) and CHA2DS2-VASc score.
O
Outcome
Area under the receiver operating characteristic curve (AUC) for predicting the presence of left atrial appendage thrombus (LAT).surrogate
Main Result
Effect estimate: AUC 0.85 (95% CI 0.82-0.89)
p-value: p=< .0001
A machine learning model combining clinical and TTE features accurately predicts left atrial appendage thrombus, potentially allowing 40% of patients on chronic oral anticoagulation to safely avoid transoesophageal echocardiography before cardioversion or ablation.
Cite This Study
Pieszko et al. (2023) conducted a cohort in Left atrial appendage thrombus (n=4,111). LAT-artificial intelligence (AI) model vs. LVEF and CHA2DS2-VASc score was evaluated on Prediction of left atrial appendage thrombus (LAT) presence (AUC 0.85, 95% CI 0.82-0.89, p=< .0001). A machine learning model (LAT-AI) predicted left atrial appendage thrombus with an AUC of 0.85 (95% CI 0.82-0.89), outperforming LVEF and CHA2DS2-VASc score (P < 0.0001).