Validation of a predictive model of the left atrium appendage thrombosis in patients with nonvalvular atrial fibrillation

Abstract

Aim: To test the previously developed predictive model on a test set, including patients with non-valvular atrial fibrillation (AF).Material and Methods. A predictive model was previously developed on a training set of 638 patients with non-valvular AF. The test set included 175 randomly selected patients with non-valvular AF hospitalized at the Tyumen Cardiology Research Center for radiofrequency ablation (RFA) or electrical cardioversion in 2018-2021. The quality of the predictive model was assessed using ROC analysis.Results. Arterial hypertension, coronary heart disease, congestive chronic heart failure, and persistent AF were more common in patients of both sets with left atrial appendage (LAA) thrombosis. Patients of the training and test sets with LAA thrombus had more pronounced structural changes in the heart cavities and similar changes in the geometry of the heart: normal left ventricle (LV) geometry was less common and eccentric LV hypertrophy was more common. According to the results of a previous retrospective analysis of the data, independent predictors of LAA thrombosis were persistent type of AF, left atrium size, and eccentric LV hypertrophy. Based on the data gathered, a predictive model LAA thrombosis was developed as an equation that includes 3 variables. The cut-off point for calculating the probability of LAA thrombosis is 0.07. Applying this model on a test set showed the good quality of the model: the area under the curve obtained using ROC analysis was 0.750 (p < 0.001). At the same time, the sensitivity and specificity of this model for the detection of LAA thrombosis were 72.3% and 71%, respectively.Conclusion. The evaluation of the quality of the LAA thrombosis predictive model developed on the training set confirmed its good quality on a similar test set of patients with non-valvular AF hospitalized for RFA or electrical cardioversion.