Machine Learning Approaches to Predict Major Adverse Cardiovascular Events in Atrial Fibrillation

Abstract

The increasing prevalence of atrial fibrillation (AF) and its association with Major Adverse Cardiovascular Events (MACE) presents challenges in early identification and treatment. Although existing risk factors, biomarkers, genetic variants, and imaging parameters predict MACE, emerging factors may be more decisive. Artificial intelligence and machine learning techniques (ML) offer a promising avenue for more effective AF evolution prediction. Five ML models were developed to obtain predictors of MACE in AF patients. Two-thirds of the data were used for training, employing diverse approaches and optimizing to minimize prediction errors, while the remaining third was reserved for testing and validation. AdaBoost emerged as the top-performing model (accuracy: 0.9999; recall: 1; F1 score: 0.9997). Noteworthy features influencing predictions included the Charlson Comorbidity Index (CCI), diabetes mellitus, cancer, the Wells scale, and CHA2DS2-VASc, with specific associations identified. Elevated MACE risk was observed, with a CCI score exceeding 2.67 ± 1.31 (p < 0.001), CHA2DS2-VASc score of 4.62 ± 1.02 (p < 0.001), and an intermediate-risk Wells scale classification. Overall, the AdaBoost ML offers an alternative predictive approach to facilitate the early identification of MACE risk in the assessment of patients with AF.