Researchers find that eXtreme Gradient Boosting (XGBoost) machine learning (ML) model is useful in predicting clinical outcomes in patients undergoing left atrial appendage occlusion (LAAO). Kamil F. Faridi, MD, MSc, of the Yale School of Medicine, New Haven, and colleagues, reported the corresponding data in a manuscript published online Monday and in the February print issue of JACC: Advances. Rare outcomes are difficult to predict with standard cardiovascular risk models, though risk models help with prognosis, clinical decision-making and quality of disease management. ML is becoming a common technique for risk prediction in cardiology. ML may be able to better predict adverse outcomes in patients undergoing LAAO. Investigators in this study used logistic regression (LR), least absolute shrinkage and selection operator (LASSO) and XGBoost to predict major adverse events (MAEs) in patients who underwent transcatheter LAAO when patients were in the hospital. Data were taken from the National Cardiovascular Data Registry LAAO Registry. The ML model was created using randomly selected 70% development and 30% validation cohorts plus assessment with 16 variables taken from the previous LAAO risk model, as well as another set of 51 variables. Some risk model variables included age, sex, body mass index (BMI), hypertension and vascular disease. A composite of all in-hospital MAE was the primary outcome in this study, including death, cardiac arrest, myocardial infarction, pericardial effusion requiring intervention, systemic arterial embolism, device embolization, adjudicated stroke, transient ischemic attack, intracranial hemorrhage, major vascular complications and major bleeding. Data were collected from 81,703 LAAO procedures (mean age of patients=76.3 years, 41.1% female; 93.3% White, 4.1% Black) and included in this study. The overall MAE rate was 1.39% as a composite (individual event rates ranges from 0.02%-1.13%). Original model variables allowed the XGBoost to work best (validation area under the receiver operating characteristic curve [AUC]=0.648, 95% confidence interval [CI]=0.626-0.670; LR=0.630, 95% CI=0.608-0.642); LASSO=0.638, 95% CI=0.626-0.670). In the expanded variables group, performance was better in XGBoost (AUC=0.653) compared with LASSO (AUC=0.644) in predicting MAE. LR did not perform well here (AUC=0.515). All models did not perform well for events that happened infrequently. For individual events, XGBoost performed best, especially with the expanded variables set. This was not applicable for rare events. XGBoost best predicted mortality compared with all other models. Overall, the XGBoost ML model was great at predicting the composite in-hospital MAE and most individual events. XGBoost was better at predicting rare events, but rare outcomes were still inconsistently predicted. Source: Faridi KF, Mortazavi BJ, Huang S, et al. Using machine learning to predict adverse events in patients undergoing transcatheter left atrial appendage occlusion. JACC Adv. 2026;5. doi.org/10.1016/j.jacadv.2025.102540 Image Credit: Елена Бутусова – stock.adobe.com