Abstract 12030: Predicting Atrial Fibrillation in Critically Ill Patients: A Machine Learning Approach

Abstract

Introduction: Atrial fibrillation (AF) is the most prevalent arrhythmia in critically ill patients, with significant implications for morbidity and mortality. Early prediction of AF risk is crucial for proactive management. This study aims to develop and validate a machine learning-based risk prediction model for AF in intensive care unit (ICU) patients. Methods: Using the Medical Information Mart for Intensive Care (MIMIC-IV) database, we conducted a retrospective cohort study that included adult ICU patients and collected 47 clinical and laboratory variables. The primary outcome variable was the development of AF within the first 48 hours of ICU admission. Various machine learning models were evaluated for their performance in predicting AF, and the best-performing model was further optimized using hyperparameter tuning. A compact model, incorporating only 15 variables and two novel features, was also developed to facilitate potential clinical implementation. The first new feature, ‘ older septic ’ indicated patients aged 70 or older with sepsis, capturing the combined effect of age and sepsis as a risk factor for AF. The second new feature, ‘ cardiac risk score ’ was a composite score considering patient’s pre-existing cardiac risk factors. Model interpretability was assessed using Shapley Additive exPlanations (SHAP) analysis. Results: The cohort comprised 46,266 ICU patients, with 4.6% developing AF within 48 hours of admission. The CatBoost Classifier model demonstrated superior performance, achieving an area under the receiver operating characteristic curve (AUC) of 0.850 (95% CI: 0.811-0.889) on the test set. The compact model with new features yielded an AUC of 0.820 (95% CI: 0.782-0.858), which is a comparable performance using a fraction of the variables. SHAP analysis highlighted the importance of the newly created features as key predictors, indicating the significance of age, sepsis, and cardiac risk factors in AF development. Conclusions: This study underscores the potential of machine learning models in predicting AF development in ICU patients. The compact model, with a satisfactory AUC, can be a valuable tool for identifying high-risk patients, enabling timely interventions to mitigate the adverse outcomes associated with AF.