TITLE:
Prediction of Mortality in Patients with Atrial Fibrillation: Analysis of the AFRICA Registry
AUTHORS:
Amadou Diabagate, Yazid Hambally Yacouba, Doffou Jérôme Diako, Katienefowa Sekou Koulibaly, Awa Fofana
KEYWORDS:
Atrial Fibrillation, Mortality Prediction, Machine Learning, Clinical Decision Support, AFRICA Registry, Artificial Intelligence
JOURNAL NAME:
Open Journal of Applied Sciences,
Vol.15 No.9,
September
24,
2025
ABSTRACT: Atrial fibrillation (AF) is a leading cardiac arrhythmia associated with elevated mortality risk, particularly in low-resource settings where early risk stratification remains challenging. This study investigates the potential of supervised machine learning models to predict all-cause mortality in patients with AF, using real-world clinical data from Côte d’Ivoire. Six classification algorithms, namely Decision Tree, Random Forest, XGBoost, Support Vector Machine, K-Nearest Neighbors, and Neural Network, were evaluated using key metrics such as accuracy, F1-score, precision, recall, MCC, and AUC-ROC. Among these models, XGBoost achieved the highest overall performance, demonstrating strong calibration, robust predictive capacity, and balanced handling of class imbalance. Random Forest also performed competitively, while the Decision Tree offered a viable trade-off between interpretability and efficiency, making it suitable for clinical deployment in resource-constrained environments. The integration of SHapley Additive exPlanations (SHAP) analysis further enhanced model transparency by identifying key predictors such as the EHRA score, heart failure status, blood pressure, and left atrial diameter—variables aligned with current clinical knowledge. Despite promising results, the study acknowledges limitations, including a modest sample size, single-center design, and absence of external validation. Nevertheless, these findings underscore the feasibility of applying explainable AI methods to support early identification of high-risk AF patients and inform personalized care strategies. This work contributes to the growing body of evidence supporting AI-driven clinical decision-making and highlights the need for further validation studies, integration with real-time workflows, and enhanced model interpretability to foster trust and adoption in diverse healthcare settings.