Do machine learning models accurately predict mortality within six months of hospital discharge in atrial fibrillation patients treated with DOACs?
Machine learning models, particularly LightGBM, can accurately predict six-month mortality in atrial fibrillation patients treated with DOACs, offering a potential improvement over traditional risk scores for personalized care.
Background/Objectives: Atrial fibrillation (AF) is a common arrhythmia linked to increased mortality and significant healthcare burden, especially in the elderly. Direct Oral Anticoagulants (DOACs) are crucial for stroke prevention in AF, offering benefits over traditional vitamin K antagonists. Despite scoring systems like HATCH and CHA2DS2-VASc, their predictive ability for mortality in AF patients is limited. This study aims to use machine learning to predict mortality within six months of hospital discharge in AF patients treated with DOACs. Methods: Using the MIMIC-IV database, data from 6431 AF patients were analyzed. Feature selection was done with the LASSO algorithm. Five machine learning models were built: Logistic Regression, Random Forest, XGBoost, LightGBM, and AdaBoost, using 27 features. The top two models were tested on a separate dataset. SHAP values explained model predictions and feature importance. Results: The best model, LightGBM, achieved an AUC of 0.886, accuracy of 0.862, sensitivity of 0.913, and specificity of 0.859. SHAP values highlighted the importance of length of hospital stay, ICU duration, and comorbidities. The model’s interpretability allows for identifying individual patient risk factors, applicable in clinical practice. Conclusions: This study demonstrates that machine learning models effectively predict mortality in AF patients treated with DOACs, potentially enhancing personalized patient care.
Ledziński et al. (Sun,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: