Machine learning model for predicting in-hospital cardiac mortality among atrial fibrillation patients.

This study developed and validated a machine learning (ML) model to predict in-hospital cardiac mortality in 18,727 atrial fibrillation (AF) patients using electronic medical record data. Four ML algorithms-random forest, extreme gradient boosting (XGBoost), deep neural network, and logistic regression-were applied to 79 clinical variables, including demographics, vital signs, comorbidities, lifestyle factors, and laboratory parameters. The XGBoost model achieved the best performance, with an area under the curve of 0.964 ± 0.014 in the training set and 0.932 ± 0.057 in the validation set, alongside precision, accuracy, and recall of 0.909 ± 0.021, 0.910 ± 0.021, and 0.897 ± 0.038, respectively. Shapley Additive Explanations identified key predictors such as thyroid function indices (e.g., total triiodothyronine, total thyroxine), procalcitonin, N-terminal pro-brain natriuretic peptide, and international normalized ratio. This interpretable model holds promise for improving early risk stratification and individualized care in AF patients. Prospective, multi-center validation is needed to confirm its generalizability.