Prediction of In-Hospital Atrial Fibrillation After Acute Myocardial Infarction

Abstract

Atrial fibrillation (AF) is a relatively frequent complication of acute myocardial infarction (AMI). While AF prediction has been extensively studied, the identification of risk factors for early, new-onset AF (NOAF) after AMI in the intensive cardiac care unit (ICCU) remains less explored. Specifically, to our knowledge, there are no reported attempts at predicting in-hospital NOAF after AMI using machine learning. In this study, we developed a machine learning model to predict in-hospital NOAF following AMI. The dataset used for model development included 2445 consecutive AMI patients admitted to the ICCU of Centro Cardiologico Monzino, out of which 241 (9.9%) developed NOAF prior to ICCU discharge. Fifty-six features encompassing demographic and clinical variables were retrospectively collected and analysed. Several data balancing, feature selection and classification techniques were evaluated and compared by means of area under the ROC curve (AUROC) through nested cross-validation. The best-performing model combined an undersampling step, based on the Edited Nearest Neighbors algorithm, a mutual-information-based feature selection and a logistic regression model. The model achieved an AUROC of 0.765 (95% CI: 0.732 - 0.795), exploiting both known and previously unreported markers.