Advances in Predicting Postoperative Atrial Fibrillation: A Narrative Review of the Current Literature

Postoperative atrial fibrillation (POAF) is a common adverse event after cardiac surgery. It is associated with increased risk of both morbidity and mortality. Predictive modalities that stratify patients vulnerable to POAF have demonstrated both poor efficacy and conflicting results on their predictive ability. This narrative review aims to highlight the research on the existing technological POAF predictive modalities. Left atrium reduced reservoir and contractile strain seemed to be associated with an increased risk of POAF via echocardiography imaging. Likewise, textural features and index epicardial adipose tissue volume were associated with POAF risk determined using CT radiomics and magnetic resonance imaging, respectively. Several machine learning models including support vector machines, deep learning (DL), gradient boosted machine (GBM), logistic regression, and random forest (RF) exhibited effective risk stratification based on area under the curve (AUC). However, their predictive value is less established due to no single model consistently performing well. Our findings suggest that technological predictors of POAF involving the role of imaging and machine learning models may play an important role in predicting POAF risk among patients who underwent cardiac surgery. Integrating these novel modalities into existing means of risk stratification can potentially enhance postoperative outcomes. However, further studies involving wide-scale clinical trials are needed to support their usage.