Predicting atrial fibrillation after cardiac surgery: a scoping review of associated factors and systematic review of existing prediction models

Introduction: Postoperative atrial fibrillation (POAF) is common after cardiac surgery and associated with increased hospital length of stay, patient morbidity and mortality. We aimed to identify factors associated with POAF and evaluate the accuracy of available POAF prediction models. Methods: We screened articles from Ovid MEDLINE(R) and PubMed Central(R) (PMC) and included studies that evaluated risk factors associated with POAF or studies that designed or validated POAF prediction models. We only included studies in cardiac surgical patients with sample size n > 50 and a POAF outcome group > 20. We summarised factors that were associated with POAF and assessed prediction model performance by reviewing reported calibration and discriminative ability. Results: We reviewed 232 studies. Of these, 142 fulfilled the inclusion criteria. Age was frequently found to be associated with POAF, while most other variables showed contradictory findings, or were assessed in few studies. Overall, 15 studies specifically developed and/or validated 12 prediction models. Of these, all showed poor discrimination or absent calibration in predicting POAF in externally validated cohorts. Conclusions: Except for age, reporting of factors associated with POAF is inconsistent and often contradictory. Prediction models have low discrimination, missing calibration statistics, are at risk of bias and show limited clinical applicability. This suggests the need for studies that prospectively collect AF relevant data in large cohorts and then proceed to validate findings in external data sets.

Automated summary

Postoperative atrial fibrillation (POAF) is common after cardiac surgery and associated with increased hospital length of stay, patient morbidity and mortality. The reporting of factors associated with POAF is inconsistent and often contradictory. Existing POAF prediction models have low discrimination, potential bias, and limited clinical applicability. Large cohort studies are needed to prospectively collect AF relevant data and validate the findings in external data sets.