Prediction Models for Postoperative Atrial Fibrillation After Cardiac Surgery: A Systematic Review and Critical...

Background/Objectives: Postoperative atrial fibrillation (POAF) is a frequent complication after cardiac surgery and is associated with increased morbidity, prolonged hospitalization, and higher healthcare costs. Numerous multivariable prediction models have been developed to estimate individual risk; however, their methodological robustness, validation status, and clinical transportability remain uncertain. This systematic review aimed to critically evaluate the methodological quality, validation strategies, and predictive performance of multivariable prediction models developed to estimate the risk of postoperative atrial fibrillation (POAF) after cardiac surgery. Methods: In accordance with PRISMA 2020 guidelines, we conducted a comprehensive search of PubMed, Scopus, Web of Science, and Embase from inception to July 2025. Studies that developed or externally validated multivariable prediction models for POAF in adult patients undergoing cardiac surgery were eligible. Data extraction was performed using the CHARMS checklist, and methodological quality was assessed with PROBAST. Model performance was summarized descriptively, focusing on discrimination (C-statistic/AUC), calibration reporting, and validation strategies. Results: A total of 39 studies were included. Most models were based on logistic regression, whereas a minority employed Cox regression or machine learning techniques. Reported discrimination ranged from 0.60 to 0.98, demonstrating substantial heterogeneity in predictive performance. Calibration was inconsistently reported. Six studies performed external validation. According to PROBAST, 32 of 39 studies (82%) were rated at high risk of bias, predominantly within the analysis domain due to inadequate handling of overfitting, insufficient events-per-variable ratios, and limited validation procedures. Conclusions: Existing prediction models for POAF show variable discrimination but are frequently limited by high risk of bias, inadequate validation, and incomplete calibration assessment, thereby restricting their clinical applicability. Future research should prioritize rigorous external validation, transparent reporting in accordance with TRIPOD recommendations, and methodological strategies that enhance model generalizability and transportability across diverse surgical populations.