Development and external validation of a pre-operative prediction model for new-onset atrial fibrillation after coronary artery bypass grafting
W Li, K Nabrdalik, T Hrapkowicz, K Irlik, Y Xie, T Gu, A Cole, B Kirmani, D Dhiraj, M Sudell, M Kelson, Y Zheng, Y Meng, G Y LipAbstract
Background
New-onset atrial fibrillation (NOAF) is a frequent complication following coronary artery bypass grafting (CABG) and is associated with prolonged hospitalisation and adverse outcomes. Existing clinical risk scores have shown variable performance in surgical populations. Improved pre-operative risk stratification may facilitate targeted preventive strategies.
Purpose
To develop and externally validate a transportable pre-operative prediction model for NOAF after CABG. As a secondary objective, we assessed whether adding intraoperative variables improved predictive performance.
Methods
Consecutive patients undergoing isolated CABG were identified from a Polish CABG cohort (development set). Patients with prior atrial fibrillation, active endocarditis during index hospitalisation, or concomitant mitral valve repair/replacement were excluded. The primary outcome was NOAF, defined by documented atrial fibrillation during postoperative monitoring during index hospitalisation. Candidate predictors included routinely available pre-operative demographic, clinical, laboratory, and echocardiographic variables. Model discrimination was assessed using cross-validated area under the receiver operating characteristic curve (AUC) and compared with established clinical risk scores (POAF score and CHA₂DS₂-VASc). The final model was refitted on the full development cohort and externally validated in an independent external CABG cohort. In a secondary analysis, intraoperative variables(off-pump CABG status, extracorporeal circulation time, and aortic clamping time) were added to assess incremental predictive value.
Results
The development cohort included 4,810 patients (mean age 65.2 ± 8.6 years; 26.1% female), of whom 17.8% developed NOAF. A penalised logistic regression model with 10 pre-operative predictors demonstrated the highest cross-validated discrimination (AUC 0.651 ± 0.030), outperforming the POAF score (AUC 0.61) and CHA₂DS₂-VASc score (AUC 0.59). In the final model, increasing age (OR 1.06, 95% CI 1.05–1.08), left atrial dimension (OR 1.49, 95% CI 1.25–1.77), heart failure (OR 1.44, 95% CI 1.08–1.91), and male sex (OR 1.29, 95% CI 1.08–1.55) were independently associated with POAF. External validation demonstrated preserved discrimination (AUC 0.692). In secondary analysis, adding intraoperative variables provided modest incremental improvement (internal AUC 0.655±0.029; external AUC 0.697).
Conclusion
A pre-operative prediction model for NOAF after CABG outperformed established clinical risk scores. Discrimination was preserved in external validation. The addition of intraoperative variables provided only modest incremental improvement, supporting the clinical utility of pre-operative risk stratification.Forest Plot for NOAFPerformance and Clinical Utility