CONGRESS ABSTRACT

Background Risk stratification in cardiac surgery is of utmost importance for informed consent and for the heart team decision. While EuroSCORE II is widely used, it overlooks intraoperative factors such as ischemic and reperfusion times that can substantially affect outcomes. This study evaluates a Bayesian ensemble approach that integrates operative duration variables to refine perioperative risk assessment. Methods We retrospectively analyzed over 12,000 adult cardiac surgery patients undergoing cardiopulmonary bypass. The model incorporated EuroSCORE II alongside aortic cross-clamp time and cardiopulmonary bypass time, providing risk estimates that adapt dynamically to intraoperative conditions. Performance was assessed through discrimination (area under the receiver operating characteristic curve), calibration (expected-to-observed ratios), and clinical utility (decision curve analysis). Risk visualization employed stratified surface plots across EuroSCORE II quartiles. Results Overall mortality was 4.8% (n = 594). The Bayesian ensemble achieved superior discrimination (AUC 0.815, 95% CI: 0.795-0.835) compared to both a simple GAM (0.805, 95% CI: 0.784-0.825) and EuroSCORE II alone (0.760, 95% CI: 0.740- 0.780). The ensemble demonstrated excellent calibration (expected-to-observed ratio = 1.000) and superior net benefit across all clinically relevant risk thresholds. For patients already at high risk (based on EuroSCORE II), longer operative times had a much stronger negative impact on outcomes than for low-risk patients. The operative time affected patients differently depending on their baseline risk. Conclusion By integrating operative time factors into established risk prediction, the Bayesian ensemble approach provides a clinically meaningful refinement of mortality estimation in cardiac surgery. This enables surgeons to better anticipate risk during complex or prolonged procedures, tailor intraoperative strategies, and enhance discussions within heart teams. The framework is readily applicable to daily practice, using routinely collected surgical variables.