Beyond EuroSCORE II: is artificial intelligence ready to redefine risk stratification in cardiothoracic surgery?

Risk stratification is central to contemporary cardiothoracic surgical practice, guiding patient selection, perioperative planning, informed consent, and benchmarking of outcomes across institutions. Established models such as European System for Cardiac Operative Risk Evaluation II and the Society of Thoracic Surgeons risk score remain widely used because they are validated, interpretable, and embedded within routine clinical workflows. However, their static structure and reliance on predefined variables may limit performance in increasingly complex and heterogeneous surgical populations. Artificial intelligence (AI) and machine learning have emerged as promising adjuncts capable of analyzing nonlinear relationships and high-dimensional data, with several studies reporting improved predictive discrimination in selected cohorts. Despite this potential, important barriers remain, including limited interpretability, risks of dataset bias, inconsistent external validation, and uncertainty regarding real-world implementation. Current evidence supports augmentation rather than replacement of traditional models. A practical pathway forward is the development of hybrid frameworks in which conventional scores provide baseline risk estimation while AI contributes individualized insights from dynamic clinical data. Successful translation will depend on prospective validation, seamless integration into electronic health record systems, clinician-friendly decision-support interfaces, and continued surgeon oversight. The future of cardiothoracic risk prediction is, therefore, likely to combine established clinical tools with responsible AI deployment to improve precision, workflow efficiency, and patient-centered care.