Artificial Intelligence for Perioperative Risk Prediction and Prevention in Cardiac Surgery: A Narrative Review and Proposed Conceptual Framework.

Cardiac surgery remains a high-risk, resource-intensive domain in which perioperative complications significantly influence clinical outcomes, institutional performance, and healthcare expenditure. Despite advances in technique and protocol standardization, contemporary perioperative management largely relies on static risk stratification and reactive quality assessment. This narrative review synthesizes the current evidence on artificial intelligence (AI) and machine learning for perioperative risk prediction in cardiac surgery, spanning acute kidney injury, mortality, prolonged mechanical ventilation, postoperative atrial fibrillation, and intensive care unit deterioration, and critically appraises the methodological limitations, validation gaps, and fairness concerns that constrain clinical translation. Across these applications, predictive models have demonstrated incremental discrimination over conventional risk scores, yet remain predominantly endpoint-specific, single-institution, and disconnected from prospective clinical implementation. Building on this evidence, we propose Preventive Cardiovascular Intelligence (PCInt) as one possible organizing framework that integrates predictive analytics, dynamic risk trajectory modeling, and structured quality improvement methodologies, and we outline how such a framework might be operationalized across the surgical lifecycle. PCInt is presented as a conceptual proposal requiring prospective validation rather than as a validated system. We conclude by discussing implementation barriers, regulatory and ethical considerations, and priorities for future research toward anticipatory, value-based perioperative cardiovascular care.