Künstliche Intelligenz in Anästhesiologie und Intensivmedizin : AI × AI

Introduction: Artificial intelligence (AI) has the potential to improve decision-making in anesthesiology and intensive care, yet its clinical integration remains limited. Although large volumes of perioperative and critical care data are routinely collected, these data are rarely leveraged to support individualized risk assessment and prevention of adverse outcomes. Materials & Methods: This dissertation combines a scoping review and an analysis of methodological challenges. With two retrospective studies based on electronic health record data. The review examines current applications of machine learning in anesthesiology and intensive care, with a focus on patient blood management. Models were developed to predict massive perioperative red blood cell transfusion and deterioration of renal function following transfusion. Model performance was assessed using discrimination and predictive value metrics, and interpretability was evaluated using feature importance and Shapley’s Additive Explanations (SHAP)-based analyses. Results: Models based on structured perioperative variables showed strong predictive accuracy, identifying patients at high or low risk for transfusion and renal complications. High discrimination was achieved using a concise set of clinically meaningful features rather than complex, high-dimensional inputs. Model explanations corresponded with established risk factors, confirming clinical relevance and practical usability. Conclusion: AI-based prediction can complement clinician expertise, improve perioperative planning, and support safer, more efficient care. To translate these tools into routine use, prospective validation, standardized data models, and clinician education are required, with attention to transparency, accountability, and equity in implementation.