Generative Artificial Intelligence for Clinical Anesthesia Decision Support

Abstract

Generative artificial intelligence, built on large language model architectures, has moved rapidly from novelty into systematic evaluation across preoperative, intraoperative, and postoperative anesthesia practice, offering a fundamentally different capability than the discriminative machine-learning models anesthesiology has used for hemodynamic and risk prediction over the past decade. This review synthesizes the emerging literature on generative AI applications in clinical anesthesia decision support, including preoperative anesthetic plan drafting and risk stratification, patient-facing communication, intraoperative documentation support, and simulated case generation for trainee education. Early case-control and randomized evaluations show that generative models can produce plausible, often accurate preoperative plans and explanations, and can generate measurable efficiency gains in perioperative documentation, but performance is inconsistent for complex comorbidity profiles and numerical dosing tasks, and equity and bias evaluation remains largely aspirational. We argue that the more rigorously validated discriminative models, particularly for intraoperative hypotension prediction, remain the appropriate tool for narrow numerical tasks, while generative AI's clearest near-term value lies in documentation, communication, and education rather than autonomous clinical recommendation. We conclude by outlining the evidentiary infrastructure, including multi-center prospective validation and demographic-stratified performance reporting, that the field still needs before generative AI can be safely integrated as more than an assistant to an accountable clinician.