Improving ASA-PS classification accuracy using private-deployment-compatible large language models: a multilingual evaluation

The study evaluates the use of advanced large language models to improve the consistency of ASA-PS classification in preoperative assessment. The most recent models achieved high accuracy and results comparable to expert evaluations. However, some high-risk cases, particularly ASA IV, were underrepresented. Therefore, AI may support clinical judgement, but it should be combined with specific safety checks and validated on real-world clinical data.