Efficacy and Efficiency of Artificial Intelligence-based Preoperative Anesthesia Evaluation: A Randomized Controlled Non-Inferiority Study

Study Objective: To evaluate the clinical effectiveness and feasibility of a novel AI-assisted anesthesia evaluation system compared with conventional preoperative anesthesia assessments. Design: Single-center, randomized, parallel-group, open-label, non-inferiority trial. Patients: 600 adult patients scheduled for elective non-cardiac surgery under general anesthesia were randomly assigned to AI-assisted assessment group or traditional in-person anesthesia evaluation group. Measurements: The primary outcome was the accuracy of ASA (American Society of Anesthesiologists) physical status classification. Secondary outcomes included the quality and completeness of medical history documentation, assessment duration, and patient satisfaction. Main Results: The AI-assisted group achieved non-inferior accuracy in ASA classification compared to the traditional group (93.3% [280/300] vs. 91.7% [275/300], P = 0.033). The difference in accuracy was 1.6% (95% CI: -2.6% to 5.9%), with the lower bound above the predefined non-inferiority margin of -5%. The AI system demonstrated significantly higher documentation quality, including fewer missing items (4.3% vs. 21.7%) and incorrect entries (7.0% vs. 18.0%). Assessment time was markedly shorter in the AI group (3.0 [2.0-5.0] vs. 7.0 [6.0-8.0] minutes, P < 0.001), and overall patient satisfaction was significantly higher (87.3% [262/300] vs. 76.0% [228/300], P < 0.01). Conclusions: The AI-assisted anesthesia evaluation system was non-inferior to traditional assessment in ASA classification accuracy while offering superior efficiency, documentation quality, and patient satisfaction. This AI-based approach may represent a scalable and effective alternative for preoperative anesthesia evaluations across diverse clinical settings. Registry: chictr.org.cn, TRN: ChiCTR2400086869, Registration date: 12 July 2024. Retrospectively registered