Comparison of preoperative education by artificial intelligence versus traditional physicians in perioperative management of urolithiasis surgery: a prospective single-blind randomized controlled trial conducted in China

Background Preoperative anxiety is common in patients awaiting urolithiasis surgery. Although adequate preoperative education can reduce anxiety and improve outcomes, time and resource constraints often limit the depth and personalization of such interventions. Recent advances in AI, particularly large language models like GPT-4o and Ernie Bot, offer potential tools to supplement traditional education. However, their comparative effectiveness remains unclear. Purpose This randomized controlled trial compared the effectiveness of AI-based preoperative education (ChatGPT, Ernie Bot) to that provided by attending urologists in reducing anxiety, enhancing satisfaction, and improving information quality. Methods Seventy-five adult patients scheduled for urolithiasis surgery were randomly assigned (1:1:1) to ChatGPT, Ernie Bot, or urologist-led education. All received a standardized consultation 12 h preoperatively. The AI groups then had a 30 min Q&A with their assigned AI, and the physician group received an additional face-to-face session. Anxiety was measured via the State–Trait Anxiety Inventory (STAI) at baseline, immediately post-intervention, 3 h pre-op, and 3 h post-op. Secondary outcomes included patient satisfaction, query count, evaluation of information quality (safety, accuracy, empathy, readability, detail), and postoperative (3 h) pain (VAS). Results All participants completed the study. Anxiety decreased significantly after intervention in all groups ( p < 0.05). The physician group achieved the greatest anxiety reduction, followed by ChatGPT and then Ernie Bot. At 3 h pre-op, physician and ChatGPT groups maintained lower anxiety, while Ernie Bot showed a non-significant rebound. Postoperatively, all groups had significantly lower anxiety than baseline. ChatGPT excelled in empathy, readability, and detail, and elicited twice as many patient questions. Satisfaction was high in both the physician and ChatGPT groups. Ernie Bot’s responses, though safe and accurate, were conservative and less detailed, leading to fewer inquiries and lower satisfaction. Postoperative pain was lowest in the physician group, followed by ChatGPT ( p < 0.05). Conclusion AI-assisted education, particularly via ChatGPT, can effectively reduce preoperative anxiety and improve patient engagement, though not to the level of physician-led education. Ernie Bot showed modest benefits. Further refinement of AI may enhance its role as a supplemental educational tool.