Large Language Model–Assisted Training And Objective Assessment For Smile Pro With The Visumax 800

To evaluate the effect of a large language model (LLM)–based interactive training system (ChatGPT-5.1 Thinking; OpenAI, San Francisco, California) on surgical performance in Keratorefractive Lenticule Extraction (KLEx) platform (SMILE Pro, VisuMax 800; Carl Zeiss Meditec, Jena, Germany) compared with conventional training alone in novice surgeons.

 Prospective, randomized, controlled, parallel-group clinical trial.

Eligible surgeons were randomly assigned (1:1) to receive either standard SMILE Pro training (didactic lectures, instructional videos, and at least one proctored wet-lab session) or the same program supplemented with access to an LLM-based tutor (ChatGPT-5.1 Thinking). The tutor provided step-by-step explanations of the SMILE Pro workflow, interactive question-and-answer sessions, case-based scenarios, checklists for preoperative planning, and brief formative quizzes with immediate feedback. Before and after the training period, all surgeons completed a standardized theoretical examination and a structured assessment of SMILE Pro technique using wet-lab or model-eye procedures. 

A total of 24 surgeons were randomized (12 to the LLM group, 12 to the control group); baseline demographic characteristics, prior refractive surgery exposure, and theoretical scores were similar between groups (all P > 0.05). After training, mean step-based performance scores were higher in the LLM group than in the control group (92.58 ± 8.9 vs 88.68 ± 9.8; P < 0.05), and a greater proportion of surgeons in the LLM met the predefined competency threshold 83.3% vs 75%; P < 0.05). Improvement in theoretical knowledge from baseline was greater in the LLM group (Δscore 92.58 ± 8.9 vs 88.68 ± 9.8; P < 0.05). 

In novice SMILE Pro surgeons, an interactive training program based on a large language model (ChatGPT-5.1 Thinking) resulted in superior step-based surgical performance and greater gains in theoretical knowledge compared with conventional training alone, without compromising early patient outcomes. Integration of an LLM-driven tutor into a standardized procedural framework may offer a scalable, data-informed approach to refractive surgery education that complements emerging AI-based methods for surgical supervision and assessment.