Artificial Intelligence in Patient Education on Osteotomy: An Assessment of Response Reliability
R.E. De Paula, J.H.D.S. Araujo, P.A.V. Maia, N.T. Cavanellas, J.L.R. De Faria, T. LemosObjectives:
To assess the adequacy and clarity of responses provided by large language models (LLMs) to common questions about osteotomy.
Methods:
Ten questions frequently asked by patients and laypeople about osteotomy, including its definition, most common procedures, clinical characteristics, and surgical recovery process, were submitted to an LLM (ChatGPT, version 5.1, on December 10th, 2025). The responses were evaluated by a field specialist and classified for adequacy (excellent, satisfactory, or inadequate) and the need for additional clarification (none, minimal, moderate). The classification of each question and the aspects needing clarification, when applicable, were analyzed.
Results:
Regarding adequacy, 7 out of 10 responses received an "excellent" rating, while 3 were classified as "satisfactory." In responses rated as "satisfactory," a moderate need for clarification was identified, due to inaccuracies, information omissions, and/or superficiality. For example, in Question 1 ("What is knee osteotomy to correct alignment?"), the LLM incorrectly stated that high tibial osteotomy procedures involve fixation with plate and screws only, ignored the possibility of corrections for both varus and valgus in both bones, and mistakenly indicated that the deformity must always be reducible.
Conclusion:
When answering questions about osteotomy, the analyzed LLM provided 70% of responses considered excellent, with a moderate need for clarification in some questions. Supervised use of this type of tool remains important.