Methodological Considerations in Evaluating Large Language Models for Anatomy Education

Recent studies evaluating large language models (LLMs) in anatomy and medical education have provided valuable evidence regarding their potential role in examination-based knowledge assessment, clinical education, and healthcare-related decision support. Assessing large language models with non-English anatomy questions is important, given the predominance of English-language datasets in the current literature. However, comparative LLM studies also raise broader methodological concerns that may affect reproducibility, fairness, and interpretation. In this letter, we highlight three major categories of methodological factors to consider in future LLM evaluation studies: technical factors, session-related factors, and experimental design factors. Technical factors include model version, interface type, access date, browser, device, operating system, internet connection, and timing method. Session-related factors include memory, personalization, chat history, custom instructions, and the use of sequential versus independent testing sessions. Experimental design factors include dataset source, question selection, exact prompt wording, input language, response format, number of attempts, question order, scoring procedures, expert evaluation, and inter-rater agreement. We suggest that future LLM comparison studies in anatomy, education, and clinical fields should provide detailed methodological reporting and align with emerging frameworks such as TRIPOD-LLM, CONSORT-AI, and MI-CLEAR-LLM. Improved transparency in these areas will strengthen reproducibility and comparability across studies.