Evaluating Source-Based Large Language Models for Preclinical Dermatology Education: Comparative Study

Large language models (LLMs) have gained increasing popularity in medical education, with evidence supporting their educational value when framed through the lens of cognitive load theory. Source-based LLMs, which explicitly ground responses in user-uploaded material via retrieval-augmented generation algorithms, may offer additional educational value by using student-developed materials to conceptualize new areas of learning within a familiar framework. This has applications for areas like medical education in dermatology, which could benefit from inclusive sources and enhanced education to alleviate health care gaps. However, no prior studies have examined whether the inclusion of student-authored notes alters the response characteristics of a source-based LLM when responding to medical questions.

This study aims to conduct an observational, comparative performance evaluation study assessing the accuracy, response reproducibility, and intermodel response similarity of freely available LLMs on text-only step 1 dermatology questions, and to explore whether providing extensive student-generated notes to a source-based LLM alters these performance characteristics.

In December 2024, 4 LLMs were evaluated: NotebookLM (NLM) with uploaded preclerkship study guides (NLM w/ Notes), NLM with an uploaded blank sheet of paper (NLM w/o Notes), ChatGPT-4o Mini, and Google Gemini 1.5 Flash. Each model completed 3 trials of 121 text-based United States Medical Licensing Examination (USMLE) step 1 dermatology questions from the AMBOSS question bank. They were evaluated for overall majority-consensus accuracy, accuracy by question difficulty, intertrial reproducibility, and agreement in answer choice selection between models. Differences were analyzed through a Cochran Q omnibus test and subsequent pairwise McNemar tests with Benjamini-Hochberg correction. Response reproducibility and intermodel agreement were analyzed through Fleiss κ statistics with 95% CI.

ChatGPT-4o Mini achieved the highest overall majority-consensus accuracy (102/121, 84.3%). NLM w/ Notes demonstrated the highest intertrial reproducibility (Fleiss κ=0.927, 95% CI 0.875-0.978) and strong performance on lower-difficulty questions but comparatively reduced accuracy on higher-difficulty items. NLM w/o Notes exhibited significantly higher omission rates (38/363, 10.5% vs ≤7/363, 1.92% for other models) than other tested LLMs. Sensitivity analysis excluding omissions increased NLM w/o Notes' accuracy from 66.9% (81/121) to 77.8% (77/99), matching NLM w/ Notes' accuracy of 74.4% (90/121). Intermodel agreement was significantly higher between NLM w/ Notes and ChatGPT-4o Mini compared to NLM w/o Notes and Gemini 1.5 Flash.

Provision of student-generated notes substantially increased response reproducibility in a source-based LLM, likely reflecting consistent retrieval of similar source excerpts across trials. However, note-grounding appeared to constrain performance on higher-difficulty questions, suggesting a retrieval-augmented generation algorithm retrieval error when question stems excluded characteristic "keywords" present in lower-difficulty items. The results highlight potential challenges of a student-level, cognitive load theory-grounded educational LLM that must deal with notes not curated by experts, balance source use and internal reasoning, and meaningfully appraise uploaded sources to assess a student's individual learning gaps.