Evaluating large language models vs residents in cataract and refractive surgery: comparative analysis using the American Academy of Ophthalmology Self-Assessment Program
Bibliographic record
Abstract
PURPOSE: To assess the accuracy of leading large language models (LLMs) in answering cataract and refractive surgery questions, determine whether prompt complexity affects performance, and compare their accuracy with ophthalmology residents. SETTING: Not applicable. DESIGN: Randomized questionnaire-based study using 100 questions from the cataract and refractive surgery section of the Basic and Clinical Science Course (BCSC) Self-Assessment Program. METHODS: 5 LLMs (ChatGPT-4, ChatGPT-4o, Gemini, Gemini Advanced, and Copilot-Precise Mode) were tested. Data from 1983 BCSC Self-Assessment Program users served as a comparison. Each LLM underwent 2 sessions: one with a simple prompt and another with a contextualized prompt. Accuracy was defined as the proportion of correct answers. RESULTS: Using the simple prompt, ChatGPT-4o achieved the highest accuracy at 84% (95% CI 77%-91%), followed by Gemini Advanced at 82%, Copilot at 78%, ChatGPT-4 at 77%, and Gemini at 62% ( P < .05). With the contextualized complex prompt, ChatGPT-4o again led (86%; 95% CI 79%-93%). Performance differences between simple and complex prompts were not statistically significant ( P > .05). Except for Gemini, all models' lower 95% CIs exceeded the 60% passing threshold. The mean resident score was 77% (95% CI 75%-79%). Only ChatGPT-4o significantly ( P = .04) outperformed residents, while Gemini Advanced and Copilot trended higher ( P ∼ .10). CONCLUSIONS: ChatGPT-4o consistently outperformed ChatGPT-4, Gemini, Gemini Advanced, and Copilot, and was the only model to significantly surpass residents. Prompt complexity did not affect LLM performance. All models except Gemini exceeded the 60% accuracy threshold, indicating the potential of LLMs as tools for knowledge assessment in refractive surgery.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.005 | 0.001 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.002 | 0.000 |
| Bibliometrics | 0.001 | 0.002 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.001 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.001 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".