Scaling Clinical Judgment to Evaluate Medical AI
Thomas A. Buckley and colleagues at Harvard Medical School fine-tune PrecepTron, a 32B judge trained with LoRA on a small number of physician-scored examples, and release GRAND-ROUNDS, a benchmark of physician scores.
Ask this paper
Benchmark: GRAND-ROUNDS holds 9,217 scores from 11 physicians over 5,250 response-rubric entries from seven published studies, covering responses from 160 clinicians and 9 AI models.
Frontier judges disagree: Frontier LLMs used as judges often disagree with physicians and with each other.
Small-data fine-tune: Fine-tuning a 32B model on a small number of physician examples gives physician-level consistent scoring across tasks, and it can run locally inside a health system.
Reproductions: PrecepTron reproduces headline findings from five studies in JAMA, Science and Nature Medicine without new human grading.
New question: It also measures frontier diagnostic accuracy when cases are revealed piecemeal, down to token by token, which would be impractical with human grading.
Abstract
Blinded physician evaluation has been considered by many to be the gold standard for assessing clinical reasoning in large language models (LLMs). This is difficult to scale; thus, prior studies typically rely on small physician panels, often from a single institution or specialty, which both limits the scientific questions investigated and makes it unclear whether findings would be reproduced with a different set of evaluators. To more rigorously and scalably study clinical reasoning in AI models, here we introduce PrecepTron, an LLM fine-tuned for physician-level evaluation of open-ended responses. PrecepTron was trained using low-rank adaptation (LoRA) of a 32-billion-parameter model on a small number of physician examples. We also release GRAND-ROUNDS, a new large-scale physician-annotated benchmark of 9,217 scores by 11 physicians across seven studies. We show that frontier LLMs in typical "LLM-as-a-judge" approaches often disagree with physicians and with each other, but fine-tuning PrecepTron on a small number of cases enables physician-level consistent scoring across tasks. We use PrecepTron to reproduce headline findings from five influential studies assessing LLMs for clinical care in JAMA, Science, and Nature Medicine without new human grading. Using PrecepTron, we then pose new questions about how LLMs reason in medicine that would have been infeasible with human grading alone, including measuring the diagnostic accuracy of frontier LLMs when clinical cases are provided piecemeal, even token by token. Together, PrecepTron and GRAND-ROUNDS provide a foundation for reproducible, large-scale study of how LLMs reason in medicine. All code, data, and labels are made freely available for researchers.