Turn on cookies to sign in
Signing in saves your progress to your Curie account. We can only do that with cookies on — turn them on to continue.
When it comes to deciding whether to biopsy a thyroid nodule, AI models are all over the map — and so are human specialists. A new study found that large language models (LLMs) recommended biopsy in 46%–74% of cases, while clinicians landed at 58%. The takeaway: LLMs can be a helpful reference tool, but shouldn't have the final say in clinical decisions.
Deciding whether to biopsy a thyroid nodule is genuinely complex — for both humans and AI. A new study published in The Journal of Clinical Endocrinology & Metabolism tested several leading large language models (LLMs) against 94 thyroid cancer specialists using 100 hypothetical clinical vignettes. The result? Significant variability across the board.
Clinicians recommended biopsy in 58% of cases, while LLM recommendations ranged from 46% (Claude Sonnet 3.5) to 74% (Llama 3.2-90B). Both groups agreed on the basics — larger nodules and higher-risk features prompted more biopsy recommendations — but diverged on edge cases. Some LLMs disagreed with unanimous clinician decisions, and 8 of 23 discordant LLM responses were flagged as seriously flawed by expert reviewers.
By the Numbers:
Why it matters: As AI tools inch closer to clinical workflows, this study is a timely reminder that LLMs aren't ready to be the final word on nuanced medical decisions. They can help clinicians look up guidelines or think through a case — but for ambiguous thyroid nodules, human judgment still needs to lead.
