The authors built ChronoEvo, a benchmark of 1,709 recommendation-level revisions spanning 12 forms of medical knowledge change, such as strengthened or weakened recommendations, changes in patient eligibility, and withdrawn interventions.
LLMs recognize updated medical guidelines but often fail to apply the revisions
Across 1,709 guideline changes, the authors found a persistent gap between models' awareness of revised guidance and their ability to translate it into clinically correct recommendations.
Academic
Min Zeng · Zaifu Zhan · Kai Yu · Rui Zhang
University of Minnesota
Research Digest··2 min read
The authors introduce ChronoEvo, a benchmark of 1,709 medical guideline revisions, and show that large language models frequently recognize when clinical guidance has changed yet fail to incorporate that change into their recommendations.
§