184P Comprehensive comparison of on-premise large language models (LLMs) for data abstraction in rare cancers from unorganized non-English medical documentation

Abstract

LLMs have proven to be extremely powerful tools for information extraction across multiple fields. In the medical domain, however, most available benchmarking data focus on cloud-based models and carefully curated text sources, such as radiology reports. This raises privacy concerns and requires additional preselection of clinical notes. In this study, we evaluated open-source, on-premise LLMs for automated data abstraction from free-text records derived from electronic health records (EHRs) of patients with rare bone sarcomas.

Ähnliche Arbeiten