Tag
This paper introduces an evidence-driven framework for automatic hospital discharge summarization using Abstract Meaning Representation and semantic graphs, aiming to reduce hallucinations in LLMs and ensure traceability in clinical documentation.
This paper characterizes the lost-in-the-middle effect in clinical LLM applications for EHR processing and introduces Query-Conditioned Clinical Suppression (QCCS) to mitigate it, demonstrating improved accuracy over existing retrieval methods.