A new framework links narrative medical events to source spans to maintain identity through text-only estimation and structured retrieval. Using the GAVEL LLM judge, the GLM 5.2 multimodal model improved temporal agreement in clinical summaries compared to text-only versions.
HOW THIS AFFECTS YOU
●
researcherThe findings suggest multimodal inputs are critical for maintaining temporal accuracy in longitudinal EHR modeling.
●
healthThis could improve the reliability of automated clinical summaries and treatment-window analysis.