Tag
This feasibility study compares a standalone LLM with a pre-specified agentic pipeline for explaining ICU mortality predictions, finding that the agentic approach improves guideline grounding and patient-specific detail but requires attribution-based checks for safety.