Auditable Clinical Support with Decision Agents

Impact 7 arXiv cs.AI

Summary

  • Researchers introduced Decision2Care, an auditable clinical decision agent that operates without generating free text.
  • The system assigns probabilities to permitted electronic health record actions and uses fixed rules to execute the most probable ones.
  • Evaluated on 546 emergency admissions, Decision2Care achieved an 86.5% diagnostic accuracy while requiring a median of 2.2 seconds per work-up.
  • The agent tracks considered options, probabilities, and execution rules at each step to provide transparent clinical audits.

Why it matters

Decision2Care offers a highly efficient, text-free approach to clinical support that enhances auditability and execution speed compared to traditional free-text LLM agents.

Who should care