Prospective evaluation of a large language model clinical decision support system in the emergency department.

A large language model-based clinical decision support system in the emergency department showed declining clinician engagement and no significant impact on patient length of stay over 4 weeks.

• Why it matters: Limited prospective evidence exists for AI-based decision support in emergency settings, and understanding barriers to sustained use is crucial for safe integration.
• What they did: Researchers conducted a DECIDE-AI stage 1 evaluation of SHAKED with 1,138 patients across two units, measuring adoption, clinician preferences, safety, and workflow metrics.
• The result: Clinician engagement decreased from 68% to 30% due to workload sensitivity, with no adverse events and a non-significant trend toward shorter consultation times, indicating that engagement, not accuracy, may limit AI effectiveness.