LLM clinical decision support deployed in live ER: adoption fell from 68% to 30% over 4 weeks despite 99/100 expert-rated outputs
dvir_a · x · 2026-09-04
A team publishing in Nature Medicine prospectively evaluated SHAKED, an LLM-based clinical decision support system, in a tertiary emergency department over 4 weeks with 1,138 patients (DECIDE-AI stage 1).
Key findings:
- Clinical adoption declined from 68% to 30%, driven by workload-sensitive disengagement (OR = 0.72 per shift hour)
- Physicians favored SHAKED for radiology consultations (OR = 2.98)
- No adverse events detected; expert review rated 99 of 100 sampled outputs favorably
The study exposes the core deployment tension: high output quality doesn't guarantee physicians will integrate the tool into busy workflows.
More from Apps
- Claude mobile app has a hidden Memory toggle in Settings — here's how to enable it — HankYeomans · 2026-09-04
- Immunologist builds research-grade flow cytometry software with GPT-6 Astra, cancels all commercial subscriptions — DeryaTR_ · 2026-09-04
- AI agent churns out magazine spreads for $0.50 vs $2,500 — sell them at $500 a pop — NathanWilbanks_ · 2026-09-04
- Wispr Flow hits $8.3M MRR with a 250-creator short-video growth network — FinanceYF5 · 2026-09-04
- Redditor uses Opus 5 to code a drop-in Windows tool that cleans Excel files — 5_Dollars_Of_BayLeaf · 2026-09-04
- Dev builds a small tool to query multiple AIs at once and compare answers side by side — nankezhishi · 2026-09-04