LLMs overorder ECGs and thyroid ultrasounds but rarely withhold care, Annals study finds
EricTopol · x · 2026-10-20
A benchmark study in Annals of Internal Medicine tested six commercial LLMs (via OpenAI, Anthropic, and Google production APIs, Feb–Mar 2026) on 225 clinical scenarios built from 15 Choosing Wisely recommendations, yielding 4,050 responses.
- Overuse rate was 5.4% (95% CI 0.6%–12.2%), driven almost entirely by two items: thyroid ultrasound for incidental subcentimeter nodules (42.8%) and routine ECGs in low-risk adults (28.3%); the other five restraint-appropriate items saw 0–3.3% overuse.
- False restraint was rare at 0.6% — LLMs almost never withheld treatment when action was appropriate, with no false restraint in 11 of 15 recommendations.
- Limitations: single prompts, benchmark rather than real-world use, LLM-judge classification with physician adjudication.
More from AGI Musings
- Economist: don't trust high-confidence predictions about the future of the labor market — robseamans · 2026-10-20
- Economist Guy Berger: AI won't cause persistent mass unemployment, but disruption is coming — robseamans · 2026-10-20
- Would you trade 20 extra healthy years for 2x humanity's extinction risk? — inductionheads · 2026-10-20
- Stardock CEO: local AI will dominate most AI work in 18 months, datacenters pointless — draginol · 2026-10-20
- Fields Medalist John Pardon: "The AI understands. 100%" — it now catches subtle errors in his notes — thesaraharminta · 2026-10-20
- Aaron Levie says agent swarms will unlock compute-bound work, but humans can't review them — dosco · 2026-10-20