← DecisionIndex home

Draft Medical care · orthopaedics

Orthopaedic AI: what has the evidence actually published?

Type a question. See every matching article across the JBJS journals by level of evidence, newest first, and where the evidence runs out. Paste what an AI told you, and each citation and quote it gives is checked against the same record by fixed rules.

Never type patient information. The question goes from your browser straight to PubMed; nothing is stored, and nothing passes through this site.

How it scores, and what it does not

No AI writes or judges anything here. Titles, years, journals and DOIs are copied from PubMed records; counts are PubMed's; verdicts are fixed rules in open code. Only the “Most relevant” order is PubMed's Best Match. A citation not found in PubMed is a reason to look, not a finding of fabrication. Quotes are matched word for word against the paper's PubMed abstract; a proportion check flags a sentence that claims cause from a non-trial, prediction with no accuracy reported, or “strongly” where the abstract says modest. A flag is a reason to read the paper, never a finding: whether the quoted words mean what the answer says still takes a person.

The set behind it

The same rules run as an open harness for orthopaedic AI (Apache-2.0, no dependencies): staged cases revealed one step at a time, two yardsticks (the guideline and the treating surgeon) scored apart and never averaged, “not enough information” as a real answer, and omissions counted alongside inventions. Its seven demo cases are synthetic and its weights are placeholders: a draft until surgeons write and approve the key. Read the code.

General information, not medical advice. In an emergency, call 911.