Content pillar

On-Call & Triage

AI in acute orthopaedics — fracture classification, imaging triage, referral decision support.

Posts in this pillar answer: how does this change what I do tonight?

The on-call shift is where a significant part of orthopaedic training happens — and it is exactly where AI tools are beginning to operate. Fracture detection algorithms, imaging triage systems, and decision-support tools for acute presentations are being deployed in emergency departments right now, in some cases in hospitals where trainees are already rotating.

This pillar covers what those tools actually do, what the published evidence shows, and what their limitations are. It also covers the harder questions: what happens when the algorithm is confident and wrong? Who is responsible when an AI-assisted decision turns out to be a missed diagnosis? These are not hypothetical concerns — they are governance and medicolegal questions that every orthopaedic surgeon working in an AI-augmented environment needs to be able to answer.

For trainees

This is your world right now. The tools covered in this pillar are being deployed in the departments you are rotating through. Understanding what they do — and what they can’t do — is directly relevant to decisions you are making on-call.

For consultants

Governance and supervision implications — what should a consultant expect a trainee to know about these tools? What are the documentation requirements when an AI-assisted triage decision is made, and who carries the clinical responsibility?

Published in this pillar

The 3am fracture: what AI can and can’t do on-callA Saturday-night trauma scenario as the starting point. Covers current AI tools for fracture detection and triage, the evidence behind them, what automation bias looks like in practice, and where clinical judgement remains the only answer.Read →When an algorithm can’t wait: ML and the necrotizing fasciitis problemA deployed web tool that distinguishes necrotising fasciitis from osteomyelitis using 10 blood biomarkers. A close look at how it was built, what the evidence shows across 3,415 patients, and what it means for acute clinical decisions.Read →Better than the middle grade: AI detection of cervical cord compression on MRIA deep learning system that outperforms mid-level physicians at detecting cervical cord compression on MRI, with two-centre external validation. What the evidence actually shows and what it means for the on-call registrar.Read →Confident and wrong: what a hand fracture study reveals about AI’s most dangerous failure modeGPT-5, Gemini, and Mistral tested on hand fracture detection. Mistral was confident and wrong. What the results reveal about how LLMs fail — and why this particular failure mode is dangerous in a clinical context.Read →