The Trenowin Digest
May 2026 — First Edition
The Trenowin Digest is a monthly round-up of AI in orthopaedics — what’s new on the site, what the evidence actually shows, and what’s coming next. This page always holds the current edition; past editions are archived on the blog, and registered members get it by email each month.
From the editor
May was a sprint. Trenowin launched with five posts in the first week, then accelerated — and by the end of the month nineteen articles were live. That was faster than planned, partly because the research kept turning up papers worth writing about, and partly because the AI Skills series took on a life of its own once the first few posts went up.
The early feedback has been positive, which is encouraging, but the most useful signal has been the questions: what are the categories? Where should I start? How do I find articles on a particular topic? Those questions have driven the changes to the site this month — the pillar pages, this digest, and the orientation strip on the blog. The site is now easier to navigate for someone arriving without context, which was the gap.
June will be slower and more deliberate. The AI Skills series is complete for now. The focus shifts back to the clinical evidence pillars — On-Call & Triage and Clinic & Planning in particular have a backlog of strong papers that haven’t been covered yet.
Published this month
The AI Skills pillar launched in the second half of May as a structured series on using AI as a personal productivity tool — distinct from the clinical evidence posts. The eight posts cover setup, model selection, voice and memory, the different Claude modes, building AI personas, and workflow automation. They are practical and first-person in a way the clinical posts are not.
If you have been thinking about using AI seriously in your own work and have not found a way in, the 15-minute setup guide (Post 16) is the most useful single post on the site right now.
A study testing GPT-5, Gemini, and Mistral on hand fracture detection. Mistral was confident and wrong — and the paper named the failure mode cleanly. The issue is not that AI gets things wrong; it is that it can do so with high expressed confidence. This is the one to read if you are thinking about AI-assisted diagnosis in any acute setting.
A real published AI study with near-perfect reported accuracy — used as a vehicle to explain overfitting, small datasets, and the gap between internal and external validation. The argument: a paper reporting very high accuracy from a small dataset is not a reason for confidence. It is a reason to ask harder questions. This post is the companion piece to the AI evaluation framework in Post 1.
A deep learning system detecting cervical cord compression on MRI with performance above mid-level physicians — and two-centre external validation. A good example of a study that does things right: tested in a new population, with a clinical scenario that justifies the development effort.
What’s coming in June
The clinical evidence backlog. There are several strong papers in draft — on outcome prediction after proximal humerus fracture, LLM performance on fracture classification from radiology text, and patient-facing AI in the consent process. These are all Clinic & Planning and On-Call & Triage material.
One longer essay is also planned — around 1,200 words. The topic will be the clinical governance question: when an AI tool influences a clinical decision and it goes wrong, what does accountability actually look like? This is the post that keeps coming up in conversations with registrars and is worth doing properly.
The AI Skills series is paused for now. June and July will be weighted towards the evidence-based clinical pillars.
Site update
The site navigation now includes a Pillars section — six pages, one per content pillar, each explaining what it covers, who it is for, and listing the published articles with brief context. If you have found the blog hard to navigate without knowing what the categories mean, these pages are the answer. Start at trenowin.co.uk/pillars.
The Imaging & Diagnostics category has been retired. The sole post in that pillar (cervical cord compression AI) has moved to On-Call & Triage, which is a better clinical fit. The category count is now six: Foundations, On-Call & Triage, Theatre & Robotics, Clinic & Planning, Career & Practice, and AI Skills.
Trenowin is free to read. Registration takes two minutes and gives you New For You and Recently Read on your account page — so you never lose your place.