THE MED-AI BRIEF
November 3, 2026 · 5-MINUTE READ · Read online
Good morning, {{first_name|Doctor}}. On Monday The Lancet Digital Health published a study in which an AI read 27,382 heart MRIs and predicted death slightly better than the clinicians who reported them. The coverage I read made a point of saying the NIHR and the BHF funded it, not a vendor. True, and seven of the authors also own shares in the company that sells the tool. It says so at the end of the paper.
In today's Med-AI Brief:
🩺 Research: AI edges clinicians on heart MRI. Read to the end
🌍 Use cases: 29 trusts "realising benefits" from Palantir say it isn't live
⚖️ Regulation: the Budget orders a review of AI in the NHS
🛠️ Practice: find the competing interests in two minutes
RESEARCH
🩺 AI vs clinician heart MRI: which predicts mortality?
The brief: On 2 November The Lancet Digital Health published a study from UCL, Barts and colleagues comparing ejection fraction measured by an AI with the figure in the original clinical report, as predictors of death.
The details:
27,382 consecutive patients: 26,106 scanned at Barts Heart Centre in London and 1,276 at the University of Pittsburgh Medical Center. Median follow-up 5.6 years; 3,458 (13%) died.
The AI predicted death marginally better, the authors' word: a C index (how well a measure ranks who dies first; 0.5 is chance) of 0.669 against 0.637. After adjusting for age and sex, 0.781 against 0.777.
The AI took 3 to 4 seconds a patient. Errors in about 1% of a random sample, but 23% in scans with extreme ejection fractions, all at the low end.
On average the AI read ejection fraction 4.1 points lower than clinicians. In hypertrophic cardiomyopathy, clinicians did better.
The algorithm is sold as 1CMR by MyCardium AI. One author is its chief executive, seven own shares and eight are paid consultants.
Why it matters: This is solid retrospective evidence that AI-measured ejection fraction predicts death at least as well as the figure in the report, in seconds rather than minutes. It is not evidence that patients do better, and the authors call for prospective studies. Two points for clinicians. A systematic 4-point difference matters when a 35% or 40% cut-off decides a device or a drug. And the interests were declared in full, at the end of the paper, where few readers get to.
USE CASES
🌍 29 trusts "realising benefits" from Palantir say it isn't live
The brief: On 2 November the Guardian reported that 29 hospital trusts NHS England lists as "realising benefits" from the £330m federated data platform (FDP) say it is not live, or that they use none of its applications.
The details:
The answers came from more than 100 freedom of information requests by Medact, a medical charity campaigning against Palantir's NHS contract.
NHS England says nearly 140 trusts have reported benefits, and that its list includes trusts using the tools for back-office data analysis and submission rather than patient care.
One trust says it is "not currently operationally using the FDP". It is recorded as live because the infrastructure for submitting data is in place.
The Office for Statistics Regulation says it is "taking this forward under our casework function". This summer NHS England conceded it could not prove that 117,539 extra operations were caused by the platform.
Why it matters: NHS England's "live" and "realising benefits" turn out to be counting rules: a trust can qualify because data is being submitted. Marc Farr, who chairs the NHS chief data and analytics officers' network, said benefits must be reported "in a way that can be properly verified". When you hear a tool is live in your trust, ask whether that means data is flowing or clinicians are using it. Gulf buyers who hear "deployed across the NHS" in a sales pitch can ask the same question.
REGULATION
⚖️ The Budget orders a review of AI in the NHS
The brief: In the 28 October Budget, the Chancellor announced an independent "review into NHS productivity and the opportunities presented by AI and technology", so that staff can spend time "with patients and not on admin".
The details:
Led by Mark Bailie, a non-executive director of NHS England. It reports in early 2027, ahead of the next spending review.
A new prize of up to £2.5m for UK firms building AI for public services will focus first on the NHS.
On data centres, the Chancellor said that when the NHS uses AI to treat patients, "we want that data here".
The Health Foundation welcomed the review and said early priorities should include the infrastructure to test and evaluate AI. The RCN said AI can free nursing time but cannot replace growth in the workforce.
Why it matters: A review is not regulation, and nothing changes on your ward this year. But it will shape how much money follows AI in the next spending round, and it will lean on benefit figures like the ones in this week's Palantir story. The useful test is whether it counts what tools already in use have actually delivered.
PRACTICE
🛠️ Two minutes: find the competing interests
The heart MRI coverage I read missed a disclosure the paper printed in full. Before you share a study, check it yourself:
Scroll to the end. Look for "Declaration of interests" (Lancet), "Competing interests" (BMJ) or "Conflict of Interest Disclosures" (JAMA).
Press Ctrl+F (Cmd+F on a Mac) and search for shares, founder, consultant and employee.
Read the data sharing or funding note for licence. It tells you whether the tool is a product.
A declared interest doesn't make a result wrong. It tells you who would like it to be right.
QUICK HITS
📰 Everything else
York and Scarborough: on 30 October the trust's board heard that its ambient voice technology will extend to same day emergency care, Copilot licences have risen to 4,000, and a new AI and Data Ethics committee will oversee new uses of AI.
Abu Dhabi: on 2 November the government signed four agreements with OpenAI, including one to make regulations machine-readable, starting with food import permits. None is in health, yet.
Riyadh: the Global Health Exhibition closed on 29 October. As of today neither the SFDA nor the Ministry of Health has published an AI announcement from it. We will report one if it appears.
Watching: the statistics regulator's casework on NHS England's FDP figures; MHRA guidance on updating AI devices after approval (December); the Bailie review (early 2027).
Know a colleague who has been told an AI tool is "live" on their ward? Forward them this before their next meeting about it. They can subscribe at themedaibrief.com.
Until next week,
Saeed