AI in NHS Acute Care: Patient Flow and Triage

Why this briefing

NHS England records, for every Martha's Rule call about a patient getting worse, whether the early warning score would have raised the alarm on its own. Between June 2025 and August 2026, in 83 per cent of those calls where a score had been taken, it would not have. In 69 per cent of those cases the patient then needed a change in treatment or a transfer of care. Seven in ten Martha's Rule calls come from families and carers.

AI is now arriving at every point where patients are ranked. AI triage in the NHS App is due to reach every App user by April 2028; the evidence given for it is a 29 per cent cut in the phone queue at one practice, where an NIHR-funded evaluation runs until August 2027. An AI tool already forecasts A&E demand for 50 NHS organisations. Nothing published measures who these tools miss.

The rules are being written now. The government has accepted all 44 recommendations of the National Commission on regulating AI in healthcare, and neither the Commission's report nor the response mentions triage. Meanwhile the clinical safety standards every trust works to sit under a power that changed in July 2025 from "have regard to" to "must comply", and now reaches IT suppliers directly.

The human moment

The patient the score did not flag, and the family who did. A woman on a ward is getting worse. Her observations are taken and the early warning score does not trigger. Her daughter can see something is wrong and uses Martha's Rule to ask for an urgent review. NHS England's own data shows that in most such calls the score had not raised the alarm, and that most of those patients then needed their care changed.

What's inside

Sixteen pages, written for an acute trust executive who knows the NHS well and AI less well:

• A note on the numbers — what is included, what was excluded, and why, including a supplier's 94 per cent triage accuracy claim
• More patients, more published data, and more ways to ask for a second look
• The pressure point — what the early warning score misses and the family catches
• Two queues, and four coroners' reports on the triage systems already in use
• Where AI actually is in patient flow, and four risks specific to acute care
• The threat side — a front door that sets the size of the queue
• The UK legal position, including Article 22B on automated decisions using health data
• The change nobody has framed — clinical safety standards that became a duty to comply
• Beyond the UK — the EU names emergency triage AI as high-risk
• Seven things a trust can do now, none needing new law
• Three levels defined by where accountability sits, three horizons, and an eight-workstream way in

The discipline behind it

Every figure carries a source: NHS England's official statistics or its own published record, the Care Quality Commission, the Health Services Safety Investigations Body, a coroner's report, a government department or the statute book. Where the briefing reports an absence — such as the word "triage" in the National Commission's report — the source was searched for it.

The most quotable numbers in the sector are excluded on the record: a supplier's "94% triage accuracy" for AI triage, attributed to an NHS evaluation that has not been published, and NHS England's £41bn benefits estimate for its technology programme, published without a method. The people who died in the coroners' cases are described, not named.

Who it's for

A chief executive, medical director, chief clinical information officer or chief information officer at an acute trust who has just picked up an AI programme, or a board member asked whether the trust's scores and tools can be shown to catch the patients they should. It assumes acute care is well understood and AI is not.