A 72-year-old man with high blood pressure sits through a resting electrocardiogram at an outpatient clinic. The trace prints, a clinician glances at it, sees nothing alarming and files it. Tracings like this are produced in vast numbers across the NHS, and most follow the same path. The US Food and Drug Administration has now cleared software that treats such a trace as a source of evidence about a leaking mitral valve that neither patient nor clinician knows is there.

The product, Tempus ECG-MR, has 510(k) clearance for patients aged 65 and over who carry cardiovascular risk. It returns a binary flag for moderate or severe mitral regurgitation, a condition that often produces only vague breathlessness, or nothing at all, and is frequently found late. The software is a triage aid and makes no diagnosis, and it excludes paced rhythms. That restraint is sensible, and it also marks out where the real argument lies for Britain.

The value of the algorithm depends on what happens after the flag. A positive result means an echocardiogram and, for some patients, specialist referral. Echocardiography is among the tests where NHS capacity is most strained. Sonographer shortages are long recognised and waits routinely breach the six-week diagnostic standard, with well over a million people in England queuing for some form of test at any time. A screening tool that raises the number of positive flags without adding imaging capacity turns hidden disease into visible queues. That may still be the right outcome, since a known problem can be managed and an unknown one cannot. It is nonetheless a cost that commissioners and trusts must plan for, and the planning rarely features in business cases written around the software itself.

Regulation adds a further layer. FDA clearance carries no standing in the UK, where the MHRA runs its own route for software as a medical device and NICE judges whether a technology offers value to the health service. The 510(k) pathway rests on equivalence to a device already on the market, a lower bar than proof of better patient outcomes. NHS leaders have seen this gap before. The AI Diagnostic Fund placed chest imaging tools in trusts from 2023, and evaluations found slow deployment, hampered by IT integration and information governance. Clinicians also raised doubts over who carries responsibility when an algorithm and a human reader disagree. An ECG tool must sit inside cardiology pathways rather than a single radiology workflow, so these frictions are unlikely to shrink.

The government's 10 Year Health Plan commits to moving care from hospital to community and from analogue to digital, and it presents earlier detection as a route to financial sustainability. Tracings are taken in primary care and in hospital clinics alike, which makes them a ready upstream sensor. A prevention strategy without a funded downstream pathway is a slogan, however, and the test for ministers is whether the diagnostic capacity the plan assumes exists where older and poorer populations carry the heaviest burden of undiagnosed heart disease.

Data is the other pressure point. Tempus is a company built on large clinical datasets, and the NHS holds some of the most continuous population-level records in the world, linking ECGs, echo results and outcomes over many years. That is precisely the material needed to test whether a model trained on American patients performs on a British population with a different demographic profile and case mix. Local validation is a condition of safe adoption. The terms on which NHS data is used to carry it out, including any commercial return to the health service, have been politically sensitive since the row over the Federated Data Platform. Leaders who open access to developers without clear terms will answer for it to patients and to the public.

The FDA decision is a modest milestone with no direct effect on British practice. Its usefulness to NHS leaders lies in what it previews. The next wave of clinical AI will resemble this one, cheap to run on data already collected and expensive to act upon. Systems that settle in advance how many extra scans they can absorb, and who holds the validation and data terms, will adopt such tools to good effect. The rest will acquire a new way of finding patients that they cannot yet treat in good time.

‍