Clinical AI & Assurance
Ambient scribes in the NHS: what “safe” actually means
In short: an ambient scribe that drafts clinical notes can save time, and it can also insert, omit or reshape safety-critical information. Safe adoption starts with a written intended use, then clinical safety (DCB0129/0160), a DPIA, and a pilot that measures errors — not only minutes saved.
Write the intended use before you buy
Most ambient products record a consultation (or take a transcript) and produce a draft note. That is documentation support if a named clinician remains responsible for every word that enters the record. It stops being “just a note” if the product proposes diagnoses, suggests codes that change the care plan, or ranks urgency. The intended-use statement is what later decides whether the software is a medical device, which hazards belong in the safety case, and what you tell patients.
The two clinical-safety duties
In UK health IT, DCB0129 and DCB0160 are routinely confused. DCB0129 is the manufacturer’s duty: hazard log, safety case, and a clinical safety officer for the product. DCB0160 is the deploying organisation’s duty: local hazard review, training, business continuity if the scribe fails, and a named owner on the trust side. A supplier safety pack is an input to DCB0160, not a substitute for it. See our ambient scribe evaluation service for how those artefacts are tested together.
Data protection is not an add-on
An ambient scribe processes special-category health data, often including a recording or a near-verbatim transcript. That almost always needs a DPIA, a lawful basis, a retention rule for audio versus the approved note, and a clear answer to “where does inference run, and who is the processor?” Patients should know a tool is in the room before it starts listening. Consent theatre — a poster no one reads — is not a DPIA.
What a useful pilot actually counts
- Edit burden — how much of the draft the clinician rewrites, and which sections.
- Omissions — allergies, doses, safety-netting and “what we decided not to do” are the first things generic models drop.
- Hallucinated content — statements that were never said, especially diagnoses and examination findings.
- Equity — accent, language, overlapping speech, and noisy clinics. A tool that works in a quiet teaching clinic can fail in urgent care.
Time saved is a secondary metric. A faster wrong note is not a productivity gain.
Claims you should not make
Do not describe an ambient scribe as diagnosing, triaging or “replacing documentation.” The clinician owns the record. Marketing copy that over-claims clinical capability raises both the SaMD bar and the clinical-safety bar, and it is the first thing a cautious buyer will test.
Meds Global Health evaluates ambient documentation tools for intended use, clinical safety and NHS deployment. See Ambient Scribe Evaluation and buyer readiness. General information, not a regulatory determination.
Answers
Frequently asked questions
Is an ambient scribe a medical device?
It depends on intended purpose. A tool that only drafts a note for a clinician to edit is usually documentation support. A tool that infers diagnoses, suggests codes that change care, or triages is more likely to be software as a medical device. Write the intended use first, then classify.
Do we need DCB0129, DCB0160, or both?
The manufacturer (or supplier acting as manufacturer) holds DCB0129. The deploying NHS organisation holds DCB0160. They are separate duties. Buying a product with a safety case does not discharge the trust’s own deployment safety work.
What should a pilot measure?
Time saved is not enough. Measure note completeness and error rates (including omissions of safety-critical information), clinician edit burden, patient consent and awareness, and whether the tool changes what gets documented versus what was said.
Evaluating an ambient scribe?
We independently test intended use, safety artefacts and workflow metrics before you deploy.