Technology AI

DCB0160 and DPIA for AI scribes: a walkthrough

A step-by-step walkthrough of the DCB0160 clinical safety case and the UK GDPR DPIA a GP practice needs before using an ambient AI scribe.

Ava research teamPublished 6 min read

On this page
  1. Two documents, two purposes
  2. What to get from the supplier first
  3. The DCB0160 walkthrough
  4. An example hazard log
  5. Choosing a risk matrix
  6. Gaps reviewers often find
  7. The DPIA walkthrough
  8. Risks people actually worry about
  9. Linking the two documents
  10. Why some front-desk tools wait for this work
  11. Example: a pilot that changed the plan
  12. Frequently asked questions
  13. References

Key takeaways

  • DCB0160 is the mandatory standard for deploying health IT safely; the supplier covers DCB0129 and you cover your own setting[1].
  • A DPIA is required before high-risk processing starts, and must describe the processing, test necessity, assess risks and set out safeguards[2].
  • The hazard log is the working heart of DCB0160: list what could go wrong, how likely and how serious it is, and what controls reduce it.
  • Review both documents whenever the product changes; new features can even change its medical device status[3].

Two documents, two purposes

Before a GP practice uses an ambient AI scribe with patients, it needs two pieces of paperwork that are often confused. The clinical safety case, under DCB0160, asks whether the tool could harm a patient and how that risk is controlled. The data protection impact assessment, under UK GDPR, asks whether the processing of personal data is necessary, proportionate and protected. They overlap, and you can draft them together, but each answers a different question for a different audience.

How the DCB0160 safety case and the DPIA differ for an AI scribe deployment
AspectDCB0160 clinical safety caseDPIA
Legal basisMandatory NHS information standard for deploying health ITUK GDPR Article 35
Question it answersCould this system contribute to patient harm, and are the risks acceptable?Is this processing of personal data necessary, proportionate and safe?
Key peopleClinical Safety Officer, clinical leads, IT leadData protection officer, Caldicott guardian, practice manager
Main outputsClinical risk management plan, hazard log, clinical safety case reportDPIA document with risks, measures and sign-off
Supplier inputSupplier's DCB0129 safety case and hazard logSupplier's data processing terms, hosting and security information

What to get from the supplier first

NHS England describes DCB0129 and DCB0160 as mandatory standards that protect patients by ensuring health IT systems are assessed for clinical risks before they are used in patient care: DCB0129 sets requirements for manufacturers, and DCB0160 covers how health and care organisations deploy and use these systems[1]. Your work builds on the supplier's, so ask for:

  • The supplier's DCB0129 clinical safety case report and hazard log for the product version you will use.
  • A clear statement of intended purpose and regulatory status. MHRA guidance from July 2026 says products solely for transcription, summarising, drafting letters or suggesting codes for clinician review are not medical devices, but those supporting diagnosis or treatment, or taking automated action, are[3].
  • Data flow diagrams: where audio goes, where it is processed, what is stored and for how long.
  • Data processing terms, sub-processor list, hosting locations and security certifications.
  • Whether customer data is used to train or improve the product, and how to opt out.

The DCB0160 walkthrough

  1. Appoint a Clinical Safety Officer

    At present a CSO must be a registered health or care professional[1]. Many practices use a PCN or ICB CSO rather than training their own.

  2. Write the clinical risk management plan

    A short document: scope (which product, which clinicians, which consultations), roles, the risk matrix you will use and how hazards will be reviewed.

  3. Run a hazard workshop

    Bring a GP, a nurse, an admin lead and the CSO. Walk through a consultation using the tool and ask at each step what could go wrong.

  4. Build the hazard log

    For each hazard record cause, effect, existing controls, likelihood, severity and the resulting risk rating.

  5. Add controls and re-score

    Add training, configuration, process and technical controls until each risk is acceptable, and record the residual rating.

  6. Write the safety case report

    Summarise the system, the hazards, the controls and the CSO's conclusion that the risk is acceptable for go-live.

  7. Sign off and maintain

    The CSO signs off. Review after the pilot, after incidents and whenever the product or its use changes.

An example hazard log

Example hazard log entries for an ambient AI scribe in a fictional practice; ratings are illustrative
HazardPossible effectControlsResidual risk
Draft note includes something not saidWrong information in the recordMandatory clinician review before saving; training on common errors; monthly auditLow
Negative finding transcribed as positiveIncorrect diagnosis recordedTemplate prompts for key negatives; review step; error reporting routeLow
Wrong medicine or dose in a letterPrescribing error by another clinicianClinician checks medicines line by line; no automatic sendingLow
Tool unavailable mid-sessionNotes incomplete, delaysClinicians able to write notes manually; downtime guidanceVery low
Supplier adds diagnostic suggestions in an updateProduct use moves outside assessed scopeChange control; feature switched off until reassessedLow

A hazard log is never finished. Add every near miss and incident found in audits, and re-score when controls change.

Choosing a risk matrix

The hazard log needs a consistent way to rate risk. Most organisations score each hazard on two scales, how likely it is to happen and how severe the harm would be, and combine them into a rating that tells you whether the risk is acceptable. Use the matrix your ICB or CSO already uses, so your ratings mean the same as everyone else's, and write its definitions into the clinical risk management plan.

  • Define each likelihood level in plain terms, such as "could happen in most weeks of use" or "not expected in the life of the system".
  • Define each severity level by patient impact, from minor inconvenience to death or permanent harm.
  • State which combined ratings are acceptable, which need further controls and which stop deployment.
  • Score before and after controls, so the log shows the effect of each control.

Gaps reviewers often find

  • Scope that says "all consultations" without excluding safeguarding, mental health crisis or patients who decline.
  • Hazards copied from the supplier's log without thinking about how your practice works.
  • Controls that rely on "clinician vigilance" alone, with no training, audit or template support behind them.
  • A DPIA that describes the product but not the actual retention settings chosen.
  • No owner or date for review, so documents go stale after launch.
  • Patient information that promises things the configuration does not do, such as immediate deletion of audio.

Ask your CSO or a neighbouring practice to read both documents before sign-off. A fresh reader finds these gaps in minutes.

The DPIA walkthrough

Where processing, in particular using new technologies, is likely to result in a high risk, UK GDPR requires the controller to assess its impact before the processing starts, and processing special category data on a large scale is named as a case that needs one[2]. Health data is special category data[4]. Article 35(7) sets the minimum contents[2]; use them as the structure of your DPIA:

  1. A systematic description of the processing and its purposes: what is recorded, by whom, where it goes, what is produced and retained, and why.
  2. An assessment of necessity and proportionality: why ambient scribing is needed, what less intrusive options were considered, how data is minimised, the lawful basis and the Article 9 condition.
  3. An assessment of the risks to people: unauthorised access, retention beyond need, inaccurate records, use of data for other purposes, patients not understanding what happens.
  4. The measures to address those risks: contract terms, encryption, access controls, retention and deletion settings, patient information, the right to object, and audit.

Involve your data protection officer and Caldicott guardian, record their advice, and keep the signed DPIA with your records of processing.

Risks people actually worry about

Privacy is not an abstract concern. Patient privacy and data security worried 82% of GPs not using AI tools[5], and professional liability concerned 89%[6]. Public trust is also sensitive: trust in the NHS to use data to improve services fell from 78% to 73% between two waves of the government's tracker[7]. A clear, honest DPIA and patient information that matches it are part of keeping that trust.

Trust in the NHS to use data to improve products or servicesDSIT public attitudes to data and AI tracker
Trust in the NHS to use data to improve products or servicesComparison. Tracker wave 3: 78%; Tracker wave 4: 73%.78%Tracker wave 3Tracker wave 3: 78%73%Tracker wave 4Tracker wave 4: 73%Trust in the NHS to use data to improve products or servicesComparison. Tracker wave 3: 78%; Tracker wave 4: 73%.78%Tracker wave 3Tracker wave 3: 78%73%Tracker wave 4Tracker wave 4: 73%

Source: GOV.UK (Department for Science, Innovation and Technology), 2024[7]

Show the data
ItemValue
Tracker wave 378%
Tracker wave 473%

82%

of GPs not using AI were worried about patient privacy and data security[5]

Linking the two documents

  • Use the same description of the system and data flows in both, so they cannot drift apart.
  • Cross-reference hazards that are also data risks, such as inaccurate records or retention of audio.
  • Agree one review schedule and one owner for updates, usually the practice manager with the CSO and DPO.
  • Record the patient information and objection process once and refer to it from both.
  • Keep both with the contract and the supplier's DCB0129 documents in a single folder.

Why some front-desk tools wait for this work

The same standards apply to other AI tools that touch patient records, such as phone assistants that would book into EMIS Web or SystmOne. That is why some, including Ava, keep health details out of calls and list clinical-system booking as Planned until clinical safety assurance, a DPIA and a data processing agreement are complete. If a supplier offers record access without that paperwork, treat it as a warning sign.

Example: a pilot that changed the plan

Example: a fictional practice's hazard workshop found that clinicians sometimes dictated a summary for the patient after the consultation, and the scribe merged it into the clinical note. The practice added a configuration control, separate sessions for patient summaries, and a training point. The DPIA review at the same time found the supplier kept audio for a default period longer than the practice needed, so retention was shortened before go-live. Both documents were updated after the pilot with the audit results.

Frequently asked questions

Who can be a Clinical Safety Officer?

At present a CSO must be a registered health or care professional, such as a registered clinician[1], with clinical risk management training. Many practices share one through the PCN or ICB.

Is the supplier's safety case enough?

No. The supplier's DCB0129 case covers the product; DCB0160 requires your own assessment of how it is deployed and used in your practice[1].

Do we need a DPIA for an AI scribe?

Yes in practice. UK GDPR requires one before high-risk processing using new technologies, including large-scale processing of health data[2].

How often should we review the documents?

After the pilot, after any incident, when the product changes and at least yearly. New features can change the product's regulatory status[3].

References

  1. NHS England (2026). National review of clinical risk management standards DCB0129 and DCB0160: consultation questions (opens in a new tab) ↑

  2. legislation.gov.uk (The National Archives) (2026). UK GDPR Article 35: Data protection impact assessment (opens in a new tab) ↑

  3. GOV.UK (Medicines and Healthcare products Regulatory Agency) (2026). MHRA clarifies regulatory status of ambient voice technologies used in the NHS (opens in a new tab) ↑

  4. legislation.gov.uk (The National Archives) (2026). UK GDPR Article 9: Processing of special categories of personal data (opens in a new tab) ↑

  5. Nuffield Trust (2025). How are GPs using AI? Insights from the front line (opens in a new tab) ↑

  6. Nuffield Trust (2025). How are GPs using AI? Insights from the front line (opens in a new tab) ↑

  7. GOV.UK (Department for Science, Innovation and Technology) (2024). Public attitudes to data and AI: Tracker survey (Wave 4) report (opens in a new tab) ↑

Book a demo

See Ava handle your calls.

Book a short demo. We’ll walk through your calls and workflow, and agree what Ava should take care of.

Book a demo