Clinical safety

NeuroScribe is a clinical documentation tool. It does not diagnose, score, triage or recommend treatment on its own authority, and it is not a medical device. Every output is a draft that a registered clinician must read, correct and approve before it is used or shared. We state this plainly because it defines the safety boundary of the product.

DCB0129 applies to manufacturers of health IT systems and DCB0160 to the organisation deploying them. We manage NeuroScribe’s development against the intent of DCB0129: hazards are identified before release, controls are implemented in the product rather than in guidance alone, and residual risk is recorded and reviewed. Where a customer requires a signed DCB0129 clinical safety case, we provide our current hazard log and safety documentation on request to support their DCB0160 process.

The Clinical Safety Officer role for NeuroScribe is held by a registered clinician appointed by NeuroScribe Ltd, contactable via clinicalsafety@neuroscribe.co.uk. The named holder and their registration details are supplied in the safety documentation pack. Deploying organisations must appoint their own Clinical Safety Officer for DCB0160 purposes; ours cannot fulfil that role for you.

Unapproved reports are visibly labelled as drafts throughout the workspace and in exports. Assessment notes cannot be saved without a recorded consent decision, and AI drafting is refused unless AI-drafting consent is present. Patient identifiers are removed before any text leaves the platform for drafting, and a request that still contains identifiers is blocked rather than sent. Approval locks a report against silent change, so a later amendment is a new, attributable version. Consent, edits, approvals, dispatches and exports are written to an append-only audit trail in the same transaction as the clinical change, so a record cannot exist without its audit entry. Clinical access requires an authenticated clinician account and owner-scoped record permissions.

The most significant residual risk is that an AI-generated draft contains an inaccuracy, an omission, or plausible content not supported by the clinical evidence. This is mitigated by mandatory clinician review and approval, draft labelling, and the fact that source notes remain visible alongside the draft — but it cannot be eliminated, and the reviewing clinician remains accountable for the content of the final report. Secondary residual risks include mis-selection of the wrong patient record during a busy clinic, mitigated by patient reference display and confirmation on destructive actions; loss of access during an outage, mitigated by export functions that allow records to be held outside the platform; and pseudonymisation missing an unusual free-text identifier, mitigated by pattern-based redaction plus clinician review before drafting.

We maintain a hazard log covering identified hazards, causes, existing controls, risk rating and residual risk, reviewed before each safety-relevant release. Where a change affects clinical safety we notify account holders directly rather than relying on release notes alone. If you believe you have found a safety issue, report it to the address above with the date, patient reference and what you observed; we treat suspected patient-safety issues as our highest-priority incident class.

Before entering live patient data you should complete your own DPIA and DCB0160 clinical risk assessment, appoint a Clinical Safety Officer, confirm that your clinicians understand that all output is draft until approved, and define local procedures for report approval, sharing and retention. We will supply the technical and safety information you need to do this; we cannot sign it on your behalf.