AI review
Review flagged EDFlow AI conversations as part of the safety review group: what gets flagged, reason codes, conflict rules, what you see, and resolving safety events.
Where a site uses the patient or family assistant, some conversations are flagged for safety review. Only members of the designated safety review group can open them.
What gets flagged
- Messages the safety layer recognised as distress or self-harm (which also alerted the care team).
- Attempts to misuse the assistant (for example trying to make it ignore its rules).
- Answers that the output checks blocked.
- A random sample of other conversations, only when the privacy office has approved sampling.
Opening a conversation
- Choose a reason code. Opening is audited and counts toward monitoring rule A-1.
- You see the scrubbed conversation: identifiers the patient typed are hidden.
- You can't open a conversation for a visit where you are on the care team, hold an open grant, or worked that zone's shift. The page blocks it and records the attempt.
Resolving
Record what happened (no action, content fix needed, prompt fix needed, clinical follow-up referred to the site) and close the safety event. Content and prompt fixes go back through the governance gates.
Retention
Conversations are deleted after the site's retention days; threads with a safety event are kept for the operational window.
Related
Dernière mise à jour le
Cette page vous a-t-elle été utile?
Vendor support access
How time-boxed vendor support access to your EDFlow site works: no standing access, a ticket and reason, read-only or admin scope, at most 4 hours, notifications to managers and a full audit trail.
Integration analysts
EDFlow for integration analysts: feed health states and what each audience sees, per-source status, heartbeats, event counts, rejections by error code, roster discrepancies and feed alerts.