Human review is a workflow, not a disclaimer.
Evaluate AI pre-op screening by looking for protocol ownership, source preservation, explicit uncertainty, review coverage, escalation behavior, and evidence that unresolved items reach closure.
Safety evaluation guide · ASC clinical operations
The center owns the protocol
The question set, patient population, timing, urgent route, and clinical boundaries should be approved by the center. A vendor's generic script is not a substitute for the center's clinical governance.
- Named protocol owner
- Version and effective date
- Approved patient population
- Defined exclusions
- Approved language and branching
- Change-control and testing process
Source responses and uncertainty survive the handoff
A summary can hide the exact wording that made a response important. The reviewer should be able to see what the patient said and whether a field came from the patient, a record, staff observation, or a clinical decision.
- Verbatim or source-preserving response
- Structured value linked to its source
- Unknown and incomplete states
- Conflict between sources
- Reason the item entered review
Deference and escalation are testable
Ask the system questions it should not answer. Test medication instructions, medical advice, symptoms outside scope, urgent statements, unclear language, and explicit requests for a nurse. Review both the conversation and the resulting handoff.
- Does it avoid inventing an answer?
- Does it state the boundary clearly?
- Does it preserve the patient's question?
- Does it use the approved route?
- Does the receiving person see enough context to act?
Every required assessment reaches closure
Human in the loop should specify who reviews what. Some centers may require review of every assessment, not only items flagged by software. The design must support the center's chosen policy and make overdue work visible.
- Review population defined
- Reviewer role defined
- Due time defined
- Corrections recorded
- Escalation receipt visible
- Resolution and final ownership documented
Evidence used for this guide
These sources inform the workflow framework. They do not establish Incerto customer outcomes or a universal clinical protocol.
Bring one questionnaire and escalation rule set.
We will map what can be gathered, what needs licensed review, how the handoff closes, and which measures belong in a pilot.