Use this review before approving one AI use case in home or
community care.
For training and workflow-design purposes, use fictional or
appropriately protected information. Do not enter names, health details
or other sensitive personal information on the webpage.
1. Name the use case
- What bounded problem are we trying to solve?
- Who will use the output?
- Who could be affected if the output is wrong or unavailable?
- What does success look like without promising a clinical
outcome?
2. Classify the role
Assist
- The task is limited and reversible.
- A person can readily verify the result.
- The output does not determine care, rights, eligibility or safety
action.
Review
- The output may influence care, prioritization, staffing, monitoring,
documentation or communication. - A qualified person has time, evidence and authority to
disagree. - Escalation and override are built into the workflow.
Stop
Do not delegate final authority for:
- emergencies;
- diagnosis or treatment;
- medication changes;
- consent or information-sharing decisions;
- care-plan approval;
- service restriction, eligibility, discipline or employment action;
or - serious or life-changing communication.
3. Define the data boundary
- What information may enter the system?
- What information is prohibited?
- Is identifiable information necessary?
- Is the data retained or used for model improvement?
- Where is it stored and processed?
- Who may access it?
- How is it deleted?
- Which contracts, laws, policies and permissions apply?
4. Require useful evidence
- Intended purpose is documented.
- Relevant populations, settings and languages are described.
- Performance measures and limitations are available.
- Fairness or subgroup evaluation is available when relevant.
- Updates and model changes are communicated.
- Medical-device status and intended use are verified when
applicable.
5. Design human review
| Requirement | Decision |
|---|---|
| Named reviewer | |
| What the reviewer must check | |
| Source information available | |
| Conditions requiring escalation | |
| How disagreement is recorded | |
| Who makes the final decision | |
| How the affected person reaches a human |
6. Test failure conditions
- Missing information
- Conflicting information
- False alert
- Missed alert
- Different language or communication need
- Reviewer unavailable
- System outage
- Person declines the recommendation
- Request for correction or complaint
7. Approve with an exit plan
- Owner for the use case is named.
- Monitoring measures are defined.
- Review date is scheduled.
- A pause trigger is defined.
- A safe manual workflow exists.
- The organization can stop the use without losing essential care
information.
Final decision
- Approve as Assist
- Approve with required Human Review
- Revise and retest
- Stop—do not delegate this use case to AI
Important: This tool provides general educational
and operational information. It is not medical, legal, privacy,
compliance, procurement or emergency advice. Requirements depend on the
tool, intended use, organization and jurisdiction.