Minimize disruptions from decommissioning
Why does this matter?
AI products may fail or need to be stopped abruptly. This can cause disruptions in clinical workflow and healthcare operations. If the AI product is used in a critical task, sudden changes to the product could lead to patient harm.
How to do this?
Step 1: Assess preparedness
- Maintain protocols at the time of AI product rollout for what to do in the event of decommissioning.
- Have a clear process whereby the healthcare delivery setting can turn off the AI product.
- Maintain communication and data streams related to the AI product with developers to monitor performance post-rollout
- Maintain communication with end-users and affected clinicians about decisions to decommission AI products and incorporate feedback.
- Follow steps to decide when to decommission an AI product.
Step 2: Keep stable components of the solution running
- Work with the AI product developer to understand which components of the solution are unstable and need to be decommissioned.
- Maintain stable components of the solution, which may continue to be helpful, even if an AI model is no longer meeting performance expectations.
- Consider turning off isolated, unstable components of the solution, while continuing to run more stable components of the solution. Work with the AI product developer to uncouple unstable components from stable components.
Step 3: Empower the clinical lead to transition to a new workflow
- Ask the clinical lead for the AI product to help identify an alternate workflow optimized for patient care.
- Rely upon artifacts and documentation from prior work to rapidly develop this new workflow.
- Consider shifting data synthesis or analysis tasks from a faulty AI system to a human clinical expert.
- Ensure that incentives are aligned to compensate and promote the additional labor burden that may be added to a clinician role in the new workflow.
- Ensure that the clinical lead is empowered with support personnel and technical assistance to transition to the new workflow.
“Obviously, you know, the stakeholders… especially for clinical solutions, who are dependent on those solutions have to be given an alternative that they see as more beneficial than what you’re decommissioning.”Technical Leader
Step 4: Generate and dissemination information about decommissioning
- See how to disseminate information about updates to end users for details.
Step 5: Closely monitor the care delivery setting and work environment
- Continue to monitor clinical and process measures and the work environment after the AI product is decommissioned.
- Assess improvements in quality of care.
- Return to problem assessment and consider new potential approaches to address the problem if the clinical problem persists and remains an organizational priority.