The FDA has shared its first draft guidance on how sponsors should assess the credibility of artificial intelligence (AI) models to support FDA decisions regarding drug safety, effectiveness or quality.
The agency noted an “exponential increase” in the use of AI in drug development and regulatory submissions in the last several years.
A Risk-Based Credibility Assessment
Like the FDA’s draft guidance for AI-enabled medical devices, this guidance envisions a risk-based framework to evaluate a model’s credibility for a particular context of use.
In the context of AI models, the FDA considers “credibility” to refer to trust established through collecting “credibility evidence.” That evidence can support the credibility of a particular AI model output for a specific context of use.
The activities used to establish credibility should be commensurate with the model risk and tailored to a specific context of use, the agency said.
A Framework for Assessing Risk
The FDA recommends sponsors follow these seven steps to for assess an AI model’s credibility:
- Define the question the AI model will address.
- Define the context of use.
- Assess the AI model risk.
- Develop a plan to establish the credibility of AI model output within the context of use.
- Execute the plan.
- Document and discuss the plan results.
- Determine the model’s adequacy for the context of use.
Engaging Early
The FDA encourages sponsors to engage early with the agency to:
- Set expectations regarding the credibility assessment based on the proposed model’s risk and context of use.
- Help identify potential challenges and how to address them.
What’s next?
The FDA is seeking public comments on the draft guidance until April 7.
