The level and type of IRB review of research involving artificial intelligence (AI) or machine learning (ML) technologies depends largely on what stage of research the project is currently in.  Pre-review of AI/ML technology should therefore focus on ensuring:

1. That the information provided by study teams is sufficient to determine whether the project is in the training stage (stage 1), the validation stage (stage 2), or the deployment stage (stage 3).  This will determine the appropriate application type, level of review (certified NHSR, exempt, expedited, convened board), and regulatory determinations that will need to be made;
2. That the risks and benefits of the AI technology to subjects (including privacy and confidentiality concerns and risks of bias) have been described, including approaches to mitigate such risks and;
3. As a project progresses, any intended future uses of the AI/ML technologies will be submitted as a change of protocol or new application.  In other words, if a study is currently in Stage 1 and being tested on a retrospective dataset, the protocol or application clearly outlines parameters that it would not be deployed in medical records without undergoing additional IRB review.

Studies involving health-related AI/ML should use this flowchart to assist in determining the appropriate application type, review pathway and regulatory oversight based on the stage the study is in.  While staging information is broadly applicable to non-health-related AI/ML projects, these projects likely qualify as minimal risk even in stage 3, and therefore an nPBA may be sufficient for all stages.  All projects involving AI/ML technology, regardless of whether or not the technology is health-related, should include the information outlined in HRP-337 - Artificial Intelligence (AI) / Machine Learning (ML).  For health care related AI/ML software functions, see also HRP- 307 – WORKSHEET – Devices.

Stage 2 studies involving an exempt diagnostic device may be eligible for expedited review under category 1b.  These studies may also qualify for a waiver of informed consent under revised FDA regulations and Common Rule waiver criteria.

Finally, see IM: Artificial Intelligence for additional guidance for study teams, descriptions of stages applicable to both health-related and non-health-related AI/ML projects, and risk considerations.