The FDA has issued draft guidance with recommendations for submitting medical devices that include AI-enabled software functions.
“The recommendations reflect a comprehensive approach to the management of risk throughout the device total product life cycle (TPLC),” the FDA said. “To support the development of appropriate documentation for FDA’s assessment of the device, this draft guidance also proposes recommendations for the design, development, and implementation of AI-enabled devices that manufacturers may wish to consider using throughout the TPLC.”
The FDA said its guidance isn’t intended to provide a complete description of what may be necessary to submit an AI-enabled device. It recommends early engagement to guide product development and submission preparation.
In the guidance, the FDA outlines its current thinking on addressing transparency and bias throughout the TPLC of AI-enabled devices. It says transparency ensures the accessibility and functional comprehensibility of important information with both the sharing of information and device usability. AI bias can produce incorrect results in a systematic, but sometimes unforeseeable way, the FDA said. That can impact device safety and effectiveness.
With AI-enabled devices, the FDA says consideration for transparency and bias addresses the fact that such devices can be hard for users to understand, given the opacity of models and model reliance on data correlations that may not map directly to biologically plausible mechanisms of action.
Throughout the TPLC, the FDA says controlling bias for AI-enabled devices can include addressing representativeness in data collection for development, testing and monitoring. Developers can also evaluate performance across subgroups of intended use.
Finally, the guidance includes recommendations for AI-enabled devices in the post-market setting. That includes highlighting sensitivities to differences in input data, such as input data used during development compared to input data in actual deployments. In addition to data drift, the devices can also be susceptible to changes in performance due to other factors, the FDA said.
