FDA Issues Draft Guidance on AI-Enabled Medical Devices

The FDA has shared draft guidance on the design and development of AI-enabled medical devices and marketing submissions for those devices. 

The Importance of Engaging Early

Sponsors of AI-enabled devices should engage with the FDA early to ensure that the testing to support the marketing submission for an AI-enabled device reflects the agency’s total product lifecycle, risk-based approach. 

The draft guidance has recommendations relating to these content areas for marketing submissions (organized by stage in the product lifecycle): 

• Development: Risk assessment, data management, and model description and development
• Validation: Data management and validation
• Description of the Final Device: Device description, model description and development, user interface and labeling, and public submission summary
• Postmarket Management: Device performance monitoring and cybersecurity.

Emphasizing Transparency and Combatting Bias

The draft guidance emphasizes the importance of transparency. It also seeks to reduce bias. The FDA says companies should keep those imperatives in mind from the earliest stage of device design throughout product marketing authorization. 

Transparency: The FDA says transparency can ensure “that important information is accessible and functionally comprehensible.” In labeling content, the FDA recommends companies:

• Include information on how AI helps achieve a device’s intended use.
• Detail and explain the model inputs and outputs and development.
• Describe the performance validation data, device performance metrics, performance monitoring tools and known limitations of the AI-enabled device, AI-device software function or model(s).

Bias: The FDA describes bias as “the potential tendency to produce incorrect results in a systematic, but sometimes unforeseeable way.” That “can impact safety and effectiveness of the device within all or a subset of the intended use population.”  To guard against bias, the FDA recommends that companies:

• Ensure representativeness in data collection in developing, testing and monitoring the device throughout the product lifecycle, as well as in the evaluation of performance, across all relevant demographic groups of intended use (e.g., race, ethnicity, sex, and age).
• Collect evidence to evaluate whether a device benefits all relevant demographic groups similarly to help ensure that such devices are safe and effective for their intended use.

Covering the Entire Product Lifecycle

The guidance makes recommendations regarding product performance that span the product lifecycle, including in the postmarket setting.

The goal is to address issues such as data drift, which occurs when inputs for AI-enabled devices change over time from development to actual deployment in ways that may impact performance. 

For example, the FDA encourages sponsors to consider using a predetermined change control plan (PCCP).

As discussed in FDA final guidance in December 2024, companies can use a PCCP to seek premarket authorization for intended modifications to an AI-enabled device software function. They can do that without first submitting additional marketing submissions or obtaining further FDA authorization. 

One Piece in an Array of AI-Related Guidance

The FDA does not intend for this draft guidance to operate on its own.  Rather, the agency envisions companies also consulting prior guidance related to AI-enabled devices and specific technologies, including guidance on:

• Content of Premarket Submissions for Device Software Functions (June 2023).
• Technical Performance Assessment of Quantitative Imaging in Radiological Device Premarket Submissions (June 2022).
• Technical Considerations for Medical Devices with Physiologic Closed-Loop Control Technology (September 2023).

What’s next?

The FDA is requesting comments on the draft guidance by April 7, so it can consider the comments as it crafts a final version. 

Our team can help companies prepare and submit comments. We also can assist sponsors seeking the FDA’s feedback on marketing submission documentation through the agency’s Q-submission program. 

Want to know more? Contact one of the authors.