The Food and Drug Administration (FDA) recently issued draft guidance clarifying how manufacturers should approach artificial intelligence (AI) in medical devices. The document provides recommendations for the design, development, and maintenance of AI-enabled devices to ensure their safety and effectiveness. It specifically addresses transparency, bias, and the need for postmarket monitoring.
Troy Tazbaz, director of the FDA’s Digital Health Center of Excellence, emphasized the importance of recognizing the unique considerations involved in AI-enabled devices. The FDA has already authorized over 1,000 AI-enabled devices, indicating the growing significance of this technology in healthcare.
One key aspect of the guidance is the recommendation for device sponsors to consider using a pre-determined change control plan (PCCP) to allow for alterations to devices post-authorization. This flexibility can help improve model performance without requiring a new submission to the FDA.
Transparency and bias are crucial considerations for AI-enabled devices, as many models can be opaque or difficult for users to understand. Patients have expressed a desire for more information about when AI is used in their care and how decisions regarding AI-enabled devices are made. The FDA recommends that developers provide detailed information about the use of AI in their devices, including the type of model used, datasets used for development and validation, and plans for updating and maintaining the model over time.
To address bias, the FDA advises manufacturers to ensure that validation data represent the intended target population of the device. Monitoring devices throughout their lifecycle and evaluating performance across different subgroups can help mitigate bias and ensure the generalizability of results across diverse patient populations.
Postmarket performance monitoring is also recommended by the FDA, as the performance of AI-enabled devices can change over time due to changes in input data or the context in which the device is used. Developers should proactively monitor and address any changes in device performance, as well as changes in inputs or usage context that could impact performance.
In conclusion, the FDA’s draft guidance on AI-enabled devices underscores the importance of transparency, bias mitigation, and postmarket monitoring to ensure the safety and effectiveness of these innovative technologies in healthcare. Stakeholders have until April 7 to provide feedback on the guidance, and a webinar is scheduled for February 18 to further discuss these recommendations.
