Sign up for the QMED & MD+DI Daily newsletter.
Navigating FDA’s Draft Guidance for AI-Enabled Medical DevicesNavigating FDA’s Draft Guidance for AI-Enabled Medical Devices
A look at how FDA's new draft guidance outlines recommendations for lifecycle management and marketing submissions of AI-enabled medical devices, emphasizing a total product lifecycle (TPLC) approach to address transparency, bias, data management, and risk management throughout the device’s development, deployment, and ongoing monitoring.
4 Min Read
_(3).png?width=1280&auto=webp&quality=95&format=jpg&disable=upscale)
At a Glance
- FDA's new draft guidance emphasizes a Total Product Lifecycle (TPLC) approach to managing AI-enabled medical devices.
- The guidance outlines specific submission requirements for AI-enabled devices, including details on functionality.
- Manufacturers must proactively manage evolving datasets to address the unique risks of AI in medical devices.
Earlier this month, FDA released a new draft guidance document outlining recommendations for lifecycle management and marketing submissions of AI-enabled device software functions. The guidance emphasizes a total product lifecycle (TPLC) approach, detailing the necessary documentation and information required for FDA review.
Building on insights from the November 2024 public meeting of the Digital Health Advisory Committee (the Committee), the document reflects FDA’s thinking on strategies to address transparency and bias throughout the TPLC of AI-enabled devices. A TPLC approach is positioned to address key challenges associated with AI in medical devices, such as data collection and preparation, training and validation, evaluation and testing, deployment and integration, and ongoing monitoring.
For adaptive systems that evolve through use, the TPLC framework is particularly focused on evaluating and monitoring real-world performance. This includes identifying and mitigating bias while potentially enhancing transparency for end users.
While the new guidance contains few surprises for medical device manufacturers, it does provide specific recommendations tailored to AI-enabled devices. These recommendations align with FDA’s ongoing concerns about ensuring safety, effectiveness, and transparency in this rapidly evolving domain. The guidance also specifies the types of information to be included in submissions for AI-enabled medical devices.
Information to be included in submissions for AI-enabled medical devices:
A statement confirming that the device uses AI.
A description of the device’s inputs (manual or automatic) and outputs.
An explanation of how AI supports the device’s intended use.
Details about the intended users, including their characteristics, training level, and required training.
A description of the intended use environment(s) (e.g., clinical or home setting).
An outline of the intended workflow for using the device.
A comparison of the device’s automation level to the current standard of care.
An explanation of the clinical scenarios for use and how outputs fit into the clinical workflow.
Details on installation and maintenance procedures.
Information on calibration or configuration tasks required to maintain performance.
Most of the items listed above are relatively standard items that are included in non-AI devices. However, the guidance also recommends that sponsors should include the following:
A description of the AI-enabled device’s user-configurable elements.
A list of all configurable elements.
An explanation of how users can adjust these elements and settings.
An analysis of how these configurations may influence user decision-making.
A total product lifecycle approach to address risk management
The guidance highlights that a total product lifecycle approach is encouraged to address risks. This includes evaluating the interaction between AI-enabled devices and clinicians or patients. What do users need to understand before using the device? During the advisory meeting, it was noted that while FDA has not yet cleared any generative AI devices, nearly 1,000 "AI-enabled" devices had been cleared as of May 2024. The TPLC approach aims to identify specific technical risks associated with AI-enabled devices, many of which remain poorly understood.
The emphasis on data management in the draft guidance
FDA emphasizes the critical role of data management for AI-enabled devices, given their reliance on data for performance. Manufacturers must improve tools and processes for data management, ensuring datasets are representative of the problem being addressed. Submissions to FDA should detail how validation datasets were assembled and the rationale behind the validation tests.
Data management will need to be proactive, treating data not as a static input but as a resource requiring ongoing maintenance. Transparency is essential, including clear documentation of data sources (e.g., institutions), limitations, quality control, generalizability across populations, and the use of synthetic data. Simple organizational methods, like labeled folders, will no longer suffice; more robust systems are necessary.
Additionally, as the dataset grows with device use, manufacturers must address the temporal impact of evolving data to maintain accuracy and reliability.
Why medical device companies may find the guidance familiar
The guidance builds on the existing infrastructure for device development while highlighting additional considerations for AI-enabled devices. TPLC management is emphasized as a cornerstone for the development and management of these devices. Risk management must address the unique challenges posed by AI, particularly in systems capable of “learning” or reconfiguring. Transparency remains critical, requiring manufacturers to fully document their approaches to data and ensure proper integration into a quality management system.
Certain specialties, such as radiology, are already at the forefront of AI-enabled applications. This is largely due to radiology’s history with connected systems (e.g., the DICOM standard) and the compatibility of image analysis techniques with AI.
Addressing the challenges presented by AI for medical devices
The scope of AI’s impact presents challenges for medical device companies, forcing them to choose between developing in-house expertise or forming partnerships. Long-term collaborations are likely to play a key role in navigating the complexities of integrating AI into traditionally well-defined and stable medical products. This guidance extends existing frameworks while addressing the specific challenges and opportunities introduced by AI.
About the Author
You May Also Like
