Sunday, July 28, 2024

Journal Club: Scott Gottlieb, JAMA: "Congress must update FDA regulations for Medical AI"

In a July 11, 2024 article at JAMA FORUM, Scott Gottlieb, former commisioner of FDA, writes that FDA's approach to AI must be updated.  He anticipates it will require Congressional action, and it could come in a model that is akin to the lab medicine VALID Act.

###

https://jamanetwork.com/journals/jama-health-forum/fullarticle/2821274



See also: From May 2024, Senate bipartisan committee releases 30 page roadmap to USA "Innovation in AI."   Here or here or here.

See a quite-long article at 360Dx by Greg Cima on regulatory trends in AI diagnostics, here.

AI CORNER

The article doesn't have an abstract, a Chat GPT 4 summary is below.

Congress Must Update FDA Regulations for Medical AI

In a recent op-ed, former FDA Commissioner Scott Gottlieb advocates for modernizing FDA regulations to keep pace with the rapid advancements in medical artificial intelligence (AI). Highlighting the successful regulation of the Apple Watch's heart rhythm detection feature, Gottlieb emphasizes the need for a firm-based regulatory approach for AI medical devices.

Key Points:

  1. Firm-Based Regulation:

    • The FDA's shift from product-focused to firm-based regulation emphasizes the quality systems of manufacturing firms over the individual mechanics of products.
    • This approach involves routine inspections and audits of firms’ processes to ensure consistent production quality and safety of medical devices.
  2. The VALID Act:

    • The Verifying Accurate Leading-edge IVCT Development Act (VALID Act), currently before Congress, seeks to codify this firm-based regulatory approach.
    • The Act aims to modernize the oversight of in vitro diagnostics, similar to the challenges faced in regulating medical AI devices that undergo constant innovation.
  3. AI Medical Devices:

    • AI models, such as OpenAI's ChatGPT, can manage patient interactions and assist in clinical decision-making, requiring regulatory oversight.
    • The complex, opaque nature of AI models makes it difficult for the FDA to fully understand their functionality, posing regulatory challenges.
  4. Risks and Challenges:

    • AI systems trained on unreliable datasets can yield incorrect answers, highlighting the need for robust regulatory frameworks.
    • Developers are cautious, often classifying AI tools as non-device clinical decision support software to avoid stringent FDA regulations.
  5. Proposed Reforms:

    • Gottlieb suggests a modern approach where the FDA focuses on the developer’s processes and systems rather than dissecting the AI models.
    • Third-party certification for lower-risk AI devices and creation of large, reliable datasets for AI development and validation are recommended.
  6. Legislative Action Needed:

    • The FDA cannot fully implement firm-based regulation without new laws.
    • Congress must enact legislation to adapt FDA policies to the unique characteristics of AI medical devices, ensuring efficient and effective oversight.

Conclusion:

Gottlieb calls for Congress to update FDA regulations to effectively manage the rapid evolution of AI in healthcare. A firm-based regulatory approach, codified through legislation like the VALID Act, is essential for ensuring the safety and efficacy of AI medical devices while fostering innovation.

Published in JAMA Health Forum, July 11, 2024.