Gottlieb Calls for Congress’ Help with AI Device Regulation
Former FDA commissioner Scott Gottlieb (2017-2019) says Congress needs to design a framework for the modern, efficient regulation of medical artificial intelligence that could be based on the regulatory scheme articulated in the pending Verifying Accurate Leading-edge IVCT Development Act (VALID Act — bills S 2209 and HR 4128) for diagnostics. Under the VALID Act, FDA would take a firm-based approach to modernize regulatory oversight of IVCTs – including in vitro diagnostics and laboratory-developed tests.
Writing in a 7/11 online JAMA essay, Gottlieb says this same approach is “uniquely suited to the regulation of artificial intelligence (AI) medical devices that can augment patient care.” FDA’s traditional regulatory approach, which relies on its capability to “meticulously examine a product’s construction, might prove infeasible” with AI medical devices in this context, he says.
An FDA firm-based approach to regulating medical devices “means that the regulatory oversight focuses on the quality systems and processes of the manufacturing firm, rather than just on the mechanics of the individual products,” Gottlieb explains. “This approach emphasizes the importance of a company’s quality management system to ensure consistent production and control of medical devices, and involves routine inspections and audits to verify the firm’s procedures and operational controls.”
Just like advanced diagnostics and genomic tests, Gottlieb writes, “AI used in medical devices is designed to analyze vast amounts of data to generate clinical insights. The challenges Congress is attempting to solve with the VALID Act are similar to those confronted by the FDA’s evaluation of medical AI. One of the challenges involves the rapid cycles of innovation inherent to these products because they undergo constant modification as new information becomes available.”
Gottlieb notes that AI medical device developers are intentionally limiting their functionality to “avoid classifying these software tools as medical devices…. Product developers are apprehensive that if the regulatory process remains uncertain, the FDA will become entangled in the boundless task of deciphering the product’s underlying code. Under a firm-based approach to regulation, innovators could bring certain new products to market and update their existing ones without undergoing the same premarket review in every case. This is a modern approach to regulation that would make the introduction of new innovations far more efficient. It is well suited for products like AI medical devices, for which the goal is to undergo near-constant evolution of the model to continuously improve its performance.”
Additionally, Gottleib says other Congressional reforms could create a framework for third-party certification of some lower-risk AI devices. “Independent organizations meeting criteria set by the FDA would assist in evaluating and certifying these devices,” he says. “In the complex realm of AI (an area where the FDA might struggle to recruit and retain top experts), this flexibility to rely on certified evaluations would enable the agency to tap into external expertise. Such an approach could streamline the FDA’s review process while enhancing the depth of expert oversight the agency can provide.”
Gottleib also encourages Congress to help create large datasets for AI development, training, validation, and postmarket monitoring, noting that private organizations already help the agency aggregate data from electronic health records and payer claims to support medical product postmarket surveillance. “Similar entities could be empowered to create a federated network of data required to adhere to FDA standards for accuracy and representativeness, ensuring that the data meet a regulatory grade for inclusion in the development and training of AI medical devices,” he says.