FDA Proposes Regulatory Framework for AI- and Machine Learning-Driven SaMD
Posted 02 April 2019 | By
The US Food and Drug Administration (FDA) requested feedback Tuesday on a new discussion paper that proposes applying a “focused review” approach to premarket assessments of software as a medical device (SaMD) technologies that are powered by artificial intelligence (AI) and machine learning (ML).
The agency said it may conduct a “focused review” in cases where proposed SaMD pre-specifications (SPS) and algorithm change protocols (ACP) “can be refined based on the real-world learning and training,” offering options for manufacturers to engage agency staff for such purposes. Options include contacting FDA review divisions to see if modifications fit under the approved or cleared model, and submitting pre-submissions for early discussions on modifications or premarket submissions or applications.
The discussion paper from the digital health team at FDA’s Center for Devices and Radiological Health (CDRH) set forth on Tuesday the proposed framework, which is centered on adapting to the iterative and autonomous nature of such SaMD tools to better leverage their ability to continuously learn from and improve performance based on real-world experience.
The move coincides with the focus of CDRH’s Digital Health Software Pre-Certification (PreCert) program, which is currently limited to first-of-its-kind AI/ML-based SaMD and the nine companies FDA selected for PreCert’s creation in 2017. PreCert has yet to become fully operational. The concept that the pilot explores—and now the discussion paper—came in response to the SaMD-specific challenge of determining a threshold for when such continuously learning tools should undergo premarket reviews.
Under the proposed framework, AI/ML-based SaMD would require a premarket submission when a software change or modification “significantly affects device performance or safety and effectiveness; the modification is to the device’s intended use; or the modification introduces a major change to the SaMD algorithm.” This approach was developed based on harmonized SaMD risk categorization principles—established via the International Medical Devices Regulators Forum (IMDRF) about five years ago—FDA’s benefit-risk framework, risk management principles in FDA’s 2017 guidance on submitting new 510(k)s for software changes to existing devices, PreCert’s envisioned organizational-based TPLC approach, as well as the 510(k), de novo classification request and premarket application pathways.
From lowest to highest risk, AI/ML-based SaMD would be placed into one of four categories using IMDRF’s two-pronged approach to risk categorization based on the clinical situation and intended use.
Applying a total product lifecycle (TPLC) approach, as envisioned in PreCert, is “particularly important for AI/ML-based SaMD due to its ability to adapt and improve from real-world use,” CDRH said. The AI/ML framework proposes a four-part TPLC approach. Its first three principles rest on establishing quality systems and good ML practices expectations, conducting premarket reviews for SaMD that require premarket submission and expecting continued patient risk management and risk management approaches to algorithm changes.
The fourth principle of the TPLC approach for AI/ML-based SaMD seeks to enable greater transparency to users and FDA on the part of manufacturers using real-world performance reporting to maintain the level of assurance of safety and effectiveness such tools promise with market entry via approvals or clearances.
“Through this framework, manufacturers would be expected to commit to the principles of transparency and real-world performance monitoring for AI/ML-based SaMD,” CDRH said.
The charge for PreCert has been led by Bakul Patel, associate director for digital health at CDRH. In reaction to the discussion paper, Patel noted on Tuesday: “FDA needs help exploring this concept.
“We need your feedback as experts/stakeholders in the AI/ML space to help inform how we regulate these devices and improve patient care,” Patel said, pointing to the 3 June deadline for public comments
on the discussion paper. The input will inform the development of a forthcoming FDA draft guidance.
Like PreCert, however, the proposed framework could face certain roadblocks. The pilot and the framework both describe approaches that may require additional statutory authority to fully implement.
“As with all of our efforts in digital health, collaboration will be key to developing this appropriate framework,” FDA Commissioner Scott Gottlieb said
of the discussion paper, which was issued in conjunction with the launch of a new AI/ML-based SaMD webpage
. Gottlieb also highlighted the agency’s ongoing work to build a Digital Health Center of Excellence to create efficiencies.
At Health Datapalooza 2019 last week, FDA principal deputy commissioner Amy Abernethy offered
her take on CDRH’s unique position to develop continuous learning practices set for agency-wide adoption. Abernethy was reportedly
just tapped to also serve as FDA’s chief information officer.