AI: Regulatory Framework for SaMD – Part Two

Editor’s note: This is the second part of a two-part series. The first installment can be found here.

The first part installment in this series examined the benefits of Artificial Intelligence/Machine Learning (AI/ML) and noted the considerations that regulatory bodies are studying for use with AI/ML algorithms. The second and final installment explores the past and current regulations, and summarizes the latest framework proposed by the U.S. Food and Drug Administration’s (FDA).

Regulations Past and Current

While current guidance around AI/ML implementation in medical devices is lacking, the FDA is working to solve the problem. They are in the process of creating a framework for software as a medical device (SaMD) that is practical to implement and covers the requirements for safety and effectiveness.

In April 2019, the FDA published a discussion paper on an outline of a regulatory framework titled, “Proposed Regulatory Framework for Modifications to Artificial Intelligence/Machine Learning (AI/ML)-Based Software as a Medical Device (SaMD) – Discussion Paper and Request for Feedback.” ⁽¹⁾

Also in 2019, the Association for the Advancement of Medical Instrumentation (AAMI) and the British Standards Institution (BSI) released a position paper that identified key considerations in the use of machine learning with medical devices. These centered on autonomy introduced by AI/ML technologies, effects of continuous learning systems that can change their output in response to new and different data, and finally explainability and understandability of the output. As part of the paper, the two groups coordinated with other bodies and suggested further work to standardize the terminology and categorization of AI/ML. Their suggestions included the recommendation to map existing and potential standards with the intent to standardize activities related to AI/ML in healthcare ⁽²⁾.

In February 2020, the FDA held a public workshop to get stakeholder feedback on best practices on AI‑automated radiological imaging software and image acquisition devices ⁽³⁾.

In February 2020, the FDA announced a marketing authorization via the De Novo pathway for the first cardiac ultrasonic software that uses AI to guide users. The manufacturer used a Predetermined Change Control Plan in its application to obtain authorization ⁽⁴⁾.

In September 2020, the FDA instituted the Digital Health Center of Excellence within the Center for Devices and Radiological Health (CDRH). The center’s focus is on “getting high quality digital health technologies to patients.” ⁽⁵⁾

Most recently in January 2021, the FDA published Artificial Intelligence/Machine Learning (AI/ML)-Based Software as a Medical Device (SaMD) Action Plan ⁽⁶⁾. The action plan consists of:

• Issuing draft guidance on the Predetermined Change Control Plan.
• Working with other regulatory agencies and the community to create harmonized Good Machine Learning Practice.
• Supporting a patient-centric approach via transparency on AI/ML-based devices (e.g. though labeling).
• Encouraging manufacturers to drive improvements to AI/ML algorithms, for instance to prevent bias and improve consistency.
• Advancing real-world performance pilot studies.

FDA Proposed AI/ML Total Product Life Cycle Framework

The proposed FDA framework for SaMD essentially seeks to address conditions where the AI/ML is adaptable over time via a controlled process.

As stated in the 2019 position paper, adaptable AI/ML can be expected to be used for:

• Optimizing performance within a specific environment (e.g., based on the local patient population).
• Optimizing performance based on how the device is being used (e.g., based on preferences of a specific physician).
• Improving performance as more data are collected.
• Changing the intended use of the device.

The adaptation process follows two stages: “learning” through the use of new data and “updating” via deployment of a new algorithm.

Based on the expected and planned modifications to performance, inputs, or intended use for the device, the FDA framework proposes to have manufacturers submit a “Predetermined Change Control Plan” that is reviewed during the initial premarket review of an AI/ML-based SaMD. This Predetermined Change Control Plan consists of a:

• “SaMD Pre-Specifications (SPS),” which describes the expected and planned modifications for the device.
• “Algorithm Change Protocol (ACP),” which describes specifics on the methods the manufacturer will take to control the risks of these modifications.

Ultimately, this approach is intended to emphasize that manufacturers will need Good Machine Learning Practices (GMLP) that are used to ensure quality is integrated into and maintained within the device over the product lifecycle. Additionally, the framework expects manufacturers to capture real-world data and maintain transparency of that data with the FDA, though the implementation of this is being defined further as part of the FDA action plan.

Importantly, non-device software functions are not subject to the FDA regulations and are not covered by the proposed framework. Specifically, functions intended ⁽¹⁾ for administrative support of a health care facility, ⁽²⁾ for maintaining or encouraging a healthy lifestyle, ⁽³⁾ to serve as electronic patient records, ⁽⁴⁾ for transferring, storing, converting formats, or displaying data, or ⁽⁵⁾ to provide certain limited clinical decision support are not medical devices and are not subject to FDA regulation.

This is in-line with FDA’s approach to SaMD which is regulating software used for medical purposes that is not part of a hardware medical device. The gap that potentially exists is that a large number of medical devices that could utilize AI/ML will likely involve hardware in some manner, for instance radiology or cardiology devices. Manufacturers can encapsulate the AI/ML algorithm portion as a SaMD separate from the hardware medical device used to obtain the data. But there are cases where the hardware and software are intertwined as the single medical device. For those devices essential elements of the guidance can still be used to establish a plan, such as the Predetermined Change Control Plan, and clearly state the objectives of the AI/ML within the device and whether it is adaptable or not.

The Future

The FDA has stated a goal of providing additional guidance for AI/ML based SaMD during 2021. In the meantime, many new medical devices using AI/ML continue to be cleared and manufacturers have begun integrating elements of the framework, such as a Predetermined Change Control Plan, into their submittals, likely expecting this will be an integral part of the upcoming guidance.

References:

1. “FDA Outlines Proposed Framework for Regulating Artificial Intelligence Software,” FDA.
2. “Digital health standards,” BSI.
3. “Public Workshop – Evolving Role of Artificial Intelligence in Radiological Imaging,” U.S. Food and Drug Administration, February 25-26, 2020.
4. “FDA Authorizes Marketing of First Cardiac Ultrasound Software That Uses Artificial Intelligence to Guide User,” U.S. Food and Drug Administration, February 7, 2020.
5. “FDA Launches the Digital Health Center of Excellence,” U.S. Food and Drug Administration, September 22, 2020.
6. “Artificial Intelligence/Machine Learning (AI/ML)-Based Software as a Medical Device (SaMD) Action Plan,” U.S. Food and Drug Administration, January 2021