– U.S. FDA regulatory policy for digital health and software products
The U.S. Food and Drug Administration (FDA) digital health regulatory policy has come a long way in the past five years since the passage of the 2016st Century Cures Act in 21. The Cures Act regulates FDA's regulatory policies on Real World Evidence, Premarket Review of Medical Device Accessories, Expedited Drug Development, and Breakthrough Device Designation. provided legal authority to modernize the software, and relaxed some FDA regulations that apply to software products. In particular, changes in regulations for software products enacted the policies previously announced by the FDA as guidelines, which facilitated the development of software products and provided a large framework for the future development of FDA's regulatory policies. .
In particular, a number of meaningful policies were announced in 2021, and it is expected that there will also be important policy changes in 2022. While we cannot discuss all of them, this article will address some of the FDA's policies for digital health products that were announced this year.
Artificial Intelligence/Machine Learning (AI/ML) Action Plan
In January 2021, the FDA announced an action plan for AI/ML devices. Action Plans do not have legal force, but they are meaningful in that they contain FDA's plans and priorities. In this action plan, FDA stated that it would (1) continue to evolve regulatory policies specific to AI/ML software products, (1) promote the development of current Good Machine Learning Practice, (2) increase transparency about AI/ML software, and (3) prepare a pilot program to monitor real world performance. Indeed, the FDA has already made some progress during 4, such as holding a workshop on transparency in AI/ML devices, and will continue to advance the points discussed in this action plan in 2021.
Good Machine Learning Practice – Guiding Principles
The FDA published the Guiding Principles for Good Machine Learning Practices in October 2021. This includes recommendations such as that the development data be different from the test/verification data, and that the development data reflect the nature of the actual user base (eg age, gender, race, etc.) and the UK regulatory body Health Canada and Medicines & Healthcare Products Regulatory Agency. It seems that the FDA is promoting harmonization with foreign organizations, and in terms of Good Machine Learning Practice, it is an area to pay attention to in the future as the regulatory policies of other countries are highly likely to follow the policies of these three countries.
Content of Premarket Submissions for Device Software Functions
In November 2021, FDA issued guidance on the information needed to prepare a software product for a premarket review, such as a 510(k). This guidance, updating existing policies in 2005, was an area of high interest in the industry even before its publication. It should be noted that this new guideline adopts the 'Document Level Review Method', which is determined according to the risk level of the product, which is simpler than the previously used 'Level of Concern method'. There are two levels of the Document Level review approach: 'Basic' or 'Enhanced', which category determines the amount of information a developer must submit to FDA. Therefore, it is one of the guidelines that companies developing software products should look closely at.
2021 was a year of great progress, while also setting the direction for the FDA to regulate future digital health products. In addition, the following year 2022 is also expected to see some significant additional policy changes and developments. For example, there will likely be announcements regarding Clinical Decision Support software, current Good Machine Learning Practices related policies. In the next article, we will look at the changes that will take place in 2022.
※ This manuscript is information prepared by external global regional experts and is not an official opinion of KOTRA.