The Food and Drug Administration (FDA), the US regulating authority in the sphere of medical devices and healthcare products has published an action plan on Software as a Medical Device (SaMD) based on the Artificial Intelligence / Machine Learning (AI/ML) technologies. The concepts addressed in the document are in line with the ideas initially outlined in the regulatory framework for AI/ML-based SaMD published earlier by the FDA

Regulatory Background 

The FDA acknowledges the importance of artificial intelligence and machine learning technologies applied in the sphere of medical devices. In particular, these technologies could be used to improve the processing of healthcare-related data. More and more medical devices based on these technologies are being placed on the market. One of the most important aspects of AI/ML-based software is connected to its ability for continuous improvement on the basis of the information processed. At the same time, this creates certain regulatory concerns, since the software incurs continuous changes after being placed on the market. Thus, such software requires a special regulatory approach to be applied. In this context, the lifecycle-based regulatory oversight becomes more important. Nowadays the Agency intends to establish an optimal balance between expanding the availability of innovative medical devices and ensuring the safety of the patients. 

In order to commence public discussions on the matter, the FDA has already published the “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” earlier in 2019. The document addressed the regulatory approach suggested by the Agency in the context of the premarket review of the submissions related to such software. In this document, the FDA also described the proposed procedures and processes related to ongoing surveillance and oversight to be conducted during the post-market stage by both medical device manufacturers and regulating authority, especially in terms of continuous changes to the software. The Agency also requested all the parties involved in operations with medical software to provide their feedback and suggestions. As a result of public consultations, the FDA received an extensive response from industry representatives and healthcare professionals. The present action plan on AI/ML-based SaMD incorporates the ideas and suggestions received and processed by the FDA. 

For AI-based software, the FDA applied the De Novo pathway – a special framework for entirely new medical devices employing innovative technologies. It is important to mention that a special approach should be applied not only in the context of the initial premarket review but also with regard to further modifications. Now the FDA intends to create a special framework that could meet the specific needs related to the AI/ML-based SaMD. Some of its elements could be also used in the context of Software in a Medical Device (SiMD). 

AI/ML SaMD Action Plan in Detail 

As it was already mentioned before, the present action plan on the SaMD employing the artificial intelligence and machine learning technologies is based on existing special regulatory frameworks, and also on the feedback received by the FDA from industry representatives during the public consultations. As the result, the Agency has created the action plan paying special attention to the following elements:

  1. Tailored Regulatory Framework for AI/ML-based SaMD, 
  2. Good Machine Learning Practice (GMPL),
  3. Patient-Centered Approach Incorporating Transparency to Users, 
  4. Regulatory Science Methods Related to Algorithm Bias & Robustness, 
  5. Real-World Performance (RWP).

The Agency states that in the course of public consultations it has found out that one of the most important aspects the industry representatives emphasized relates to the Predetermined Changes Control Plan dedicated to the continuous surveillance and monitoring of the safety and performance of the AI/ML-based SaMD placed on the market. The point to be considered here is the particular aspect of the software that will be changing in the course of learning. Moreover, the manufacturer shall also provide the details regarding the Algorithm Changes Protocol (ACP) describing the way the algorithm will learn and change while remaining safe and effective. In other words, when assessing the software in question, the Agency needs to have sufficient information regarding the expected changes the software will incur when being used for the intended purpose. It was also suggested to distinguish different types of modifications depending on their importance and actual impact on the safety and performance of the software. The Agency states that it is going to issue draft guidance dedicated to these matters and initiate public consultations later in 2021. 

The industry representatives agreed on the importance of Good Machine Learning Practice (GMLP) based on the consensus standards recognized by the FDA. As a response, the FDA intends to encourage the development of the GMLP. The aspects it would cover include, inter alia, the following ones:

  • Data management,
  • Feature extraction,
  • Training,
  • Interpretability,
  • Evaluation and documentation.  

The approach to be utilized in this context should be similar to one applied in the sphere of good software engineering practices or quality system practices. 

With regard to transparency to users, the Agency intends to hold a public workshop dedicated to the labeling of the AI/ML-based SaMD and the way it could improve the transparency. According to the present FDA action plan, promoting transparency is a key aspect of a patient-centered approach. It is also mentioned that the labeling of such products requires special consideration. In particular, it is important to clearly describe the following aspects:

  • The data used to train the algorithm,
  • The relevance of its inputs, 
  • The logic the software employs,
  • The role intended to be served by its output, and 
  • The evidence of the device`s performance. 

The Agency additionally emphasizes the importance for the patients to be fully aware of all known and identified benefits, risks, and limitations of the AI/ML-based SaMD. In this regard, the FDA has already commenced public consultations and workshops dedicated to various aspects related to the software utilizing towel technologies.

Key Considerations of the New Regulatory Approach 

The present FDA action plan on AI/ML-based SaMD also contains a detailed summary of all the information processed by the Agency and the findings it made. First of all, the FDA acknowledges the important role played by the feedback and suggestions submitted by the industry representatives in response to the public consultations initiated by the regulating authority. The Agency intends to use this information in the course of the development of a new regulatory framework for the software based on artificial intelligence and machine learning technologies. The FDA states that it would take all the measures necessary to facilitate further development and broad application of these novel technologies in medical devices. 

Summarizing the information provided here, the action plan dedicated to the AI/ML-based SaMD highlights the most important aspects related to the specific features of such products – the ones associated with the ability of continuous improvement in the course of use for the intended purpose. The FDA outlines the points requiring the most attention, as well as the potential approaches to be applied. 

How Can RegDesk Help?

RegDesk is a next-generation web-based software for medical device and IVD companies. Our cutting-edge platform uses machine learning to provide regulatory intelligence, application preparation, submission, and approvals management globally. Our clients also have access to our network of over 4000 compliance experts worldwide to obtain verification on critical questions. Applications that normally take 6 months to prepare can now be prepared within 6 days using RegDesk Dash(TM). Global expansion has never been this simple.