The Food and Drug Administration (FDA), the US regulating authority in the sphere of medical devices, has published a discussion paper dedicated to the proposed regulatory framework for modifications to the software as a medical device (SaMD) based on the artificial intelligence/machine learning technologies. The Agency encourages medical device manufacturers to provide their feedback and suggestions regarding the matter.

Regulatory Background 

The FDA acknowledges the important role the artificial intelligence (AI) and machine learning (ML) technologies are playing in the further development of the healthcare industry in general and medical devices in particular. The applications based on the aforementioned technologies could be used to detect diseases, improve the accuracy of diagnosis, and also further personalization of diagnostics and treatment. The Agency additionally emphasizes the importance of the ability of AI/ML-based software to learn from real-world feedback and improve its performance accordingly – this unique feature creates new opportunities for the SaMD. Hence, the regulatory framework should be flexible enough to make these novel technologies available for healthcare professionals and patients. At the same time, rigorous regulatory oversight is needed to ensure the safety of all software products placed on the market. Thus, it is important to establish a proper balance between the unneeded regulatory burden and the protection of patients. 

The paper provides the definition of the Software as a Medical Device (SaMD) initially suggested by the International Medical Device Regulators Forum (IMDRF), which defines SaMD as software intended to be used for one or more medical purposes that perform these purposes without being part of a hardware medical device. The Agency also refers to the scope of medical purposes outlined by the Federal Food, Drug, and Cosmetic (FD&C) Act, which considers medical purposes as those purposes that are intended to treat, diagnose, cure, mitigate, or prevent disease or other conditions.  

According to the current regulations, SaMD manufacturers may use the following pathways:

  • 510(k) notification, 
  • De Novo request (a special application which includes classification request in case of novel medical devices having no predicates – similar devices already placed on the market), and
  • Premarket approval application (PMA).

Mandatory Change Notification

The authority also mentions that in certain cases a 510(k) application should be submitted with regard to modifications to the SaMD already approved by the FDA and placed on the market. In particular, this could be necessary in case of significant changes impacting the safety and performance of the SaMD in question. 

According to the document, when determining the applicability of 510(k) change notification, the main aspect to be considered relates to the risks associated with the modifications in question. For instance, a premarket submission would be required in case of the following changes: 

  • A change that introduces a new risk or modifies an existing risk that could result in significant harm,
  • A change to risk controls to prevent significant harm, and
  • A change that significantly affects the clinical functionality or performance specification of the device. 

With regard to the software, the abovementioned points would be applicable in case of modifications to the software affecting its safety and performance, including the changes to its intended use or the algorithm the software is based on. 

The document also outlines the cases when an additional application would be required for the SaMD approved for marketing under the PMA. For instance, it would be necessary in case of:

  • New indications for use, 
  • Significant modifications to the technology the SaMD is based on, providing that such changes affect performance characteristics,
  • New clinical effects. 

However, the general approach is not always applicable in the case of AI/ML-based SaMD that is continuously learning, hence, there are certain changes to the algorithm. Thus, the FDA has to develop a new approach, that would be flexible enough to consider such minor continuous changes. Nowadays the Agency assesses the changes to AI/ML-based software using the same approach as in the case of initial premarket review. 

There are some examples of AI/ML-based SaMD already approved by the FDA, but in most cases, these SaMDs have their learning algorithms “locked” in a way preventing the algorithms from further changes during the actual use. At the same time, in most cases, it is important to leave the possibility to learn open, since it is the main feature of AI/ML-based SaMD allowing it to improve its performance using real-world data. This creates a situation when in some time after the marketing authorization the SaMD actually uses the algorithm that is slightly different from the one subject to the initial review. In such a case, a traditional approach could not provide an effective solution. Consequently, there is a need for an entirely new total product lifecycle (TPLC) regulatory approach that would take into account the continuous self-improvement of the software, at the same time ensuring its safety, effectiveness, and performance. 

The new regulatory framework proposed by the FDA in the present discussion paper is actually based on the risk categorization principles developed by the IMDRF, existing FDA`s benefit-risk framework, the principles of risk management with regard to the software products and modifications thereto, and also the TPLC approach introduced by the Digital Health Software Precertification (Pre-Cert) Program. The new framework also employs certain elements applied in the course of the initial assessment under the 510(k) premarket notification, De Novo request, and PMA pathways. 

It is expected that the suggested regulatory approach would successfully address the aspects related to the continuous changes to the SaMD algorithm taking place due to the nature of the underlying technology while ensuring the safety of the patients.

AI/ML-based SaMD: Regulatory Approach 

As it was mentioned before, the suggested framework is actually based on the concepts developed by the IMDRF. In particular, it employs a risk-based approach similar to one provided under the current regulations. According to the document, the main factors to be considered when assessing the SaMD include:

  1. Significance of information provided by the SaMD to the healthcare decision (i.e. the way the information provided by the SaMD would be used), and
  2. State of health situation or conditions (i.e the particular state of the patient in the situation when the SaMD in question should be used). 

The aforementioned factors create a basis for the whole classification system which divides AI/ML-based software into four categories depending on the risk associated thereto. 

Another important concept described in the document issued by the FDA refers to the “locked” algorithms. As it was already mentioned before, some of the algorithms are “locked”, providing that the output they provide would be the same in case if the same input is provided, while the adaptive algorithm would change its behavior on the basis of the data processed. 

Summarizing the information provided here above, the new regulatory framework suggested by the FDA is intended to improve existing regulations in the sphere of AI/ML-based SaMD. The document describes specific features of such software and outlines the approaches to be used to ensure the safety of patients. 

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.