The US Food and Drugs Administration (FDA) released a five-part action plan, called 'Artificial Intelligence/Machine Learning (AI/ML)-Based Software as a Medical Device Action Plan', on 12 January 2021. The plan, released by the Digital Health Centre of Excellence (DCE), lists out short-term steps that would help in regulating products that are based on artificial intelligence (AI) and machine learnings (ML). The DCE was created in September 2020 by the FDA's Centre for Devices and Radiological Health to advance FDA's purview into digital health technologies. 

Back in 2019, the FDA had published a discussion paper on AI/ML-based devices called the, 'Proposed Regulatory Framework for Modifications to Artificial Intelligence/Machine Learning-Based Software as a Medical Device (SaMD).' The paper invited feedbacks, on the basis of which the Action Plan has been created. The plan is divided into five main categories: 

1) Tailored Regulatory Framework for AI/ML-based SaMDs:

The paper had proposed a framework that would allow modifications to AI/ML-based medical devices by following the principle of a 'Predetermined Change Control Plan.' The plan lays down a controlled approach to the types of modifications and the methodologies to implement changes that can be done to the devices. This move allows the FDA and the manufacturers to have a robust monitoring system from pre-market development up to the post-market performance. While this is an initial document, the FDA has announced that a proper guidance document for the same will be released later this year. 

2) Good Machine Learning Process (GMLP)

The consolidated comments on the 2019 paper unanimously emphasised on the need of 2) Good Machine Learning Process (GMLP). the GMLP refers to AI/ML best practices that ensure quality system practices and software engineering practices. Protocols for data management, validation, documentation, and algorithm training, amongst others will be covered under GMLP. The FDA will join hands with the Medical Device Cybersecurity Program will support the development of a GMLP that would be used to evaluate and improve AI/ML algorithms. To continue to improve the GMLP principles, the FDA has said it will participate in global working groups. 

3) Patient-Centred Approach to Incorporating Transparency to Users:

A public workshop will be held to explain the importance of device labelling in supporting clarity to enhance end-user transparency. The move, FDA says, will increase trust in AI/ML-based SaMDs. Information on what to add to the labelling will be gathered by the FDA from manufacturers so that users understand the risks and benefits of the products they would be using. 

4) Regulatory Science Methods Related to Algorithm Bias & Robustness:

The majority of comments received on the 2019 paper were focused on highlight the need for improved methods to evaluate and address algorithmic bias. The FDA will continue its efforts to eliminate bias by working with its research partners to build a scientific approach. 

5) Real-world performance

Real-world data is pertinent for the monitoring of safety and effectiveness of AI/ML-based SaMDs. The 2019 paper suggested that the modifications to the SaMDs could be done by studying the real-world data. The FDA can study this information to keep abreast of all product changes and to evaluate algorithm behaviour. The FDA aims to start a voluntary program to develop a framework that can be used to gather and validate the real-world metrics and parameters.