Software as a Medical Device (SaMD) – FDA’s AI/ML Plan
The greatest advantage of artificial intelligence/machine learning (AI/ML) is its ability to learn, reason, and resolve problems from real-world use and experience and add to its capability to improve performance. This ability for AI/ML to train itself from real-world feedback and improve its performance makes it unique.
In February 2020, the FDA for the first time approved an AI-assisted cardiac ultrasound guidance software named ‘Caption Guidance’. This device is notable not only for its pioneering AI but also for its flexibility to incorporate future modifications. Following this approval, multiple workshops and programs were held by the FDA with an aim of seeking relevant opinions and feedback from stakeholders who are active in the healthcare domain in order to improve its activities in this space.
The FDA has recently launched the Digital Health Center of Excellence (DHCE) to regulate digital health products. On Jan 18, 2021, through this center, FDA presented the ‘AI/ML-based Software as a Medical Device (SaMD)’ action plan in 5 sections that address stakeholders’ feedback. Although completely focused on SaMD, it is expected that it would also be relevant to other medical device areas like Software in a Medical Device (SiMD).
The Action Plan identifies several opportunities for SaMD developers to engage the FDA and its regulatory framework as the field of SaMD gains momentum. While this Action Plan is the roadmap for future regulations, an operational framework is expected to be released soon.
Highlights of the Action Plan
Tailored Approach
It is expected that the regulatory framework would allow for post-market modifications in SaMD based on the establishment and utilization of Pre-specifications (SPS) and an Algorithm Change Protocol (ACP). This is included in the action plan as ‘Determined Change Control Plan’. SaMD or AI/ML developers are expected to submit comments on the draft guidance that is expected to be released later in 2021.
GMLP – Good Machine Learning Practice
The FDA aims to encourage harmonization of the GMLP development through consensus of standards and community initiatives. It supports the development of GMLP in evaluating and improving the algorithms used in ML tools. The GMLP program of the FDA would also be pursued in collaboration with the Medical Device Cybersecurity Program within the FDA.
Patient-centric approach and transparency
- Manufacturer and user transparency: FDA proposes to hold public workshops to decide how device labeling would be able to support ease and transparency to enhance users’ trust in AI/ML-based devices. Additionally, it also plans to analyze the manufacturer’s transparency to the users when it comes to the functioning of AI/ML-based tools so that they can understand the benefits, risks, and limitations of their use
- Labeling of Devices: FDA intends to gather input to identify the type of information that would be recommended to be included in the labeling of AI/ML-based healthcare tools
Algorithm transparency
FDA supports scientific efforts in developing methods that evaluate the robustness of software algorithms especially when it comes to ensuring that AI/ML devices are suitable for all types of population irrespective of racial and ethnic diversity. These efforts include research collaborations with the Centers for Excellence in Regulatory Science and Innovation (CERSI) at UCSF, Stanford University, and Johns Hopkins University.
Real-world performance
Real-world data collection, its monitoring, and analysis are important concepts in the proposed regulatory framework. Through its framework, the agency has introduced mechanisms that the manufacturers would be able to leverage while working on software modifications in order to mitigate the risks involved and build a favorable benefit-risk profile of the device during marketing authorization submission.
There are also plans to monitor real-world performance by utilizing pilot projects on a voluntary basis and use learnings to validate the real-world performance parameters.
Next steps for developers of SaMDs
- The Action Plan currently appears modest in its initial objectives. Though the agency aims for a multipronged approach, the initial plan for 2021 is to publish draft guidance on Predetermined Change Control Plans.
- This can be a continued opportunity for AI/ML developers to engage and influence FDA’s thinking on AI/ML key concepts that can eventually be incorporated into the final guidelines.
#SaMD #artificialintelligence #machinelearning #FDA #SiMD
Sources: