Discover SaMD and AI regulatory impacts, challenges, mistakes and readiness
"Diverging trends of different jurisdictions on requirements to, and qualification classification of SaMD. For one, different risk classes induce a different rigor of scrutiny. That of course depends, if your software is regulated or not. Here, manufacturers face fractured market definitions which can be a real hassle when going to markets beyond their own.
Essential principles are written to be applicable for all devices, including software. Some of them are hard to transpose for software, e.g. the ‘single fault condition’. Other challenges look on economic operators. This concept originates from the blue guide, where CE marked software was not in focus."
"Apart from the given introduction date late further guidance, lack of Notified Bodies and massive migration from class I to class IIa, I'd say the barriers it brings to SME's, while the EU has specified in the Digital Single Market transformation of health and care that it does want to facilitate SME's to participate in this market as well."
"One of the most significant regulatory challenges for digital technologies is not having a well-established regulatory framework, and the pace of new technologies is exposing gaps in the current regulatory system. Regulatory bodies are challenged to develop a framework that will provide the least burdensome path to assess the safety and effectiveness of SaMDs that can support the rate of technological advancement.
Another regulatory challenge is the ability of the SaMD manufacturer to implement a quality management system (QMS) that will be useful throughout the SaMD’s lifecycle. This includes appropriate and timely risk assessment and well-structured verification and validation activities, including clinical evaluation, to ensure that the SaMD continues to meet safety and efficacy claims throughout its lifetime in the marketplace."
"The whole medical device industry currently agrees that compliance with MDR and IVDR for all products is the biggest regulatory challenge at the moment. This is not different for SaMD. However, I am also convinced that every change brings along new opportunities: MDR and IVDR force companies to optimise their current products, processes and documentation.
Software and AI can help companies standardize the way their products are being used in the field, reduce the training and support effort, and increase the reliability of the result. This creates a proven measurable process as well as a competitive edge."
"One of the issues with the MDR/IVDR coming into force is that it’s often difficult to even get software and AI on the agenda. Indeed, much of the oxygen in the room (perhaps rightly so) is taken up the serious lack of Notified Body capacity, lack of guidance at the EU level, and Brexit uncertainty.
However, this means that the careful attention and debate that would otherwise be spent considering how new technology should be regulated is being spent elsewhere. In my eyes, the biggest challenge is having concrete plans for software and AI in the context of this uncertainty."
"Some AI algorithms pose special challenges for the medical device framework. For instance, black box models may find it difficult to link their model to established ground truth if we don’t know what the model finds significant. The MDR/IVDR and harmonised standards do not fully deal with this challenge. Moreover, adaptive algorithms retrain, they may be difficult to reconcile with change management processes. In responding to these challenges, it is important not to be exceptionalist. AI is not new, sometimes it offers new challenges for our regulatory framework but sometimes it represents a marginal improvement over and above hand-coded models."
"Further standardisation and operationalisation, at least in the area I focus on in health apps."
"With an increasing understanding of molecular biology and focus on personalized medicine, the complexity of medical diagnosis, prognosis and therapeutics will only increase. Software and AI provide an opportunity to manage this complexity by providing the necessary information and context to the physician. Medical devices will therefore increasingly rely on software and AI as part of the system and the regulatory landscape will have to reflect this."
"Some might call for a dedicated understanding of software as ‘in silico’ device. Besides raising the requirements on cyber security, AI as a technology field with all its methods exposes another dimension of variance in how to understand a product lifecycle and essential principles. Especially on ethical considerations of AI, and for medical devices, we should look on a by-sector approach as the systems are designed to help humans, not to kill them."
"Common mistakes companies make with SaMD development are:
"Being in touch with startups, I can say they start with the technology first, not with the intention of what problem to solve. They create a powerful tool, or software, but then struggle with what its intended use should look like. Not to mention the following problems with clinical validation. Second, documentation requirements are highly underestimated."
"According to a piece of Dutch research many companies are not aware of it or else are aware of it, but do not taken appropriate steps. In a sample of 271 health apps, 21% proved to be medical devices, yet more than half did not have a CE-mark."
"A number of pitfalls for SaMD are addressed in the recent guidance for classification of SaMD, recently published by the European MDGC, such as the fact that not all software used in healthcare facilities is considered a MD. This MDGC guidance document also recognizes that nowadays software often consists of non-medical device and medical device modules. It is important that manufacturers clearly describe which part of the software is defined as SaMD.
For SaMD that is used in combination with other devices, it is important to define the intended use and the regulatory roles and responsibilities of all parties at an early stage in the development process, so that each party can take his responsibility into account during the course of the product development. Such a proactive approach prevents surprises or missing information at the time of regulatory submission."
"SaMD has been in the market for quite some decades. And so has AI. It’s a matter of understanding that AI represents a vast field of technologies, which can be quite simple - such as decision trees, or complex - such as neural networks. We should respect that trained models are already out there. They do not necessarily change themselves without manufacturer interaction and exactly this aspect of continuity raises most questions.
I believe they are ready, but what matters most is to have all stakeholders on the same perspective looking at this field of technology. It’s practically worthless if manufacturers bring out new technology that is not adopted by healthcare providers because of the lack of trust in manufacturers and regulators."
"Most of the regulatory bodies around the world have started several progressive initiatives to resolve gaps in the current regulatory landscape and to propose new strategies to address the rise of digital health technologies, but it will take several years before there is a concurrence on regulations between regulatory bodies and industry. FDA started the Software Precertification (Pre-Cert) Pilot program, which is intended to streamline the premarket notification process and shorten the overall review time. But time will tell if this process is worth the effort. Currently FDA is working with well-established companies to evaluate this program, and may not be able to address issues raised by smaller companies."
"Until a year ago, I would have responded ‘no’ on this question, as it was hard to find any regulatory guidance documents on SaMD and AI. But over the last year, my opinion has changed. The European Commission recently submitted a guidance document on Classification of SaMD under the new MDR and IVDR. The FDA also published a discussion paper ‘Proposed Regulatory Framework for Modifications to Artificial Intelligence/Machine Learning (AI/ML)-Based Software as a Medical Device (SaMD)’ in April 2019 .
These efforts demonstrate that regulators do care about SaMD and they want to have a dialog with the industry about this topic. This dialog between all stakeholders, regulators, the industry and patients is very important to define a regulatory framework that works in practice and supports the launch of safe and effective innovative products on the market that allow improved treatment of patients. The fact that regulators are certainly embracing AI and providing pathways for use in therapies, diagnostics and sensors, can be seen by the exponential growing number of FDA approvals since 2015."
"The UK has new strategies in the works. There is a new regulator in the mix: NHSX. Regulatory bodies are working together to consider how the new world of digital health will look, how it will be regulated, and how existing regulatory frameworks may need to change. We’ve seen some of the fruits of this new collaborative spirit already but plans for SaMD and AI (as a medical device) in particular are yet to unfurl."
"With millennials we all see and experience a change in usage of wearables and apps and the inherent expectation of ease of access and ease of use of this technology spans across all industries, from consumers to medical devices.
Digital Technologies will impact patients especially in the area of precision medicine. They allow easy collection and transmission of data and will be a key requirement for future technologies and special care needs to be taken for user-groups which do not interact as tightly with digital technologies.
Reimbursement is an interesting topic. We see Belgium, France and Germany moving forward to actively promote digital health technologies. Since health systems are a matter of EU member states, a friction can now already be seen. Such Digital Technologies, on a global basis, can only succeed if the results of assessments such as NICE are accepted worldwide. This would basically call for a convergence of separate health systems in that specific area."
"My focus is in oncology and monitoring symptoms has proven to add 5-7 months survival and more quality of life, because clinicians are made aware of declines in functioning sooner, enabling them to intervene sooner, when the patient is still functioning relatively well. Worthwhile also from an efficiency point of view.
Again in my country [The Netherlands] 1 in 7 now work in healthcare. That is expected to be 1 in 4 by 2040 if nothing changes. We already can't find enough people now, we would not be able to afford 1 in 4 in healthcare and we need these people for other jobs as well. AI can be very promising, however does require an ethics by design approach to not harm or exclude patients."
"Digital technologies (SaMD & Al) have significant potential to provide reliable data, early disease prediction, and prognosis evaluation, which can ultimately provide high-quality care to patients. It can empower patients to make informed decisions regarding their health.
Providers can monitor their patients much more closely and provide more actionable and better feedback. This will eventually help improve the relationship between the patient and healthcare provider."
"AI promises to change healthcare across the board. However, its impact upon patients depends upon where AI is introduced first and how it is introduced into practice.
In regards to where AI is being introduced, the sector picks low-hanging fruit first: automated image segmentation for radiological analysis, scheduling of appointments, and so on. Further, these initial uses will typically be assistive only: the system being a second reader or being used as a clinical decision support tool. Consequently, the near-term should bring appreciable impacts of quality and speed of service but not revolution in care.
In regards to how AI is introduced, predictive accuracy is often the primary measure of an AI’s performance. However, for use in healthcare, we should also focus on the system’s usability – are healthcare professionals able to contextualise the system’s outputs, can clinicians interpret what the model found significant and communicate a simplified explanation to their patient? Usability of systems – how the system integrates into practice will make a great deal of difference for patient satisfaction and quality of care."
"SaMD and AI are playing an increasingly key role in the accuracy and efficiency of diagnosis and treatment across various specializations. Reimbursement protocols currently do not sufficiently reflect this role. We believe that the move towards value based and outcome based healthcare will increase the adoption of SaMD to drive quality and efficiency across the health continuum."
