The introduction of artificial intelligence (AI) for medical devices, including in vitro diagnostics (IVD) medical devices, has created an extra level of complexity for regulating them.
That is according to Therapeutic Goods Administration (TGA) Devices Emerging Technology Section Director, Fiona McCormack, who spoke at Accreditation Matters 2024 about some of observations the government authority has made when regulating AI-based devices defined as therapeutic goods.
In addition to complexity caused by AI being a Cloud-based technology, McCormack said the mix of different software components, sourced from different manufacturers and suppliers, can be difficult to evaluate, assesses, and monitor.
“They’re often incorporating open-source software into products, but as soon as you’re a manufacturer of software, your name is on that product,” she said. “You take responsibility for the outputs of that software, even if it incorporates open-source software.”
More than just a device
A problem that TGA has encountered is “function creep,” when changes are made to software to augment the functionality of a product.
McCormack said this can often change the intended purpose of the product, and when the suppliers or the vendors of those products fail to inform the TGA about the changes, they may no longer be complying with their regulatory requirements.
“The incorporation of AI is not always immediately visible in the design of the product,” she said. “Not to the TGA and not to the user.”
When the altered function of a product changes its intended purpose, it includes the risk classification of the product. This has often been the case with laboratory information systems, specifically in the pathology sector.
“It takes us quite a long time, and a lot of back and forth between the sponsors and the manufacturer, to get the information we need to get these products approved,” McCormack said.
Understanding regulations
Another challenge when regulating software-based devices is that developers are often not used to working in a regulated environment.
“They’re new to the regulation and don’t know the requirements,” McCormack said. “So there’s a large amount of education included in our role, just to get those products through the regulatory process and approved for supply.”
Another obstacle is that data that is used for the training and testing of AI-based products is often not provided or well explained. That means that the AI models used for devices are frequently developed on data sourced from overseas.
“There’s often no commentary or relevance to Australia and how it’s being used,” McCormack said. “This is introducing quite a lot of bias into the products, and it’s really challenging for us to unpack.”
Supporting the framework
The regulation of the AI-enhanced medical devices is a big task for the TGA, so it relies on its partners such as NATA to better make sense of and manage it.
The TGA and NATA have a memorandum of understanding which sets out joint undertakings for assessment and information sharing on the manufacture and use of in-house medical devices.
McCormack said NATA assessors stand out for their detailed knowledge of the IVD regulatory framework the TGA overseas.
“They undertake their assessment in conjunction with the ISO 15189:2012 assessment and its accreditation,” she said. “It emphasises the responsibilities a laboratory has around the requirement for the validation the verification and the documentation for those IVDs that they are using.
Full more insights by McCormack, view the full presentation.