Digital Health Devices raised a lot of attention to the MedDevice community for the challenges that need to be addressed to fully meet regulatory compliance. Particular attention have been given to those devices for which the software is based on machine learning algorithm and artificial intelligence.
After the discussion paper published by FDA and discussed in one of the previous post , another similar position paper has been published in conjunction by British Standard Institute (BSI), the Association for the Advancement of Medical Instrumentation (AAMI), in collaboration as well with Medicines and Healthcare products Regulatory Agency (MHRA).
The discussion paper entitled The emergence of artificial intelligence and machine learning algorithms in healthcare: Recommendations to support governance and regulation can be downloaded from AAMI website and it is the results of a series of workshop in the field of Digital Health devices and the related challenges in regulatory compliance.
The main topics of discussion and recommendation mentioned in the AAMI/BSI/MHRA discussion paper can be summarised below.
Software Validation for AI-systems
Of course software validation plays a fundamental role and the paper suggested to have a system for declaration of each versioning of a continuous learning systems. Although this looks very good as per intention, how this will be performed from practical point of view remains unclear.
Quality of Data Inputs
Quality of the data used to train the algorithm will play a fundamental role in ensuring safety and effectiveness of the devices. Variety of the data shall be large enough to ensure to avoid errors in the AI output: this is of fundamental importance also during clinical trials.
The use of AI-based solutions is going to introduce a new paradigm in the MedDevice sector, where human decisions are going to be influenced by machine. It is however important that clinicians remain accountable for the decisions they take and, on the other hand, regulation related usability for AI-related device shall be addressed, to continue to ensure safety of the devices on the market.
Last but not least, cybersecurity plays also an important role, although no specific other challenges than those ones already identified for standard medical devices have been identified for AI-based devices.
Software regulations need to be amended and updated to take in consideration devices based on artificial intelligences. In this contest, the update of ISO 62304 will be of great importance and it is already known that the next version of the standard will have impact on all the “healthcare software”.
Particular emphasis shall then be posed on risk management: there is the need to evaluate the gap between AI-based medical device solutions and current risk management regulations, to ensure an appropriate development of new standards and guidelines which covers AI-based medical devices.
The last point I want to highlight is the validation and verification approach to be used for AI-based systems: also in this case, at the moment there is a lack of specific guidelines and standard that cover this topic. This is a gap that should be fulfil quickly by the regulator as validation and verification activities have big impact on quality and safety of the device.
In conclusion, regulatory panorama for digital health product is still under full development. There is an urgent need of new standards and guidelines to cover the main concerns related to the application of AI-based systems in the medical device world.