Consultation on new guidelines for the application of ISO 14971 to artificial intelligence and machine learning

BS AAMI 34971 Guidance on the Application of ISO 14971 to Artificial Intelligence and Machine Learning is a guidance document for applying a risk management process when evaluating medical technologies that utilise artificial intelligence (AI) and machine learning (ML). It is intended to be used with ISO 14971. It does not modify the risk management process but provides guidance to inform the application of ISO 14971 to AI and ML systems. This guidance document will address the same types of risk that are already covered in ISO 14971, but it focusses on those risks that are elevated or unique to AI-based medical devices.

AI describes the capability of a computer to perform tasks or functions that are normally associated with human intelligence. ML is a type of AI where a system can learn from being exposed to new data, rather than following a strict set of rules. AI and ML can bring many benefits to healthcare, including earlier detection of diseases, more personalised approaches to treatment and improved care pathways. However, the implementation of these innovative technologies also presents a number of enhanced risks to patient safety and workflow management.

In the last three years, a joint venture between BSI and the Association for Advancement of Medical Instrumentation (AAMI) has explored the safety and effectiveness of AI and ML in medical technology. It concluded that data-driven AI medical devices differ from traditional rules-based software for a number of reasons:
- The level of autonomy that these systems can introduce to clinical and management practices,
- Their ability to adapt and change their output over time in response to new data,
- The sophisticated computational nature of AI makes it difficult to explain or understand how an AI-generated output has been reached.

BSI AAMI 34971 is the first standard to be developed under this joint venture for AI and medical technology. The draft guidance is currently available for a public consultation.

The guidance follows the risk management process set out in ISO 14971. It covers points that could be considered for AI/ML-based solutions in relation to the general requirements for the risk management system, risk analysis, risk evaluation, risk control, evaluation of the overall residual risk, risk management review, and production / post-production activities. The guidance explores particular challenges associated with AI/ML safety, such as data quality and security, bias, over-trust, and adaptive learning systems.

BSI AAMI 34971 provides users with an informative overview of the ISO 14971 risk management process. It also provides some examples of AI/ML-related hazards and some further considerations for autonomous systems.

The public consultation for the draft guidelines is open until 8th August 2021 and they can be accessed via the BSI Standards Development portal (registration required): https://standardsdevelopment.bsigroup.com/projects/2020-02770#/section 

The final version of BSI AAMI 34971 is expected to be published in late 2021. In future it is expected that this standard will be used as a basis for further global standardisation through the International Organisation for Standardisation (ISO), either through incorporation into existing standards and guidance, or as a standalone document.

Author: Rob Turpin is Head of Sector (Healthcare) at BSI Knowledge Solutions.  His key areas of focus cover standardisation for medical device regulation, digital healthcare, hospital services and population health matters.