COCIR Artificial Intelligence in EU Medical Device Legislation

COCIR Artificial Intelligence in EU Medical Device Legislation

COCIR published an analysis in 2020 on AI-based devices in the scope of MDR and IVDR.

The purpose of this analysis is to provide an update on several topics/issues related to AI-enabled devices as well as to show how an AI device can be compliant with regulations in order to also provide better user confidence in AI enabled devices.

COCIR makes concrete proposals throughout this analysis and offers recommendations vis-à-vis the regulatory environment, including possible approaches to software change management.

It is important to note that since the publication of this guide predates the publication of the proposed European regulation on AI (AI act of 2021), the proposals made by the COCIR regarding AI legislation have been partly taken up in the AI act.

In the COCIR recommendations, a point of honor is given to the implementation of general basic principles related to the use of AI systems such as ethics, trust, transparency, explicability or confidentiality and data protection according to European regulations.

Thus, the MedTech association through these recommendations places risk management at the center of the processes of design, development and use of AI in the health field.

The COCIR recommendations also address the need to update standards used in the healthcare field to reflect AI-enabled devices:

  • IEC 82304-1 so that manufacturers describe in user information:
    • The ability of the device to provide human control as well as describe the provisions under which the device evolves over time (fixed or evolving AI, description of possible changes and how the human can control them),
    • Provisions regarding the possibilities of modification of the device by the user,
    • Inform about the performances of the AI-based device to better discern the results that can be expected from the AI system.
  • IEC 62304 to incorporate documentation of an “Algorithm Change Protocol (ACP)” for learning AI devices that change by learning. Predetermining the type of change allows a better adaptation of the patient or user and also allows the device itself to adapt to the patient. This protocol (ACP) will therefore be incorporated into the technical documentation for review during the conformity assessment.

Note that the MDR and IVDR do not currently allow manufacturers to market AI-based devices intended to be modified outside of a predetermined envelope of modifications or to suggest claims, intended uses, or conditions of use of the device for which a conformity assessment has not been conducted.