The Medical Device and Health Technology Evaluation Committee (CNEDiMTS) is the French National Authority for Health (HAS) Committee which evaluates, in particular, medical devices in view of their reimbursement by the French health insurance scheme. Its task of scientific assessment kicks in once CE marking has been obtained.  

In 2018, HAS introduced an "e-health" section in its work program and announced the preparation of a Guide to the submission of a connected medical device-specific dossier. This Guide was published in January 2019. The specific features of connected medical devices are therefore taken into account during their clinical evaluation, in particular their very rapid technological development and the algorithms based on machine learning processes.

Nevertheless, the CNEDiMTS has noted the difficulties encountered by manufacturers to describe the artificial intelligence component of medical devices equipped with it. The CNEDiMTS has therefore developed a draft descriptive grid of machine learning algorithms in order to have the necessary information to evaluate the control of the clinical decision-making process. Through this grid, the CNEDiMTS' objective is "to make the Committee's examination of applications for inclusion on the list of products and services qualifying for reimbursement more fluid, to strengthen the confidence necessary for their use and their appropriation by professionals and users and, in fine, not to delay access to useful innovation" (HAS framework note of February 2019).

The adoption of the grid was preceded by a public consultation phase between November 2019 and January 2020 during which the players involved in the development or use of medical devices incorporating learning algorithms were able to submit their opinions and suggestions.

In a press release dated October 14, 2020, HAS announced the update of its submission guides with the introduction of the grid definitively validated by the CNEDiMTS. This grid assists applicants in the constitution of their application dossier for access to reimbursement or innovation package. It enables them to describe the medical device functions relying on machine learning processes. In the case of a medical device including several functions of this type, the manufacturer should complete one grid per function.

The descriptive grid, which is identical for a request for access to reimbursement or an innovation package, includes four categories of evaluation criteria:

• Purpose: describes the use of the medical device, in particular the role of the "smart" function, the characteristics of the target population and its operating environment.
• Data: describes input and output data. Real-life data is accepted.
• Model: describes algorithm training, validation, and testing, before and after deployment of the medical device.
• Functional characteristics: allows to describe the behavior of the model, in particular the performance and qualification, the system robustness, the system resilience, the explainability and the interpretability.

Manufacturers must fill in a total of 42 items, divided into these four categories and accompanied by information to help complete the grid. In the event that some of these items are not adapted to the type of technology in question, it is specified that the manufacturer must then expressly specify this and provide a justification. Conversely, it is also possible to mention any additional information deemed useful for the evaluation and not listed in the grid. The descriptive grid is accompanied by a glossary. Finally, HAS specifies that the grid will be updated to adapt to future technological upgrades.

It should be noted that HAS integrated the specific questions inherent to connected medical devices into the submission guide for inclusion on the list of products and services qualifying for reimbursement. HAS therefore chose to have a single guide insofar as “medical device connectivity is now featured in many dossiers submitted to the CNEDiMTS for evaluation”.

On November 18, 2020, HAS organized a webinar entitled "MD & Artificial Intelligence: what specificities for evaluation? ». The recording of this webinar, as well as a Frequently Asked Questions section are available on its website.

Next steps

The next step will be the publication by HAS of a "functional classification grid of digital solutions according to their purpose of use (screening, diagnostic, prevention aid, aid to understanding of hygienic and dietetic measures, etc.)". The draft was submitted for public consultation between April and June 2020 and should be finalized by the end of the year. The President of the CNEDiMTS, Isabelle Adenot, also indicated, during the webinar on November 18, that HAS will soon publish a guide on the organizational impact of medical devices and a joint guide for the drug, medical device and medico-economic commissions on post-registration studies.

Finally, over the next few months, we will closely monitor the final adoption of the French bioethics law, which includes provisions relating to the implementation of guarantees surrounding the use by healthcare professionals of tools enabling the algorithmic processing of massive data in healthcare.

Authored by Mikael Salmela