Working for a company that is at the frontline of digital innovation in medical technologies, I believe that with continued advances in artificial intelligence (AI)-enabled medical devices, the possibility for improving patient lives can reach an entirely new level. AI in healthcare has the potential to solve pressing challenges for European healthcare systems. However, there are several barriers that need to be addressed in order to further foster AI development and AI adoption.
The unexplored opportunities ahead of AI in health
With the increased development and availability of telemedicine services, a considerable part of healthcare delivery can take place at the healthcare facility level. Combined with the increased use of AI with our healthcare technologies, we are at the beginning of a shift to more predictable and personalized healthcare. Increasingly, the amount of data generated from patients will be too vast and will not be possible to be analysed in a meaningful manner by other means than AI.
How is AI really used in medical technologies today?
While there is a large discussion around it, today, the real footprint of AI in healthcare is still limited and only at the beginning of its full potential. For example, when you are driving your car with navigation, you receive more support from algorithms than a surgeon in a surgery room. Yet, currently, AI already supports surgeons in multiple ways. Some impressive examples I have witnessed in the operating room are:
- AI for increasing patient safety: AI can alarm a surgeon if surgical tools are out of the surgeon’s sight during minimally invasive surgery.
- AI for training purposes: AI can remove non-relevant information during surgery training videos;
In order to further advance development of AI-enabled medical technologies, it is critical that the regulatory requirements be calibrated to the risk profile of the product. For example, under the EU AI Act it is proposed that any AI-enabled technology that requires a notified body for a conformity assessment under EU MDR is considered “high risk” under the EU AI Act, but this includes products with very different clinical risk profiles.
For this reason, we were surprised that under the proposed AI Act low- risk medical technologies would almost automatically qualify as high-risk AI, triggering additional essential requirements, pre- and post-market assessment and surveillance obligations that will significantly slow down release to market. We support MedTech Europe in its strive for a system where the risk profile of the AI components of a MD is assessed as part of the risk profile of the overall device consistent with MDR.
Challenges from the R&D room
There are several barriers, which need to be addressed to roll out AI in healthcare:
- A key challenge in the research and development phase is the availability of high-quality health data. Firstly, this is due to the fact that historically, health data has not systematically been recorded electronically, or not the same data for the same treatment, or are not interoperable. Secondly, this is partly due to the General Data Protection Regulation (GDPR). With health data being a sensitive data type under GDPR, players in the healthcare space prefer to take a conservative approach when it comes to interpreting the rights to re-use or share health data for AI development, training and testing purposes.
- In addition, there are challenges related to the proposed way that AI in medical devices will be regulated. The horizontal requirements set out in the proposed AI Act, combined with the sectoral requirements from the Medical Device Regulation and the In vitro Diagnostics Medical Devices Regulation, risk duplications of and possibly conflicting requirements, that will lead to potential delays in patient access to these important technologies.
- Finally, there is a need for great trust from patients and physicians in AI use in healthcare because it cannot work if they do not adopt it. For that, education, multi-stakeholder cooperation, patient empowerment and smart regulation are key.
How can the EU help address these barriers
First of all, policymakers and all stakeholders need to recognize that the use of health data can bring social benefits. A balanced regime, that balances the protection of patients’ data rights with the use of health data for certain purposes that are considered to be legitimate, subject to safeguards, can allow to unlock those social benefits. This is an important consideration with regulating health data.
This is partially being addressed by the European Health Data Space (EHDS), but we are yet to see how and whether it will solve the specific GDPR issues. To stimulate innovation, also on the side of those generating and holding health data in the current ecosystem, when building the new EHDS, policymakers need to engage with industry and consider the implications of Intellectual Property and Trade Secrets (IP) for manufacturers both as contributors and recipients of the EHDS. Finally, the future horizontal AI Act should be better aligned with the EU MDR/IVDR, to allow for a regulatory framework that takes a fit-for-purpose risk-based approach, taking into account the specificities of AI development while acting in favour of patients and in the interest of rolling out AI in healthcare in the EU. For example, the conformity assessment processes within the AI Act need to be clarified and aligned with the EU MDR/IVDR, and it needs to be ensured that the competent authorities (Notified Bodies for medical technologies) are ready to conduct AI assessments according to the AI Act, to avoid bottlenecks when bringing solutions to patients and health care systems.
I believe that AI in medical technologies can truly contribute to better patient outcome, improved and sustainable access to healthcare and the global competitiveness of the EU, but the right policies, and a balanced regulatory ecosystem needs to be put in place.1