How AI shapes healthcare’s ‘now’ and ‘then’

  • 4 minutes
  • Posted on 10.01.2023

How AI shapes healthcare’s ‘now’ and ‘then’

Giovanni Briganti

Chair of Artificial Intelligence and Digital Medicine, University of Mons

As a Lead of AI4Health at AI4Belgium (Belgium’s federal ecosystem for AI), holder of the Chair of Artificial Intelligence and Digital Medicine at the University of Mons and a Medical Doctor at the Brugmann University Hospital in Brussels, AI solutions have profoundly changed my work in healthcare. AI can impact the day-to-day of healthcare professionals (HCPs) in several ways: by removing unnecessary administrative burdens and helping the choice of diagnosis and treatment, supporting research, and helping clinicians conceptualise better recommendations to inform evidence-based policymaking. At the same time, AI also improves the quality of care and citizens’ health through tailored analytics and predictions. Moreover, it supports the logistics and functioning of healthcare institutions that face financial and organisational challenges.

What holds Europe back from using AI in healthcare

Some may see AI solutions as futuristic, but we need to understand that these technologies already exist in hospitals today. We need to shift society’s thinking towards increased adoption of effective AI technologies. Most challenges come from the fact that AI solutions are developed by engineers alone, instead of collaborating with clinicians. Normally, if an algorithm is tested and proves to outperform a certain standard set for an HCP’s activity, it can start being produced. Yet, reimbursing the technology, requires that it undergoes clinical validation like any other medical solution. For this purpose, what is needed is the right ecosystem, including, hospitals, clinicians, engineers, set KPIs and centralised entities that provide clinical validation.

A further challenge to AI adoption is the upfront cost it currently entails for hospital institutions. When approaching hospitals, most AI start-ups do not possess information on a hospital’s return on investment (ROI) to help with the decision of adopting the AI solution. We need to prove that AI solutions are worth it. Using a certain solution free of charge for 3-6 months during clinical studies can back its financial and clinical ROI. This mirrors the concept adopted by pharmaceutical companies that provide medication for free for a limited time. I firmly believe medtech start-ups need to learn from this habit of hospital co-creation of AI solutions. For instance, the targeted AI programmes that I lead are an example of how AI department champions can be formed among clinicians to see AI projects very pragmatically, which is the best path towards AI adoption.

Setting standards as the key to improving AI adoption in healthcare

While we acknowledge the intentions of current EU initiatives such as the European Health Data Space and the AI Act, there is still a lot to be done, for example filling in the discrepancies between all legislative proposals. AI standardisation will bring the use of AI in healthcare to the next level and improve adoption. It is important to build on existing international standards for data interoperability (such as for instance HL7 FHIR®) which have been issued by standard developing organisations (SDOs). These standards are helpful for decisions on how to treat and exchange data, but more initiatives are needed specifically for medical terminology. In my opinion, EU Member States should start thinking about harmonised standards for AI implementation in the health space. 

The EU legislative eco-system as an AI-enabler

I am delighted to see that the proposed EU framework of the AI Act includes provisions on the conformity for AI use. The criteria for post-marketing monitoring of the performance of AI systems are commendable and will support AI adoption. In terms of the proposed risk classification for AI use in healthcare, we need to keep in mind that some technologies can pertain to healthcare.  Still, in practice, the risk they pose can be very low, such as in the case of automated billing or natural language processing systems. The proposed AI Act should provide clearer provisions in these situations.

The vast opportunities ahead of AI in health in the EU

There are so many AI solutions in daily use in healthcare already: personalised surgical apps following a patient’s recovery after a heart attack, wearables predicting future seizures, virtual assistants, diagnostic software detecting abnormal structures of cancer from radiological images, platforms making recommendations on how to switch from one kind of chemotherapy to another based on patient’s data. The list is long. In my view, the first step towards increased AI adoption in healthcare is a European mapping of existing AI solutions that includes the way they function, how they perform and what their actual risk is. This can help Europe build up its capacity to transform healthcare through targeted and adequate AI technology and a targeted building of an AI stakeholder and legislative ecosystems.

The comments are closed.