No one involved in healthcare, including the patient, can ignore it any longer. Artificial intelligence (AI) is deeply embedded in modern healthcare and is evolving rapidly. AI has already provided added value in the clinical interpretation of medical imaging, identification of chronic conditions based on algorithms, and much more. In short, thanks to AI, our quality of life can be improved. AI also provides a potential answer to the government’s objectives to deal with limited financial resources in an efficient way.

However, despite innovative joy, one needs to remain critical. In theory, AI trains itself based on the availability of clinical data. Suppose your company is the driving force in the development of a break-through drug. In a world without AI, clinical outcome measures will be thoroughly analysed to assess whether the drug offers clinical benefit compared to a predefined comparator or setting. Once approved by the FDA and/or the EMA, the developer can consider market access.

Imagine that clinical expertise is partially or entirely taken over by AI. It would not surprise us if AI manages to define subgroups that benefit more or less from the use of a specific drug. This would be fine as long as no data bias leads to potentially incorrect conclusions, which would lead to restriction of access to the innovation for a certain subgroup, due to an over-confidence in AI-generated conclusions.

A less highlighted aspect is the deliberate manipulation of source data. It seems plausible that virtual ‘real-life’ data can be created by money-hungry entrepreneurs, after which AI potentially uses these virtual data when drawing conclusions. The opposite can also occur: correct information may not be adequately recognized by AI because competing companies place ‘wrong’ data on the internet.

In short, the use and integration of AI in current regulation and legislation seems to be a valuable addition to assessing and interpreting a solid database. However seems prudent to only use certified data while ensuring individual privacy. For example, individual patient data could be provided with a digital stamp to prove its authenticity.

To limit data bias, when setting up clinical study protocols, the target group should be described in such way that AI driven decisions are reliable and reproducible. The set up of optimal inclusion and exclusion criteria can be tailored so that sufficiently reliable data become available for each potential subgroup without favouring a subgroup due to data bias.

And what about uncertainty? AI can lead to artificial decisions where (un)certainty is not reproducibly measurable but where the whole dataset itself is significantly better versus the selected comparator and/or setting (e.g., predicting the chance of developing disease conditions in relation to the patients lifestyle). Additionally, there are cultural and region-specific influences that can impact the formation of AI conclusions (e.g., treatment guidelines, expertise,  reimbursement criteria, …). To cover this variability at international level, governments and regulators need to collaborate in a consolidated manner.

In anticipation of the forthcoming EU Joint HTA, a regulatory framework is currently being developed for ‘Joint Clinical Assessments’ (JCA) via Regulation 2021/2282. ATMPs and cancer drugs will be subjected to this regulation as of January 1, 2025, followed by orphan drugs in 2027. To date, there is no specific discussion about the potential impact and/or integration of AI within these JCA. Nevertheless, it seems important to include AI within the regulatory framework at EU level.

As a market access consultant, we noticed several companies are preparing for the EU Joint HTA by setting up international working groups. We feel our clients’ need for an external partner to serve as a conduit for tracing risks and opportunities and are happy to discuss the potential impact of AI with you.