AI solutions for health will thrive and benefit patients in Europe only through a coherent and ethical strategy that will prepare governments’ health systems and mitigate any unintended consequence of AI’s use. AI will play an important role as diseases like cancer are expected to rise due to Europe’s ageing population.
A study predicts AI alone has the potential to add 0.2–1.3 years onto the average life expectancy.AI can improve and accelerate the development of safe and effective medicines, support chronic disease management (such as diabetes), enhance the information available for screening and treatment decisions as well as provide continuous monitoring tools supporting diagnosis or tracking disease progression. They also have the potential to deliver tools to better understand the risk for future diagnoses, stratify patient populations for more precise treatment options and patient management, improve adherence to treatments and ultimately improve clinical and patient outcomes. DIGITALEUROPE calls on relevant policymakers, such as the European Commission, Member States’ governments and elected representatives, to:
Prioritise risk assessments and avoid one-size-fits-all approaches in the design of any future AI policy framework for health. The risk of deploying AI in this domain greatly varies under the specific application considered. Predicting hospital attendance is very different from life-saving AI solutions that diagnose a certain disease. Any future framework should also consider existing international standards, legislation and ethical principles.
Empower AI development for a health data ecosystem, by earmarking more resources to foster the accessibility and interoperability of health data and addressing its provenance as well as curation. In addition, accelerate health data sharing through FAIR (Findable, Accessible, Interoperable and Reusable) principles.
Promote policies that encourage the development of AI solutions for clinicians integrating Electronic Health Records (EHRs).
Tackle the creeping fragmentation and regulatory divergences of health data-processing across the EU, while accelerating the creation of a European health data space. Even if the GDPR allows each Member State to introduce further data-processing conditions for genetic and health data, the Commission and national data protection authorities should strive for a harmonised framework of rules.
Foster the adoption of initiatives on the secondary use of health data for AI research. They should be deployed at scale across the EU.
Commit sufficient resources to the development, adoption and implementation of AI in health applications.
Recognise the benefits of all AI applications for health systems and society. The technology holds indeed potential in a variety of patient treatment solutions. It is equally helpful in areas away from pure patient care. For example, Natural Language Processing, a branch of AI, can free up precious time for practitioners by dramatically speeding up EHR documentation, one of the most time-consuming tasks in today’s health care context.
Adjust the implementation of the EU health regulatory framework to allow new emerging AI solutions on the market. The overall objective and requirements of existing legislation are generally fit for purpose to cover existing AI products and services in health. The Medical Device Regulation and subsequent guidance will require market assessment bodies who are enforcing the regulation to be regularly trained to understand thoroughly the inner workings of AI, including new mhealth solutions in the EU. Training should also clarify how to exploit the self-learning capabilities of AI systems in health. If practitioners, patients and society are to benefit from AI fully, our regulatory approach to health technology must move in parallel to market innovation.
Proportionality is the key in the discussion on the explainability of AI decisions specifically in respect to the intended use of the technology, and the corresponding potential risk for the patient. Explaining how the algorithm works should be understood as an occasion for the user to obtain meaningful information. It should not be about sharing source code or training data.
Recognise that developing completely bias-free algorithms will never be possible. However, policymakers can incentivise research, training and increase data availability to tackle and reduce potential unintended or discriminatory bias in AI algorithms.
Advance ambitious upskilling and reskilling programmes tailored to the specific needs of all relevant health stakeholders, be them patients, doctors or health market authorities certifying AI technology. Training should be complemented by awareness-raising campaigns on the role of AI in health.
Firmly consider cybersecurity as an essential element for trust in AI in health. We reiterate our support to strengthen cybersecurity across the EU through a framework for the development of voluntary cybersecurity certification schemes based on already internationally recognised schemes.