The latest My City-Lab Talk Series on AI took place on 18 November in the framework of the digital Annual Congress of the European Health Management Association (EHMA). The purpose of the debate was to offer a dynamic discussion among key stakeholders and policymakers on the ways in which AI can support the screening and prevention of cardiovascular diseases (CVDs).
Cardiovascular disease is the leading cause of death in Europe. It causes over 1.8 million deaths each year or 37% of all deaths in the EU. Across Europe, deaths from CVD exceed those from cancer. Currently 60 million Europeans live with CVD and each year over 6.1 million new cases are diagnosed. Annually, the cost of CVD on EU healthcare systems is as high as EUR 210 billion, making cardiovascular disease one of the heaviest burdens on EU healthcare systems
The discussion aimed to bring together a range of stakeholders and policymakers for a debate on the ways in which AI can support the screening and prevention of cardiovascular diseases as well as on the benefits and risks of such technologies and the barriers to their deployment. Participants also reflected on some of the already existing strategies for CVD screening by means of AI and on the role of EU countries and the European Union in boosting patient safety through the uptake of AI-assisted solutions.
The debate was opened by MEP Maria Carvalho (Co-Chair of the MEP Heart Group, European Parliament), who stressed the great prevalence of CVD in Europe and the significant burden that it imposes on European healthcare systems. As stressed by MEP Carvalho, Successful deployment of AI technologies in the fight against CVD would be an important step towards advancing public health.
The debate also featured insights from:
- Professor Alan Fraser – Former Chair of the Regulatory Committee and Chair of the Biomedical Alliance in Europe Task Force on Medical Devices, European Society of Cardiology
- Birgit Berger – CEO, European Hearth Network
- Ed Harding – Director, The Heart Failure Policy Network
- Erik R. Ranschaert – AI Project Coordinator Radiologie, ETZ Hospital, The Netherlands
CVD is an area in which AI-based algorithms have shown great promise. Thanks to innovative solutions and big data, cardiologists have better tools to analyse images, detect early signs of CVD and perform more robust risk assessments thus allowing patients to be directed to the right type of care based on their individual symptoms and medical history. New algorithms based on machine learning enable healthcare professionals to predict patient outcomes more accurately and to offer increasingly personalised treatment.
While all of those applications can go a long way in improving the quality of the care offered to patients suffering from CVD, there are a number of limitations in the utility of AI algorithms that should be recognised. Even though there has been much optimism for boosting the efficiency of image diagnostics through AI, a deeper dive beneath the hype reveals that the gains in diagnostic reliability and accuracy are more modest than often suggested. Furthermore, while some indications exist that machine learning can meaningfully improve the accuracy of disease progression prognosis, the efficacy of such tools should not be overstated. Data remains insufficient to determine the actual reliability of those AI technologies. When dealing with big data one needs to also be aware of the danger of chance correlations as opposed to actual causation.
In this context, it is essential that AI is not seen as a panacea on its own but as a tool which when utilised properly by healthcare professionals can greatly enhance quality of care and patient safety. Delegating too much of the treatment decisions to AI algorithms when understanding of the technology is still underdeveloped, risks leaving practitioners unable to critically evaluate the way in which such tools perform in real-life situations. This could further exasperate the lack of trust both by patients and healthcare professionals, who often lack the necessary skills and digital literacy to fully comprehend the way in which AI algorithms make their decisions. To counter this issue, any large-scale deployment of AI in healthcare provision needs to be preceded by the establishment of clear and concrete EU guidelines for evaluating the efficacy of AI tools in medical devices. In addition, requirements to make machine learning code and data utilised by AI algorithms open source would greatly boost the ability of researchers to identify sources of potential bias as well as to validate the results of developed models.
Furthermore, the upscaling of AI technologies in healthcare systems in the European Union would require addressing the fact that many European healthcare systems are currently not prepared to offer the technological infrastructure necessary for AI algorithms to work properly. The progress in the implementation of electronic patient records varies widely across the EU. The state of the IT infrastructure in many European healthcare systems is often inadequate. Even frontrunner countries still face considerable challenges linked to facilitating data sharing and access.
Similar divergence across countries exists also when it comes to the national systems for reimbursement of costs linked to the use of AI-based healthcare services and telehealth provision more generally. Deploying AI without addressing those inconsistencies risks exasperating health inequalities as those who are less able to bear the additional costs or lack the necessary digital skills to make use of innovative devices will simply be left behind.
While AI’s potential to improve the quality of care for patients with CVD is clear, its successful large-scale deployment requires the development of a robust European AI ecosystem incorporating reliable IT infrastructures, health education frameworks providing healthcare professionals with the right digital literacy as well as a comprehensive European health data space, which is standardised, transparent and accessible for all. The establishment of such an ecosystem is only possible if paralleled by the development of a uniform ethical and legal framework for AI in healthcare that tackles issues of data protection, consent, and liability for misdiagnosis and data leakage.
In short, the successful deployment of AI in CVD care and management and healthcare more broadly, requires further technological advances but more than that it requires systemic political leadership. It calls for a cooperative and unified European approach which brings together the full range of relevant stakeholders from engineers to doctors to patients and promotes a common vision of innovative and patient-centred care. The European Union has an important role to play in facilitating this process.