From mandate to mechanism: Closing the delivery gap in cardiovascular screening

From mandate to mechanism, Parker J.,(1,2,3), Jackson T.,(1,3) and Eriksson O (1,4), discuss artificial intelligence and the delivery gap in population cardiovascular screening

Cardiovascular disease is the leading cause of death and disability in the European Union (EU), causing around 1.7 million deaths a year and costing over €282 billion annually. (1) The European Commission’s ‘Safe Hearts Plan’, published on 16th December 2025, is the first comprehensive EU-level strategy to address it. (2,3) Its ambition has been widely welcomed, but commentators have questioned whether it can be delivered in the absence of a clear implementation pathway and sustained funding. (4,5) Here we offer the perspective of a developer of artificial intelligence (AI) technology for cardiovascular image analysis.

The Plan rests on three pillars: prevention; early detection and screening; and treatment, care and rehabilitation. In doing so, it targets the right levers to change cardiovascular disease statistics and correctly frames digital innovation and AI-enabled infrastructure as essential to all three. Yet, a mandate to screen does not equate to a system capable of delivering cardiovascular screening, much less the long-term outcomes intended. Applied well, we believe technology can address the structural gaps that have decided the success or failure of past screening programmes. Here we examine five.

Population-scale screening will create tens of millions of additional clinical data points, and Europe neither possesses nor can train the clinical workforce within the relevant timeframe to interpret them by conventional means. The World Heart Federation has observed that the Plan does not acknowledge the shortages in cardiology, nursing and primary care that directly constrain prevention, screening and treatment capacity across many Member States. (5) This constraint is fundamental and is not relieved by policy ambition alone. The capacity issue is reflected in calls for automated tools to augment and upskill the existing workforce to meet radiology demand. (6)

Established screening principles state that a programme benefits patients only where a care pathway can act on its findings. (7) A clinical finding identified without defined next steps tends to generate patient anxiety, cost and unnecessary investigation rather than fewer events. Therefore, any technology that increases detection must also be judged against the system’s capacity to act on it. Implementing care pathways, however, can transform what screening achieves: in the VIVA trial, men aged 65 to 74 invited to combined screening for abdominal aortic aneurysm, peripheral arterial disease and hypertension, with referral for preventive drug treatment, had a significant reduction in all-cause mortality (hazard ratio 0.93) attributed primarily to that treatment, unprecedented in population screening. (8)

Cardiovascular risk evolves over time, and screening that genuinely improves outcomes must track the individual longitudinally. Yet patient data remains fragmented across imaging archives, laboratory systems and electronic records, so accurate detection at a single point in time cannot readily become sustained prevention. Building the integrated infrastructure to close that gap, such as a platform that unifies cardiovascular data points across modalities and tracks them over time, is a large and multifaceted undertaking; but one these authors are actively pursuing.

A screening programme benefits only those it reaches, and inequities in access to care are well documented. (9) Screening programmes tend to reach people already engaged with health services, who are better resourced. While underserved groups, whether defined by geography, socioeconomic status, sex, ethnicity or age, are reached last or not at all. This pattern, long described as the inverse care law, is typically seen in populations with the highest cardiovascular risk. (10) The Plan’s impact will therefore be limited by how much of the at-risk population its pathways reach. Additionally, because AI tools are typically trained on data from those already in contact with services, they can widen this gap unless representation in training and validation are treated as design requirements rather than assumptions.

The population risk instruments in current use, such as SCORE2 in Europe (11), QRISK in the United Kingdom (12) and nationally derived models such as the Icelandic Heart Association’s Reykjavik Study risk estimator (13), rest on a limited set of conventional variables and treat broad demographic groups as homogeneous. They make little use of an individual’s own physiology and typically underutilise existing imaging. A screening programme is constrained by the risk model that directs it, and existing models lack the resolution the Plan’s ambition requires.

AI-enabled technology can address these gaps, and we will explore each, in turn, in future contributions to this journal. We also highlight an untapped resource not considered in the Plan, pre-existing medical imaging. Cardiovascular disease is often visible on scans acquired for unrelated reasons, such as lung-cancer screening. Therein lies an opportunity to ‘make every image count’ and opportunistically screen for cardiovascular disease in all future and historical imaging.

AI makes this a possibility. It can detect but, more importantly, quantify clinically meaningful cardiovascular findings from imaging, at a scale no human workforce can achieve. This longitudinal assessment can enrich and inform an individual’s risk profiling and is a real step toward personalised medicine. Extending this capability across imaging modalities is a focus of our own development.

AI plays a major role in the Plan, and although we believe it is necessary to achieve the desired outcomes, we must acknowledge its current limitations. They correspond to the same five gaps discussed here. An algorithm that increases detection without an actionable pathway exacerbates the detection-benefit gap; one validated on an unrepresentative dataset exacerbates the equity gap; and one optimised on a narrow sample misstates individual risk, undermining both precision and generalisability. AI contributes meaningfully only when developed to the standard these gaps demand; that is the responsibility of developers, but must equally be insisted upon by the Member States adopting these solutions.

This principle informs our own work. ‘AI-CARE’, an international, multi-centre validation study, evaluates the generalisability of an AI tool across heterogeneous populations and equipment, rather than optimising for a narrow sample (the most efficient route to market but one often flawed in real-world deployment). AIATELLA’s tools are produced under a certified quality management system in accordance with the EU Medical Device Regulation, and embedded in longitudinal cohort science. While the authors would advocate for accelerated regulatory initiatives to bring mass screening into practice as soon as possible, the same robust principles of medical device development must not be compromised.

The ‘Safe Hearts Plan’ has made cardiovascular screening a European priority and secured the commitment to pursue it. Translating that into measurable reductions in cardiovascular events is now an engineering, evidentiary and infrastructural challenge as much as a political one. Our subsequent contributions, developed with field experts, will take these gaps in turn, exploring how technology can address them. The aim is to shorten the distance between a policy mandate and a prevented cardiovascular event, and to argue that, for the first time, that distance is no longer a matter of ambition but of execution.

1. AIATELLA Oy, Helsinki, Finland. 2. Northumbria University, Newcastle upon Tyne, UK. 3. Arizona Heart Foundation, Phoenix, U.S. 4. Aalto University, Espoo, Finland.