Beyond the algorithm: health technology assessment frameworks for AI in cardiology under the European Union Health Technology Assessment Regulation: a systematic review

Introduction

Artificial intelligence (AI) is rapidly transforming cardiovascular medicine, powering applications across electrocardiogram (ECG) interpretation, echocardiography, coronary computed tomography (CT) analysis and remote monitoring. These technologies offer the promise of earlier diagnosis, more precise risk stratification and improved clinical efficiency (1). However, their rapid adoption also presents challenges to traditional health technology assessment (HTA) processes, which were originally developed to evaluate static interventions such as pharmaceuticals and conventional medical devices. This mismatch reflects a broader methodological tension between the rapid pace of AI innovation and the comparatively slower processes of evidence generation and HTA decision-making (2-4).

In the context of cardiology, several barriers to effective HTA have emerged. These include limited involvement of clinicians, particularly cardiologists, in the design and development of HTA methodologies, fragmented implementation of digital health technologies (DHTs), and inconsistent assessment practices across European countries (4,5). Cardiologists, while central to determining clinical endpoints and navigating real-world application, are often underrepresented in HTA activities. Furthermore, the integration of AI into clinical workflows has outpaced the development of evaluative tools tailored to its unique features, leading to reimbursement misalignment and regulatory uncertainty across jurisdictions. This gap has raised increasing concern among regulators and policymakers regarding the readiness of current HTA frameworks to support safe, effective, and equitable implementation of AI-based technologies (3).

Crucially, AI systems differ fundamentally from conventional medical technologies. They are adaptive, data-driven, and deeply embedded within clinical processes. Many cardiovascular AI tools, such as real-time ECG analysis or AI-enhanced echocardiography, function continuously and are retrained over time, often in response to changing datasets, patient populations or clinical guidelines (6). These characteristics pose specific lifecycle challenges, ranging from initial pre-clinical development to iterative updates, validation of comparative effectiveness, and determination of appropriate reassessment thresholds (7). Such lifecycle complexity is rarely captured by traditional HTA, which generally assumes a one-time evaluation of fixed, non-adaptive interventions. As a result, conventional HTA approaches may underestimate uncertainties related to performance drift, generalizability, and long-term clinical impact (2,3).

Cardiovascular AI tools often operate in continuous, ambulatory, or embedded clinical workflows, such as AI-assisted ECG interpretation, echocardiographic triage, or home-based monitoring, which differ significantly from the episodic or diagnostic imaging models common in oncology or radiology AI. These systems must adapt to diverse signal inputs and changing physiological contexts, making static, one-time evaluations inadequate. Therefore, cardiovascular AI presents a compelling case for domain-specific, lifecycle-aligned HTA that accounts for longitudinal performance, usability, and integration in clinical decision pathways. Despite this, there remains limited consensus on how such adaptive characteristics should be systematically incorporated into HTA methodologies (2,3).

To address fragmentation in HTA across Member States, the European Union (EU) adopted the Health Technology Assessment Regulation (HTAR; Regulation 2021/2282), which entered into force in 2022 and will become mandatory in phases through 2030 (8). The EU HTAR introduces a coordinated framework for joint clinical assessments (JCAs) of selected high-risk technologies, including certain medical devices, with the objective of harmonizing the evaluation of clinical effectiveness and safety at the European level. The EU HTAR places emphasis on methodological transparency and structured stakeholder engagement and will support voluntary cooperation towards the progressive use of real-world evidence (RWE) relevant for HTA. However, AI is still not explicitly within the scope of the EU HTAR, so it has not provided operational guidance on AI-specific challenges such as algorithmic updates, adaptive performance or digital biomarkers, creating a potential gap between regulatory ambitions and evaluative tools. This creates uncertainty regarding how AI-based cardiovascular technologies will be assessed within forthcoming JCAs, particularly in relation to lifecycle management and evidence requirements (9).

In parallel, initiatives such as EDiHTA (10) and ASSESS-DHT (11) have proposed more agile, modular approaches to the assessment of DHTs. However, these initiatives remain under development, and empirical evidence demonstrating improved agility or effectiveness compared with established HTA frameworks is not yet available. While promising, its specific relevance and adaptability to cardiovascular AI tools, including AI-enabled ECG platforms, remote monitoring systems and decision support for echocardiography, require further investigation. Additionally, the DECIDE-AI reporting guidelines have sought to bridge the translational gap between early-stage development and clinical evaluation of AI-based interventions, offering structured guidance for pilot studies and feasibility testing. These efforts highlight an evolving but still fragmented landscape of methodological approaches to AI evaluation (7).

The relevance of aligning AI-based cardiovascular technologies with European HTA processes is further underscored by the recent publication from the HTA Coordination Group’s Emerging Health Technologies (EHT) Subgroup. Their 2025 report identifies two cardiology-related medical devices as potential candidates for JCA under the EU HTAR in 2026 (12). This early prioritization signals an increasing regulatory focus on cardiovascular innovations and reinforces the need for methodological readiness. Ensuring that current HTA frameworks are adaptable to the evolving nature of AI-enabled cardiology tools is therefore both timely and critical for shaping future EU-level evaluations.

This systematic review therefore aims to assess how current HTA frameworks address the evidentiary, lifecycle, and regulatory requirements specific to AI-based cardiovascular technologies and whether they align with the EU HTAR, by synthesizing existing methodological guidance and applied examples to identify gaps in current HTA approaches and outline pathways to develop AI-adaptive, cardiology-specific HTA tools. We present this article in accordance with the PRISMA reporting checklist (13) (available at https://atm.amegroups.com/article/view/10.21037/atm-2026-0052/rc).

Methods

The primary objective was to assess how current HTA frameworks address the evidentiary, lifecycle, and regulatory requirements specific to AI-based cardiovascular technologies and evaluate their alignment with the EU HTAR (Regulation 2021/2282) (8).

The review protocol was registered in PROSPERO (CRD420251150487) prior to the commencement of the study.

A structured literature search was conducted on 5^th December 2025 across three major databases: PubMed, Scopus, and ScienceDirect. The search strategy was designed to capture peer-reviewed publications from 1^st January 2020 to 1^st December 2025, reflecting a period of significant growth in AI development and progressive regulatory implementation under HTAR. In addition to database searches, relevant grey literature sources were consulted, including publications and documentation from the European Society of Cardiology (ESC) and the European Commission, to capture policy and methodological guidance related to AI and HTA. Search terms were developed to encompass key concepts across three domains: AI, cardiovascular medicine, and HTA or related policy frameworks. Boolean operators were adapted to database constraints while maintaining semantic consistency. The full search strategies, including database-specific Boolean operators, are provided in Appendix 1.

Records were eligible for inclusion if they reported original research or conceptual work related to the evaluation, regulation or implementation of AI-based technologies relevant to HTA. Relevance was determined based on inclusion of cost-effectiveness analysis, policy evaluation, lifecycle considerations, use of RWE or engagement with regulatory or HTA frameworks. Studies focused solely on clinical or technical performance, without clear links to cost-effectiveness, policy relevance, lifecycle considerations, or integration with HTA frameworks, were excluded.

All records identified through the database search were imported into Rayyan.ai, a web-based platform for systematic review screening and collaboration. Duplicate entries were removed prior to screening. Titles and abstracts were screened independently by a junior reviewer under the supervision of a senior researcher. This approach reflects a single-reviewer screening process with supervisory oversight. Discrepancies regarding eligibility were resolved through consensus. Full-text articles were subsequently retrieved and evaluated using predefined inclusion and exclusion criteria. Reasons for exclusion were documented to ensure transparency and methodological rigor.

A structured data extraction template was developed in Excel to chart relevant information from included studies. Extracted data included citation details, country of origin, publication year, AI application type, cardiovascular focus area, study design, use of HTA frameworks, incorporation of RWE, stakeholder involvement, lifecycle considerations, and reference to HTAR or related regulatory standards.

To support interpretation, studies were organized thematically based on their primary focus. Thematic clusters included clinical and technical evaluations, HTA- and policy-oriented studies, and conceptual or methodological frameworks. This clustering informed the narrative synthesis and highlighted distinct approaches to evaluating AI technologies across the evidence base.

In addition to mapping the included literature, a comparative analysis of HTA frameworks was undertaken to assess their relevance for AI-based cardiovascular technologies. Frameworks were selected based on their use in included studies, citation frequency in relevant literature, and prominence in European HTA practice. Each framework was assessed across six analytical dimensions: cardiovascular specificity, lifecycle integration, use of RWE, stakeholder engagement, regulatory alignment with HTAR, and adaptability to AI system updates. The findings of this comparative analysis are summarized in Table 1. Scores reflect the extent to which each initiative explicitly addresses the evaluated dimensions in publicly available documentation, rather than their level of validation, maturity, or implementation readiness.

Given the heterogeneity of study designs, a flexible domain-based appraisal approach was adopted. Elements from established critical appraisal frameworks were considered, although no single validated tool was fully applicable across all included study types. This tool was adapted from standard critical appraisal frameworks and evaluated six domains: clarity of aims, methodological transparency, external validation or generalizability, use of structured HTA frameworks, stakeholder engagement and transparency regarding funding or conflicts of interest. A single reviewer conducted the appraisal under the supervision of a senior researcher, with any uncertainties resolved through discussion. Risk levels were categorized as low, moderate or high for each domain. Each domain was assessed qualitatively based on the extent to which the study explicitly addressed the criterion, rather than using predefined quantitative thresholds. A summary of this appraisal is provided in Table 2.

The domain-based appraisal was chosen due to the heterogeneity of the included studies, which spanned clinical evaluations, economic modelling, and conceptual frameworks. Traditional tools such as ROBIS or AMSTAR 2 were deemed unsuitable for this purpose, as they are typically designed for homogeneous interventional or review-based literature. We acknowledge that this approach does not replace validated risk-of-bias tools and may limit comparability across studies. By assigning domain-specific scores (low, moderate, high), this flexible yet structured approach enabled comparative assessment across studies while maintaining consistency with HTA-relevant criteria.

Results

A total of 223 records were screened, of which 33 were retrieved for full-text review. After applying eligibility criteria, six studies were included in the final synthesis. The selection process is outlined in the PRISMA flow diagram (Figure 1). The included articles spanned publication years from 2021 to 2025 and represented a range of settings, including the United Kingdom, Finland, the Netherlands, Austria, and Brazil.

Figure 1 PRISMA 2020 flow diagram of study selection. Search conducted on 5^th December 2025.

Table 3 summarizes the characteristics of the six included studies, which were grouped into three thematic clusters: clinical and technical evaluations (n=3), economic and policy-oriented studies (n=2), and conceptual and framework design (n=1). Although none of the studies explicitly referenced formal HTA frameworks in their methodology, each was included based on its contribution to HTA-relevant domains, such as economic modelling, real-world evaluation, regulatory alignment, or lifecycle considerations.

Clinical and technical evaluations

Three studies were categorized under clinical and technical evaluations. These papers were selected not solely for their clinical focus, but because they incorporated dimensions relevant to HTA, such as model validation using real-world data, consideration of implementation challenges and early-phase lifecycle features.

For instance, one cross-sectional study assessed AI classifiers for stroke outcome prediction using real-world electronic health records, indirectly informing clinical effectiveness evaluation (14).

Another study employed machine learning to identify cost-related risk factors in a cardiac rehabilitation program, offering insights into resource allocation and economic implications (15).

The third study, a protocol paper, introduced the DETECT project: a machine learning-enabled wearable wristband for real-time cardiac arrest detection. Although still in development, the study incorporated key lifecycle considerations such as system responsiveness, integration into emergency response workflows, and real-time monitoring. These elements resonate with the lifecycle model proposed in DECIDE-AI and represent an example of anticipatory evaluation in the design phase of DHTs (16).

Together, these studies highlight technical feasibility and clinical applicability of cardiovascular AI, but reveal limited interaction with established HTA frameworks or regulatory alignment.

HTA, economic, and policy-oriented studies

Two studies were explicitly designed around cost-effectiveness and health economic evaluation in cardiovascular AI contexts.

The first assessed the economic impact of a novel AI tool that quantifies coronary inflammation via coronary CT angiography. Conducted in collaboration with the National Institute for Health and Care Excellence (NICE), the study incorporated quality-adjusted life-years (QALYs), National Health Service (NHS)-specific cost parameters, and probabilistic modelling to support real-world reimbursement decision-making. It provides one of the few direct examples of HTA-congruent evaluation in cardiovascular AI (17).

The second evaluated a machine learning algorithm used to identify undiagnosed atrial fibrillation in primary care through the PULsE-AI trial. Using data from a randomized controlled trial, the study demonstrated the integration of AI-based risk stratification with robust cost-effectiveness modelling, although lifecycle adaptability and regulatory foresight were not explicitly addressed (18).

Both studies underscore the potential of empirical economic evaluations to generate HTA-relevant evidence, particularly for reimbursement and coverage, while also illustrating ongoing methodological gaps related to RWE and adaptive system evaluation.

Conceptual and framework design

A single study contributed primarily to conceptual advancement, proposing a patient-level computational model for optimizing atrial fibrillation management. Through individualized simulations and decision support logic, the study emphasized transparency, personalization, and clinical interpretability. While it did not explicitly reference regulatory frameworks, it aligns conceptually with HTA principles related to explainability and shared decision-making (19).

This contribution reflects an emerging trend toward incorporating AI explainability and personalization into decision-support systems, an area of increasing interest to HTA bodies evaluating digital technologies.

Cross-study observations

To facilitate interpretation across a heterogeneous evidence base, we applied a lifecycle-oriented analytical lens adapted from the DECIDE-AI framework. This model, originally designed to guide early-phase clinical evaluation of AI-based decision support tools, was modified to reflect HTA-relevant stages spanning development, deployment and post-market evaluation (Figure 2). This adapted lifecycle framework helped situate each study’s contributions within the broader trajectory of AI implementation and highlighted gaps in post-deployment monitoring, RWE integration and iterative evaluation, all critical for alignment with the EU HTAR.

Figure 2 Alignment of HTA and AI lifecycle. AI, artificial intelligence; HTA, health technology assessment; RWE, real-world evidence.

Despite diversity in geographic origin, clinical focus and AI technique, several common features and limitations were observed across the corpus.

RWE was incorporated in four of the six studies, most commonly through the use of retrospective electronic medical records (EMRs) or data from prospective trials. However, few studies extended this to long-term implementation outcomes or post-deployment monitoring (14,15,17,18).

Lifecycle evaluation, emphasized in frameworks such as DECIDE-AI, was largely absent. Only the DETECT study considered integration, real-time responsiveness, and early-stage deployment planning (16).

Engagement with stakeholders, particularly clinicians or cardiologists, was notably underreported. While clinician relevance was implied in several study designs, formal engagement processes, such as co-design, usability testing, or consensus development, were not documented in any of the studies reviewed.

Moreover, while economic modelling and cost-effectiveness analysis were robust in two studies, use of structured HTA frameworks was limited. No study referenced the EUnetHTA Core Model (20) and only one was explicitly linked to NICE-congruent processes (21).

Comparative analysis of HTA frameworks

A comparative analysis of three current HTA frameworks, EUnetHTA Core Model (20), NICE Evidence Standards Framework (21), and ASSESS-DHT (11) is presented in Table 1.

The EUnetHTA Core Model remains the most widely referenced, offering detailed coverage of clinical and organizational domains. However, its structure is optimized for static technologies and lacks provisions for adaptive performance, real-time updates, or retraining. It does not address software as a dynamic intervention and has limited capacity for lifecycle-based assessment.

The NICE Evidence Standards framework introduces tiered evidence thresholds and accommodates early-stage evaluations, particularly for DHTs. Nonetheless, its application to embedded AI in cardiology remains limited. It does not fully accommodate continuous-use AI or support frameworks for retraining and versioning.

ASSESS-DHT, a newer and still pilot-phase initiative, offers a modular structure designed for adaptability. It incorporates planned and exploratory pilot activities, stakeholder feedback mechanisms, and explicit attention to lifecycle considerations. At present, these pilots do not encompass full HTA processes, and their capacity to inform routine HTA decision-making remains to be demonstrated.

Among the frameworks, NICE’s tiered structure offered greater flexibility for early-stage AI evaluations, while EUnetHTA remained stronger in clinical evidence domains but less agile. ASSESS-DHT, though promising in lifecycle and stakeholder components, remains under development and lacks cardiovascular validation. None of the frameworks fully address real-time adaptability or retraining of AI tools.

As illustrated in Figure 2, most included studies generated evidence concentrated in early lifecycle phases, with limited attention to post-deployment monitoring, iterative updates, or reassessment triggers.

Across all three frameworks, gaps remain in addressing the full spectrum of HTAR-aligned priorities, particularly in integrating RWE, stakeholder co-design and lifecycle adaptability.

Discussion

This systematic review highlights both the progress and persistent gaps in how AI-based cardiovascular technologies are being evaluated within health systems. The six included studies span a spectrum of clinical, economic and conceptual contributions, reflecting a growing interest in the potential of AI to support cardiovascular diagnosis, monitoring and care optimization. However, none of the studies explicitly applied established HTA frameworks, and alignment with the evolving requirements of the EU HTAR was limited or indirect.

A central finding concerns the limited consideration of lifecycle dynamics in current evaluation practices. Cardiovascular AI tools are frequently deployed in real-world clinical environments, rely on continuous data streams, and may undergo iterative updates after deployment. These features contrast with the static assumptions underlying many HTA models originally designed for pharmaceuticals and fixed medical devices. Although a small subset of studies acknowledged lifecycle complexity, this perspective was rarely operationalized within formal evaluative frameworks. The absence of structured mechanisms to assess algorithm updates, performance drift, or contextual adaptation may limit the accuracy and relevance of HTA conclusions over time (16,17,19).

The limited use of RWE represents a closely related challenge. Although the EU HTAR explicitly promotes the provision of RWE relevant for HTA, only a minority of included studies incorporated prospective implementation data or evidence generated in routine care. Most relied on trial-based, retrospective or simulated datasets, which may inadequately capture performance variability, workflow integration and patient heterogeneity encountered in routine cardiovascular practice. This suggests a methodological lag between EU HTAR expectations regarding RWE and current HTA literature for AI-based health technologies used in cardiovascular care (14,17,18).

The comparative analysis of HTA frameworks further illustrates these tensions. The EUnetHTA Core Model remains comprehensive and widely adopted but is primarily oriented toward conventional technologies, offering limited guidance for adaptive software systems. The NICE Evidence Standards Framework introduces a more modular structure and explicitly addresses DHTs, yet provides only partial coverage of post-deployment monitoring and algorithmic retraining. Emerging initiatives such as ASSESS-DHT point toward more flexible and iterative evaluation approaches, but remain under development and have not yet been systematically validated or widely adopted in cardiovascular AI contexts. Similarly, the DECIDE-AI reporting guidelines provide important principles for early-phase evaluation, particularly around usability, safety and human factors, but are not designed to function as HTA frameworks and have yet to be formally integrated into HTA processes (21).

Stakeholder involvement, particularly engagement with clinicians, was inconsistently reported across the included literature. While several studies implicitly relied on clinical expertise, structured engagement of cardiologists or end-users was explicitly documented in only a limited number of cases. Given the interpretive role clinicians play in cardiovascular AI deployment, insufficient stakeholder engagement may undermine both the clinical relevance and legitimacy of assessment outcomes. This gap contrasts with the emphasis placed on early and meaningful stakeholder involvement within the HTAR (17-19). Future HTA guidelines should prioritize formal mechanisms for clinician involvement, including cardiologist co-design and early validation.

The structured risk of bias assessment conducted in this review supports these observations. Most studies demonstrated clear objectives and adequate methodological transparency. However, recurring limitations were identified in external validation, structured use of HTA frameworks and formal stakeholder engagement. These weaknesses reinforce the need for more consistent methodological standards as AI-based cardiovascular technologies move closer to routine clinical and regulatory decision-making.

Given that ASSESS-DHT is still under development, its inclusion in this comparative analysis should be interpreted as an examination of conceptual direction rather than operational performance. Ongoing feedback from HTA bodies and pilot experience will be critical in determining whether proposed methodological advances translate into practical HTA utility.

The limited number of included studies (n=6) also reflects important structural challenges in the current evidence landscape. Many studies evaluating AI in cardiology remain focused on technical performance or diagnostic accuracy, without incorporating HTA-relevant dimensions such as cost-effectiveness, policy implications, lifecycle considerations, or real-world implementation. As a result, relatively few studies meet the criteria required for HTA-oriented evaluation. This highlights a key gap between the development of AI technologies and their readiness for assessment within established HTA frameworks.

Addressing this gap will require closer alignment between AI development, reporting standards and HTA evidence requirements. Future research should prioritize the integration of lifecycle evaluation, prospective RWE generation and structured stakeholder engagement. In addition, clearer methodological guidance is needed to support the reassessment of adaptive algorithms over time and to ensure that AI-based cardiovascular technologies can be evaluated in a manner consistent with the requirements of the EU HTAR.

Taken together, these findings suggest that while existing HTA methodologies provide a useful foundation, they do not yet fully capture the evidentiary, lifecycle and contextual dimensions of AI in cardiovascular care. The HTAR offers a promising regulatory opportunity to address these gaps, but methodological adaptation will be required to ensure alignment between regulatory ambition and practical evaluation.

Limitations

This systematic review has several limitations that should be considered when interpreting its findings. The number of included studies was limited (n=6), reflecting the early and fragmented nature of the evidence base at the intersection of AI and HTA in cardiovascular care. This narrow evidence base may limit the generalizability of the findings and underscores the need for cautious interpretation.

Although the search strategy was comprehensive and transparently reported, many of the included studies were published prior to the full implementation of the EU HTAR. Consequently, the literature may not yet reflect emerging regulatory expectations related to lifecycle adaptability, structured stakeholder engagement or systematic integration of RWE.

Moreover, despite refinement of Boolean search strategies and the use of multiple databases, variability in terminology across AI, HTA and cardiovascular medicine may have led to the omission of relevant studies employing alternative descriptors or indexing practices. This challenge is inherent to interdisciplinary fields and may have affected completeness.

Additionally, the review was restricted to English and Spanish language publications. While common in international HTA research, this may have resulted in underrepresentation of studies in other languages describing European contexts, particularly in regions where HTA processes and digital health adoption are evolving but less frequently reported in English-language journals.

Furthermore, the review prioritized conceptual and methodological synthesis rather than empirical evaluation of HTA implementation or regulatory decision-making. The included studies were heterogeneous in design, encompassing economic evaluations, clinical trials, modelling studies and conceptual frameworks. Given this diversity, a single standardized critical appraisal tool was not appropriate. Instead, a tailored domain-based risk of bias assessment was applied to evaluate transparency, validation, stakeholder engagement and HTA relevance.

Finally, some initiatives discussed are ongoing European research projects, and findings should therefore be interpreted with particular caution regarding implementation readiness. These newer initiatives and frameworks for digital health assessment continue to emerge, and their application to cardiovascular AI remains limited and uneven. As a result, this review may not fully capture ongoing pilot activities or framework co-development efforts that have not yet been published. These limitations highlight the need for continued empirical research, longitudinal evaluation and inclusive stakeholder engagement as HTA methodologies evolve alongside AI in cardiovascular care.

Conclusions

AI is increasingly integrated into cardiovascular care, offering new capabilities for early diagnosis, dynamic risk stratification and continuous patient monitoring. However, this systematic review underscores a persistent misalignment between the characteristics of AI-based technologies and the assumptions embedded in traditional HTA methodologies.

Important principles for HTA outlined in the EU HTAR, including the progressive integration of RWE, structured stakeholder engagement, and the capacity to update JCAs across the technology lifecycle, are only partially reflected in current peer-reviewed literature reporting assessments of AI-based cardiovascular technologies. Most studies continue to rely on trial-based or simulation data, and few explicitly incorporate discussion of what should trigger reassessment or how algorithmic updates should be monitored over time. Stakeholder input, particularly from clinicians such as cardiologists, remains inconsistently reported.

While foundational frameworks such as the EUnetHTA Core Model continue to provide valuable structure, there is limited guidance available for evaluating adaptive, data-driven technologies. Emerging models such as ASSESS-DHT suggest possible future directions, while reporting standards like DECIDE-AI suggest new directions for more iterative, transparent and stakeholder-informed evaluations, but their adoption in cardiovascular domains remains limited.

To align HTA processes with the evolving landscape of AI-enabled cardiovascular technologies, several priorities emerge. First, the routine integration of real-world data will be essential to assess clinical effectiveness, safety and implementation. Second, HTA methodologies must evolve to accommodate the full lifecycle of AI tools, including post-deployment updates, performance monitoring and dynamic clinical integration. Third, stronger and earlier involvement of domain experts, particularly clinicians, will be critical to ensure that assessment criteria reflect frontline realities and support practical implementation.

As the HTAR enters its operational phase, there is an important opportunity to strengthen methodological and regulatory convergence. Advancing HTA tools to meet the demands of AI in cardiovascular care will require targeted investment, multidisciplinary collaboration and sustained commitment to methodological innovation.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the PRISMA reporting checklist. Available at https://atm.amegroups.com/article/view/10.21037/atm-2026-0052/rc

Peer Review File: Available at https://atm.amegroups.com/article/view/10.21037/atm-2026-0052/prf

Funding: This work was supported by the European Commission under the Horizon Europe Program, as part of project ASSESS-DHT (No. 101137347) via funding to B.V., F.B., J.C.R.P., D.K., and L.O.

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://atm.amegroups.com/article/view/10.21037/atm-2026-0052/coif). R.C.A. serves as an unpaid editorial board member of Annals of Translational Medicine from July 2024 to June 2026. B.V., F.B., J.C.R.P., D.K., and L.O. report receiving funding from the European Commission under the Horizon Europe Program, as part of project ASSESS-DHT (No. 101137347). B.V. reports receiving funding by UK Research and Innovation (UKRI) grants 10108522 (University of Oxford). P.M. has received funding from Horizon 2020 program through University of Oxford for research, scientific meetings and travel purposes. S.G. has received funding for research related to the following Horizon 2020 projects: ASSESS-DHT and CYMEDSEC, and from the Federal Ministry of Education and research: BMBF 16KISA100K. S.G. declares the following competing financial interests: he has or has had consulting relationships with Una Health GmbH, Lindus Health Ltd., Flo Ltd., ICURA ApS, Rock Health Inc., Thymia Ltd., FORUM Institut für Management GmbH, High-Tech Gründerfonds Management GmbH, the European Commission Directorate General for Research & Innovation, Prova Health Ltd., Haleon Plc., Ada Health GmbH, and the Saudi Arabia Food and Drug Authority (FDA) funded through the United Nations Development Programme (UNDP); and holds share options in Ada Health GmbH. S.G. is a News and Views editor of NPJ Digital Medicine. R.C.A. has received small amounts for educational meetings from Abbott and Boston Scientific. The authors have no other conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Svennberg E, Caiani EG, Bruining N, et al. The digital journey: 25 years of digital development in electrophysiology from an Europace perspective. Europace 2023;25:euad176. [Crossref] [PubMed]
2. Vervoort D, Tam DY, Wijeysundera HC. Health Technology Assessment for Cardiovascular Digital Health Technologies and Artificial Intelligence: Why Is It Different? Can J Cardiol 2022;38:259-66. [Crossref] [PubMed]
3. Casado-Arroyo R, Osoro L. Health Technology Assessment and Cardiology: A Review of the Present and Future of Innovation. J Mark Access Health Policy 2025;13:30. [Crossref] [PubMed]
4. Boriani G, Svennberg E, Guerra F, et al. Reimbursement practices for use of digital devices in atrial fibrillation and other arrhythmias: a European Heart Rhythm Association survey. Europace 2022;24:1834-43. [Crossref] [PubMed]
5. Frausing MHJP, Nielsen JC, Westergaard CL, et al. Economic analyses in cardiac electrophysiology: from clinical efficacy to cost utility. Europace 2024;26:euae031. [Crossref] [PubMed]
6. Finlayson SG, Subbaswamy A, Singh K, et al. The Clinician and Dataset Shift in Artificial Intelligence. N Engl J Med 2021;385:283-6. [Crossref] [PubMed]
7. Vasey B, Nagendran M, Campbell B, et al. Reporting guideline for the early-stage clinical evaluation of decision support systems driven by artificial intelligence: DECIDE-AI. Nat Med 2022;28:924-33. [Crossref] [PubMed]
8. European Commission. Regulation on Health Technology Assessment. 2017 [cited 2024 Aug 22]. Available online: https://health.ec.europa.eu/health-technology-assessment/overview_en
9. Toumi M, Falissard B, Jouini A, et al. EU-HTA Guidance for Clinical Validity: Misconceptions and Flawed Processes. J Mark Access Health Policy 2025;13:33. [Crossref] [PubMed]
10. European Digital Health Technology Assessment (EDiHTA). Available online: https://edihta-project.eu/
11. ASSESS-DHT. [Cited 2024 Apr 4]. Available online: https://assess-dht.eu/
12. HTA CG. Subgroup for Identification of Emerging Health Technologies. Report on Emerging Health Technologies - 2025. 2025 [cited 2026 Dec 1]. Available online: https://health.ec.europa.eu/document/download/4c5f6b6a-f28c-4226-9ab3-e1bc236273a9_en?filename=hta_hatcg_2026-eht-rep_en.pdf
13. Page MJ, McKenzie JE, Bossuyt PM, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ 2021;372: [Crossref] [PubMed]
14. Zanotto BS, Beck da Silva Etges AP, Dal Bosco A, et al. Stroke Outcome Measurements From Electronic Medical Records: Cross-sectional Study on the Effectiveness of Neural and Nonneural Classifiers. JMIR Med Inform 2021;9:e29120. [Crossref] [PubMed]
15. Hautala AJ, Shavazipour B, Afsar B, et al. Machine learning models for assessing risk factors affecting health care costs: 12-month exercise-based cardiac rehabilitation. Front Public Health 2024;12:1378349. [Crossref] [PubMed]
16. Edgar R, Ebrahimkheil K, Scholte NTB, et al. Design of the DETECT project: automated cardiac arrest detection and activation of the emergency medical chain integrated into a wristband. Resusc Plus 2025;24:100977. [Crossref] [PubMed]
17. Tsiachristas A, Chan K, Wahome E, et al. Cost-effectiveness of a novel AI technology to quantify coronary inflammation and cardiovascular risk in patients undergoing routine coronary computed tomography angiography. Eur Heart J Qual Care Clin Outcomes 2025;11:434-44. [Crossref] [PubMed]
18. Hill NR, Groves L, Dickerson C, et al. Identification of undiagnosed atrial fibrillation using a machine learning risk prediction algorithm and diagnostic testing (PULsE-AI) in primary care: cost-effectiveness of a screening strategy evaluated in a randomized controlled trial in England. J Med Econ 2022;25:974-83. [Crossref] [PubMed]
19. Cai M, Barrios-Espinosa C, Rienstra M, et al. Optimizing atrial fibrillation management using a novel patient-level computational model. Med 2026;7:100896. [Crossref] [PubMed]
20. European Commission. EUnetHTA21 service contract. [Cited 2025 Mar 7]. Available online: https://health.ec.europa.eu/health-technology-assessment/regulation-health-technology-assessment/eunethta21-service-contract_en
21. NICE UK. NICE health technology evaluations: the manual. 2025 [cited 2025 Aug 21]. Available online: https://www.nice.org.uk/process/pmg36/chapter/introduction-to-health-technology-evaluation#:~:text=The%20health%20technology%20evaluation%20methods,screening%20tools