The STRATIFYHF project aims to develop an AI-based decision support system for the risk stratification, early diagnosis, and management of heart failure, utilising comprehensive data from European clinical centres to enhance patient care and outcomes.
Heart failure (HF) is a global pandemic currently affecting up to 15 million people in Europe. It is a complex clinical syndrome associated with impaired heart function, poor quality of life for patients, and high healthcare costs. The STRATIFYHF project addresses this global challenge with an AI-based Decision Support System for HF.
The STRATIFYHF aims to develop, validate and implement the first artificial intelligence (AI)-based, Decision Support System (DSS) for assessing and predicting the risk of HF, its early diagnosis, and progression. STRATIFYHF project integrates 1) patient-specific data, i.e. demographic, clinical, genetic, lifestyle and socio-economic, 2) an AI-based digital patient library and algorithms for risk stratification, early diagnosis, and disease progression, and 3) a highly innovative multifunctional AI-based and computational modelling DSS and mobile app for informing a patient-centred, personalised, prevention and treatment strategies (Fig. 1).
Clinical study
Eight European Clinical Centres of Excellence in the diagnosis and management of heart failure, which are part of the STRATIFYHF project, have delivered retrospective and prospective clinical studies since the beginning of the project. In the retrospective phase, Centres managed to collect demographic and clinical data for 5,624 patients with a confirmed diagnosis of heart failure, and 4,465 patients with suspected heart failure. In addition, access to Clinical Practice Research Data link allowed access to 11 million patient records who are at risk of developing heart failure, and 680,000 patients with confirmed heart failure. These data have been used to develop risk stratification, diagnostic, and prognostic models. We have also analysed and disseminated findings on multimorbidity in heart failure using retrospective data (Fig. 2).
The prospective phase of the clinical study recruited 350 patients. All patients underwent a standard heart failure diagnostic clinical investigation, including medical history, physical examination, blood test for NTproBNP, electrocardiography, and echocardiography. In addition, investigations included the two novel technologies, such as cardiac output response to stress test, and voice recognition and voice biomarkers assessment. The prospective phase of the clinical study is used to validate and further refine the accuracy of the risk stratification, diagnostic and prognostic models developed using data from the retrospective phase of the clinical study.
Digital patient library and AI-driven analytics
Substantial progress has been made toward establishing a comprehensive data-driven framework for early diagnosis, risk assessment, and disease progression of heart failure. By combining large-scale clinical data integration, advanced data augmentation, and validated machine learning models, this work delivers a coherent pipeline that transforms real-world healthcare data into clinically actionable decision support.
This work is grounded in a centralised, harmonised patient data repository integrating heterogeneous retrospective data from five European medical centres, later complemented by primary care data from the Clinical Practice Research Datalink (CPRD). The repository combines multimodal clinical data from more than 6,000 patients from medical centres and more than 500,000 patients from CPRD. The data is interoperable and machine-learning ready. Building on this foundation, advanced methods were developed to address key limitations of real-world clinical data, particularly class imbalance, through a complete pipeline for data quality assurance, class balancing, and synthetic data generation. State-of-the-art generative and probabilistic techniques were benchmarked and validated using statistical similarity and model-based criteria, demonstrating the creation of clinically plausible virtual patient populations suitable for robust and fair predictive modelling.
The curated and augmented datasets were used to develop and validate machine learning models addressing three core clinical objectives: early diagnosis of heart failure, risk stratification of individuals without a confirmed diagnosis, and prognostic assessment of disease progression in patients with established heart failure. Early diagnosis was addressed using both a single-stage model tailored to primary care data and a hybrid, cost-aware cascaded approach that balances diagnostic accuracy with efficient use of clinical resources, while risk stratification was formulated as a time-to-event prediction problem producing individualised risk estimates and clinically interpretable survival curves. Disease progression models focused on prognostic assessment, including mortality risk estimation. Across all modelling tasks, rigorous validation strategies were applied. Deployment protocols were defined to enable seamless integration of the models into a clinical decision support system (Fig. 2) and support their evaluation in real-world settings.
Decision support system
STRATIFYHF cloud-based Decision Support System (DSS) is a platform designed to assist healthcare professionals in the risk stratification, early diagnosis, and management of heart failure. A milestone has been reached with the finalisation of the formal Software Architectural Design, which provides the structured roadmap necessary for the system’s technical realisation. The STRATIFYHF Conceptual Multi-layer Hierarchical Framework is shown in the figure below. As illustrated in the Conceptual Multi-layer Hierarchical Framework below, we have defined a five-layer architecture—spanning hardware, security, workflow, back-end, and front-end layers. The functionality of the STRATIFYHF DSS is categorised into specialised modules: The Data Management Module (DMM) serves as the gateway, utilising a Data Quality Control Engine to clean, normalise, and assess the reliability of incoming patient records. Central to the system’s operation is the Workflow Manager Module (WMM), which acts as an orchestrator, employing a Docker Engine to deploy and manage high-computation tasks like the 3D Computer Modelling (PAK) tool for biomechanical heart simulations. Analytical power is provided by the Risk Stratification and Early Diagnosis modules, which apply machine learning algorithms to identify high-risk patients and detect heart failure markers that might be missed during routine exams. Finally, the Visual Analytics Module (VAM) and the User Access Management Module (UAMM) ensure that these complex outputs are translated into intuitive, secure, and interactive dashboards, allowing clinicians to visualise disease progression and voice biomarker trends directly within their workflow (Fig. 4).
Development of the STRATIFYHF mobile application
This part of the project focuses on the design and specification of the STRATIFYHF mobile application, entitled mAIHeart, laying the foundations for a user-centred, secure, and scalable digital solution for heart failure management and end-user adherence to prevention and treatment strategies. The work conducted in WP4 spans from the systematic elicitation of end-user requirements to the definition of the core architectural components that enable the application’s implementation and integration with the wider STRATIFYHF ecosystem.
The first phase of WP4 concentrated on understanding the needs of the two primary end-user groups: heart failure patients or individuals at high risk of developing heart failure, and healthcare professionals operating in primary and secondary care settings. This process combined (i) a literature review on heart failure risk factors and (ii) management strategies with a structured analysis of the existing mobile health software landscape. In parallel, dedicated questionnaires and interviews were conducted with patients, carers, and healthcare professionals, ensuring early and continuous user involvement (Co-Design/Co-Create). The collected insights were translated into functional and non-functional requirements, addressing usability, security, performance, and regulatory considerations, and were further articulated through detailed user stories capturing real-world clinical and self-management scenarios.
Building on the validated requirements, the second phase of WP4 focused on the definition of the mAIHeart reference architecture and its core architectural components. A modular client–server architecture was designed to support scalability, interoperability, and secure data handling. The architecture clearly separates patient-facing and healthcare professional-facing functionalities, while introducing dedicated layers for data management, business logic, communication, and security. Emphasis was placed on the integration of wearable devices, external storage systems, and Decision Support Systems, enabling continuous data collection, AI-driven analytics, and real-time risk assessment. Detailed data flow diagrams and interface specifications were produced to describe key operational scenarios, including sensor data ingestion, manual data entry, reporting, goal management, alerts, and asynchronous communication. Robust security and compliance mechanisms, such as role-based access control, OAuth-based authentication, AES encryption, and GDPR-aligned data governance, were embedded in the architecture by design.
Clinical study delivery and regulatory development
This part of the project has successfully established and sustained the regulatory, clinical, and quality framework underpinning the development, evaluation, and implementation of the STRATIFYHF Decision Support System (DSS) and mobile application. Through a coordinated and standards-driven approach, the work package has ensured that all technical and clinical activities remain aligned with European regulatory requirements, international standards, and best clinical practice, while laying the foundation for future validation, regulatory approval, and exploitation.
A clear regulatory strategy was defined and implemented for the DSS and mobile application in accordance with EU MDR 2017/745 and applicable guidance for software as a medical device. This included the establishment of an appropriate regulatory classification, development roadmap, and regulatory positioning that support long-term conformity assessment and market readiness. Regulatory alignment has been maintained throughout the project as the technical scope and clinical use cases evolved.
Beyond task-specific activities, WP5 acted as the central regulatory and clinical coordination point across the consortium, ensuring coherence between technical development, clinical evaluation, and regulatory expectations. Through sustained cross-work-package interaction, WP5 enabled consistent interpretation of regulatory and clinical requirements, supported project-level reviews and strategic discussions, and reinforced audit-ready, regulator-facing project governance.
Quality by Design has been applied across the full lifecycle in line with EU MDR, international standards, and best clinical practice.
Overall, WP5 has established a solid foundation for large-scale clinical validation, final clinical evaluation, and future regulatory submissions. The work package has positioned STRATIFYHF for conformity assessment, deployment, and sustainable exploitation beyond the project lifetime.
Finally, WP5 has embedded Quality by Design principles across regulatory strategy, risk-based software development, and clinical evidence planning, ensuring that safety, performance, and regulatory compliance of the STRATIFYHF DSS were built in from early design through lifecycle readiness (Fig. 6).
Health economics and cost-effectiveness analysis
Work Package 6 (WP6) addresses the health economic evaluation and value assessment of the STRATIFYHF clinical decision support system (CDSS) and mobile application, ensuring that the project’s technological innovations are supported by robust evidence on their economic impact, affordability, and potential for sustainable adoption within European healthcare systems. WP6 activities are structured to progressively inform decision-makers, payers, and health technology assessment (HTA) bodies on the expected value of STRATIFYHF solutions in both primary and secondary care settings.
The first major achievement of WP6 was the systematic characterisation of the current standard of care for heart failure (HF) and its associated burden across Europe. This work included a comprehensive systematic literature review on the epidemiological, clinical, and economic burden of HF. The analysis provided a harmonised overview of HF prevalence, hospitalisation rates, mortality, and direct healthcare costs, establishing a robust baseline against which the potential impact of STRATIFYHF can be assessed.
Conclusions
The STRATIFYHF consortium is a well-balanced combination of partners with complementary knowledge and skills. The consortium consists of 15 partners, comprising universities and research institutions, including eight clinical centres, as well as three SMEs and two not-for-profit organisations, from 11 countries across Europe.
STRATIFYHF will change the ways in which HF is managed today, thereby improving the quality and length of patients’ lives. A solution in STRATIFYHF will lead to an efficient and sustainable healthcare system by reducing the number of HF-related hospital admissions and deaths, and unnecessary referrals from primary to secondary care.
Project Consortium Co-Authors:
Djordje Jakovljevic, UK
Zoran Bosnic, SI
Bogdan Milicevic, RS
Dimitrious Boucharas, GR
Petros Malitas, CH
Meritxell Ascanio, ES
Marija Gacic, RS
Please Note: This is a Commercial Profile
Please note, this article will also appear in the 25th edition of our quarterly publication.
Source link
