Key takeaways:
- Trustworthy Artificial Intelligence for Personalised Risk Assessment, or AI4HF, is a 4-year project funded by the European Union's Horizon Europe Framework
- AI4HF aims to co-design, develop and evaluate the first trustworthy artificial intelligence (AI) tool for personalising the care and management of patients with heart failure.
- The project brings together patients, clinicians and other stakeholders from Europe, South America and Africa with regular workshops to allow input from all
- The workshops have produced many ideas for refining AI for heart failure treatment, including increasing health and digital literacy, personalised recommendations on improving lifestyle, and social components such as involvement of carers. A new round of workshops is ongoing
The "Trustworthy Artificial Intelligence for Personalised Risk Assessment" (AI4HF) project, funded by the European Union's Horizon Europe Framework, has successfully engaged clinicians, patients and other stakeholders across Europe, South America and Africa to achieve its target to co-design, develop and evaluate the first trustworthy artificial intelligence (AI) tool for personalising the care and management of patients with heart failure.
Heart failure (HF) is a major global public health issue, and more than 64 million patients worldwide1 have HF. The prevalence of HF has increased, mainly due to the aging and longer survival of patients receiving guideline-directed medical therapy (GDMT) for HF. The reported incidence of HF is 1–4 per 1000 person-years. The prevalence of HF is 1–3% in the overall adult population in developed countries; however, HF prevalence increases with aging affecting more than 10% and 30% of individuals older than 70 and 85 years, respectively.
"Heart failure is a complex syndrome with diverse causes, overlapping symptoms, and a progressive course that often makes early diagnosis and optimal treatment difficult.," explains Prof Folkert W. Asselbergs, Amsterdam UMC, coordinator of AI4HF and also Chair of the European Society of Cardiology (ESC) Digital Cardiology and AI committee, "Traditional methods can struggle to process the vast and varied data generated by heart failure patients—such as electronic health records, imaging, laboratory results, and wearable device data. AI excels at integrating and analysing these large, high-dimensional datasets, uncovering subtle patterns and relationships that may not be apparent to clinicians. AI therefore empowers clinicians to deliver more precise, timely, and individualised care for heart failure patients, ultimately improving survival and reducing the burden of this growing global health problem."
Work Package 1 of the AI4HF project (the Multi-stakeholder engagement and social innovation ) is led by SHINE 2Europe, a Portuguese organisation that connects research, education, and technology to create impactful solutions.
Other partners including the European Heart Network (EHN) and the European Society of Cardiology (ESC) have provided support to link to the patients and professionals, respectively.
Workshops and meetings first established the list of stakeholders (not limited to patients and clinicians, but also including, for example, hospital administrators and advocacy organisations). On the 12th month of the project, the consortium organised a one-day international multi-stakeholder workshop titled "Towards trustworthy AI-driven tools for personalised treatment of heart failure: workshop on the present and future of heart failure treatment", with 28 people representing various stakeholders. The key issues raised included increasing health literacy,; personalised recommendations to improve lifestyle,; identifying the specific type of heart failure experienced by the patient at an early stage to improve treatment,; increasing patient motivation to engage in their own care, and fostering shared decision making.
Local Clinical/Patient Working Groups (WGs) were established at the project's five clinical sites: Utrecht/Amsterdam (Netherlands), Barcelona (Spain), Brno (Czech Republic), Dar es Salaam (Tanzania), and Lima (Peru). Each group consists of 6-8 healthcare professionals specialising in heart failure care and 6-8 patients. Their primary role is to assess the cultural and organisational appropriateness of the materials and tools developed within the project and provide centre-, country- and context-specific input and feedback. Additionally, the Clinical/Patient WGs are also being consulted to guide the development of the AI4HF tools following a human-centred design methodology.
Key findings from these working groups, for which two rounds of sessions have been completed, included that the 'typical' heart failure patient does not exist – all have their individual traits. It is important to adjust the care system and the AI tool so that it works for all patients. The WGs agreed the project should be as individualised for the patients as possible, and that patients want more than just a tool for assessing their chance of hospitalisation or death (such calculators already exist).
Furthermore, patients wanted the AI4HF to be able to advise them if they have any problem (for example dyspnoea, oedema) and for it to identify the best medical plan and intervention for them, aiming to maximise the benefit and minimise the cost or side effects. And while patients mostly feel comfortable using cell phone technology, there is still no acceptable level of trust for the use of artificial intelligence in health, both for patients and health personnel. Most encouragingly, the involvement of both patients and healthcare workers in the discussion was very engaging, indicating enthusiasm to for this work to be successful, to improve patient care and for assist clinicians' decision making. With more than 200 stakeholders engaged during the first two years, the project has been hailed for its inclusive approach, especially appreciated by patients and healthcare professionals who feel they have a real voice in shaping innovation for heart failure treatment.
The third round of Local Working Group (WG) sessions is currently underway across the five clinical sites. In this phase, clinicians and patients are providing feedback on the initial version of the risk prediction tool, with a particular focus on its visual design and its potential impact on patient-clinician communication. The sessions also explore context-specific sources of bias that may influence both the development of the AI models and the tool's implementation in real-world clinical settings. Findings from these sessions are expected in Autumn 2025.
