throMboembolic Risk Associated To High atrIal Fibrillation riSk

Overview

Cardiovascular diseases are the leading cause of mortality from treatable conditions in the European Union and the second from preventable causes, with a standardized mortality rate of 257.8 deaths per 100,000 inhabitants. In 2022, more than 1.11 million deaths in individuals under 75 years could have been avoided. Atrial fibrillation (AF) and major adverse cardiovascular events (MACE) are highly prevalent in the elderly and generate substantial healthcare costs. AF significantly increases the risk of MACE and is projected to rise markedly in the coming decades. In Europe, AF prevalence is expected to increase 2.5-fold over the next 50 years, with a lifetime risk of 1 in 3-5 individuals after age 55. AF-related strokes are projected to increase by 34%, and ischemic strokes in individuals over 80 are expected to triple between 2016 and 2060. Additionally, a 27% increase is anticipated among stroke survivors who subsequently develop AF or related conditions. AF substantially impacts morbidity, mortality, and disease progression, and early detection and treatment are crucial to prevent severe outcomes. European action plans (2018-2030) and the 2024 ESC/ESO guidelines emphasize early detection and management of AF in primary care. Although several AF prediction models exist, their integration into clinical practice remains challenging. AF represents a clinical continuum, with thrombotic risk present even before arrhythmia onset. High-risk patients for AF also show a high incidence of MACE, defined as a composite of myocardial infarction, stroke, systemic embolic events, and cardiovascular death. The proposed strategy involves developing and clinically validating an Artificial Intelligence (AI) model to improve early thrombotic risk prediction in patients at high risk of AF, using MACE as the primary outcome. This model aims to outperform the traditional CHA₂DS₂-VASc score by incorporating both classical and emerging clinical factors. The estimated timeline from clinical validation to commercialization is approximately 48 months. AI-based prediction is expected to enable personalized treatment, reduce the incidence of MACE, hospitalizations, and disability, and improve cost-effectiveness, ultimately decreasing the social and economic burden of AF and stroke in Europe.

Atrial fibrillation and its thromboembolic complications represent a growing clinical and socioeconomic challenge in Europe. AF is strongly associated with stroke, major adverse cardiovascular events (MACE), disability, and mortality, disproportionately affecting older adults. As the European population ages, the prevalence of AF and AF-related stroke is projected to increase substantially, leading to escalating healthcare expenditures and societal burden. Stroke care alone costs an average of €22,605.66 in the first year, largely driven by hospitalization and long-term dependency, with 45-50% of survivors experiencing residual disability. Preventing AF-related thromboembolic events therefore represents both a clinical and economic priority. From a clinical standpoint, there is a critical unmet need for improved upstream thromboembolic risk stratification in individuals at high risk of AF. Although several AF prediction models can estimate the likelihood of incident AF over 5-10 years, and systematic screening of adults aged ≥65 years has demonstrated cost savings through stroke prevention, a validated tool to guide anticoagulation initiation in high-risk individuals without established AF is lacking. Current standard practice relies on the CHA₂DS₂-VASc score once AF is diagnosed; however, this score has recognized limitations. It does not incorporate several relevant risk modifiers such as chronic kidney disease, cancer, biomarkers, electrocardiographic abnormalities, or ethnicity, and it may inadequately discriminate risk in certain subgroups, including women and patients with multimorbidity. Consequently, clinical decision-making often extends beyond the score, reflecting the need for more comprehensive and precise tools. Emerging evidence supports the concept of AF as a clinical continuum. A prothrombotic atrial substrate may precede overt arrhythmia, creating a "pre-AF" stage during which thromboembolic risk is already elevated. The 2023 ACC/AHA/ACCP/HRS guidelines formally recognize "at-risk" and "pre-AF" stages, highlighting an opportunity for earlier preventive intervention. However, practical tools to identify and stratify this population in routine primary care remain limited. Artificial intelligence (AI) offers a promising strategy to address these gaps. By leveraging high-dimensional electronic health record (EHR) data, AI models can capture complex, non-linear interactions among classical and emerging risk factors, potentially providing more accurate individualized thromboembolic risk prediction than traditional scores. Early machine learning (ML) approaches have demonstrated improved discrimination for AF and cardiovascular events compared with conventional models. The MATHIAS project (throMboembolic risk Associated To High atrIal fibrillation riSk) aims to develop and prospectively validate an AI-based model to estimate thromboembolic risk in adults aged ≥65 years at high risk of AF using real-world primary care EHR data. This model will be integrated into a digitally enabled care pathway incorporating targeted, risk-guided photoplethysmography screening and individualized anticoagulation decisions. The objective is to enhance early detection, refine anticoagulation strategies, and personalize rhythm control and comorbidity management. Preliminary retrospective analyses based on validated AF risk stratification cohorts (AFRICAT NCT03188484 and PREFATE NCT05772806) and five pilot ML models demonstrated promising results. The Adaboost model significantly outperformed CHA₂DS₂-VASc in predicting MACE (AUC 99.99% vs. 81.71%; p = 0.0034). While these findings are encouraging, prospective, multicenter evaluation is required to assess generalizability, optimal follow-up intervals, patient selection (including high-risk, TIA/stroke, and varying CHA₂DS₂-VASc strata), and impact on patient-important outcomes such as stroke, bleeding, quality of life, and cost-effectiveness. The project also incorporates a Markov decision-analytic model to estimate MACE, stroke, disability, quality-adjusted life years (QALYs), and costs from both healthcare payer and societal perspectives. Scenario analyses will evaluate whether integrating the AI model into routine care is cost-effective compared with usual care, opportunistic screening, or wearable-first strategies. Special attention will be given to sex-specific outcomes and potential inequities in benefit distribution. The anticipated clinical impact includes reduction of MACE incidence to below 50 per 1,000 person-years in high-risk populations, lowering AF-related stroke proportion to under 10%, improving anticoagulation appropriateness in up to 65% of eligible patients, and ensuring that at least 90% of high-risk individuals receive appropriate oral anticoagulant therapy. Adherence to structured AF care pathways (e.g., AF-CARE) has already been associated with significant reductions in all-cause mortality and MACE; integrating AI-driven risk stratification may further enhance these benefits. Economically, although screening and broader anticoagulant use may initially increase direct costs, stroke prevention is expected to generate substantial long-term savings. In Catalonia alone, estimated annual savings for high-risk AF populations range from €12.3 million to €79.2 million. Reductions in stroke severity, disability, and hospitalizations will mitigate both direct medical costs and indirect societal costs related to loss of productivity and long-term dependency. In summary, this project addresses a major gap in cardiovascular prevention by developing and implementing an AI-driven thromboembolic risk stratification model for individuals at high risk of AF. By aligning with contemporary European guidelines and precision medicine strategies, the initiative seeks to improve early identification, personalize therapeutic decisions, reduce MACE and stroke burden, preserve patient autonomy and quality of life, and ensure sustainable healthcare resource utilization.