Rethinking Stroke Prevention in Atrial Fibrillation: One Size Does not Fit All

The cornerstone of stroke prevention in atrial fibrillation (AF) lies in thromboembolic risk stratification, typically assessed through simple clinical risk scoring systems such as the CHA2DS2-VASc score [1-4]. The latest 2024 ESC guidelines for the management of AF proposed the sexless CHA2DS2-VA score (Level of Evidence: C) to assist the decisions on anticoagulation therapy “in the absence of other locally validated alternatives,” as “inclusion of gender complicates clinical practice both for healthcare professionals and patients” and “omits individuals who identify as nonbinary, transgender, or are undergoing sex hormone therapy” [2, 5]. Long-term oral anticoagulation (OAC) is recommended for patients with CHA2DS2-VASc scores of 2 (three in females) or higher, while it should be considered in those with CHA2DS2-VASc score of 1 (two in females) [1, 2]. Risk scores like CHA2DS2-VA(Sc) should be regarded as simplifications and are reductionist, to aid decision-making. Indeed, not all elements within the CHA2DS2-VA(Sc) score carry an equivalent risk of stroke. It is essential to clarify that female sex should not be regarded as an independent risk factor but rather as a risk modifier, as its contribution to stroke risk becomes significant only when at least one additional stroke risk factor is present [6, 7]. The risk of stroke varies notably among patients with a single non-sex-related risk factor, with the greatest risk observed in individuals aged 65–74 years with atrial fibrillation and those diagnosed with diabetes mellitus, followed by vascular disease and congestive heart failure [8]. For instance, in a male patient with arterial hypertension, the hazard ratio (HR) of ischemic stroke was 1.681 (95% CI: 1.333–2.120; p < 0.001), while in a male patient aged 65–74 years, the HR was 3.085 (95% CI: 2.790–3.410; p < 0.001). Both patients have a CHA2DS2-VA(Sc) score of 1, but they exhibit a completely different risk of ischemic stroke [8]. Also, it is simple common sense that a man with AF age 65 is going to be at substantially lower risk of stroke compared to a man with AF age 74, yet both have one point on the CHA2DS2-VA(Sc) score. Also, simple clinical risk scores like CHA2DS2-VA(Sc) do not include every conceivable stroke risk factor, the dynamic nature of risk factors [9], nor do they account for ethnic differences in stroke risk [10, 11]. Another important issue is the AF burden [1, 2]. It is difficult not only to define, but also to assess the AF burden in our daily clinical practice as it requires long-term rhythm monitoring. Until now, the AF burden can be precisely assessed only in patients with cardiac implantable electronic devices or implantable loop recorders. External long-term cardiac rhythm monitoring devices present a low compliance rate among patients, while the data from smartwatches and smartphones should be interpreted with caution as their sensitivity and specificity in the detection of AF are still debated. Recent advancements in wearable technology and artificial intelligence (AI) have improved the accuracy of AF detection and burden assessment. For example, the FDA-cleared AF History algorithm on the Apple Watch enables researchers to analyze AF burden for clinical and research purposes. These innovations hold promise for integrating consumer-grade wearables into clinical practice, enhancing the accuracy and utility of rhythm monitoring. The current ESC guidelines (2024) [2] and ACC/AHA/ACCP/HRS (2023) [1] for the management of AF recommend the administration of OAC in all patients diagnosed with AF irrespective of the arrhythmia recurrence profile, burden, or type (paroxysmal, persistent, permanent). However, recent studies have shed light on the relevance of AF burden, defined as the extent and frequency of atrial fibrillation episodes, in refining stroke risk prediction. Additionally, some patients, after the first symptomatic episode of AF, have a low recurrence rate of the arrhythmia or even no recurrence [12]. The absence of a temporal link between the detection of AF and stroke indicates that other factors might play significant roles in causing strokes, and having AF may not be essential [13]. Although device-detected AF frequently lacks a clear temporal relationship with stroke, some studies have identified a significant temporal association, indicating potential causality in certain instances [14, 15]. Importantly, the thromboembolic risk is not solely affected by AF but also by the patient's clinical characteristics and the atrial myopathy. In other words, AF per se might not cause the stroke. In 1997, Zipes used the term “atrial myopathy” for the first time to explain how AF causes a tachycardia-induced atrial cardiomyopathy (myopathy) that results in electrophysiological and anatomical remodeling of the atria [16]. The development of atrial myopathy is driven by pathophysiological mechanisms involving atrial tissue damage caused by various factors such as aging, hypertension, obesity, diabetes mellitus, and ischemic heart disease. Additionally, numerous genetic mutations have been identified that contribute to atrial myopathy [17]. These factors and genetic predispositions increase oxidative stress, activate inflammatory pathways, and stimulate the renin-angiotensin-aldosterone system, leading to greater atrial wall stretch [18]. These changes ultimately initiate atrial tachyarrhythmias, such as AF [19]. On the other hand, some patients, particularly those with minimal inflammatory comorbidities, may experience AF without the functional impact of an underlying myopathy [19]. While atrial myopathy is a frequent underlying condition associated with AF, it is crucial to acknowledge that AF can also occur independently in specific circumstances, such as paroxysmal “lone AF” triggered by pulmonary vein activity. Instances of successful treatment with pulmonary vein isolation (PVI) highlight that not all AF originates from atrial myopathy. Atrial myopathy can form the substrate for AF progression and leads to endothelial dysfunction and stasis, thereby creating a prothrombotic state [20]. The question of whether AF causes atrial myopathy or if atrial myopathy leads to AF remains unanswered. However, large cohort studies utilizing transthoracic echocardiography (TTE) and cardiac magnetic resonance (CMR) have demonstrated that increased left atrial (LA) volumes and reduced LA function are linked to the development of new-onset AF. Until now, anatomical characteristics of the left atrium have not been taken into account in the assessment of patients with newly diagnosed AF, and the decision for prescribing OAC is based only on clinical risk factors such as the presence of congestive heart failure, hypertension, age > 65 years, diabetes mellitus, stroke, and vascular disease. However, the underlying pathophysiology of thrombosis in AF is complex and multifactorial [21]. At present, no specific AF-treatment decision is based solely on cardiac imaging. Recent data from the ARCADIA trial [22] for individuals diagnosed with embolic stroke of undetermined source (ESUS) and atrial cardiopathy indicate that apixaban did not significantly reduce the likelihood of another stroke recurrence compared to aspirin in patients with cryptogenic stroke and evidence of atrial cardiopathy but without AF. Nevertheless, in the ARCADIA study, atrial cardiopathy was characterized by a P-wave terminal force surpassing 5000 μV × ms in electrocardiogram lead V1, a serum N-terminal pro-B-type natriuretic peptide level exceeding 250 pg/mL, or a left atrial diameter index of 3 cm/m2 or higher on echocardiogram. The majority of patients met the enrollment criteria based on the N-terminal pro-B-type natriuretic peptide level, though it's worth noting that this biomarker might have limitations due to its lack of specificity and its potential association with comorbidities, such as a patient with underlying illness or cardiac issues. In our daily clinical practice as far the management of AF is concerned, we should distinguish the terms “clinical” from “subclinical AF.” “Clinical AF” refers to symptomatic or asymptomatic AF documented by 12-lead ECG or other ECG devices, usually arbitrarily defined as at least 30 s of an ECG tracing [2]. On the other hand, subclinical AF refers to individuals without symptoms attributable to AF, in whom clinical AF has not previously been detected by entire 12-lead ECG or 30 s of an ECG tracing. The term “subclinical AF” mainly refers to device-detected AF episodes—atrial high-rate episodes (AHRE) detected by cardiac implantable electronic devices and insertable cardiac monitors [1, 2]. It is crucial to distinguish between the two entities (clinical and subclinical AF) as this significantly affects our decision-making for prescribing OAC. Specifically, only patients with “clinical AF” have a clear indication for long-term OAC [1, 2]. In the case of subclinical AF, a further analysis of the ASSERT study has demonstrated that only episodes of duration > 24 h increased significantly the risk of stroke [23]. However, other studies have reported shorter AF duration thresholds that are associated with an increased risk of stroke. In the MOST study (MOde Selection Trial) [24], the risk of stroke or death was found to be 2.5 times higher in individuals who experienced at least one atrial high-rate episode lasting longer than 5 min. Similarly, the TRENDS study (A Prospective Study of the Clinical Significance of Atrial Arrhythmias Detected by Implanted Device Diagnostics) [25] demonstrated that thromboembolic risk correlates with the overall burden of atrial fibrillation. The current ESC guidelines recommend personalization of the OAC according to subclinical AF burden and the patient's thromboembolic risk. Two recently published trials (NOAH-AFNET 6 and ARTESIA) attempted to answer the question of whether the administration of OAC in patients with subclinical AF is beneficial or not. The NOAH-AFNET 6 [26] study tested the effectiveness and safety of oral anticoagulants in patients with subclinical AF/atrial high-rate episodes (AHRE) lasting 6 min or longer and having at least one additional stroke risk factor. The study showed that OAC with edoxaban did not reduce the occurrence of stroke, systemic embolism, or cardiovascular death compared to no AT, while notably increased the risk of major bleeding. These findings remained consistent regardless of AHRE duration and CHA2DS2-VASc score. This study indicated that even AHRE episodes lasting longer than 48 h should not automatically prompt physicians to start OAC therapy. Conversely, the ARTESIA [27] study demonstrated that among patients with subclinical atrial fibrillation/AHRE, apixaban reduced stroke or systemic embolism compared to aspirin, although it carried an increased risk of major bleeding but not fatal bleeding. In the referenced studies, patients with device-detected AF often exhibited a stroke risk that was lower than what would be predicted by the CHA2DS2-VASc score. In the modified intention-to-treat population of the NOAH-AFNET 6 trial, the annual incidence of ischemic stroke was 1.1% in patients receiving aspirin or placebo and 0.9% in those treated with edoxaban. Similarly, in the intention-to-treat population of the ARTESiA trial, the annual rates of ischemic stroke were 1.0% for aspirin and 0.6% for apixaban [28]. A meta-analysis combining data from NOAH-AFNET 6 and ARTESiA demonstrated a significant reduction in ischemic stroke with oral anticoagulation (RR: 0.68; 95% CI: 0.50–0.92; I² = 0%). The estimated absolute risk reduction for ischemic stroke was three fewer events per 1000 patient-years (95% CI: 5 fewer to one fewer) based on trial data, and 6 fewer events per 1000 patient-years (95% CI: 10 fewer to 2 fewer) based on baseline estimates from observational study meta-analyses [28]. Consequently, these observations raised doubts about the previously established approach of initiating long-term OAC therapy solely based on the CHA2DS2-VASc score and AF burden in individuals with AHRE. Notably both studies did not answer the question of whether OAC is beneficial in subclinical AF and if long-term OAC is definitely indicated in this population. Considering the above observations, it is clear that one size does not fit all in the case of long-term anticoagulation in AF patients, especially in subclinical AF/AHRE. A personalized approach to long-term OAC in AF seems attractive. A question to be answered is “what has changed during the last years to reconsider our therapeutic approach?.” The answer is that now we have anticoagulation therapies with a rapid onset and offset of their action and plenty of options for long-term monitoring in our patients. In light of the evolution of artificial intelligence, advancements in remote rhythm monitoring technologies, including implantable loop recorders, smartwatches, and smartphones, have enabled continuous surveillance of AF burden, facilitating timely interventions. Consequently, the combination of new oral anticoagulants and remote rhythm monitoring devices in the era of artificial intelligence makes tailored AT in AF promising [29]. A further question is whether tailored OAC management in AF is feasible, effective, and safe. The IMPACT [30] study was the first study of tailored anticoagulation management in AF, which failed to demonstrate the superiority of tailored OAC over conventional OAC. The results of the IMPACT study can be partly attributed to the use of vitamin K antagonists in a significant portion of the patients. The IMPACT study also had a significant delay between AF onset and achieving therapeutic anticoagulation, which likely contributed to its outcomes. Subsequent trials, such as REACT.COM [31] and TACTIC-AF [32], embraced a personalized approach by adjusting anticoagulation based on individual AF recurrence profiles. These studies revealed promising reductions in anticoagulation duration without compromising stroke prevention efficacy, particularly when employing NOACs. However, it is worth noting that REACT.COM and TACTIC AF were single-arm studies not powered to assess stroke prevention. Their findings highlight potential benefits of tailored anticoagulation but require validation in larger, randomized trials. The REACT-AF [33] trial (NCT 05836987) will compare standard long-term OAC with a smartwatch-guided tailored OAC approach for AF in 5350 patients. Participants in the treatment group will wear smartwatches that detect AF episodes and receive notifications to take anticoagulant medication when necessary. The trial aims to demonstrate the safety and efficacy of this approach, with primary endpoints including stroke prevention and mortality, and secondary endpoints focusing on major bleeding. If successful, this method could offer new stroke prevention strategies for AF patients. One of the most challenging populations for AT in AF are patients with a single non-sex-related stroke risk factor (CHA2DS2-VA(Sc) = 1 or 2 in females), in whom long-term OAC is not strictly indicated. Guidelines recommend that OAC should be considered in these intermediate thromboembolic risk patients (ESC, IIa, C; ACC/AHA/ACCP/HRS, 2a, A) [1, 2]. However, taking into account the patient's preferences, AF burden, AF recurrence profile, and patient's bleeding risk, a personalized approach may be considered in selected patients [34]. More specifically, AF patients with intermediate thromboembolic risk and high bleeding risk [35] (e.g., HAS-BLED Score ≥ 3) or low bleeding risk and low burden of the arrhythmia (episodes with duration < 24 h) may be potential candidates for tailored anticoagulation therapy (pill-in-pocket) [34]. Despite these advancements, challenges persist in implementing “pill-in-pocket” personalized anticoagulation strategies (Figure 1). The precise threshold for defining high AF burden remains unknown, necessitating further research to define its impact on thromboembolic risk. Moreover, the complex interplay between AF, atrial myopathy, and thrombogenesis underscores the need for holistic patient assessment beyond conventional risk scores. Looking ahead, large-scale randomized trials are warranted to validate personalized AT strategies and assess the utility of interventions such as left atrial appendage occlusion and pulmonary vein isolation. Furthermore, the challenges of data management, patient compliance, and cost-effectiveness necessitate careful consideration in the integration of advanced technologies into routine care. The growth in data science and machine learning approaches raises the possibility of “real time“ dynamic stroke risk stratification and management strategies [36, 37]. As we navigate this complex terrain, collaboration between clinicians, researchers, and technology developers will be pivotal in realizing the full potential of personalized medicine in AF management. P.E.P. reports consultancy for Boehringer Ingelheim and was an investigator in the Bayer-sponsored OCEANIC-AF study. G.Y.H.L. is a consultant and speaker for BMS/Pfizer, Boehringer Ingelheim, Daiichi-Sankyo, Anthos. No fees are received personally. He is a National Institute for Health and Care Research (NIHR) Senior Investigator and co-PI of the AFFIRMO project on multimorbidity in AF (grant agreement No. 899871), TARGET project on digital twins for personalized management of atrial fibrillation and stroke (grant agreement No. 101136244) and ARISTOTELES project on artificial intelligence for management of chronic long term conditions (grant agreement No. 101080189), which are all funded by the EU's Horizon Europe Research & Innovation program. The data that support the findings of this study are available from the corresponding author upon reasonable request.