Atrial fibrillation (AF) and heart failure (HF) are two closely interconnected cardiovascular conditions that frequently coexist due to shared risk factors and pathophysiological mechanisms.1,2 The number of affected patients has been steadily increasing in recent years.1,2 Their coexistence has an adverse effect on overall health and disease-specific outcomes.3,4 Importantly, their coexistence significantly increases the risk of thromboembolic events, particularly ischemic stroke,5 and also elevates the risk of major bleeding in anticoagulated patients.6 Proper risk stratification in this complex and high-risk patient subgroup remains crucial for optimal disease management. Both the CHA2DS2-VASc score, widely used to assess the risk of stroke, and the HAS-BLED score, used to estimate bleeding risk, have played a key role in clinical decision-making for patients with AF for over a decade.7 However, both tools were developed based on large cohort studies when patients were primarily treated with vitamin K antagonists (VKAs) before significant demographic and therapeutic changes in AF management occurred.8, 9, 10 Importantly, those historical risk scores do not account for recent advances in pharmacotherapy, such as the widespread adoption of both direct oral anticoagulants (DOACs) and sodium-glucose cotransporter-2 (SGLT-2) inhibitors, as well as other novel therapeutic options that have significantly improved the prognosis for patients with AF and HF.7 Furthermore, currently, patients are generally older, have more comorbidities and are substantially more likely to experience polypharmacy compared with the baseline populations of the above cohorts.7 This highlights the limitations of existing risk stratification models and presents a need to reassess and refine our understanding of stroke/systemic thromboembolism and bleeding risk factors in the modern-day population of patients with AF and HF. This review aims to summarize evidence on novel and potentially modifiable risk factors for stroke and major bleeding in this specific patient subgroup.

Atrial fibrillation (AF) and HF frequently coexist due to overlapping risk factors and shared pathophysiological pathways. Structural and electrical atrial remodeling, characterized by fibrosis, left atrium (LA) dilation and ionic current changes, typical of HF,22 promotes AF development and impairs ventricular filling, thereby worsening HF symptoms.23 Additionally, AF may promote the onset and progression of HF through several hemodynamic mechanisms, such as tachycardia-mediated cardiomyopathy, heart rate irregularity, the absence of effective atrial contraction, and the development of functional mitral regurgitation.23,24 Another key pathophysiological factor is neurohormonal activation, particularly the renin-angiotensin-aldosterone system and an increase in sympathetic tone, which exacerbates myocardial fibrosis as well as atrial and ventricular dysfunction.24 Additionally, systemic inflammation and oxidative stress contribute to myocardial remodeling, thereby promoting arrhythmogenesis and ventricular impairment.25 As previously noted, HF increases the risk of stroke regardless of the rhythm status.26,27 According to the 2023 ESC consensus on the interaction between stroke and HF, key contributors include blood stasis from reduced cardiac output, endothelial dysfunction, and a prothrombotic state characterized by elevated levels of coagulation factors such as fibrinogen and D-dimer. Additionally, neurohormonal activation and systemic inflammation further exacerbate these processes.28 This may be further aggravated by the high prevalence of comorbid conditions that increase the risk of stroke, such as chronic kidney disease (CKD), which affects nearly 40-50% of patients with HF and is a strongly prothrombotic condition .29

Regarding the association between HF and increased bleeding risk, the underlying pathophysiology may be explained by severe hemodynamic changes in this specific subgroup, thus leading to fluid retention.30 Anemia is a common finding in patients with HF, affecting up to 50 % of the study population.31 Furthermore, it is a well-known risk factor for major bleeding and worse outcomes.31 The pathophysiological pathways leading to high prevalence of anemia may involve iron deficiency, inadequate production of erythropoietin, hemodilution, neurohormonal activation, chronic inflammation, and renal dysfunction related to cardiorenal syndrome.31,32 Anemia may be exacerbated by gastrointestinal (GI) abnormalities such as vitamin/iron malabsorption due to gut wall edema or GI bleeding.31,32 Endothelial dysfunction, polypharmacy, frailty, renal and/or hepatic impairment, commonly observed in HF patients, further increase bleeding risk.33

The coexistence of AF and HF, regardless of ejection fraction (EF), has an adverse effect on overall health, as it significantly increases the risk of HF hospitalization,21 thromboembolism/stroke,5,21,34,35 major bleeding events,6,36 major adverse cardiac events (MACE)37,38 and all-cause mortality.20,21,37, 38, 39 The incidence of AF in patients with HF increases the risk of stroke nearly 3-fold compared to patients with HF and sinus rhythm.5 The risk of stroke/thromboembolism varies, depending on HF subtype, although the evidence has been inconclusive. Findings from the HF substudy (PREFER in AF) showed that annual stroke incidence decreases linearly with improving LVEF, with the highest rates observed in patients with HFrEF and the lowest in those with HFpEF.40 A large cohort study of 8,358 individuals from Taiwan reported a higher risk of stroke in HFpEF (HR 1.40, 95% Cl 1.16-1.69) and heart failure with mildly reduced ejection fraction (HFmrEF) subgroups (HR 1.184, 95% Cl 1.075-1.303) compared to patients with HFrEF.41 In their meta-analysis, Zhang et al. showed that HFrEF was associated with a decreased risk of stroke and systemic thromboembolism when compared to HFpEF.42 However, most studies included in the meta-analysis did not report any statistical difference between HFpEF and HFrEF regarding stroke/thromboembolism risk.42 Conversely, in patients with HF, the presence of AF has the most severe adverse prognostic impact on those with HFrEF, who exhibit a significantly increased risk of HF hospitalization and HF-related mortality compared with patients with HFpEF.6

A report from the ESC-EHRA AF long-term general registry showed that HFrEF was associated with the highest risk of MACE and all-cause mortality compared to HFmrEF and HFpEF.38 Interestingly, even though stroke/thromboembolism is usually considered the most severe complication of AF, HF is the most common cause of death in patients with AF.43 Interestingly, the correlation between HF and major bleeding is complex, since the findings are inconclusive. Most evidence suggests that HF significantly increases the risk of major bleeding in patients with AF.10,44,45 This correlation seems to be true regardless of HF subtype, as patients with HFrEF46 and HFpEF47 appear to experience higher rates of GI bleeding. Furthermore, although it has been reported that the HF subtype does not influence the risk of major bleeding,6 it may nonetheless affect clinical decisions regarding DOAC administration. Notably, one meta-analysis showed that patients with HFrEF may not benefit from the typical reduction in major bleeding risk with DOACs when compared to warfarin.48 Furthermore, their study showed no significant difference between direct oral anticoagulants (DOACs) and warfarin in reducing the risk of stroke or systemic embolism in patients with HFmrEF and HFpEF.48 The risk of major hemorrhage in patients on anticoagulation increases much more drastically with the severity of HF than the risk of stroke.49 Conversely, a subgroup analysis of the ENGAGE-AF TIMI 48 trial demonstrated no statistically significant difference in the risk of major bleeding between patients with AF and concomitant HF and those without a history of HF.6 Similarly, in their study, Mentias et al. investigated the bleeding risk in patients with AF, depending on their HF subtype. They found no significant differences in admissions for GI bleeding or intracranial hemorrhage not only between HF subtypes but also compared to patients with no history of HF.50 Interestingly, major bleeding risk increases nearly 3-fold in patients with HF and comorbid CKD requiring hemodialysis, and nearly 1.5 times in patients with HF, CKD, and no renal replacement therapy, further increasing the clinical significance of this highly prevalent comorbidity.51

Considering the above evidence, it is clear that effective risk stratification is necessary in this high-risk subpopulation with concomitant HF and AF. The CHADS₂/CHA₂DS₂-VASc and HAS-BLED scores remain the most widely used clinical tools for stratifying stroke and bleeding risk, respectively.52 The CHA₂DS₂-VASc score refines the earlier CHADS₂ model by incorporating additional risk factors to improve risk discrimination, particularly in patients at lower risk. The scoring system assigns 1 point for congestive HF, hypertension, age 65-74 years, diabetes mellitus, vascular disease (e.g., prior myocardial infarction or peripheral artery disease), and female sex; and 2 points for age ≥75 years and previous stroke, transient ischemic attack (TIA), or thromboembolism. The maximum score is 9, with a score of ≥2 in men or ≥3 in women indicating the need for oral anticoagulation therapy.9,52 To evaluate the risk of bleeding in anticoagulated patients, the HAS-BLED score is commonly used. It includes 1 point for hypertension, abnormal renal or liver function (1 point each), prior stroke, bleeding history or predisposition to bleeding, labile INR (if on VKA therapy), age >65 years, and the use of drugs (e.g., antiplatelets or nonsteroidal anti-inflammatory drugs) or alcohol (1 point each). The total possible score is 9. While a HAS-BLED score ≥3 indicates a higher risk of bleeding, it should prompt closer monitoring rather than exclusion from anticoagulation.52,53 The 2024 ESC guidelines for the management of AF recommend the initiation of oral anticoagulants (OAC) in patients with an elevated risk of thromboembolism. Although they introduce a novel modified risk score (CHA₂DS₂-VA), they do not endorse the use of any specific score due to the modest predictive values of all risk scores. Clinicians should instead use scores validated for their local population or apply CHA₂DS₂-VA if such tools are not available. Those guidelines provide several reasons for it, e.g. the lack of proper discrimination and classification of adverse events, various risk factors specific to different populations, and limited data on how to implement risk scores on a larger scale to improve patient outcomes.7 Furthermore, the 2024 ESC guidelines do not endorse a single bleeding risk score for similar reasons. It is emphasized that clinical evaluation, including the assessment and management of bleeding risk factors, remains crucial for risk stratification.7

CHADS₂/CHA₂DS₂-VASc are historical risk scores developed based on non-anticoagulated subpopulations from large cohort studies, which entails significant limitations, as those studies had low levels of randomization and substantial underreporting of thromboembolic events.8,9,54 Furthermore, notable evolution has occurred in baseline patient characteristics, available therapeutic options, and treatment goals, raising questions about the continued applicability of these risk scores in clinical practice. The introduction of DOACs, the availability of non-pharmacological stroke prevention via left atrial appendage occlusion (LAAO), increasing life expectancy, substantial multimorbidity, and polypharmacy are among the factors that influence stroke risk assessment in patient population.7 Notably, this crucial change may be even more pronounced in patients with AF and HF due to important novelties in HF treatment with the sudden widespread adoption of SGLT2 inhibitors or angiotensin neprilysin-receptor inhibitors (ARNIs). It has been reported that SGLT-2 inhibitors may have antiarrhythmic properties.55,56 Introduction of SGLT2 inhibitors in patients presenting with HF and type 2 diabetes lowers the incidence of AF/AFl,57, 58, 59 stroke,58 and intracardiac thrombosis.59 However, the evidence for potential AF benefits in non-diabetic patients is scarce and insufficient to draw conclusions.58,60 Additionally, ARNIs have shown some promising results regarding atrial arrhythmias, with reports of lower recurrence rates.61,62 However, in their meta-analysis, Mujadzic et al. showed no significant difference in the incidence of AF between ARNI and ACEI/ARB.63 Additionally, significant changes have also occurred regarding the diagnosis, management, and outcomes of various comorbidities, such as coronary artery disease64 and hypertension.65 Both are significantly intertwined with AF pathophysiology and clinical outcomes.

Over the years, numerous additional stroke risk factors in patients with AF, beyond those included in the CHADS₂/CHA₂DS₂-VASc scores, have been identified in various cohort studies. Cancer, hyperlipidemia, and smoking are associated with an increased risk of stroke.7 Ethnicity is also identified as a significant factor influencing the prevalence of AF and the associated risk of stroke. Studies showed that Caucasian patients had a higher incidence of AF but a lower risk of stroke, whereas Black and Hispanic patients showed a lower prevalence of AF with a higher risk of stroke.7,66 Interestingly, a recently published nationwide US study seems to support these observations.66 However, a similar nationwide study conducted in the UK population showed no significant differences in stroke risk among ethnic groups.67 This may be partially explained by a range of US-specific factors, such as socio-economic determinants of access to healthcare.68 Similarly, both obesity7 and low body weight (<50kg)69 are independently correlated with a higher incidence of ischemic stroke. Malnutrition may also play a crucial role in the elderly and multimorbid population, as it is associated with adverse outcomes in patients with various chronic diseases such as HF,70 coronary artery disease,71 diabetes,72 CKD,73 and most notably, AF.74, 75, 76, 77, 78 The prevalence of malnutrition in the AF population ranges from 5% to 60 %, depending on age and the method of assessment.76, 77, 78, 79 In patients with AF aged ≥ 80 years, malnutrition significantly increases the risk of stroke (sHR 1.37, 95% CI 1.10-1.69).76 The Controlling Nutritional Status (CONUT) score is a widely used screening tool that assesses nutritional status based on serum albumin, total cholesterol levels, and lymphocyte count. It is primarily used to identify malnutrition in hospitalized patients.80 A CONUT score of 5 or more, indicating moderate to severe malnutrition, was associated with the risk of ischemic stroke/TIA (aHR 2.25, 95% CI 1.11−4.56).74 Recently, metabolic-associated fatty liver disease (MAFLD) has been associated with a higher prevalence of stroke compared to the general population.81 A prospective cohort study including 325,129 patients from the UK reported an increased risk of stroke and myocardial infarction in patients with versus without MAFLD.82 Furthermore, MAFLD has deleterious effects on cardiovascular prognosis as it is associated with an increased risk of AF, ventricular arrhythmias, HF with preserved ejection fraction (HFpEF), atherosclerotic cardiovascular disease, and CKD.83, 84, 85 Obstructive sleep apnea (OSA) is another comorbidity closely associated with AF and obesity at both the pathophysiological and clinical levels.86 The presence of OSA in patients with AF has been reported to increase the risk of stroke.87,88 A retrospective study of 22,760 patients from the ORBIT-AF I and II registries showed that OSA was independently associated with a higher risk of stroke/systemic embolism (HR 1.38, 95% CI 1.12-1.70) and major adverse cardiac and neurologic events (MACNE) (HR 1.16, 95% CI: 1.03-1.31).87 CKD increases the risk of stroke by 5-30-fold, with the highest risk observed in patients requiring hemodialysis.89 Additionally, proteinuria and reduced glomerular filtration rate are independent stroke risk factors in patients with AF.90 A large cohort study by Ocak et al., including 12,394 patients, showed that CKD was associated with a more than 2-fold increase in the risk of ischemic stroke in patients with AF.91 Although a vast body of literature has evaluated the risk of stroke in patients with AF, leading to the identification of many risk factors and the development of multiple validated risk scores, the amount and quality of evidence addressing stroke risk assessment in HF populations remain limited. However, a recently published meta-analysis that included data from 30 studies and 75,357 patients has shed some light on this underexplored topic. Higher NYHA class, lower systolic blood pressure, diuretic use, and diabetes were associated with an increased risk of stroke.92

An emerging group of stroke risk factors based on serum levels of various biomarkers, such as troponin, natriuretic peptides, growth differentiation factor-15, cystatin C, and interleukin-6, has recently shown promising results in improving stroke risk stratification.7 Elevated levels of B-type natriuretic peptide (BNP) and N-terminal pro-B-type natriuretic peptide (NT-proBNP) play a central role in the pathogenesis, diagnosis, and management of HF.93 However, numerous studies have shown that elevated levels of these peptides may be equally valuable for predicting the risk of stroke, as higher levels of NT-proBNP increase the risk of ischemic stroke in the general population.94,95 Elevated BNP levels are independently associated with a higher risk of stroke in patients with HFpEF,96 HFrEF97 and in those recently hospitalized for HF.98 Furthermore, elevated levels of NT-proBNP were positively correlated with an increased risk of stroke in patients with AF.99, 100, 101 Notably, adding NT-proBNP levels to various risk scores significantly improves stroke risk stratification.94,100,102 Incorporating NT-proBNP levels into the CHA₂DS₂-VASc score increased C-statistics from 0.62 to 0.65 (p = 0.0009)100 and 0.68 (p = 0.047)102 for stroke and/or systemic embolism, depending on the study. Interestingly, the data from the Fushimi AF Registry suggests that elevated levels of BNP and NT-proBNP in patients with AF were positively correlated with the incidence of stroke,103 whereas a history of HF was not.103 Another study from the same registry showed that elevated levels of NT-proBNP were predictors of stroke/systemic embolism even in patients with AF and without HF.104 A post-hoc analysis of the ARISTOTLE trial demonstrated that elevated levels of both high-sensitivity troponin I (hs-TnI) and high-sensitivity troponin T (hs-TnT) were independently associated with an increased risk of stroke in patients with AF.105,106 Similarly, findings from ENGAGE AF-TIMI 48 also demonstrated that hs-TnI had a significant prognostic value for stroke in patients with AF.101 In their meta-analysis, Leonie et al. included 96,702 patients and compared high versus low levels of high-sensitivity cardiac troponin (hs-cTn), showing that high levels of hs-cTn were associated with an increased risk of stroke in both patients with AF and the general population.107 Similar to NT-proBNP, the addition of hs-cTn levels to the CHA₂DS₂-VASc risk score has also improved the C-statistic for predicting stroke and/or embolism.105,106 Several commonly used laboratory tests have also been shown to be independently associated with stroke/thromboembolic events. Markers of hypercoagulability, such as increased levels of D-Dimer,101,108 von Willebrand factor (vWf),109 and fibrinogen,110 have all shown promising results in stroke prediction. Oxidative stress (OS) and inflammation play a key role in the pathophysiology of AF and its outcomes.111,112 Understandably, numerous markers of OS and inflammation are associated with increased stroke/thromboembolic risk in the population with AF, including C-reactive protein (CRP),113 interleukin-6 (IL-6),113 and growth differentiation factor 15 (GDF-15).114 Additionally, various markers of inflammation, such as red blood cell distribution width (RDW), neutrophil-to-lymphocyte ratio (NLR), and mean platelet volume (MPV), derived from a simple complete blood cell count, were associated with an increased risk of stroke/ thromboembolism.115, 116, 117

Electrocardiographic parameters, including an abnormal P-wave axis,118 as well as the type and duration of AF, have been identified as independent predictors of stroke.119 The CHA₂DS₂-VASc score fails to differentiate between patients with paroxysmal and persistent/permanent AF, despite several studies, such as the ROCKET-AF120 and the Fushimi AF registry,121 proving that paroxysmal AF is associated with a lower incidence of stroke compared to persistent AF. A large meta-analysis that included 99,996 patients across 12 different studies supported this conclusion.122 Furthermore, in their meta-analysis, Yang et al. showed that the duration of an AF episode of more than 5 min was also independently associated with the risk of stroke.123

In recent years, numerous structural and functional parameters of left atrial appendage (LAA), left atrium (LA), and left ventricle, along with other echocardiographic findings, have been implicated as potential risk factors for stroke/thromboembolism. LA enlargement (LAE) was independently associated with a higher risk of stroke in patients with AF.124,125 Furthermore, LAE was independently correlated with an increased risk of stroke, even in the general population.126 Additionally, LAE increased the risk of recurrent stroke, particularly of the cardioembolic/cryptogenic type.127 It should also be noted that the severity of LA fibrosis, assessed using novel late gadolinium enhancement cardiac magnetic resonance imaging, showed a significant correlation with ischemic stroke/TIA in patients with AF.128 It was reported that both reduced LAA flow velocity119 and specific LAA morphology were associated with a higher risk of stroke.129 There are four types of LAA morphology - Chicken Wing (CW), Cactus, Windsock, and Cauliflower, listed in descending order of prevalence.129 Non-CW morphology was associated with an increased risk of stroke in a paper by Di Biase et al., which introduced this morphological classification129 and a meta-analysis by Anan et al.130 Importantly, neither the presence of spontaneous echo contrast (SEC) nor LA/LAA thrombus is included in classic risk scores, despite their clear pathophysiological significance.131 Spontaneous echo contrast (SEC) is an echocardiographic phenomenon characterized by smoke-like echogenicity, most commonly observed in LA or LAA.132 Notably, in patients with SEC on transthoracic echocardiography (TTE), or more commonly transesophageal echocardiography (TEE), stroke severity was higher (median NIHSS 5 vs 3; P=0.004) and they more often had poor outcome at 3 months (mRS >2 in 32.3% vs. 16.1%; P<0.001).133 Conversely, the ARISTOTLE substudy focusing on patients with SEC, LA/LAA thrombus, or a complex aortic plaque on TTE/TEE showed no significant difference in the risk of stroke compared with patients with AF without such findings.134 Left ventricular (LV) parameters, including LV relative wall thickness135 and the E/e’ ratio,136 were independent stroke risk factors. Table 1 shows risk factors.

Additionally, numerous novel AF risk scores have been developed in recent years, as summarized in Table 2. Some of them are modifications of the CHA₂DS₂-VASc score that include renal dysfunction, elevated biomarkers (e.g. NT-proBNP or cardiac troponin), or imaging findings related to left atrial size. A different approach, e.g. the ABC-stroke score, integrates age, biomarkers, and clinical history, providing a computer-generated annual percentage risk.137 Furthermore, an emerging subgroup of machine-learning models shows similar or even better results than the CHA₂DS₂-VASc score.138 In 2015, a model that evaluated stroke risk in patients with HF but without concomitant AF was developed.139 It included eight variables, including age, BMI, NYHA class, creatine, a history of stroke, peripheral artery disease, coronary artery disease, and diabetes mellitus requiring insulin therapy.139 The simplified version, known as S2I2N0-3, was later validated for patients with HFrEF and HFpEF using data from various randomized control trials (RCTs), including PARADIGM-HF, ATMOSPHERE, DAPA-HF, I-Preserve, and PARAGON-HF.97,140 This version included a history of stroke (2 points), diabetes mellitus treated with insulin (2 points), and NT-proBNP levels (100-499 pg/ml- 0 points; 500-1499 pg/ml- 1 point; 1500-4999 pg/ml –2 points; 5000-20000 pg/ml- 3 points).97 The S2I2N0-3 score showed strong discriminatory power, with c-statistics of 0.8497 for HFrEF and 0.81140 for HFpEF.