Stratifying Stroke Risk in Atrial Fibrillation: Beyond Clinical Risk Scores

Background

Atrial Fibrillation (AF) is the most common atrial arrhythmia, affecting nearly 1% of the United States population.¹ This estimate may understate its true burden, given that a high proportion of patients have subclinical AF. The prevalence of AF increases with age^{1, 2} and given an aging population, the projected prevalence of AF in the United States is at least 5.6 million by the year 2050.²

AF is associated with increased risk of stroke or systemic embolism³ and death.⁴ The heightened thromboembolic risk seen in AF is substantially reduced by anticoagulation.⁵ On the other hand, long-term anticoagulation can also lead to hemorrhagic complications. The recommended approach to making decisions about anticoagulant therapy in AF is to balance the expected risks of stroke versus bleeding. In this paper, we review widely used clinical methods for predicting stroke risk in AF, and augment this discussion with an overview of more recent risk predictors beyond conventional clinical scores.

Prediction of Stroke Risk Based on Clinical Factors

Risk Factors

Several studies have reported risk factors for stroke in patients with AF. A pooled analysis of five trials with 5,956 patient-years of follow-up found that independent risk factors for ischemic stroke were age (relative risk [RR] per 10 years, 1.4), prior stroke or transient ischemic attack (TIA) (RR, 2.5), hypertension (RR, 1.8), diabetes (RR, 1.7), and congestive heart failure (RR, 1.4).⁶ In patients younger than 65 years of age who had none of these risk factors, the annual rate of stroke was 1.0% (95% confidence interval [CI], 0.3%–3.1%) with antiplatelet or no antithrombotic therapy, whereas in patients older than 75 years of age with one or more of these risk factors, the annual stroke rate was 8.1% (95% CI, 4.7%–13.9%).⁶ Another systemic review, including seven studies with over 12,000 patients, identified the following factors to be associated with stroke or systemic embolism: prior stroke or TIA (RR, 2.5; 95% CI, 1.8–3.5), increasing age (RR per decade, 1.5; 95% CI, 1.3–1.7), hypertension (RR, 2.0; 95% CI, 1.6–2.5), and diabetes (RR, 1.7; 95% CI, 1.4–2.0).⁷

A less well established risk factor for stroke in AF is female sex. A study from a Swedish hospital discharge registry, feature 100,802 patients with a median follow-up of 1.2 years, found an association between female sex and the risk of stroke (hazard ratio [HR], 1.18; 95% CI, 1.12–1.24).⁸ Women with AF and no other stroke risk factors (often termed lone AF) had a non-significantly higher stroke risk when compared to men (overall rate, 0.7% vs. 0.5%; P = 0.09).⁸ The association between female sex and stroke risk in AF has been also reported elsewhere.^{9, 10} On the other hand, some studies have found no association between female sex and stroke in AF.⁷

After publication of the pooled analyses of early trials in AF discussed above, several other risk factors of stroke in AF were reported. Several studies have shown an association of coronary artery disease,^{11, 12} peripheral vascular disease,¹³ and complex aortic plaque¹⁴ with the risk of stroke or systemic embolism in patients with AF. In addition, renal disease has recently emerged as a possible risk factor for stroke in AF. In a study with 33,165 person-years of follow-up, the risk of thromboembolism in AF was increased in the presence of proteinuria (RR, 1.54; 95%, 1.29–1.85) or an estimated glomerular filtration rate less than 45 ml/min/1.73 m² (compared to ≥60) even after adjustment for other risk factors (RR, 1.39; 95% CI, 1.13–1.71).¹⁵ On the other hand, another study with 5,912 patients and 1 year of follow-up found no association between renal impairment and stroke risk in AF after adjustment for known risk factors (HR, 1.06; 95% CI, 0.75–1.49).¹⁶

Risk Scores

Clinical scores have been developed to help predict the risk of stroke and systemic embolism in patients with AF, with the ultimate goal of determining whether that risk is high enough to warrant the bleeding risks associated with anticoagulant therapy. The CHADS₂ score assigns 1 point each for congestive heart failure, hypertension, age ≥75 years, and diabetes, and 2 points for prior stroke or TIA. The CHADS₂ score was one of the first published stroke risk prediction tools for management of AF. In the CHADS₂ derivation cohort, with 2,121 patient-years of follow-up), each 1-point increase in the CHADS₂ score was associated with a 1.5-fold higher hazard of stroke, and the score had good predictive ability with a c-statistic of 0.82.¹⁷ External validation studies showed fair-to-good predictive ability of the CHADS₂ score, with c-statistics ranging between 0.56 and 0.82.¹⁸ Despite these validation studies, major limitations of the CHADS₂ score have become apparent as its use has become more widespread. First, the CHADS₂ score has also been shown to predict hemorrhagic complications in patients with AF.¹⁹ This means that patients with a high predicted risk of thromboembolic events are also the same patients with a high predicted risk of hemorrhagic complications, which complicates decisions about anticoagulant therapy. More importantly, the score suffers from significant heterogeneity in actual stroke risk in patients labeled as “low risk” based on their CHADS₂ score. In the overall group of patients assigned a CHADS₂ score of 0, the annual stroke risk can range up to 3%.²⁰ This is problematic if the score is used to withhold anticoagulant therapy in those with low scores, a purpose for which it is often used.

To better risk stratify patients considered to be “low risk” based on the CHADS₂ score, investigators designed the CHA₂DS₂-VASc score. This score builds on the CHADS₂ score by adding an extra point each for female sex and vascular disease (which includes both coronary heart disease and peripheral vascular disease), and dividing age into three categories (<60 years, 60–74 years, and ≥75 years) instead of the two categories in the original CHADS₂ score.²¹ It appears that the CHA₂DS₂-VASc score outperforms the CHADS₂ score in discriminating stroke risk in the group of patients with a CHADS₂ score of 0 or 1.^{21, 22} The CHA₂DS₂-VASc score is especially helpful in that patients classified as “low risk” truly do appear to be at low risk of stroke and may safely be managed without anticoagulant therapy.¹⁸ In addition to the CHADS₂ and CHA₂DS₂-VASc scores, several other clinical scores have been proposed.^23–26

A shortcoming shared by all of these clinical risk prediction scores is that they include general risk factors for stroke that apply to patients with and without AF. It is intuitive that stroke risk increases with the general burden of vascular disease. In fact, studies have shown that the CHADS₂ and CHA₂DS₂-VASc scores predict stroke risk even in patients without AF.²⁷ On the other hand, specific biomarkers reflecting dysfunction of the left atrium or left atrial appendage (LAA) may be more predictive of the type of stroke that is most likely amenable to prevention with anticoagulant therapy.

Prediction of Stroke Risk Based on Electrocardiographic Markers

Type of Atrial Fibrillation

There is conflicting evidence on whether stroke risk varies based on the type of AF. In the ACTIVE-W trial, which enrolled 6,706 patients and compared anticoagulant therapy versus the combination of aspirin and clopidogrel, the risk of stroke or systemic embolism was similar in patients with paroxysmal as compared to persistent AF (RR, 0.87; 95% CI, 0.59–1.30).²⁸ More recently, an analysis of data from the ENGAGE trial showed that the risk of stroke or systemic embolism was significantly lower in patients with paroxysmal AF (1.49% per year) when compared to persistent AF (1.83% per year) or permanent AF (1.95% per year).²⁹ A meta-analysis including 99,996 patients in 12 studies found that the rate of thromboembolism was higher in patients with non-paroxysmal AF as compared to those with paroxysmal AF (HR, 1.38; 95% CI, 1.19–1.61).³⁰ Overall, there is a growing consensus that the risk of thromboembolism increases as patients progress from paroxysmal to sustained to permanent AF.

Duration of Atrial Fibrillation

Studies investigating the association between these subclinical episodes of AF and stroke have yielded mixed results. The AFFIRM study randomized 4,060 patients with AF to rate- versus rhythm-control strategies. After 5 years of follow-up, the stroke risk among the 481 patients (12%) with asymptomatic AF was similar to the stroke risk in the remaining patients with symptomatic AF (3.8% vs. 4.4%, P = 0.52).³¹ However, since only 4% of asymptomatic AF episodes were of less than 6 hour duration, no conclusions can be made on stroke risk in patients with brief episodes of AF.

Several studies investigated stroke risk in relation to the duration of AF. The ASSERT study enrolled 2,580 patients 65 years of age or older with hypertension and no known history of AF who had recently undergone pacemaker or defibrillator implantation. During the first 3 months after device implantation, subclinical atrial tachyarrhythmias lasting at least 6 minutes were captured in 261 (10.1%) patients and these were associated with increased stroke or systemic embolism risk over a mean follow up of 2.5 years (HR, 2.49; 95% CI, 1.28–4.85). Similarly, the SOS AF study, which included 10,106 patients with implanted cardiac devices who underwent at least 3 months of follow-up, found an association between AF episodes of duration 5 minutes or longer and stroke (HR, 1.76; 95% CI, 1.02–3.02).³² There appeared to be a proportional relationship between AF burden and stroke risk (HR per hour of AF duration, 1.03; 95% CI, 1.00–1.05; P = 0.04).³² On the other hand, the RATE study, which included 5,379 patients with pacemakers and a mean follow-up period of 22.9 months, found that patients with episodes of AF <20 seconds duration had a similar risk of stroke or TIA compared to patients with no AF (HR, 0.87; 95% CI, 0.58–1.31). On the other hand, episodes of AF ≥20 seconds were associated with stroke or TIA (HR, 1.51; 95% CI, 1.03–2.21).³³ A common limitation of these studies is that patients included were already at high risk for AF and perhaps stroke, due to the confounding cardiac diseases that necessitated implantation of such devices. Therefore it is difficult to extrapolate this result to a wider population.

Anticoagulation therapy has been proven to reduce stroke risk in patients with a sufficient burden of AF to have it detected by pulse palpation or 12-lead ECG at multiple points in time.³⁴ However, it is uncertain if anticoagulation reduces stroke risk in patients with brief and infrequent runs of subclinical AF. The ARTESiA trial (NCT01938248), for example, aims to enroll 4,000 patients with device-detected AF last 6 minutes to 24 hours and randomly assign them to apixaban or aspirin and determine if anticoagulation reduces the risk of ischemic stroke and systemic embolism in this population.

Prediction of Stroke Risk Based on Serum Biomarkers

High-Sensitivity Cardiac Troponin

Cardiac troponin (cTnT) is a serum biomarker of myocardial injury or dysfunction.³⁵ Highly sensitive cTnT assays can detect concentrations 10 times lower than the levels detected by conventional assays.³⁵ Studies have shown an association between high-sensitivity cTnT with cardiovascular and cerebrovascular events.^{36, 37}

In the RE-LY study, subjects in the highest quartile of high-sensitivity cTnT had a significantly elevated risk of stroke when compared to those in the lowest quartile (HR, 1.99; 95% CI, 1.17–3.39).³⁸ Similarly, a recent study found that high-sensitivity cTnT level ≥23 ng/L was independently associated with stroke or systemic embolism risk (HR, 1.98; 95% CI, 1.42–2.78).³⁹ In addition, adding high-sensitivity cTnT to the CHA₂DS₂-VASc score improved the c-statistic from 0.63 to 0.65.

Amino Terminal Pro-Brain Natriuretic Peptide

Amino-terminal pro-brain natriuretic peptide (NT-proBNP) is another biomarker released by the myocardium in the setting of stretch and may be increased with structural heart disease, heart failure, and ventricular strain.⁴⁰ NT-proBNP is an independent predictor cardiovascular events³⁷ and stroke⁴¹ particularly of cardioembolic subtype.⁴²

In the RE-LY trial, AF patients with the highest quartile of NT-proBNP faced a higher stroke risk when compared to those in the lowest quartile (adjusted HR, 2.4; 95% CI, 1.41–4.07).³⁸ Similar findings were demonstrated in another study, and the addition of NT-proBNP level to the CHA₂DS₂-VASc score improved its c-statistic from 0.62 to 0.65.⁴³

Inflammatory Biomarkers

Biomarkers of inflammation have been shown to be associated with ischemic stroke risk⁴⁴ and cardiovascular events.⁴⁵ Inflammation is one of the proposed factors leading to thrombus formation in patients with AF. An analysis of patients in the SPAF-III trial found an independent association between C-reactive protein (CRP) levels and cardiovascular events and cardiovascular mortality, but not specifically stroke.⁴⁶ In RE-LY, subjects in the highest quartile of levels of an inflammatory cytokine (IL-6) faced a significantly higher stroke and systemic embolism risk when compared to those in the lowest quartile (HR, 2.03; 95% CI, 1.27–3.26).⁴⁷ The addition of IL-6 improved the c-statistic of the CHA₂DS₂-VASc score from 0.62 to 0.64. However, when more standard cardiac biomarkers such as high-sensitivity cTnT and NT-proBNP were added to the model, IL-6 was no longer associated with stroke or systemic embolism risk. Another study showed increased rates of all-cause mortality in those with the highest quartile of IL-6 (HR, 1.93; 95% CI, 1.57–2.37) and CRP levels (HR, 1.49; 95% CI, 1.24–1.79), but these biomarkers were not associated with stroke or systemic embolism risk.⁴⁸

ABC Score

Based on the above data, investigators developed the Age, Biomarkers, and Clinical history (ABC) score to predict stroke and systemic embolism risk in patients with AF. This score was derived from a cohort of 14,701 patients and externally validated in a separate cohort of 1,400. In this derivation/validation study, the ABC score was superior to the CHA₂DS₂-VASc score in predicting the risk of stroke and systemic embolism in both the derivation cohort (c-statistic, 0.68 vs. 0.62; P <0.001) and the validation cohort (c-statistic, 0.66 vs. 0.58; P <0.001).⁴⁹

Prediction of Stroke Risk Based on Structural and Functional Markers

Markers Related to the Left Atrium

Left atrial (LA) enlargement can potentiate stasis and endothelial injury, thereby potentially leading to thrombus formation. In the general population of patients regardless of AF status, LA enlargement (LAE) is associated with ischemic stroke risk⁵⁰, particularly of embolic subtypes⁵¹. In patients with AF, studies showed an association between LAE and stroke and systemic embolism.^{52, 53} In another study of 2,713 patients, LA diameter >45 mm was associated with the risk of ischemic stroke even after adjustment for the CHA₂DS₂-VASc score (HR 1.74 95% CI 1.25–1.83). Furthermore, in a study of AF patients being treated with either a vitamin-K antagonist (warfarin) or a non-vitamin K antagonist oral anticoagulation (NOAC) drug, 2.7% had a LAA thrombus seen on transesophageal echocardiogram (TEE) despite compliance with anticoagulation.⁵⁴ Those with thrombus had a mean LA diameter of 47 mm compared with 41 mm in those without thrombus (P = 0.003).⁵⁵

Markers Related to the Left Atrial Appendage

The LAA is the main origin of thrombi in patients with AF.⁵⁶ Reduced LAA flow velocity on echocardiography is a marker of LAA stasis. A cross-sectional study of AF patients showed that those with prior stroke had a lower LAA flow velocity (36 cm/s) than those without stroke (55 cm/s) (P <0.001).⁵⁷ Furthermore, in a post-hoc analysis of 721 patients from the SPAF-III trial who underwent TEE, peak anterograde (emptying) LAA flow velocity <20 cm/s was independently associated with LAA thrombus (RR, 2.6; P = 0.02).⁵⁸

Spontaneous echocardiographic contrast (SEC) is a dynamic smoke-like appearance pattern not uncommonly seen on TEE and considered to be a marker of hypercoagulability, stasis, and aggregation of red blood cells.⁵⁹ In an analysis of 382 patients undergoing TEE in the SPAF trials, SEC was detected in nearly 63% of patients and was independently associated with LAA thrombus and future thromboembolic events.⁶⁰

In addition to biomarkers of LAA dysfunction, the morphology and structure of the LAA are other determinants of stroke risk. In a cross-sectional study of 932 AF patients who underwent TEE, about 8% had a history of ischemic stroke or TIA. The morphological appearance of the LAA was classified as “chicken wing” in 48% of patients, “windsock” in 19%, “cactus” in 30%, and “cauliflower” in 3%. After adjusting for known stroke risk factors, morphologies other than chicken wing were associated with a higher odds of stroke: cauliflower (odds ratio [OR], 8.0; P = 0.056), windsock (OR, 4.5; P = 0.038), and cactus (OR, 4.08; P = 0.046).⁶¹ Other studies in AF also found an inverse association between chicken-wing morphology and stroke risk and covert brain infarcts.⁶² The mechanistic basis of these associations may be related to a higher LAA flow velocity in patients with chicken-wing morphology as compared with non-chicken wing morphology.⁶² On the other hand, the association between a cauliflower morphology and stroke was confirmed by another study showing that extensive trabeculations, which are often seen with the cauliflower morphology, were independently associated with stroke.⁶³

Besides the morphological shape of the LAA, a larger orifice size⁵⁷ and higher number of lobes⁶⁴ has also been associated with thromboembolic risk.

In addition to visualizing the LAA, cardiac MRI can characterize tissue changes such as fibrosis and scarring,⁶⁵ and more recently, the presence of late gadolinium enhancement on cardiac MRI has been used to evaluate the degree of fibrosis in patients with AF.⁶⁶ In a study comparing cardiac MRI and TEE, there was an association between the degree of atrial fibrosis on cardiac MRI and the detection of thrombus or spontaneous echo contrast on echocardiography (OR, 3.6; P <0.01).⁶⁷

Clinical trials showed that in patients with AF, LAA closure was non-inferior to warfarin in reducing stroke risk, suggesting a structural and functional role of the LAA in determining stroke risk in patients with AF.⁶⁸ Much of this benefit, however, was driven by a reduction in hemorrhagic stroke, so it remains unclear whether closure is as protective against ischemic stroke as anticoagulation, particularly if compared to direct oral anticoagulants, which have a significantly lower risk of intracranial hemorrhage than warfarin.

Prediction of Stroke Risk Based on Electrocardiographic Markers

P-wave terminal force in lead V₁ (PTFV₁) on a 12-lead electrocardiogram (ECG) can be used as a gauge of underlying left atrial abnormalities.⁶⁹ PTFV1 can be calculated by multiplying the duration and amplitude of the p-wave is lead V1.⁷⁰ PTFV₁ has been found to be associated with ischemic stroke risk in patients both with and without AF, suggesting that it provides additive information about stroke risk.⁷¹ However, it has not been formally tested whether the addition of PTFV₁ to commonly used clinical scores provides additional predictive ability. Furthermore, advanced interatrial block and Bayes syndrome are arrhythmogenic syndromes associated with atypical atrial flutter and atrial fibrillation and also predict non-lacunar cerebral ischemia and, probably, of vascular dementia.⁷²

Prediction of Stroke Recurrence

Stroke recurrence is an independent predictor of in-hospital mortality in patients with cardioembolic stroke. While clinical scores have fair to good ability to predict long-term stroke risk in patients with AF, there are limited data on predictors of early stroke recurrence.⁷³ Recently, a multicenter, prospective study found that predictors of early stroke recurrence in AF patients included severe left atrial enlargement (HR, 2.05; 95% CI, 1.08–2.87), brain infarct size >1.5 cm (HR, 1.82; 95% CI, 1.00–3.33), and advanced age (HR per year, 1.06; 95% CI, 1.00–1.11).⁷⁴ Based on this cohort, the ALESSA score (age ≥80 years: 2 points and between 70 and 79 years: 1 point; ischemic lesion >1.5 cm: 1 point; severe atrial enlargement: 1 point) was derived and validated in a separate patient cohort.⁷⁴ Higher ALESSA scores were shown to predict recurrent ischemic events at 90 days in both the derivation cohort (c-statistic, 0.7) and validation cohort (c-statistic 0.65) cohorts. The ALESSA score was less predictive of the risk of hemorrhagic complications in the derivation (c-statistic 0.59) and validation (c-statistic 0.41) cohorts.⁷⁴

Conclusions

Despite the strong association between AF and stroke, our ability to predict stroke risk in individual AF patients remains moderate at best. The addition of serum, structural, and ECG markers to commonly used clinical prediction scores may improve risk stratification and personalized therapy. Clinical trials are needed to test whether therapeutic strategies based on additional markers can improve outcomes compared to the current approach of tailoring therapy based on clinical risk scores alone.