Risk Stratification for Stroke in Atrial Fibrillation: A Critique

Ammar M. Killu; Christopher B. Granger; Bernard J. Gersh


Eur Heart J. 2019;40(16):1294-1302. 

In This Article

Predicting Risk of Stroke From Currently Available Risk Scores

Despite their wide use, current stroke risk stratification scores are, at best, only modestly good at predicting an individual's risk of thromboembolism. The most widely used score is the CHA2DS2-VASc score. This provides a score between 0 and 9 [one point for congestive heart failure, hypertension, age 65–74 years, diabetes mellitus, peripheral vascular disease, and female gender (Sex category) and two points for prior stroke/transient ischaemic attack or age ≥75 years]. Males with CHA2DS2-VASc 0 and females with CHA2DS2-VASc 1 are at very low risk of stroke (<1% per year), whereas males with CHA2DS2-VASc ≥1 and females with CHA2DS2-VASc ≥2 are at higher risk. This suggests that female sex is a risk-modifier rather than a risk factor, per se.[20] The CHA2DS2-VASc score has usurped the CHADS2 score given its greater ability to identify truly low-risk individuals.[21] In a Danish cohort study of 73 538 non-anticoagulated AF patients, those at low risk (score 0) had a thromboembolism rate of 1.67/100 person-years with the CHADS2 schema and 0.78/100 person-years with the CHA2DS2-VASc schema at 1 year of follow-up. Furthermore, in those traditionally labelled as intermediate risk (Score 1), the CHADS2 stroke rate was 4.75/100 person-years compared with 2.01/100 person-years for CHA2DS2-VASc score. Importantly, both scores agree that patients with a score of 0 truly are at low risk for stroke.[22] Although development of the CHA2DS2-VASc score in 2010 helped to redefine a significant number of individuals from the moderate- to low-risk category, the C-statistic (measuring goodness of fit) is only 0.6 (score range between 0 and 1, with 0.5 representing random chance).[23] In fact, the majority of stroke risk scores have a C-statistic of approximately 0.6–0.65.[23] As such, it is clear that there is a considerable amount of simplification for an exceedingly complex and as yet not fully defined process.

Current societal AF guidelines provide recommendations regarding anticoagulation therapy on the basis of the CHA2DS2-VASc score. The American Heart Association/American College of Cardiology and Heart Rhythm Society (AHA/ACC/HRS) state that anticoagulation therapy is not indicated for those with a score of 0 and indicated for those with CHA2DS2-VASc ≥2. Individuals with CHA2DS2-VASc 1 may be treated with an oral anticoagulant, aspirin or nothing at all.[19] This guideline was partially influenced by the Canadian Cardiovascular Society (CCS) guidelines which state that aspirin is reasonable in AF patients with CHADS2score 0 if they have arterial vascular disease (coronary, aortic, or peripheral); in the absence of these, no antithrombotic therapy is recommended.[24] Yet, the more recent (2016) ESC guidelines state that anticoagulation is recommended in men with CHA2DS2-VASc ≥2 (or should be considered in those with CHA2DS2-VASc 1) and women with CHA2DS2-VASc ≥3 (or should be considered in those with CHA2DS2-VASc 2). The American College of Chest Physicians (ACCP) AF guidelines suggest oral anticoagulation (OAC) for patients with a single non-sex CHA2DS2-VASc risk factor instead of no therapy or antiplatelet therapy while the Heart Foundation and the Cardiac Society of Australia and New Zealand (NHFA CSANZ) guidelines state that anticoagulation should be considered for these patients.[25,26] In addition, both favour NOACs in place of vitamin K antagonists where feasible. Importantly, the ESC guidelines provide a recommendation against thromboprophylaxis in AF in those with CHA2DS2-VASc 0, and favour OAC for those with a score ≥1 (Figure 2). Aspirin is given a Class III, level of evidence A recommendation—it should not be used because of harm—for stroke prophylaxis. Similarly, the National Institute for Health and Care Excellence (NICE) explicitly state that aspirin monotherapy should not be offered solely for stroke prevention in patients with AF,[27] while the NHFA CSANZ guidelines state that antiplatelet therapy is not recommended for stroke prevention in AF patients, regardless of risk.[26]

Figure 2.

Main highlights of the societal guidelines pertaining to anticoagulation in atrial fibrillation based on risk score. Strong and weak refer to strength of recommendation from the respective society. *Female gender alone is excluded. ACCP, American College of Chest Physicians; AHA/ACC/HRS, American Heart Association/American College of Cardiology/Heart Rhythm Society; ATT, antithrombotic therapy; AVR, aortic valve replacement; ESC, European Society of Cardiology; INR, international normalized ratio; MVR, mitral valve replacement; NHFA CSANZ, Cardiac Society of Australia and New Zealand; NOAC, novel oral anticoagulant; VKA, vitamin K antagonist.

This begs the question: Why is there discrepancy between guidelines regarding aspirin and what has led to the recent recommendation against it? For years, aspirin has been recommended as a suitable thromboprophylactic agent for AF in individuals that are not considered to be high risk for stroke. This included patients with a CHADS2 or CHA2DS2-VASc score of ≤1 and was based on the misperception that aspirin is both efficacious (reduces stroke occurrence) and safe (low risk of bleeding).[28] Even in the Physicians Health Study, often quoted as evidence for the role of aspirin in primary prevention of cardiovascular disease, there was an increased incidence of intracranial haemorrhage with no benefit on overall mortality.[29] Prior suggestions of benefit with aspirin are largely derived from the Stroke Prevention in AF (SPAF-1) trial in which a 325 mg daily dose was used. As pointed out in the most recent AHA/ACC/HRS guidelines, the effect was highly heterogeneous.[30] Worryingly, this led to oral anticoagulants being underutilized in patients who meet criteria for them, with a significant proportion of patients on aspirin instead. For example, in an analysis of the Canadian Stroke Network the majority of patients with known AF admitted with a stroke were either sub-therapeutic on warfarin on not taking it.[31] Meanwhile, in a study derived from the PINNACLE registry, less than half of high-risk patients received an OAC prescription.[32] Part of the underutilization of anticoagulants may be driven by inherent differences in perception between patients and physicians with the latter placing more value on avoiding bleeding compared with reducing strokes.[33] As such, the increased risk of bleeding in the elderly may dissuade some providers from prescribing anticoagulants, instead favouring aspirin. However, the stroke risk is highest in the elderly and it has been shown that aspirin is least effective in this population though remains associated with a significant bleeding risk. In the Birmingham AF Treatment of the Aged Study (BAFTA), patient's ≥75 years with AF were randomized to warfarin (target international normalized ratio 2–3) vs. aspirin 75 mg daily: the warfarin group had a 2% yearly risk reduction in fatal or disabling stroke with no significant difference in major bleeding compared with those on aspirin. Other evidence against aspirin as a stroke preventive strategy comes from a large cohort study which demonstrated a similar HR for stroke amongst individuals treated with aspirin compared with untreated individuals,[34] while a recent Taiwanese database study demonstrated net clinical benefit (lower risk of ischaemic stroke with no difference in intracranial bleeding) with oral anticoagulant use in AF patients >90 years.[35] In sum, aspirin results in a non-significant reduction in (mostly less disabling) stroke despite a significant rise in major bleeding.[36] As such, it should not be seen as an alternative in anticoagulant-eligible patients. This was demonstrated by the AVERROES (Apixaban vs. Acetylsalicylic Acid to Prevent Strokes) trial which enrolled 5599 patients deemed unsuitable for warfarin and randomized them to apixaban or aspirin. The trial was terminated early after demonstrating that apixaban was superior for preventing stroke while the rates of major bleeding with similar between the two treatment arms including the same rates of intracranial haemorrhage.[37] Given the above, aspirin has limited role for the treatment of AF except in those who have other indications for antiplatelet therapy such as coronary or peripheral arterial disease and are not already on anticoagulants. If patients are truly unable to take anticoagulants, aspirin and clopidogrel combination therapy has been shown to reduce the stroke risk though with significantly elevated bleeding risk.[38] In such patients, LAA occlusion procedures could be considered.[39]

It is important to recognize that the risk of stroke varies between individuals irrespective of their CHA2DS2-VASc score. Stated differently, it is not only the score that is important but the individual components of the score with individual risk elevated as people age (Figure 3). In addition, the risk in specific populations varies. This can be illustrated by assessing the annual stroke rate in different cohort studies of patients with CHA2DS2-VASc 1. For example, in a Taiwanese cohort, the annual risk was around 2.5% compared with <1% in a Swedish cohort.[40,41] Oral anticoagulation is associated with a major bleeding risk of around 1% per year; hence the threshold for annual ischaemic stroke rate where the benefit of anticoagulation outweighs bleeding risk is generally around 1–2%. Complexities include that there is a strong correlation between risk of stroke and risk of bleeding, and that generally one would tolerate substantially more (non-intracranial) bleeding than stroke. Nonetheless, the risk-benefit ratio may not be in favour of anticoagulation in a low-risk cohort. Another study that suggested patients with CHA2Ds2-VASc 1 do not necessarily derive significant stroke reduction from NOACs comes from a large American insurance claims database. In this analysis of 64 661 patients, several findings were noted.[42] First, OAC adherence anticoagulation was modest (<50% 'covered' with anticoagulation for ≥80% of the time); second, adherence with NOACs was only marginally better than with warfarin; and third, amongst patients with CHA2Ds2-VASc 0 or 1, the time not taking anticoagulation was not associated with stroke. Interestingly, in these low-risk groups, not taking anticoagulation for ≥3 months was associated with a significant reduction in bleeding despite no increase risk in stroke again suggesting that the risk-benefit ratio may be in favour of not anticoagulating. Based on the totality of the above information, it would seem inappropriate to recommend OAC in all patients with CHA2DS2-VASc 1 and individualization is advisable. Although such studies provide important insights into general practice settings, they are associated with significant limitations including the potential for substantial confounding. For example, in the study by Friberg et al.[41](referenced above) patients that were ever started on anticoagulant were excluded. This could result in 'conditioning on the future' error, resulting in lower event rates than expected.[43] Conversely, however, Quinn et al.[44] performed a systematic review of patients with non-valvular AF not treated with anticoagulation and found a stroke rate of >2% per year in approximately 20% of patients with CHA2DS2-VASc 1. Yet, this included studies enrolling CHA2DS2-VASc 1 patients based on female gender alone (low risk) and those with 'conditioning on the future' error; hence pooled event rates were low. An excellent report on the effect of analytic methods is provided by Nielsen et al.[43]

Figure 3.

The variable hazard ratio associated with the individual risk factors of the CHA2DS2-VASc score is demonstrated. As illustrated, not all factors are of the same risk despite assignment of the same score. Reused with permission from Dr Edgar Argulian, Mt. Sinai St. Luke's and Roosevelt Hospitals, New York, NY, USA.94

Finally, though the CHA2DS2-VASc score is calculated based on baseline risk factors, risk is dynamic. In fact, it has been shown that most patients with AF developed an additional CHA2DS2-VASc risk factor prior to presentation with stroke, and that a change in CHA2DS2-VASc score strongly predicted stroke risk.[45] As such, periodic evaluation for development of new comorbidities is required.