Atrial Fibrillation Risk and Discrimination of Cardioembolic From Noncardioembolic Stroke

Abstract

Background and Purpose- Classification of stroke as cardioembolic in etiology can be challenging, particularly since the predominant cause, atrial fibrillation (AF), may not be present at the time of stroke. Efficient tools that discriminate cardioembolic from noncardioembolic strokes may improve care as anticoagulation is frequently indicated after cardioembolism. We sought to assess and quantify the discriminative power of AF risk as a classifier for cardioembolism in a real-world population of patients with acute ischemic stroke. Methods- We performed a cross-sectional analysis of a multi-institutional sample of patients with acute ischemic stroke. We systematically adjudicated stroke subtype and examined associations between AF risk using CHA₂DS₂-VASc, Cohorts for Heart and Aging Research in Genomic Epidemiology-AF score, and the recently developed Electronic Health Record-Based AF score, and cardioembolic stroke using logistic regression. We compared the ability of AF risk to discriminate cardioembolism by calculating C statistics and sensitivity/specificity cutoffs for cardioembolic stroke. Results- Of 1431 individuals with ischemic stroke (age, 65±15; 40% women), 323 (22.6%) had cardioembolism. AF risk was significantly associated with cardioembolism (CHA₂DS₂-VASc: odds ratio [OR] per SD, 1.69 [95% CI, 1.49-1.93]; Cohorts for Heart and Aging Research in Genomic Epidemiology-AF score: OR, 2.22 [95% CI, 1.90-2.60]; electronic Health Record-Based AF: OR, 2.55 [95% CI, 2.16-3.04]). Discrimination was greater for Cohorts for Heart and Aging Research in Genomic Epidemiology-AF score (C index, 0.695 [95% CI, 0.663-0.726]) and Electronic Health Record-Based AF score (0.713 [95% CI, 0.681-0.744]) versus CHA₂DS₂-VASc (C index, 0.651 [95% CI, 0.619-0.683]). Examination of AF scores across a range of thresholds indicated that AF risk may facilitate identification of individuals at low likelihood of cardioembolism (eg, negative likelihood ratios for Electronic Health Record-Based AF score ranged 0.31-0.10 at sensitivity thresholds 0.90-0.99). Conclusions- AF risk scores associate with cardioembolic stroke and exhibit moderate discrimination. Utilization of AF risk scores at the time of stroke may be most useful for identifying individuals at low probability of cardioembolism. Future analyses are warranted to assess whether stroke subtype classification can be enhanced to improve outcomes in undifferentiated stroke.

Keywords: atrial fibrillation; heart; humans; odds ratio; stroke.

Figure 1.

Patient flow through the study. Flow through the study is depicted.

Figure 2.

Distributions of CHA₂DS₂-VASc, CHARGE-AF and EHR-AF scores stratified by stroke mechanism The probability distributions of CHA₂DS₂-VASc (panel A), CHARGE-AF (panel B), and EHR-AF (panel C) scores stratified by stroke mechanism are depicted. Blue shade depicts non-cardioembolic events, while red shade indicates cardioembolic events.

Figure 3.

Predictive value of EHR-AF score for cardioembolic and non-cardioembolic stroke in a simulated undifferentiated ischemic stroke cohort The EHR-AF score-informed probabilities of cardioembolic and non-cardioembolic stroke are depicted along increasing true prevalence of cardioembolic and non-cardioembolic stroke in a simulated cohort of undifferentiated ischemic stroke patients. EHR-AF thresholds sensitive for cardioembolic and specific for non-cardioembolic stroke are depicted in Panels A and D, respectively, to illustrate the use of the EHR-AF score to identify patients at low risk for cardioembolism (and high risk for non-cardioembolism). EHR-AF thresholds specific for cardioembolic and sensitive for non-cardioembolic stroke are depicted in Panels B and C, respectively, to illustrate the potential use of the EHR-AF score to identify patients at high risk of cardioembolism (and low risk for non-cardioembolism). At a given true prevalence on the x-axis, the corresponding score-informed probability on the y-axis at each threshold curve is the probability of the given stroke subtype in an individual with an EHR-AF score equal to the indicated threshold. Gray dashed lines depict the performance of an uninformative test. The selected cardioembolic stroke true prevalence range (5-30%) reflects rates typically encountered in stroke cohorts.