Detection and management of atrial fibrillation after cryptogenic stroke or embolic stroke of undetermined source

Abstract

Cryptogenic stroke (CS) and embolic stroke of unknown source (ESUS) represent a major challenge to healthcare systems worldwide. Atrial fibrillation (AF) is commonly found after CS or ESUS. Independent of the mechanism of the index CS or ESUS, detection of AF in these patients offers the opportunity to reduce the risk of stroke recurrence by prescribing an anticoagulant instead of aspirin. The detection of AF may be pursued with different monitoring strategies. Comparison of monitoring strategies should take into account that AF detection rates reported in published studies, and then pooled in meta‐analyses, are not only a function of the monitoring strategy itself, but also depend on patient‐related, device‐related, and study design–related factors. Once AF is found, the decision to anticoagulate a patient should be made on the basis of AF burden and the baseline risk of the patient. Empirical anticoagulation in patients with ESUS and no evidence of AF is an intriguing but still‐unproven strategy and therefore should not be adopted outside of randomized clinical trials.

1. INTRODUCTION

Each year, >790 000 people experience a new or recurrent stroke in the United States.1 The cause of ischemic stroke remains undetermined after hospitalization in 10% to 40% of cases; this is often referred to as “cryptogenic” stroke (CS).2 Differences in the definition and diagnostic workup contribute to this variable incidence in published series. To mitigate this limitation, the term “embolic stroke of undetermined source” (ESUS) has recently been proposed to identify those cases where the etiology of nonlacunar ischemic stroke remains elusive after a standardized diagnostic pathway.3 The prevalence of ESUS was 16% to 32% in a recently published series.4 Hence, even after a standardized comprehensive evaluation, the cause of ischemic stroke remains elusive in a significant proportion of patients, thus representing a major challenge for treatment selection.

2. STROKE RECURRENCE IN PATIENTS WITH PRIOR CS AND ESUS

In a recent study, patients with CS had an ischemic stroke recurrence rate of 9.1% (range, 6.2%–12.0%), 24.0% (range, 19.1%–28.9%), and 31.9% (range, 25.2%–38.6%) at 1, 5, and 10 years, respectively.5 Recurrent stroke in these patients was classified again as cryptogenic in 63% of cases.5 Stroke recurrence rate in ESUS was also reported to be high and similar to CS.6 Possible causes of stroke recurrence, irrespective of the mechanism of the index stroke, are paroxysmal atrial fibrillation (AF), arterial sources of thromboembolism (TE), patent foramen ovale (PFO), cardiac structural abnormalities, and other less common etiologies. Detection of AF after a CS or ESUS does not imply causality but still offers an opportunity for reducing the risk of stroke recurrence by prescribing oral anticoagulation (OAC). However, AF is often asymptomatic, paroxysmal, and difficult to detect with conventional tools. Several monitoring strategies have been investigated to detect asymptomatic AF after an ischemic stroke, including electrocardiography (ECG), in‐hospital monitoring, Holter ECG of variable duration, external event or loop recorders, mobile cardiac outpatient telemetry, and implantable cardiac monitors. These studies, and their recent meta‐analyses,7, 8, 9, 10, 11, 12 contributed significantly to our current understanding of the incidence of AF after ischemic stroke.

3. AF DETECTION RATE IN DIFFERENT STUDIES

AF detection rates in published studies are the result of a complex interplay of patient‐related, device‐related, and study‐related factors (Figure 1), which may be a source of hidden heterogeneity in meta‐analyses and may lead to flawed conclusions when attempting indirect comparisons.

Figure 1.

Interaction between patient‐related, device‐related, and study design–related factors influencing AF detection rate in clinical studies. Abbreviations: AF, atrial fibrillation; TEE, transesophageal echocardiography; TIA, transient ischemic attack

3.1. Patient‐related factors

The detection rate of AF in a study population depends to a great extent on the prevalence of risk factors for incident AF in the cohort being studied. Age, sex, heart failure, hypertension, diabetes mellitus, coronary artery disease, chronic kidney disease, and sleep apnea are all important confounders in these studies. However, the single most influential risk factor for incident AF is age: the older the study population, the more prevalent AF will be during follow‐up. Clinical presentation of the patient may also influence the incidence of AF. Patients presenting with transient ischemic attack are reported to have a lower incidence and prevalence of AF13 in some studies, but opposite findings were observed in others.14 Compliance with the monitoring strategy is the result of the interaction between patient characteristics, device characteristics, and study design and can be reduced by device‐caused discomfort and dependence on patient interaction to transmit the recordings. Wearable devices have shown lower levels of patient compliance as compared with implantable cardiac monitors, with a progressive decrease in compliance as the duration of monitoring increases.15 Monitoring strategies of comparable intrinsic diagnostic accuracy but associated with a higher compliance will be associated with a higher AF detection rate.

3.2. Device‐related factors

Studies on community screening,16 in patients with implanted devices,17, 18, 19, 20, 21, 22 and after ischemic stroke14, 23 have found that AF is mostly asymptomatic and would therefore remain undetected with monitoring strategies based on symptom‐triggered recordings. Monitoring strategies based on algorithm‐activated recordings have the potential to detect both symptomatic and asymptomatic AF. Algorithms based on a heart‐rate threshold have indeed shown a low diagnostic yield.24 Conversely, studies on algorithms that are independent of heart rate have shown a high sensitivity and specificity.25 Another device‐related factor influencing the AF detection rate after CS is the minimum detectable episode duration. This factor sets a lower limit to the minimal duration of an arrhythmic episode to be flagged as an episode of AF. In a given study population, devices with a shorter minimum detectable episode duration will result in a higher AF detection rate. This device‐related factor affects a study‐related factor, which is the study definition of AF (eg, qualifying episode of 30 seconds vs 6 minutes). The overall duration of cardiac monitoring, which depends both on the technical characteristics of the monitoring device and the study design, significantly affects the AF detection rate, with longer monitoring durations associated with higher AF detection rates. Monitoring strategies based on intermittent recordings have a lower diagnostic yield compared with strategies based on continuous monitoring, and this has been confirmed by simulation studies of snapshot recordings obtained with hand‐held devices such as smartphones,26 Holter monitoring, and up to 30 days of continuous monitoring.26, 27 This is not surprising, as many episodes of AF after CS are of short duration and the probability that intermittent recordings are obtained by chance during an episode of AF is low.

3.3. Study design–related factors

The definition of AF is one of the most important confounding factors in studies reporting AF detection rates as the result of a monitoring strategy. The definition of AF as an endpoint varies in different studies from an irregular supraventricular arrhythmia without detectable P waves of 2 or 3 seconds to several minutes. A definition of AF that includes shorter arrhythmic episodes will obviously result in a higher AF detection rate. External, independent adjudication of AF episodes may also have an impact on the AF detection rate, as the possibility exists that at least some of the episodes claimed to be AF will not be confirmed. Patient selection and the preliminary workup may have an impact on AF incidence and the AF detection rate. Inclusion of patients with stroke and past history of AF is an evident bias, but it is rarely observed. In several studies, patients were enrolled after ruling out AF only with a standard ECG, others with a 24‐hour Holter, and still others with bedside monitoring of variable duration during hospitalization. In a meta‐analysis, the admission ECG revealed previously undiagnosed AF in 7.7% (95% confidence interval [CI]: 0%‐10%) of cases; further in‐hospital monitoring with serial ECGs, continuous monitoring, and additional Holter or bedside monitoring for 24 hours allowed the detection of AF in 5.1% of previously undiagnosed patients (95% CI: 3.8%‐6.5%); and ambulatory monitoring after discharge allowed the detection of AF in an additional 10.7% of patients (95% CI: 5.6%‐17.2%).9 Therefore, a less aggressive monitoring strategy to rule out AF before enrolling patients will be associated with a higher AF detection rate during the study.

Transesophageal echocardiography (TEE), which is not included in the standardized assessment required to diagnose ESUS, and which is not always performed in studies on patients with CS, changes patient management in a significant proportion of patients. As several abnormal findings described with TEE in CS and ESUS patients may be linked to undetected AF (eg, atrial thrombi, cardiac tumors, other), it is possible that inclusion of TEE in the preliminary workup may influence the AF detection rate. The overall duration of cardiac monitoring, which depends on the technical characteristics of the monitoring device and on the study design, significantly affects the AF detection rate, with longer monitoring durations associated with higher AF detection rates.

With the aforementioned caveats in mind, Table 1 provides a summary of studies that have investigated the issue of AF detection in CS patients who received implantable cardiac monitors.14, 28, 29, 30, 31, 32, 33, 34, 35 Table 2 provides a summary of studies that have explored AF detection in CS patients using external cardiac monitors.36, 37, 38, 39, 40, 41, 42 Several key factors such as patient age, the duration of the minimum AF episode, and follow‐up duration are provided to help put these results into context.

Table 1.

Studies of AF detection using implantable cardiac monitors

Authors	Year	No. of Patients	Age, y	AF Definition	Monitoring Duration	AF Detection Yield, %
Cotter et al28	2013	51	51.5 ± 13.9	2 min	229 ± 116 d	25.5
Ritter et al29	2013	60	63 (48.5–72)	2 min	382 (89–670) d	16.7
Etgen et al30	2013	22	61.6	6 min	1 y	27.3
Rojo‐Martinez et al31	2013	101	67 ± 13	2 min	281 ± 212 d	33.7
Christensen et al32	2014	85	57.6	2 min	569 ± 310 d	16.1
Sanna et al14	2014	221	61.1 ± 11.4	2 min	6 mo	8.9
					12 mo	12.4
					36 mo	30.0
Poli et al33	2015	75	66.4 ± 12.5	2 min	6 mo	28.0
					12 mo	33.3
Ziegler et al34	2017	1247	65.3 ± 13.0	2 min	579 ± 222 d	21.5
Israel et al35	2017	123	65.0 ± 9.4	2 min	12.7 ± 5.5 mo	23.6

Abbreviations: AF, atrial fibrillation; IQR, interquartile range; SD, standard deviation.

Data are presented as mean ± SD or median (IQR) unless indicated otherwise.

Table 2.

Studies of AF detection using external cardiac monitors

Authors	Year	No. of Patients	Age, y	Monitoring Technology	Monitoring Duration	AF Definition	AF Detection Yield, %
Tayal et al36	2008	56	66 ± 11	MCOT	21 d	AF <30 s	18
						AF >30 s	5
Elijovich et al37	2009	20	68 ± 15	EM	30 d	N/A	20
Gaillard et al38	2010	98	63.6	TTM	30 d	32 s	9
Bhatt et al39	2011	62	61 ± 14	MCOT	28 d	30 s	24
						≥5 min	9
Flint et al40	2012	236	64.6 ± 13.8	MCOT	30 d	5–30 s	4
						>30 s	7
Kamel et al41	2013	20	65 ± 15	MCOT	21 d	30 s	0
Miller et al42	2013	156	68.5 ± 10.9	MCOT	30 d	<30 s	12
						>30 s	5
Gladstone et al15	2014	286	72.5 ± 8.5	EM	30 d	>30 s	16.1
						>2.5 min	9.9

Abbreviations: AF, atrial fibrillation; EM, event monitor; IQR, interquartile range; MCOT, mobile cardiac outpatient telemetry; N/A, not available; SD, standard deviation; TTM, transtelephonic monitor.

Data are presented as mean ± SD unless indicated otherwise.

4. CLINICAL IMPLICATIONS OF AF DETECTION AFTER CS AND ESUS

4.1. AF burden and risk of stroke

AF is a risk factor for ischemic stroke; however, observations made when the diagnosis of AF is based on standard ECG, which biases the selection of patients with longer paroxysmal episodes and persistent AF, may not be applicable to the so‐called device‐detected AF. The availability of advanced diagnostics on monitoring devices allows us to detect even short‐duration episodes of AF. The clinical significance of short‐duration AF episodes and whether the risk of TE events in patients with CS is a function of the overall time spent in AF or AF burden is unclear. A recent meta‐analysis has even demonstrated that frequent premature atrial contractions are associated with an increased risk of all‐cause mortality and stroke.43 However, without the benefit of continuous arrhythmia monitoring in these studies, it cannot be determined if the premature atrial contractions were the cause of these poor outcomes or if they were simply a marker for undetected AF.

Studies performed in unselected pacemaker or implantable cardioverter‐defibrillator (ICD) recipients attempted to find a correlation between AF burden and the risk of TE events. However, in these studies, patients with prior stroke were underrepresented, being only 0.2% to 32% of the study population. In these unselected populations of pacemaker or ICD recipients, episode duration and AF burden were correlated with TE risk. The threshold for a significant increase of TE risk varied from 5 minutes to 24 hours.17, 18, 19, 44, 45, 46, 47

It seems reasonable to speculate that the implications of AF burden may be different in patients with different baseline risks, in particular in patients with or without prior stroke. An interesting observation supporting the hypothesis of a multivariable model and the absence of a fixed threshold predictive of increased risk is suggested from a proof‐of‐concept study in which the risk of stroke or TE events was found to be not only a function of the time spent in AF, but also of the baseline characteristics of the patients.48 In this study, in patients with a high CHADS₂ score, even short episodes of AF significantly increased the rate of TE events; whereas in patients with a very low CHADS₂ score, even long episodes of AF did not. A summary of these studies that have explored the relationship between AF durations and stroke risk is provided in Table 3.

Table 3.

Studies of AF duration/burden and TE risk

Authors	Year	No. of Patients	Age, y	Follow‐up Duration	AF Threshold	TE Rate (Below Threshold), %	TE Rate (Above Threshold), %
Glotzer et al44	2003	312	74.0	27 mo	5 min	1.3	5.0
Capucci et al45	2005	725	71 ± 11	22 mo	24 h	HR: 3.1
Botto et al48	2009	568	70 ± 10	12 mo	AF burden + CHADS2 score	0.8	5.0
Glotzer et al19	2009	2486	70.9 ± 11.1	1.4 y	5.5 h	1.1	2.4
Shanmugam et al47	2012	560	66 ± 10	370 d	3.8 h	HR: 9.4
Healey et al17	2012	2580	76 ± 7	2.5 y	6 min	0.69	1.69
Boriani et al18	2014	10 016	70 (61–76)	24 mo	1 h	HR: 2.11
Van Gelder et al46	2017	2455	76 ± 7	2.5 y	24 h	HR: 3.24 vs no AF

Abbreviations: AF, atrial fibrillation; CHADS₂, congestive HF, HTN, age ≥ 75 y, DM, prior stroke/TIA/TE; DM, diabetes mellitus; HF, heart failure; HR, hazard ratio; HTN, hypertension; IQR, interquartile range; SD, standard deviation; TE, thromboembolism; TIA, transient ischemic attack.

Data are presented as mean ± SD or median (IQR) unless indicated otherwise.

Whether OAC offers any advantage over aspirin in unselected pacemaker or ICD recipients with a burden of AF <24 hours is currently being investigated by 2 studies, Apixaban for the Reduction of Thromboembolism in Patients With Device‐Detected Subclinical Atrial Fibrillation (ARTESiA; NCT01938248) and Non–Vitamin K Antagonist Oral Anticoagulants in Patients With Atrial High Rate Episodes (NOAH; NCT02618577), but the applicability of these findings to the CS or ESUS population is questionable.

4.2. Temporal relationship between AF and stroke

Although current evidence supports the correlation between AF burden and risk of stroke, the mechanisms underlying this correlation are unclear. In post‐hoc analyses of the Prospective Study of the Clinical Significance of Atrial Arrhythmias Detected by Implanted Device Diagnostics (TRENDS),20 Asymptomatic Atrial Fibrillation and Stroke Evaluation in Pacemaker Patients and the Atrial Fibrillation Reduction Atrial Pacing Trial (ASSERT),49 and Combined Use of BIOTRONIK Home Monitoring and Predefined Anticoagulation to Reduce Stroke Risk (IMPACT)50 studies, AF was found in the 30 days before TE events only in 27%, 8%, and 6% of cases, respectively. However, it is not surprising that only a minority of cases of TE were retrospectively found to be preceded by AF, as in these unselected patients any other competitive cause could have been responsible for the observed events: stenoses of the arteries supplying the brain, aortic plaque ulcers, PFO, cardiac structural abnormalities, and any other less common etiology. The issue of the temporal relationship between AF episodes was more recently addressed with a novel approach in a larger study of pacemaker and ICD recipients. In this study, patients with AF episodes >5.5 hours showed a marked increase in the odds ratio of TE events in the first 5 days after the episode, which gradually returned to baseline after 3 to 4 weeks, thus suggesting the existence of a time dependency of risk.51 In the attempt to reconcile the findings on AF burden and the temporal relationship between AF and TE risk, it seems reasonable to make the hypothesis that in patients with device‐detected AF, the risk of TE is a function of both the AF burden and the clinical characteristics of the patient and that this risk is highest immediately after the episode and declines to baseline in 3 or 4 weeks.

4.3. AF and pathophysiology of stroke

The association between AF and stroke has been consistently demonstrated; however, it is not clear whether the classical sequence of auricular blood stasis, coagulation cascade activation, thrombus formation, and embolization is the only pathophysiological model to explain this association. Although this model can be easily invoked for long episodes of paroxysmal AF, persistent AF, or permanent AF, it seems less convincing when attempting to explain the association of stroke with just a few minutes of AF. An updated pathophysiologic model suggests that, at least in some cases, AF might be the marker of a disease of the atrial wall, causing, beyond an altered electrical activity, a dysfunction of the atrial cells, ultimately triggering platelet activation, coagulation cascade activation, and thrombus formation.

5. ANTICOAGULATION AFTER CS OR ESUS

In most cases of CS and ESUS, antiplatelet agents are recommended as the treatment of choice when there is no evidence of AF, whereas anticoagulants are recommended once AF is detected. Empirical anticoagulation with direct anticoagulants (ie, without pursuing the detection of AF) has been proposed as an alternative approach, and several randomized trials are currently investigating this study hypothesis (ClinicalTrials.gov NCT02239120, NCT02313909, and NCT02427126). However, the randomized Warfarin–Aspirin Recurrent Stroke Study (WARSS) failed to find an advantage of warfarin over aspirin in patients with ischemic and cryptogenic stroke.52 Of interest, in the WARSS no significant difference was found among treatments in the incidence in the composite primary endpoint of death or recurrent stroke, nor in the incidence of haemorrhage‐related death. This results are possibly explained by a lack of superiority, rather than by an excess of harm, of warfarin over aspirin in the prevention of recurrent stroke.

Occult arterial sources of arterial embolism are aortic plaques and noncritical stenoses of intra‐ and extracranial arteries supplying the brain. Several comparisons of antiplatelets vs OAC in both primary and secondary prevention of TE events in the setting of atherosclerotic lesions of the aorta have been published, but most of them were nonrandomized and had a limited power. A number of studies suggested a potential benefit of OAC in the presence of atherosclerotic lesions of the aorta, whereas others failed to confirm these findings. In the absence of a clear advantage of OAC over antiplatelets, current guidelines suggest that the choice between the alternative strategies is made on other indications for these treatments.53 Noncritical stenoses of intra‐ and extracranial arteries supplying the brain are often found in patients with CS or ESUS. On the basis of current evidence, current guidelines recommend that these patients receive antiplatelet therapy, statins, and risk‐factor modification, independent of an indication for surgical or interventional treatments.54 PFO is another common finding in patients with CS. Current practice advisory suggests that, in the absence of another indication for anticoagulation, clinicians prescribe antiplatelet medications to patients with CS and PFO.55

In summary, among the causes of CS and ESUS, AF is the only disease in which the benefits of OAC clearly outweigh the risks in most patients. Conversely, the efficacy of OAC in patients with aortic arch ulcers and atherosclerosis of the arteries supplying the brain is still debated. Pathophysiology of thrombus formation during AF and thrombus formation at the site of an arterial plaque are different, and their response to anticoagulants or antiplatelets may be different, accordingly. Analysis of thrombus composition in patients with CS suggested a composition more similar to cardioembolic rather than to noncardioembolic stroke, but the overlap between thrombus phenotypes in different etiologies makes this information unsuitable for clinical decisions at single‐patient level.56 Finally, the relative risk of bleeding in patients with CS on direct anticoagulants as compared with aspirin is unclear. In the Apixaban Versus Acetylsalicylic Acid to Prevent Stroke in Atrial Fibrillation Patients Who Have Failed or Are Unsuitable for Vitamin K Antagonist Treatment (AVERROES) study, in >5000 subjects with AF and a contraindication to vitamin K antagonists randomized to apixaban or aspirin, the rates of major bleeding and major or clinically relevant bleeding were not significantly different between groups.57 However, in the prespecified group of patients with prior stroke or transient ischemic attack, there was a trend for an increase in major or clinically relevant bleedings (hazard ratio: 1.49, 95% CI: 0.95‐2.25, P = 0.08)57; and whereas in this prespecified group the major bleeding rate in patients randomized to aspirin was apparently not significantly different from patients randomized to apixaban (2.89 [1.42–5.90] vs 4.10 [2.36–7.10], P = 0.73), the power of the study to detect a difference in bleeding rates was low.

In conclusion, in patients with ESUS but without evidence of AF, empirical anticoagulation is an intriguing approach to treatment, but it is currently unclear if it is more effective or safer than aspirin.

6. CONCLUSION

CS and ESUS represent a major challenge to healthcare systems worldwide. AF is commonly found after CS or ESUS. The detection of AF may be pursued with different monitoring strategies. Comparison of monitoring strategies should take into account that AF detection rates reported in published studies, and then pooled in meta‐analyses, are not only a function of the monitoring strategy itself, but also depend on patient‐related (with older age being the single most influential factor for incident AF in the study population), device‐related, and study design–related factors. Once AF is found, the decision to anticoagulate a patient should be made on the basis of AF burden and the baseline risk of the patient. Empirical anticoagulation in patients with ESUS and no evidence of AF is currently an investigational strategy and should not be adopted outside randomized clinical trials.

Conflicts of interest

Tommaso Sanna is a consultant for Medtronic and has received speaker fees from Medtronic, Boehringer Ingelheim, and Pfizer/Bristol‐Myers Squibb. Paul D. Ziegler is an employee of Medtronic Diagnostics Research and a Medtronic shareholder. The authors declare no other potential conflicts of interest.

Sanna T, Ziegler PD, Crea F. Detection and management of atrial fibrillation after cryptogenic stroke or embolic stroke of undetermined source. Clin Cardiol. 2018;41:426–432. 10.1002/clc.22876