Noninvasive Cardiac Monitoring for Detecting Paroxysmal Atrial Fibrillation or Flutter After Acute Ischemic Stroke

Atrial fibrillation/flutter (AF), a strong risk factor for stroke, is arguably the most important finding on cardiac workup in patients with ischemic stroke.^1–6 Once identified, introduction of oral anticoagulant therapy (eg, warfarin; international normalized ratio 2 to 3) provides an additional 40% risk reduction in recurrent stroke compared with antiplatelet therapy.^7–10 Furthermore, recent evidence suggests that therapeutic oral anticoagulation (international normalized ratio 2 to 3) may also be associated with reduced stroke severity, if ischemic stroke does occur in patients with AF.^11,12 Given that ischemic stroke with AF is associated with greater disability and mortality than those without AF, establishing the presence of underlying AF is of clear clinical importance.^3,13–15

In addition to baseline electrocardiogram and clinical examination, noninvasive cardiac monitoring is used to detect AF.^16–19 A previous review on this topic reported that 24- to 48-hour Holter monitoring identified AF in 1% to 5% of patients undetected by initial electrocardiogram.²⁰ Currently, there is no clear recommendation on whether routine Holter monitoring is indicated in unselected patients with acute ischemic stroke or transient ischemic attack.²⁰ Furthermore, many of the studies included in that review were conducted at a time when the clinical significance of paroxysmal AF was underappreciated.^21–25 It is now acknowledged that paroxysmal AF is associated with a comparable increase in risk of ischemic stroke as continuous AF.⁶ Moreover, since the previous review of this topic, a number of studies evaluating the use of Holter monitoring, and newer methods that allow more prolonged periods of monitoring, have been published. We performed a systematic review to determine the frequency of occult AF detected by noninvasive methods of continuous cardiac rhythm monitoring in consecutive patients with ischemic stroke.

Methods

Data Sources

Studies were identified from the PubMed and EMBASE databases between 1966 and May 2006 by crossreferencing the following MeSH terms: “monitoring, physiological” and “electrocardiography, ambulatory” with “atrial fibrillation/diagnosis,” “arrhythmia” as well as with “stroke,” “brain ischemia,” “ischemic attack, transient,” and “cerebrovascular accident.” The free text terms “cardiac event recorder” and “telemetry” were used in all searches. After finding all potentially relevant articles, reference lists were examined as well as “relevant articles” links. Science Citation Index was used to identify additional studies. Two investigators independently conducted searches. Articles were limited to those in the English language. A research librarian assisted in our search strategy.

Study Selection

Two investigators independently reviewed all studies identified on the search. For inclusion, studies were required to fulfill the following inclusion criteria: (1) study design: (a) randomized, controlled trials in which patients are randomized to receive noninvasive cardiac monitoring compared with control or (b) prospective cohort studies; (2) population: consecutive patients with a diagnosis of acute ischemic stroke or transient ischemic stroke; (3) intervention: participants underwent noninvasive rhythm monitoring to detect AF for a minimum duration of 12 hours; and (4) outcome: frequency of AF detected by noninvasive monitoring was reported.

Data Abstraction

Data abstraction was conducted independently by two investigators (J.L., Z.K.) using standardized data abstraction forms developed before searches were conducted.

Statistical Analysis

The study-specific effect estimates were combined using a precision-weighted (ie, reciprocal of the variance), random-effects model and 95% CIs were calculated. Zelen’s exact test was used to determine whether there was statistical heterogeneity between individual studies. A 95% CI that did not include one was considered to be statistically significant. Minitab 14.3 statistical software was used for the analyses. The following factors were hypothesized, a priori, to influence the detection rate of AF: duration and timing of monitoring, stroke subtype and severity, and setting.

Results

Study Identification and Selection

The initial search identified 387 studies, reduced to 60 potentially eligible studies that fulfilled our sorting criteria. After application of eligibility criteria, 5 prospective cohort studies (736 participants) were eligible^24,26–29 (Figure). No randomized, controlled studies were identified. Tables 1 and 2 describe the population, sample size and intervention in each of the individual studies.

• Download figure
• Download PowerPoint

Frequency of Atrial Fibrillation and Atrial Flutter Detected by Routine Cardiac Monitoring

Five studies (588 participants) evaluated Holter monitoring in the inpatient setting.^24,26–29 In these studies, rates of detection of new AF ranged from 3.8% to 6.1% (Table 3). When the results of these studies were combined in a random-effects model, new AF was detected in 4.6% (95% CI: 0% to 12.7%) of consecutive patients with ischemic stroke independent of baseline electrocardiogram and clinical examination (test for heterogeneity P=0.9). Two studies (140 participants) evaluated event loop recorders after Holter monitoring.^26,28 New AF was detected in 5.7% and 7.7% of consecutive patients in these 2 studies. One study (159 participants) evaluated continuous cardiac monitoring using inpatient telemetry (48 hours) and detected new AF in 2.5% of participants²⁴ (Table 3).

Stroke Subtype, Stroke Severity, and Setting

Two studies reported the frequency of AF within ischemic stroke subtypes.^26,28 In the largest of these studies (n=149), most cases of AF were detected in patients with total and partial anterior circulation ischemic strokes (68%), whereas no cases of AF were identified in patients with lacunar stroke.²⁸ In the other study, no significant association between AF and stroke subtype was reported.²⁶

Two studies reported stroke severity in participants with and without AF.^27,28 Jabaudon et al²⁸ reported more severe neurological deficit (National Institutes of Health Stroke Scale ≥10) in patients with AF (22.7%) compared with patients without AF (3.1%). (P=0.003)²⁷ Hornig et al²⁵ did not report an association between stroke severity and AF. No eligible studies included nonhospitalized patients with ischemic stroke only.^24,26–29

Duration and Timing of Monitoring

All studies reported the duration of monitoring.^24,26–28 Duration of monitoring ranged from 21 to 159 hours. Studies evaluating event loop recorder monitors reported a detection rate of 5.7% after 70 hours and 7.7% after 159 hours of monitoring.

Three studies reported on the timing of initiating of monitoring from the diagnosis of stroke.^26,28,29 One study initiated monitoring 55 days (median delay) postevent.²⁸ Two studies reported that monitoring began on admission to the ward; however, the timing from onset of stroke was unclear.^26,28 Two studies did not report the timing of initiating of monitoring.^24,27

Detection of Atrial Fibrillation Resulting in a Change in Management

One study reported the proportion of patients in which the results of noninvasive monitoring resulted in a change in antithrombotic therapy.²⁸ In that study, oral anticoagulation was started in 28.6% (2 of 7) of patients with new-onset AF detected by Holter monitoring and in all 5 patients with AF detected by event loop recorder.

Discussion

Our review suggests that routine Holter monitoring in consecutive hospitalized patients with ischemic stroke detects AF in approximately one in 20 patients, beyond that detected by physical examination and initial electrocardiogram. The range of detection rates (3.8% to 6.1%) is slightly more than that reported in an earlier review by Bell et al (1% to 5%).²⁰ Our study differs from that review in a number of respects. First, we only included prospective studies. Second, most studies in the previous review were older and conducted (>20 years ago) when the clinical importance of paroxysmal AF was not appreciated, which may have influenced whether paroxysmal AF was reported.^21–25 Third, 2 studies included in our analysis were published subsequent to the review by Bell et al.^26,28

Very limited data suggest that confining Holter monitoring to patients with nonlacunar stroke may improve its diagnostic use. In the study by Jabaudon et al,²⁸ no cases of AF were reported in patients with lacunar stroke. A potential limitation of the Jabaudon study²⁸ is the unblinded evaluation of stroke subtype, which may have been biased by knowledge of the results of cardiac monitoring. One retrospective study, not included in this review, reported that Holter monitoring identified the greatest yield of AF in patients with unexplained embolic stroke.³⁰ This is further supported by the results by Schuchert et al,²⁹ which found a relatively higher detection rate of 6.1% in patients with suspected cardioembolic ischemic stroke. Furthermore, there are 2 observational studies that suggest that oral anticoagulation may not be associated with the same benefit for prevention of recurrent stroke in patients with AF presenting with lacunar stroke compared with those presenting with other ischemic stroke subtypes.^11,31 Although based on very limited data, these observations suggest that limiting Holter monitoring to patients with nonlacunar ischemic stroke might be reasonable, but further research is required before definitive recommendations can be made.

Increased duration of monitoring appears to be associated with increased rates of detection of AF (Table 3). In the 2 studies that evaluated event loop recorders (140 participants), new AF was detected in 5.7% and 7.7% of consecutive patients in these 2 studies.^26,29 Our review is unable to determine the optimal duration of monitoring. Similarly, the best time to initiate cardiac monitoring after a stroke is uncertain. No study explicitly evaluated and reported on early monitoring (within 48 hours) from onset of stroke symptoms.

Our review has several limitations. Eligible articles were restricted to published studies in English language journals, and conference abstracts were not accessed. Thus, unpublished studies may have been overlooked. Furthermore, because only hospitalized patients were included in eligible studies, we are unable to comment on the use of Holter monitoring in the outpatient stroke population. Our strict inclusion criteria resulted in fewer numbers of eligible articles compared with a previous review. Furthermore, sample sizes of studies included were relatively small. As a result, the confidence interval about the summary estimate is wide (95% CI: 0% to 12.7%), a clear limitation of this review. However, we believe that restricting eligibility to prospective studies is an overall strength of our review.

Do the results of this review have implications for clinical practice? Should all patients with ischemic stroke be screened with Holter monitoring? If we apply the general Principles of Good Screening, we observe that Holter monitoring fulfills most criteria (Table 4).^32–36 However, in unselected patients, it remains open to argument whether AF is insufficiently prevalent to justify routine screening.

In conclusion, screening consecutive patients with ischemic stroke with routine Holter monitoring will identify new AF in approximately one in 20 patients. Extended duration of monitoring and confining its use to patients with nonlacunar stroke may improve detection rates. However, further research is needed to evaluate extended monitoring, using newer techniques, in patients with ischemic stroke.

Footnotes