Prolonged Rhythm Monitoring for the Detection of Occult Paroxysmal Atrial Fibrillation in Ischemic Stroke of Unknown Cause

The ambulatory ECG (Holter) continuously records an ECG signal for 24 to 48 hours, although at some centers recordings of up to 7 days are available.²⁸ It is battery operated and records 2 to 3 different leads. Versions that permit recording up to 12 leads are available. Although more cumbersome, these permit characterization of premature ventricular complex morphology in patients who are potential ablation candidates; the benefit in atrial arrhythmia patients is less clear. Modern devices record data onto electronic media that are then analyzed with the help of software to identify rhythm abnormalities. Many recorders include patient-activated event markers as well as time markers to allow an increased correlation between symptoms and rhythm abnormalities.²⁹ Tracings are overread by cardiologists, and Holter recordings are thus considered the gold standard. The greatest limitation is the recording period, typically 1 to 2 days.

In contrast, event recorders are used for prolonged time periods (4 to 6 weeks), with episodic recordings triggered by the patient. These devices may be worn continuously or applied only at the time of recording. Continuously worn devices record the ECG into a memory buffer loop, which is “frozen” when activated by the patient and permits capture of the ECG for several minutes before and after activation. Alternatively, a credit card–sized monitor that is applied to the chest wall at the time of symptoms of arrhythmia is also available. These generally require a symptomatic arrhythmia lasting at least 3 minutes to allow time for application of the device to the chest. Once a rhythm is recorded, the monitors transmit recordings by telephone by converting ECG data to audio signals. The audio signals are received at a central station that reconstructs the electric signal into a conventional ECG recording for interpretation.²⁹ Newer forms of transmission (wireless, Internet based) are emerging. Event monitoring necessarily only captures symptomatic events, unless patients are instructed to transmit periodic rhythm samples. Patient compliance is required. Event recorders are useful for symptomatic patients and typically tolerated up to 4 weeks in motivated patients. Rhythm strips are overread by technicians and/or caregivers.

Mobile cardiac outpatient telemetry systems provide home-based, nearly continuous real-time monitoring of symptomatic and asymptomatic arrhythmias. Patients wear a small transmitting pendant that is connected to 3 ECG electrodes. The pendant transmits real-time ECG data to a cell phone–sized monitor that analyzes the rhythm. Any detected arrhythmias are transmitted via the cellular network to a laboratory where technicians read the episodes. Arrhythmias trigger immediate alerts on the basis of predetermined physician-set monitoring thresholds and response parameters. Patients may also activate the monitor to report symptoms.³⁰ If a patient travels outside of the cell network, data are stored and transmitted when back in range. Monitoring is performed for 7 to 21 days, and patients remove and apply electrodes themselves. Newer devices composed of a “band aid” that records and transmits the ECG to a cell phone, which then analyzes and transmits the data to a back-end processor for human and/or computer processing, are currently under development and in early human trials.

As monitoring duration is lengthened, the burden of data for review increases, and automated algorithms to identify arrhythmias are increasingly deployed. Limitations of analysis of skin (and subcutaneous) recordings include the relatively small amplitude of atrial signals compared with ventricular signals and false-positives generated by myopotential and motion artifact. Published independent validation of many of these systems is limited.

The CardioNet mobile cardiac outpatient telemetry system (Conshohocken, PA) uses nonlinear statistics in a proprietary algorithm that analyzes R-R interval variability, QRS morphology, and P-wave presence to screen for AF.³¹ When tested against a standardized database of recorded and cataloged rhythms, the episode sensitivity and positive predictive values are 81% and 88%, respectively; with episodes of at least 30 seconds, the reported sensitivity and positive predictive values are 100% and 100% (Anna McNamara, Cardionet, written communication). In all noninvasive systems, real-world limitations include loss of signal because of inappropriate system use, myopotentials, artifact, and noncompliance. Thus, recording times are significantly extended beyond those available with standard Holter recordings but at the cost of noncontinuous recordings and loss of summary statistics (eg, premature ventricular contraction count). The rate of false-positives reported to clinicians is low because tracings are overread by technicians who prepare the reports that are then sent to the treating caregiver; the rate of false-negatives in clinical use is not currently known.

It is well established that the longer a patient is monitored, the greater is the likelihood of detecting symptomatic and asymptomatic AF.^28,32,33 Implantable loop recorders are small (typically 6 cm by 2 cm by 0.8 cm) devices implanted subcutaneously to overcome the limitations of skin irritation and patient compliance to allow very prolonged recording periods. They automatically record tachyarrhythmias and bradyarrhythmias with physician-programmable parameters or when a patient triggers a recording with a wireless activator. A dedicated AF detection algorithm analyzes the variability in the difference between consecutive R-R intervals (ΔR-R[i] versus ΔR-R[i−1]).³⁴ It analyzes irregularity to distinguish sinus rhythm (minor variations related to ectopy and sinus arrhythmia) from AF (uncorrelated irregularity). The duration and number of recordings are programmable, with current devices capable of recording up to 50 minutes of ECG. Data can be transmitted over the telephone or wirelessly with appropriate equipment to the Internet for Web-based physician review. The implant may be left in place up to 3 years and can be explanted once a diagnosis is made or the battery life has ended.²⁹ Automatic triggers and the lack of external electrodes minimize the need for patient compliance to capture an event. In a study in which 247 patients received an implantable loop recorder (Reveal XT, Medtronic, Minneapolis, MN) with an automated algorithm and underwent simultaneous Holter recording for validation, the sensitivity, specificity, positive predictive value, and negative predictive value for identifying patients with any AF were 96.1%, 85.4%, 79.3%, and 97.4%, respectively.³⁵ False-positives were recorded in 14.5% of patients because of premature atrial or ventricular complexes, myopotential oversensing, sinus arrhythmia, or undersensing of R waves, highlighting the importance of human overreading of automatically detected episodes. Episodes had to be at least 2 minutes in duration to permit detection, resulting in missed AF in 3.9% of patients. Considering episodes of longer duration (>6 minutes as opposed to 2 minutes) increases the positive predictive value from 73% to 80%.³⁴

Dual-chamber pacemakers and implantable cardioverter-defibrillators include an atrial lead that permits direct recording of atrial electrograms (Figure 3). If the atrial rate exceeds a programmable value for a defined number of complexes, an atrial tachyarrhythmia is declared. Thus, in contrast to surface and subcutaneous systems, dual-chamber cardiac implantable electronic devices detect atrial tachyarrhythmias with a regular ventricular response, even if the ventricular rate is in the normal range. Importantly, single-chamber pacemakers generally do not detect atrial arrhythmias, and single-chamber implantable cardioverter-defibrillators only detect them if the ventricular rate exceeds the ventricular tachycardia detection rate. At that point, R-R interval variability is assessed. Thus, for reliable detection of atrial arrhythmias, dual-chamber systems are used.

Current-generation devices automatically record intracardiac electrograms for later review when arrhythmias are detected. The sensitivity for detection ranges from 57% to 98.1%, and the specificity ranges from 85.4% to 100%.^35–37 Many of the arrhythmias may not be true AF but rather atrial tachycardias or atrial flutters with rates that exceed the programmable recording threshold. These are therefore referred to as atrial high-rate episodes. Manual overreading of recorded episodes is important because far-field R wave oversensing (Figure 4) and, less commonly, lead fracture or external electromagnetic signals may also trigger event capture. Because of intermittent undersensing, a single prolonged episode may be recorded as multiple shorter episodes so that the overall arrhythmia burden may be more reliable than number of episodes.³⁸ Cardiac implantable electronic device batteries last 5 to 12 years, depending on the device type and therapy delivered.

To determine the incidence of AF in this population, we considered the number of new cases of AF detected with the use of cardiac monitoring devices. Patients with AF diagnosed before their hospitalization or during routine inpatient screening were excluded from this analysis. Relevant data were abstracted from these studies by 2 investigators (R.C.S.S. and A.A.R.) and are summarized inTables 2 and 3. A total of 3039 subjects (mean age, 66 years; 62% male gender) were included in this analysis. The average incidence of new AF in these studies was 6.3%, the highest incidence was in patients older than 60 years evaluated with continuous inpatient ECG and 24-hour Holter monitoring (21.3%),⁵⁴ and the lowest incidence was detected in younger cryptogenic stroke patients monitored with an implantable loop recorder.⁴⁵

To determine the impact of patient selection on the detection yield of AF, we divided studies into 2 groups: an unselected population of stroke/TIA patients (11 studies;Table 2) and a population selected on the basis of age, stroke etiology, and prescreening for cardiac arrhythmias (8 studies;Table 3). The average incidence of new diagnosis of AF in an unselected population was 5.3%, whereas the average incidence in the selected population was 10.9%. When selection criteria were applied, the mean detection yield of AF with ambulatory ECG was 6.4% (range, 5.3% to 9.0%); with combined use of continuous inpatient and ambulatory ECG, it was 21.3%; and with automatic event recorder, it was 16.7% (range, 14.3% to 20.0%). Despite a longer period of monitoring (14.5 months), the implantable loop recorder did not detect significant cardiac abnormalities in younger stroke patients with fewer AF risk factors.⁴⁵

Several studies compared the incidence of AF according to stroke subtypes, locations, and severity. A higher incidence of AF was observed in patients with cryptogenic stroke compared with those with large atherothrombotic and lacunar strokes in one study⁴¹ but not in another.⁵⁸ Subsequent studies restricted to patients with cryptogenic stroke found an incidence of AF ranging between 14.3% and 27.3%.^50,57,58 In younger patients with cryptogenic stroke, however, the use of implantable loop recorders was not able to detect new cases of AF.⁴⁵ Among patients with suspected embolic strokes, the incidence of AF was 6% in a study that used a 72-hour ambulatory ECG monitoring device.⁴⁴ The wide range of reported incidences in these patients may arise from uncertainties concerning the definition of cryptogenic stroke. The term cryptogenic stroke usually refers to strokes with no clearly definable cause despite extensive workup.⁶⁰ The yield from these investigations to elucidate the etiology of stroke may vary depending on the practice of centers managing stroke patients, timing of investigations (because some causes may be reversible, such as cardiac thrombus and vegetations), and lack of expertise or manpower to routinely perform more specialized investigations such as transesophageal echocardiogram and transcranial Doppler study. In the Cryptogenic Stroke and Underlying Atrial Fibrillation (CRYSTAL AF) study, a randomized prospective study that compares standard arrhythmia monitoring (control arm) and implantation of a subcutaneous cardiac monitor (Reveal XT; Medtronic, Inc, Minneapolis, MN) (continuous monitoring arm) for AF detection, a stroke/TIA is considered to be cryptogenic if no possible cause can be determined according to the standard protocol of the participating center.⁶¹

Because an embolus from the heart may preferentially lodge in the anterior circulation or break up and lead to multiple ischemic foci, studies assessing infarcted territory and its association with AF have been performed. In 1 study, the incidence of AF was higher in patients with anterior circulation territory infarctions than in those with lacunar stroke (68% versus 0%).⁵³ Patients with AF had severe neurological deficits (National Institutes of Health Stroke Scale >10) more often than those without AF (n=5, 22.7% versus n=4, 3.1%; P=0.003).⁵³ In other studies, brain imaging parameters such as infarct size,⁵⁰ number of infarcts,⁴⁸ and stroke locations^55,59 were identified as significant predictors of AF detection after acute stroke. A combination of clinical and radiological features (comprising embolic infarct pattern, age >65 years, and preexistent coronary artery disease) has also been associated with a higher incidence of AF.⁵⁹

Limited data are available on the optimal time and duration of monitoring. Seven studies reported the interval between stroke onset and the timing of cardiac monitoring, which ranged from 72 hours to 3 months after the stroke.^{44–46,50,55,56,58} The duration of cardiac monitoring was reported in 13 studies and ranged from 22 hours to 14 months.^{41,42,44,45,47–49,53–59} Regardless of the detection method used, studies evaluating earlier monitoring reported a higher incidence of AF than those in which monitoring was pursued later.⁵⁴ Similarly, studies that monitored patients for a longer period identified a higher incidence of AF.^{50,52,55,57,58}

Concerns have been raised about the low agreement between different cardiac monitoring methods to detect AF. One study suggested superiority of ambulatory ECG monitoring over continuous inpatient ECG,⁵⁹ whereas another claimed superiority of continuous inpatient ECG over ambulatory ECG monitoring.⁵⁴ In certain studies, ambulatory ECG actually failed to corroborate the detection of AF on 12-lead and serial ECGs.^41,43 These differences highlight the practical difficulties faced by investigators and neurologists to capture the intermittent and periodic bursts of AF, which may vary depending on the device used and the timing of cardiac monitoring.