Cryptogenic Stroke and Underlying Atrial Fibrillation: Reducing Risk with ECG

Related Articles: Diagnostic ECG

Atrial fibrillation is an underlying cause of major morbidity and mortality across the globe. According to a review article in Stroke, more than five million people in the United States were estimated to have Afib in 2010, and this number exceeded eight million in the European Union. The prevalence of Afib is expected to surpass twelve million in the US by 2030 and reach 17.9 million in the EU by 2060 due to aging populations. This anticipated growth in Afib prevalence lends even greater urgency to early ECG detection in preventing complications such as cryptogenic stroke.

Cryptogenic Stroke and Underlying Atrial Fibrillation

As the Stroke review article notes, the presence of Afib quintuples the risk of stroke. A review article in the Journal of Atrial Fibrillation states that Afib is a factor in at least 15% of all strokes. Particularly insidious is the occurrence of asymptomatic Afib, which an evaluation in American Family Physician found to affect as many as 59% of participants.

The American Heart Association (AHA), American College of Cardiology (ACC), and Heart Rhythm Society's (HRS) Guideline for the Management of Patients with Atrial Fibrillation states that "Clinically unrecognized and asymptomatic AF is a potentially important cause of stroke, supporting efforts for early detection of AF in at-risk individuals." As a major cause of stroke, asymptomatic or previously undiagnosed Afib can lead to significant—and potentially preventable—morbidity, disability, and even death.

Cryptogenic stroke refers to cerebral infarction due to an obscure or idiopathic mechanism. According to a Frontiers in Physiology review article, as many as 25% to 40% of ischemic strokes are cryptogenic, and the etiology for these cryptogenic strokes is often never identified. However, additional extended ECG monitoring may be able to identify paroxysmal Afib in a subset of cryptogenic strokes.

ECG Screening for Afib

The AHA/ACC/HRS guidelines state that "Prolonged or frequent monitoring may be necessary to reveal episodes of asymptomatic Afib," and the ACC recommends considering prolonged cardiac monitoring beyond 24-hour telemetry as part of the cardiac assessment for cryptogenic stroke.

A Cochrane Review concluded that both systematic and opportunistic screening for Afib increase detection rates of asymptomatic Afib, based on existing evidence. Ambulatory ECG monitoring is the most common modality for asymptomatic Afib detection and screening, particularly following stroke or transient ischemic attacks, and various methods are currently available. These include Holter monitors, event recorders (external as well as implantable), ECG patch recorders, and most recently, three-lead and single-lead ECG monitoring.

A feasibility study in the International Journal of Cardiology evaluated the use of three-lead ECG monitoring as a screening tool for Afib and found that the three-lead technology was feasible and effective. Another study, in Europace, looked at the yield of Afib screening with a handheld, single-lead ECG monitor and found use of the single-lead device to be feasible and high-yield, resulting in a single-day yield of 1.1% new cases of Afib among a primary care population presenting for routine influenza vaccination.

The Promise of AI

Ever more promising opportunities for reducing risk of cryptogenic stroke due to asymptomatic Afib are being realized through advances in artificial intelligence (AI) for cardiac care. As this technology evolves, large data sets are necessary to obtain enough information for thorough evaluation.

A study in the Lancet assessed the ability of a newly developed AI-enabled ECG algorithm to identify patients with Afib using information gathered in their ECG during normal sinus rhythm. The study, which ran from 1993 to 2017 and included over 180,000 patients, collected nearly 650,000 normal sinus ECGs for analysis. Each ECG was obtained using a standard 10-second, 12-lead GE ECG with each patient in supine position. Many patients had multiple ECGs recorded over the study inclusion period, but the researchers only used those collected during a short, specific window of interest.

Following data collection, each ECG was initially read by the GE ECG and the interpretation was verified or corrected by a physician-supervised, trained ECG technician. Patients with at least one ECG showing Afib or atrial flutter were classified as positive for Afib.

Then, each ECG was randomly assigned to one of three groups:

  • Training: 70% of the ECGs were assigned to this group and used to train the AI-enabled algorithm.
  • Internal validation: This group included 10% of collected ECGs and was used to optimize the network and select network hyperparameters.
  • Testing data set: 20% of ECGs were used to assess the AI-enabled algorithm for its ability to detect a history of Afib.

The authors noted that 8.4% of patients in the testing data set had verified Afib before the collection of the normal sinus rhythm ECG. When analyzing the collected ECGs, the AI-enabled algorithm identified patients who had prior episodes of Afib with an overall accuracy of 83.3%. However, this study represents only one type of algorithm on the market today.

AI-enabled ECG software can improve rhythm diagnoses in a multitude of ways, offering automatic detection of Afib episodes to help prevent cryptogenic strokes. Additionally, the enhanced ability to detect occult Afib can help reduce the development of tachycardia-induced cardiomyopathy, which, as the AHA/ACC/HRS guidelines observe, can otherwise occur in untreated patients who have Afib with rapid ventricular response.

A recent review in Circulation Research suggests that larger health systems may use AI-enabled ECG algorithms as a low-cost, mass screening tool for Afib and other cardiovascular issues. And, as the authors of a review in Mayo Clinic Proceedings note, ECG technology is becoming cheaper and more accessible to the general population. AI could eventually be used to interpret ECG data collected from personal devices, predicting rhythm disorders like Afib and triaging patients to appropriate medical personnel.

While AI-enabled algorithms may be used to enhance current diagnostic methods, it is unlikely they will replace standard practices in the near-term. As the technology develops, external automatic monitoring of ECG through the use of AI will enhance cardiologists' ability to detect and treat asymptomatic Afib, improving patient outcomes related to cryptogenic stroke and underlying atrial fibrillation. Ultimately, AI will increase efficiency and effectiveness in clinical practice and enable the delivery of a more advanced level of personalized medicine.