Atrial fibrillation (AF) is known to affect at least 30 million people worldwide [Citation1,Citation2], although this may be an underestimation. AF can be asymptomatic and fleeting and often goes undetected. In fact, it has been estimated that approximately one million Americans live with unrecognized AF [Citation3]. The proportion of patients with paroxysmal AF versus persistent AF varies with age (with paroxysmal AF more common in patients <50 years), and it is estimated that about 25% of patients with AF have a paroxysmal pattern [Citation4]. Identifying patients with undiagnosed AF is important as they have a fivefold increased risk of stroke [Citation1,Citation2] and the first manifestation of AF may be a disabling stroke. Furthermore, AF-related strokes carry a particularly poor prognosis [Citation3,Citation5]. When AF is recognized, interventions including oral anticoagulation or left atrial appendage closure can lower stroke risk and mortality [Citation5,Citation6].
Due to its frequently paroxysmal nature, AF is often under detected. Currently, prolonged electrocardiographic monitoring is implemented to detect patients with suspected AF – a process that is expensive, resource intensive, and at times poorly tolerated. In nearly 5,000 patients referred for continuous 24-hour monitoring, the prevalence of paroxysmal AF was 2.5% [Citation7]. It has been estimated that even among a high-risk cohort of patients with ischemic strokes, 20% remain cryptogenic despite thorough diagnostic evaluation [Citation5]. Apart from the low yield, long-term cardiac monitoring is resource intensive, expensive, and impractical for broad-scale application.
A frequent clinical dilemma is whether or not to anticoagulate patients without documented AF based on incomplete information; studies of empiric anticoagulation following embolic stroke of uncertain source have found no benefit and harm (i.e. bleeding) [Citation8,Citation9]. Therefore, it is essential to detect paroxysmal AF to guide therapy to prevent stroke.
Recently, we developed an artificial intelligence-enabled electrocardiogram (AI-ECG) algorithm using over 500,000 normal sinus rhythm standard 10-second 12-lead ECGs from over 180,000 patients using machine learning to identify those with a high likelihood of undocumented AF [Citation10]. This work demonstrated that the application of a convolutional neural network (CNN) to a single ECG recorded during sinus rhythm could effectively identify paroxysmal AF, with an area under the receiver operator curve (AUC) of 0.87 (95% confidence interval [CI], 0.86–0.88), sensitivity of 79.0% (95% CI, 77.5–80.4%), specificity of 79.5% (95% CI, 79.0–79.9%), F1 score of 39.2% (95% CI, 38.1–40.3%), and overall accuracy of 79.4% (95% CI, 79.0–79.9%). The diagnostic yield improved when applied to patients with multiple ECGs (AUC 0.90).
With the impressive performance of the AI-ECG algorithm, the question becomes: what is the AI seeing that the human eye is missing? Due to the nature of CNNs, identification of the signal features selected by the AI is currently not possible. We presume that underlying structural changes (e.g. myocyte hypertrophy, fibrosis, chamber enlargement) precede the onset of AF and that these substrate changes result in subtle yet detectable ECG changes. It has been reported that normal sinus rhythm on a surface ECG may not accurately reflect atrial function. One report found approximately one-third of patients with AF undergoing cardioversion to lack sinus contraction of the left atrial appendage despite a surface ECG demonstrating sinus rhythm [Citation11]. Another report showed that about one-fourth of patients had a surface ECG revealing sinus rhythm despite fibrillation of the left atrial appendage documented via transesophageal echocardiogram [Citation12]. These studies suggest that there may be unrecognized patterns on the ECG associated with AF that are detectable during sinus rhythm by means of deep neural network. The exposure of CNN to over half a million ECGs enables it to extract and process subtle features not routinely noticed by the human eye.
In a recently published clinical case, we reported a patient with a cryptogenic stroke deferred anticoagulation therapy, based on the lack of documented AF on long-term cardiac monitoring, who developed another stroke a few years later [Citation13]. Retrospective AI-ECG analysis of this patient’s available ECGs in sinus rhythm demonstrated the patient had a high likelihood of undiagnosed AF years before the incident stroke events. Based on these findings, one may have considered it reasonable to initiate anticoagulation therapy earlier in the course and possibly preventing harm to the patient. This exemplifies a potential clinical role for the AI-ECG algorithm in management decisions and poses the question: could a high likelihood prediction of AF by the algorithm serve as a surrogate for AF and prompt therapy in patients with cryptogenic stroke?
Despite the optimism and tremendous promise of machine learning aiding with clinical decision-making, it is important to realize that this work is still in its infancy, and has inherent limitations. It will be important to conduct prospective validation studies since the algorithm was developed and tested in patients with multiple ECGs over a relatively short time period, and therefore may represent a different cohort than community dwellers with unrecognized AF. Cross-institutional collaboration is also needed to understand how to transfer and adapt the algorithm in a different health system and across populations.
Based on these validation studies, a risk stratification approach could be used to classify patients into low-, medium-, and high-risk subgroups using the algorithm’s prediction, which may guide subsequent management decisions. For instance, patients considered to be at high-risk could proceed with anticoagulation without extended monitoring, whereas those considered to be at medium-risk could undergo long-term cardiac monitoring and only proceed with anticoagulation if AF is detected. At that point, a shared decision-making approach will need to be instituted to allow the patient to make an informed decision based on extended monitoring and the need for anticoagulation based on the patient’s goals and preferences.
Furthermore, patients diagnosed with AF after AI-ECG screening and long-term monitoring have subclinical AF, and thus, may have a lower risk of stroke than patients with clinical AF. The benefit of anticoagulation in this population remains uncertain. Therefore, a critical next step will bet to conduct a large randomized trial to demonstrate the value of anticoagulation in patients with AF diagnosed after the AI-ECG screening.
Despite the numerous questions, limitations, and clinical implications that have yet to be answered, it is clear that the potential of AI-ECG is only beginning to emerge. A machine learning algorithm has taken a clinically ubiquitous, noninvasive, and inexpensive tool to unveil novel possibilities and potentially a new mass screening tool for AF. With the advent of smartphone and watch-based applications, the ability to capture high-quality patient data at the convenience of their home is only emerging. The capability of an AI-ECG algorithm to function using wearable obtained ECGs will require an algorithm developed and validated on single-lead device input. Single-lead ECGs contain their own inherent limitations that will also need to be addressed, such as detecting subtle morphological changes.
If such an AI-ECG algorithm becomes certified as medical device software for clinical implementation, this technology may be able to automatically identify patients at risk or those with unrecognized AF and, based on the patient’s medical record, determine their eligibility for anticoagulation. This would provide enormous benefits (i) to patient care by preventing ‘foreseeable’ catastrophic events, (ii) to clinical practice by enhancing the delivery of care, and (iii) to clinical research by identifying patients eligible for related clinical trials.
In conclusion, the AI-ECG algorithm’s performance holds tremendous promise as a rapid, inexpensive, and noninvasive tool capable of assessing AF risk. The ability to risk-stratify people with unrecognized and/or suspected AF would be extremely valuable in the community setting and may provide for an objective substitute when determining the appropriateness of initiating anticoagulation therapy in unclear circumstances.
Declaration of Interest
The authors have no relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript. This includes employment, consultancies, honoraria, stock ownership or options, expert testimony, grants or patents received or pending, or royalties.
Reviewers Disclosure
A reviewer on this manuscript has disclosed being a founder and board member of Rhythm AI Ltd. Peer reviewers on this manuscript have no other relevant financial relationships or otherwise to disclose.
Additional information
Funding
References
- Benjamin EJ, Wolf PA, D’Agostino RB, et al. Impact of atrial fibrillation on the risk of death the Framingham heart study. Circulation. 1998;98(10):946–952.
- Chugh SS, Havmoeller R, Narayanan K, et al. Worldwide epidemiology of atrial fibrillation: a global burden of disease 2010 study. Circulation. 2014;129(8):837–847.
- Turakhia MP, Shafrin J, Bognar K, et al. Estimated prevalence of undiagnosed atrial fibrillation in the United States. PloS One. 2018;13(4):e0195088.
- Zoni-Berisso M, Lercari F, Carazza T, et al. Epidemiology of atrial fibrillation: European perspective. Clin Epidemiol. 2014;6:213–220.
- Li L, Yiin GS, Geraghty OC, et al. Incidence, outcome, risk factors, and long-term prognosis of cryptogenic transient ischaemic attack and ischaemic stroke: a population-based study. Lancet Neurol. 2015;14(9):903–913.
- Sahay S, Nombela-Franco L, Rodes-Cabau J, et al. Efficacy and safety of left atrial appendage closure versus medical treatment in atrial fibrillation: a network meta-analysis from randomized trials. Heart. 2017;103(2):139–147.
- Primo J, Gonclaves H, Macedo A, et al. Prevalence of paroxysmal atrial fibrillation in a population assessed by continuous 24-hour monitoring. Rev Port Cardiol. 2017;36(7–8):535–546.
- Hart RG, Sharma M, Mundl H, et al. Rivaroxaban for stroke prevention after embolic stroke of undetermined source. N Engl J Med. 2018;378(23):2191–2201.
- Diener HC, Sacco RL, Easton JD, et al. Dabigatran for prevention of stroke after embolic stroke of undetermined source. N Engl J Med. 2019;380(20):1906–1917.
- Attia ZI, Noseworthy PA, Lopez-Jimenez F, et al. An artificial intelligence-enabled ECG algorithm for the identification of patients with atrial fibrillation during sinus rhythm: a retrospective analysis of outcome prediction. Lancet. 2019;394(10201):861–867.
- Bellotti P, Spirito P, Lupi G, et al. Left atrial appendage function assessed by transesophageal echocardiography before and on the day after elective cardioversion for nonvalvular atrial fibrillation. Am J Cardiol. 1998;81(10):1199–1202.
- Warraich H, Gandhavadi M, Manning WJ. Mechanical discordance of the left atrium and appendage: a novel mechanism of stroke in paroxysmal atrial fibrillation. Stroke. 2014;45(5):1481–1484.
- Kashou AH, Rabinstein AA, Attia IZ, et al. Recurrent cryptogenic stroke: a potential role for an artificial intelligence-enabled electrocardiogram? HeartRhythm Case Rep. 2020.