Current and future implications of the artificial intelligence electrocardiogram: the transformation of healthcare and attendant research opportunities

The application of artificial intelligence (AI) in medical practice has grown exponentially in recent years. By applying AI to the electrocardiogram (ECG), the ability of a greater-than-a-century old test to identify occult and impending cardiac disease has been turbocharged.1 The implications for clinical and basic research, and for clinical medicine itself, are transformational. Treatments can be instituted before clinical adverse events occur, and the promise of new pathophysiological insights is opened. Deep neural networks, a form of AI, have been able to detect multiple, complex, non-linear changes in the ECG to identify cardiovascular and other disease conditions, although the precise signal changes correlated with disease have yet to be identified. Often before symptoms develop, ventricular dysfunction, valvular disease, cardiac amyloidosis, hypertrophic cardiomyopathy, and other conditions can be diagnosed with the AI ECG.1,2 Since electrolytes impact the ECG, the AI ECG serves as a ‘bloodless blood test’ to identify hyperkalaemia in select, high-risk populations.3,4 As ECGs are embedded in clinical practice and AI is software-based, implementation is facilitated. We prospectively enrolled over 22 000 patients in primary care settings in 8 months during a pandemic (none of the clinicians had AI expertise) and found that that the AI ECG increased the identification of asymptomatic left ventricular dysfunction by over 30%.5 Moreover, AI improved the yield of the echocardiograms that were ordered, providing healthcare resource stewardship, and facilitated primary care clinician top of license practice. As ECGs are now widely available—via electronic stethoscopes, smartphones, and smart watches—screening and monitoring of cardiovascular disease becomes massively scalable.6,7 The widespread availability of this diagnostic prowess will transform medicine and open research opportunities. Traditionally, medical practitioners have had an ‘ice pick’ view of a patient’s health via clinic visits and hospitalizations, with no health information while a person is experiencing the fullness of life (Figure 1A). To bridge potentially important data gaps, practitioners may increase the frequency of clinic visits or secondary evaluations (i.e. laboratory visits or nursing assessments), at the expense of patient convenience and healthcare resource utilization (Figure 1B). For a hypertensive heart failure patient with renal dysfunction, for example, clinic visits and tests to assess volume status, left ventricular ejection fraction, electrolyte abnormalities, and renal function may prevent hospitalization. (A) The current ‘ice-pick’ model of healthcare with isolated, singular clinic visits. If wanting more medical data, providers may resort to increased healthcare utilization (B) with more lab visits, patient emails, or nurse visits. AI device application to the current healthcare system would allow for a continuum of care (C) with data inflow from home-monitoring systems enabling patients to start treatment or request increased level of treatment (‘hospital-at-home’) when necessary to avoid hospitalization (D). AI will allow patients to monitor and self-direct care for chronic conditions while simultaneously allowing providers to monitor what happens between episodic clinic visits. Using the AI ECG with mobile, handheld, and home-based devices enables continuous assessment of health status (Figure 1C), closing physiologic and health data gaps. This allows for at-home evaluation of arrhythmias, electrolytes, and a number of cardiac and non-cardiac pathologies such as valvular heart disease, left ventricular dysfunction, analyte abnormalities, and other conditions. Such tools transform the paradigm of healthcare delivery from resource intensive, direct clinician care to enlightened patient self-care with clinician guidance (Figure 1D). This is analogous to the shift in diabetic care from clinician administered insulin early on, to clinician guided, patient managed self-glucose measurement and insulin injection. For example, patients up-titrating an evidence-based heart failure medical regimen could potentially screen for hyperkalaemia at home using AI ECG-based technology, and then self-administer one of the new, safe potassium binders if indicated. AI in medicine will not replace in-person clinical assessments and human contact. Rather, it will identify who needs evaluation, and enrich in-person interactions by providing integrated, useful information to enable top of license care delivery and guide appropriate use of imaging tests and specialty referral. Along with the many benefits of the AI ECG, there are risks that must be mitigated. AI models may be susceptible to bias if not developed and tested on diverse populations; care must be taken to preserve individual privacy; assessment of performance in practice is required to avoid unnecessary anxiety and diagnostic testing. Medical educators must teach clinicians how to use these new tools, as with any medical test. Cross-border data movement, population screening, and insurance implications must all be considered. While the AI ECG stands poised to transform health and healthcare, critical research questions remain. What signal features—invisible to expert humans—does the AI see? Understanding this may lead to new pathophysiologic insights. The deep phenotyping offered by the AI ECG may lead to new disease classifications, genetic tests, and personalized therapies. How do we best deploy these systems to improve care, prevent disease, and appropriately channel resources? How do we optimally reimburse for the use of these systems to meaningfully drive positive outcomes? What are the regulatory implications of a perpetually learning system? We look forward to inspired global collaborations between clinicians, researchers, engineers, industry, and government to address these questions and fuel this big step in maintaining health. Conflict of interest: Dr Friedman and Dr Attia and Mayo Clinic have licensed AI algorithms to companies commercializing them and may benefit from their use. Biography: Dr David M. Harmon is an academic internist at the Mayo Clinic in Rochester, MN, USA. Dr Harmon is a member of the Alpha Omega Alpha Medical Honor Society and is directing research efforts at the Mayo Clinic through a National Institute of Health Stimulating Access to Research in Residency grant. His research in artificial intelligence and population preventative health has been recognized by the Heart Rhythm Society and the American Society for Preventive Cardiology from which he received a designated poster invitation and a Young Investigator award, respectively. Dr Harmon’s research, based in the Department of Cardiovascular disease, is focused on the clinical development and application of artificial intelligence to the electrocardiogram. He aims to broaden the use of artificial intelligence technology, validating its application in wearable and handheld devices which hold implications for both at-home medical care and in the acute, emergent settings. Biography: Zachi I. Attia is an AI researcher in Mayo Clinic Minnesota, developing AI-based medical screening tests for early detection of cardiac diseases. Three of these tests are currently in FDA approval process under ‘breakthrough designation’. As the co-director of AI in cardiology, Zachi direct a team of AI researchers, working side by side with cardiologist to find effective algorithm to help patients get the care they need, as quickly as possible. Zachi received his BSc and MSc in Electrical engineering from Ben Gurion University in Israel and his PhD in Bioinformatics and Computational Biology from the University of Minnesota Biography: Dr Paul A. Friedman is a Professor of Medicine and the Norman Blane & Billie Jean Harty Chair, Mayo Clinic Department of Cardiovascular Medicine Honoring Robert L. Frye, MD at Mayo Clinic, Rochester, MN, USA. He is ABIM board certified in cardiovascular medicine and cardiac electrophysiology and is an active participant in the MOC process. Previously, Dr Friedman served as Director of the Cardiac Implantable Device Lab, Mayo Clinic. He is a trained electrical engineer, with deep experience in innovation (>40 patents issued, named Minnesota Top Inventor), and scientific research (>250 scientific original publications). He is a committed educator, serving as a director for 5 national and international meetings, editor of 7 textbooks, and author of over 60 book chapters. He is a frequent visiting professor and lecturer at educational meetings. Dr Friedman received his BA in Plan II liberal arts and BS in Electrical Engineering from the University of Texas at Austin. He received his medical degree from Stanford University, and trained in internal medicine at the University of Washington, Seattle (internship) and Stanford University. He trained in cardiovascular medicine and cardiac electrophysiology at Mayo Clinic, Rochester, MN, USA.