ECG has long been a key diagnostic tool in clinical practice, and since the middle of the 20th century, continuous efforts have been made to automate interpretation. Researchers first figured out how to digitize the analog ECG signal in the 1950s, with the development of the computer-based ECG interpretation algorithm starting shortly thereafter, according to a recent review in the American Journal of Medicine (AJM).1
Now, after decades of technological advancements, including the emergence of artificial intelligence, more than 100 million computerized ECG interpretations are performed each year in the U.S.1
"Computerized ECG software has aided in ECG interpretation, improved clinical workflow, and advanced our understanding of electrophysiology," says Anthony Kashou, MD, in a Mayo Clinic podcast, adding that capabilities are improving.2
But as the title of the AJM review ("The Computerized ECG: Friend or Foe") suggests, care must be taken in integrating ECG data analysis software into the clinical workflow. On the plus side, algorithms quickly provide accurate values for routine measurements, which saves time in a busy clinical setting and supports decision-making. The downside, however, is that "the computer does have difficulty separating small waves from artifacts, difficulty in making distinctions between similar ECG findings in different diseases, and difficulty in dealing with the myriad other combinations that clinical ECGs may present," according to the review.1
Algorithms also can't consider the full clinical context in which the ECG is obtained. "These limitations lead to computer interpretation errors that require over-reading and correction," the review states. "But, the computer and reader complement each other and together provide for the most accurate ECG interpretations. Both appear here to stay."1
How an ECG Interpretation Algorithm Helps Readers at All Levels
Physicians at any level of expertise can use some assistance in interpreting ECGs. A systematic review in JAMA Internal Medicine encompassing data from 78 studies and more than 10,000 participants showed that the median accuracy for ECG interpretation across all training levels was just 54% on pretraining assessments and 67% on posttraining assessments.3 Pretraining accuracy was lowest among medical students (42%), increasing to 56% among residents, 69% among practicing physicians, and 75% among cardiologists. More experience improved accuracy, but deficiencies remained across all levels.
That's concerning given that research has shown misinterpretation of ECGs can be harmful. A study in the European Heart Journal, for instance, showed that a failure to recognize STEMI on ECG was associated with a lower likelihood of receiving proven treatments like aspirin, heparin, and revascularization, and greater odds of dying.4
An ECG interpretation algorithm, then, can prove useful for all physicians, though how it's used may differ depending on the expertise of the reader.
In a blog post, Ken Grauer, MD, outlines how computerized interpretation can be best used by expert and nonexpert readers.5 For experts, computer-assisted interpretation can save time because the algorithms provide accurate assessments of rate, intervals, and axis, eliminating the need for physicians to measure them. The main benefit for nonexperts is having a backup opinion that either supports the reader's initial impressions or creates the impetus to take a second, more careful look. For hurried experts and novice readers alike, the information provided by ECG interpretation algorithms can ensure that no significant findings are overlooked.
Strengths and Limitations of Computerized ECG Interpretation
Although ECG interpretation algorithms can make life easier for physicians, it's important to recognize both their strengths and their limitations.
A blog post from two emergency medicine physicians provides an overview of some of the advantages and shortcomings, noting that, in the absence of artifacts, algorithms "fairly reliably" calculate rate, axis, and intervals, and accurately identify normal ECGs and sinus rhythm.6 However, these programs have been shown to overcall atrial fibrillation, leading to unnecessary testing and interventions, and they may have difficulties in assessing certain clinical scenarios, such as STEMI, pacemaker rhythms, and prolonged QTc.
A review in the Journal of the American College of Cardiology (JACC) raised numerous concerns with algorithms as well: "Unfortunately, inexperienced physicians ordering the ECG may fail to recognize interpretation mistakes and accept the automated diagnosis without criticism. Clinical mismanagement may result, with the risk of exposing patients to useless investigations or potentially dangerous treatment," the authors write, pointing out that there are frequently incorrect readings for arrhythmias, conduction disorders, and pacemaker rhythms, and wide variation in false-positive and false-negative results for STEMI. "Systematic over-reading of [computer-interpreted ECG] is mandatory, requiring continuous education and active ECG training."7
Algorithms Improve Decision-Making
The authors make several recommendations on how ECG interpretation algorithms can be improved, calling for increased collaboration between manufacturers, standardization across algorithms, and testing of the programs using databases representative of the overall population.
Still, despite the limitations, there is evidence that algorithms improve physician decision-making. The JACC review authors note that computer-assisted interpretation has been shown to decrease analysis time by up to 28% for experienced readers.7 Moreover, ECG analysis software provides a repository for serial ECGs, giving easy access to physicians who want to compare different tests and evaluate changes over time. "Besides indicating differences between ECGs, it improves interpretation accuracy, for example, in acute coronary syndromes."7
To protect against the possible harm that could come from incorrect interpretation, the review authors underscore "the requirement for continuous development in software and systematic over-reading of the ECG."
To learn more about the power of the ECG in today's clinical landscape, browse our Diagnostic ECG Clinical Insights Center.
Training Is Paramount
The JACC review puts a spotlight on the importance of maintaining competency in ECG interpretation through continuous training and education. Because that starts with getting a high-quality recording, nurses and technicians tasked with obtaining ECG recordings should be well-schooled in placing the electrodes, with periodic retraining, the authors say. On the physician side, cardiology fellows in the U.S. are already asked to interpret 3,000 to 3,500 ECGs during training.8
According to the review's authors, there is a consensus that ECG interpretation algorithms should be used as an adjunct to a human electrocardiographer, who should systematically over-read all of the computer-based reports. That process requires an understanding of the strengths and limitations of computerized ECG interpretation, and adequate training and education.
"Significant progress was made in the development of ECG algorithms for use in the [computer-interpreted ECG]," the review's authors note. "However, limitations are still present, requiring standardization with continuous improvement in applied software and uniformization of ECG diagnostic criteria and statements." 7
References:
1. Smulyan H. The computerized ECG: friend or foe. The American Journal of Medicine. February 2019; 132(2): 153-160. https://www.amjmed.com/article/S0002-9343%2818%2930853-2/fulltext
2. Mayo Clinic. Computerized ECG interpretation software. YouTube.com. https://www.youtube.com/watch?v=80DNDURN3sI. Accessed September 9, 2022.
3. Cook DA, Oh SY, Pusic MV. Accuracy of physicians' electrocardiogram interpretations: a systematic review and meta-analysis. JAMA Internal Medicine. September 2020; 180(11): 1461-1471. https://jamanetwork.com/journals/jamainternalmedicine/fullarticle/2771093
4. Vijayaraghavan R, Yan AT, Tan M, et al. Local hospital vs. core-laboratory interpretation of the admission electrocardiogram in acute coronary syndromes: increased mortality in patients with unrecognized ST-elevation myocardial infarction. European Heart Journal. January 2008; 29(1): 31-37. https://academic.oup.com/eurheartj/article/29/1/31/2398024
5. Grauer K. ECG blog #126 – computerized ECG interpretations. ECG-Interpretation.blogspot.com. https://ecg-interpretation.blogspot.com/2016/05/ecg-blog-126-computerized-ecg.html. Accessed September 9, 2022.
6. MacDonald Z, Malette J. Computer based ECG interpretation: its use in the ED. EMOttawaBlog.com. https://emottawablog.com/2021/05/computer-based-ecg-interpretation-its-use-in-the-ed/. Accessed September 9, 2022.
7. Schläpfer J, Wellens HJ. Computer-interpreted electrocardiograms: benefits and limitations. Journal of the American College of Cardiology. August 2017; 70(9): 1183-1192. https://www.jacc.org/doi/abs/10.1016/j.jacc.2017.07.723.
8. Balady GJ, Bufalino VJ, Gulati M, et al. COCATS 4 Task Force 3: training in electrocardiography, ambulatory electrocardiography, and exercise testing. Journal of the American College of Cardiology. May 2015; 65(17): 1763-1777. https://www.jacc.org/doi/10.1016/j.jacc.2015.03.021
