Abstract
Exercise stress testing is crucial for assessing myocardial ischemia. However, artifacts from algorithmic interpretation and limitations of linked median algorithms can lead to false-positive findings. A female patient, age 63 years, underwent treadmill stress testing for atypical chest pain. Linked median analysis showed narrow complex tachycardia at 270 bpm during the stress test. Having supraventricular tachycardia (SVT) with a heart rate of 260 bpm would be extremely uncommon in a patient 63 years of age, especially without symptoms. The patient was initially planned for inpatient admission, but the raw electrocardiogram (ECG) confirmed sinus tachycardia. This case emphasizes the need for verification of synthesized ECG to avoid false-positive arrhythmia diagnoses.
- Exercise stress test
- False-positive findings
- Linked median algorithm
- Myocardial ischemia
- Supraventricular tachycardia
Exercise stress testing is a crucial tool for assessing myocardial ischemia and cardiac function under stress. However, the accuracy of this diagnostic method can be compromised by artifacts, which may lead to false-positive findings and unnecessary interventions.1,2 During exercise stress testing, various physiological and technical factors, such as patient movement, skeletal muscle activity, and respiratory motion, can introduce significant noise and artifacts into electrocardiogram (ECG) recordings. These artifacts are particularly pronounced during peak exercise when heart rates are elevated, leading to challenges in accurate ECG interpretation.3 One key tool developed to mitigate these issues is the linked median algorithm. By averaging multiple cardiac cycles, this algorithm helps establish a stable baseline for more precise ST-segment analysis, especially in scenarios where the P and T waves may merge.2 This algorithm creates a median beat from multiple ECG cycles that helps reduce noise and artifacts.4 The linked median algorithm is specifically used during exercise stress tests rather than for all ECGs. This algorithm is particularly beneficial in the context of exercise stress testing due to its ability to reduce noise and artifacts that are prevalent during physical activity. The American Heart Association (AHA) highlights that most digitized exercise ECGs use averaged updated cardiac cycles to enhance the precision of ST-segment measurements, which is crucial during the high-exercise phase when artifacts are more pronounced.2 This approach is less commonly applied to resting ECGs, where the conditions are more controlled and the need for such advanced noise reduction techniques is reduced.
Automated algorithms in ECG interpretation, while beneficial for their efficiency, have notable pitfalls. Automated measure-ment errors, especially in determining the J-point and ST-segment levels, can lead to incorrect diagnoses. This is especially critical at higher heart rates, where the T wave and P wave merge, complicating ST-segment measurement.2 This case report emphasizes verifying synthesized ECG data through a detailed review of treadmill myocardial perfusion and stress testing data from a woman, aged 63 years. Initial automated analysis suggested an implausibly high heart rate, which was later corrected to sinus tachycardia through manual review. This case under-scores the importance of reviewing both the computer-generated linked median ECG and the raw ECG during every exercise stress test. The findings align with the AHA’s recommendations for exercise laboratories, which advocate for the use of reliable ECG recording s ystems and the importance of comparing raw analog data with computer-generated data for validity.2
Case Presentation
A woman, age 63 years, with a history of hyperlipidemia presented with atypical chest pain and palpitations. The patient reported occasional episodes of central chest discomfort, dull in nature that were not associated with exertion, diaphoresis, or radiation. She had no prior history of cardiac arrhythmias, syncope, or dizziness. Her family history was unremarkable for premature cardiovascular disease. The patient was not on any medications and denied smoking, alcohol, or substance abuse.
To further evaluate her symptoms considering her age and risk factor due to hyperlipidemia, the patient underwent treadmill myocardial perfusion and stress testing using the Bruce protocol (Table 1). At baseline, her blood pressure was 124/66 mmHg, and her heart rate was 72 beats per minute (bpm). The resting ECG demonstrated a normal sinus rhythm with a ventricular rate of 72 bpm (Figure 1).
Exercise stress test summary with hemodynamic and performance parameters
Resting Electrocardiogram Showing Normal Sinus Rhythm.
During stage 3 of the Bruce protocol, at a treadmill speed of 3.4 mph and a 14.0% incline, the ECG was flagged for an abnormal finding. The linked median algorithm suggested the presence of a narrow complex supraventricular tachycardia (SVT) with a heart rate of 260–272 bpm (Figure 2A). Peak exercise blood pressure was 172/66 mmHg, and there were no symptoms of dizziness, syncope, chest pain, or hemodynamic instability. Physical examination during the stress test was unremarkable. Myocardial perfusion imaging showed no evidence of reversible ischemia or infarction (Figure 3). Ventricular function was normal with no regional wall motion abnormalities.
Misinterpretation by Linked Median Algorithm During Exercise Stress Testing. (A) Linked median EKG trace during exercise stress testing falsely detected supraventricular tachycardia (SVT) with a heart rate of 260–272 bpm, based on synthesized rhythm data. (B) Manual analysis of the raw EKG revealed sinus tachycardia at 136 bpm, consistent with physiological exertion.
Myocardial Perfusion Imaging Demonstrating Normal Perfusion. The myocardial perfusion images show normal distribution of radiotracer uptake during both stress and rest phases, acquired in upright and supine positions. The short-axis (SA), horizontal long-axis (HLA), and vertical long-axis (VLA) slices confirm homogenous perfusion without evidence of reversible or fixed defects. The stress (Str) and rest (Rst) images demonstrate preserved perfusion from apex to base and across the inferior-anterior (INF–ANT) and septal-lateral (SEP–LAT) walls, consistent with normal myocardial perfusion.
Upon closer inspection of the linked median trace (Figure 2A), the report of SVT was deemed clinically atypical. A sustained narrow complex tachycardia at 260–272 bpm in a 63-year-old patient without associated symptoms would be extremely rare. To rule out an ECG algorithmic error, the raw ECG data from the stress test was manually reviewed (Figure 2B).
Manual analysis of the raw ECG strip revealed a sinus tachycardia at 136 bpm, consistent with the patient’s exertion level during exercise. The P waves preceding each QRS complex were clearly identifiable, confirming the sinus origin of the rhythm. Furthermore, the QRS morphology remained unchanged from baseline, with no evidence of aberrant conduction, ectopy, or irregular rhythm that would suggest SVT or atrial fibrillation (AF). The linked median algorithm had misinterpreted high-frequency noise and rapid sinus activity during exercise as SVT, highlighting a critical limitation of synthesized ECG data.
The manual confirmation of sinus tachycardia instead of SVT altered the clinical course. The initial plan for inpatient admission and further evaluation was deferred, avoiding unnecessary hospitalization. The patient was reassured, and no additional interventions were required.
Discussion
Machine-generated data elements used during ECG interpretation in exercise stress tests include continuous heart rate monitoring to assess cardiovascular response and detect arrhythmias,1 automated ST-segment analysis to measure shifts for detecting myocardial ischemia,2 and QRS complex analysis using linked median algorithms to average ECG cycles and reduce noise.2 The algorithm also processes T-wave and P-wave data to identify arrhythmias and conduction abnormalities.5 Heart rate variability (HRV) analysis provides insights into autonomic function and coronary artery disease diagnosis,6 while real-time heart rate display aids in immediate decision-making.7
A linked median algorithm, used in ECG interpretation, reduces noise and artifacts by averaging multiple ECG cycles. This process creates a median beat, which provides a clearer ECG. During exercise stress testing, where motion artifacts and muscle activity introduce noise, the linked median algorithm improves signal-to-noise ratio, enabling reliable detection of significant ECG changes like ST-segment depression or elevation. A study by Mickelson et al.4 demonstrated computer interpretation using median averaged beats is a reasonable surrogate for visual interpretation, with an 88% agreement between the two methods. However, the linked median algorithm has limitations. It can produce false-positive findings, particularly in high-frequency settings like exercise stress tests. This is due to the potential misinterpretation of merged T and P waves at higher heart rates, which can lead to incorrect diagnoses of arrhythmias or ischemic changes.1,7
Studies have shown automated ECG interpretation systems frequently misdiagnose arrhythmias, conduction disorders, and electronic pacemakers. A study by Mickelson et al.4 evaluated the reliability of computer-generated exercise ECG interpretation using median averaged beats and reported an 88% agreement with visual assessment, supporting its use as a reasonable surrogate for physician interpretation. However, the study also highlighted instances of discordance—primarily false positives—underscoring the need for clinician oversight and review of raw ECG data to avoid misinterpretation.4 Panneerselvam et al.8 reported a case of pseudo–tombstone-like ST elevations during a stress test, caused by errors in the linked median averaging algorithm, emphasizing the need to verify synthesized ECG data with raw tracings. Guglin et al.9 found significant disagreements between computers and cardiologists in 9.9% of all ECGs and 15.9% of abnormal ECGs, with arrhythmias and conduction disorders being the most common errors. Another study by Hwan Bae et al.10 highlighted that erroneous computer interpretations of AF were not rare, leading to inappropriate follow-up studies or treatments in some cases.
In our case, the manual verification of the raw ECG data was crucial in avoiding a false-positive diagnosis of SVT and unnecessary interventions. Ongoing advancements in artificial intelligence and machine learning hold significant promise for enhancing ECG accuracy. Deep learning models, such as convolutional neural networks and structured state space models, have shown superior performance in capturing long-term dependencies in time series data, leading to improved diagnostic accuracy for arrhythmias and myocardial ischemia.11,12 More robust validation studies are needed for new algorithms, especially under high-noise conditions like exercise stress testing.
Conclusion
Automated ECG algorithms enhance efficiency but have limitations, especially in noisy environments. Manual verification of machine-generated ECG data during exercise stress testing is crucial to avoid false positives and unnecessary interventions.
- Received February 5, 2025.
- Revision received March 24, 2025.
- Accepted April 22, 2025.
References
- 1.↵Myers J, Arena R, Franklin B, . Recommendations for clinical exercise laboratories: a scientific statement from the american heart association. Circulation. 2009;119(24):3144-3161. doi:10.1161/CIRCULATIONAHA.109.192520
- 2.↵Fletcher GF, Ades PA, Kligfield P, . Exercise standards for testing and training: a scientific statement from the American Heart Association. Circulation. 2013;128(8):873-934. doi:10.1161/CIR.0b013e31829b5b44
- 3.↵Kaiser W, Findeis M. Artifact processing during exercise testing. J Electrocardiol. 1999;32 Suppl:212-219. doi:10.1016/s0022-0736(99)90083-3
- 4.↵Mickelson JK, Bates ER, Hartigan P, Folland ED, Parisi AF. Is computer interpretation of the exercise electrocardiogram a reasonable surrogate for visual reading? Veterans Affairs ACME Investigators. Clin Cardiol. 1997;20(4):391-397. doi:10.1002/clc.4960200417
- 5.↵Wolthuis RA, Froelicher VF, Hopkirk A, Fischer JR, Keiser N. Normal electrocardiographic waveform characteristics during treadmill exercise testing. Circulation. 1979;60(5):1028-1035. doi:10.1161/01.cir.60.5.1028
- 6.↵Bailón R, Mateo J, Olmos S, . Coronary artery disease diagnosis based on exercise electrocardiogram indexes from repolarisation, depolarisation and heart rate variability [published correction appears in Med Biol Eng Comput. 2003 Nov;41(6):746]. Med Biol Eng Comput. 2003;41(5):561-571. doi:10.1007/BF02345319
- 7.↵Paridon SM, Alpert BS, Boas SR, . Clinical stress testing in the pediatric age group: a statement from the American Heart Association Council on Cardiovascular Disease in the Young, Committee on Atherosclerosis, Hypertension, and Obesity in Youth. Circulation. 2006;113(15):1905-1920. doi:10.1161/CIRCULATIONAHA.106.174375
- 8.↵Panneerselvam A, Shankarappa RK, Nanjappa MC. Tombstone-like electrocardiographic changes during exercise stress test. J Electrocardiol. 2012;45(1):13-14. doi:10.1016/j.jelectrocard.2011.07.029
- 9.↵Guglin ME, Thatai D. Common errors in computer electrocardiogram interpretation. Int J Cardiol. 2006;106(2):232-237. doi:10.1016/j.ijcard.2005.02.007
- 10.↵Hwan Bae M, Hoon Lee J, Heon Yang D, . Erroneous computer electrocardiogram interpretation of atrial fibrillation and its clinical consequences. Clin Cardiol. 2012;35(6):348-353. doi:10.1002/clc.22000
- 11.↵Mehari T, Strodthoff N. Towards Quantitative Precision for ECG Analysis: Leveraging State Space Models, Self-Supervision and Patient Metadata. IEEE J Biomed Health Inform. 2023;27(11):5326-5334. doi:10.1109/JBHI.2023.3310989
- 12.↵Elul Y, Rosenberg AA, Schuster A, Bronstein AM, Yaniv Y. Meeting the unmet needs of clinicians from AI systems showcased for cardiology with deep-learning-based ECG analysis. Proc Natl Acad Sci U S A. 2021;118(24):e2020620118. doi:10.1073/pnas.2020620118
