Interpretation and measurement consistency of seven ECG computer programs

• Published in: Journal of electrocardiology 2019-11, Vol.57, p.S99-S99
• Main Authors: De Bie, J., Diemberger, I.
• Format: Article
• Language: English
• Subjects: Cardiac arrhythmia; Data analysis; Data processing; Electrocardiography; Pediatrics
• ISSN: 0022-0736; 1532-8430
• DOI: 10.1016/j.jelectrocard.2019.08.021

Background: Despite final diagnostic responsibility lying with physicians, computer interpretations of electrocardiograms (ECGs) influence management of patients. The quality of interpretations, however, has been poorly assessed. We compared directly seven ECG programs (table) to assess measurement stability and interpretation consistency in a large set of representative adult and pediatric ECGs obtained from hospital and acute-care settings. Methods: From pre-recorded ECGs we created analog continuous loop samples for replay, replacing data between the last T wave and first P wave with a cubic spline and allowing limited stretching or compression to achieve an exact 10 s sample length. After exclusion of distorted samples, 2155 eligible records remained for playback. These were played with a Whaleteq MECG 2.0 Multichannel ECG Test System (WHALETEQ Co Ltd, Taipei City, Taiwan) connected to a laptop PC. Electrocardiographs were connected with their original patient cables in an environment free from electric and magnetic interference. Three repeat captures were made of each ECG per electrocardiograph, yielding 45,255 captures. Differences in PR interval, QRS duration, and QT interval across repeat captures were used to assess measurement stability, and interpretation text, conclusion statements, and P wave detection were used to assess interpretation consistency. Overall consistency for programs was judged by the number of "normal" vs. "abnormal" conclusion statements over the whole dataset. Results: Differences across repeat captures in PR interval >30 ms varied from 0.7% to 2.6%, in QRS duration >20 ms from 0.2% to 2.1%, and in QT interval >40 ms from 0.5% to 3.1%. Only half of computer interpretations were consistent across all three repeat captures (range 26–64%). One of three readings differed in a mean of 34% of cases (range 29–44%) and all three differed in a mean of 15% of cases (range 8–30%). Interpretation conclusion changed from normal to abnormal or vice versa in an average of 3.4% of 1642 adult cases (range 0.5–7.5%). P-waves were detected inconsistently in a mean of 3.6% of cases (range 2.1–5.3%). Proportions of normal and abnormal conclusions differed substantially for adults from 8–29% (normal) to 48–77% (abnormal). Differences were smaller for pediatric ECGs, 37% to 60% (normal), 22% to 46% (abnormal). Conclusions: The programs varied substantially in measurement stability and interpretation consistency.