Validity of computer interpretation of the electrocardiogram

Area of Science:

• Cardiology
• Medical Informatics
• Artificial Intelligence in Medicine

Background:

• Standard 12-lead electrocardiograms (ECG) are crucial for diagnosing cardiac conditions.
• Computerized interpretation of ECGs aims to improve efficiency and consistency.
• Assessing the accuracy of current computer algorithms is essential for clinical adoption.

Purpose of the Study:

• To evaluate the diagnostic accuracy of a computer algorithm for interpreting 2019 standard 12-lead ECGs.
• To compare computer interpretation against cardiologist diagnoses for both normal and abnormal ECGs.

Main Methods:

• A dataset of 762 normal and 1257 abnormal 12-lead ECG records, as classified by cardiologists, was used.
• The accuracy of a computer interpretation system was assessed by comparing its readings to the cardiologist's classifications.
• Disagreement rates were analyzed across different categories of ECG abnormalities.

Main Results:

• For normal ECGs (n=762), the computer agreed with cardiologists in 41% of cases, identified minor abnormalities in 49%, and gross abnormalities in 10%.
• For abnormal ECGs (n=1257), the computer misclassified 43 as normal, including five cases of infarction and three of left ventricular hypertrophy (LVH).
• Agreement rates for specific abnormal categories varied widely (31%-85%), with persistent disagreement in 33% of cases even after review.

Conclusions:

• The computer interpretation of 2019 standard 12-lead ECGs demonstrates significant limitations in accuracy when compared to expert cardiologist interpretation.
• The algorithm frequently misclassifies normal ECGs as abnormal and, more critically, misses significant abnormalities like infarction and LVH.
• Further development and validation are required before widespread clinical implementation of this computer interpretation system.