The most common errors in automatic ECG interpretation.

Introduction: The 12-lead electrocardiogram (ECG) is one of modern medicine's most important and useful diagnostic tests. It is a non-invasive, widely available and relatively cheap method used mainly in the diagnostics of arrhythmias and a variety of other pathologies, including conduction disturbances and ischemia. ECG is also used to evaluate the activity of a pacemaker or other implantable devices. Medical personnel worldwide routinely use the automatic analysis and interpretation of ECGs to support their medical evaluation of patients' electrocardiograms. Methods: In the current study, we performed a precise analysis of electrocardiograms from 526 different patients and compared it to the automatic interpretations to determine the frequency of false-positive (overinterpretation) and false-negative (underinterpretation) incorrect interpretations. Results: It was found that about 39% of ECGs were interpreted incorrectly and amongst the misinterpreted 193 ECG cases, 58% were false-negative, while 57% were false-positive. Additionally, it was revealed that incorrect diagnosis of ischemia (false-positive) was correlated with the body mass index (BMI) of the subjects as well as with the undiagnosed chamber enlargements/hypertrophies. Moreover, we found that in elderly people (>60 years old) a larger number of incorrect diagnoses occurred. The diagnosis of ischemia, which is clinically the most important role of ECG, in our study occurred in 16.1% as a false-positive diagnosis, while in 22.3% ischemia remained unrecognized. Conduction abnormalities were overdiagnosed in 21.8% of cases, while underdiagnosis occurred in 14.5%. Arrhythmias were overdiagnosed in 28% of cases and 17.1% of cases were underdiagnosed. Discussion: In conclusion, we support the statement that relying on the automatic ECG analysis may lead to misinterpretations, which may mislead the medical staff. Automatic analysis of ECG may contain valuable data, although it requires verification and additional knowledge of electrocardiography in every case to achieve a correct and complete interpretation. Results of our studies should increase the caution among clinicians not to rely fully on the automated analysis. Future perspectives should include the application of AI in algorithms used in ECG analysis by manufacturers and paying more attention to accessing proper feedback from clinicians to device manufacturers.