Diagnostic accuracy of automated computerised electrocardiogram interpretation compared with a panel of experienced cardiologists

Introduction

Computerised electrocardiogram (ECG) interpretation is widely applied, especially within the clinical settings of primary care and surgical preadmission. Concerns have been raised over the accuracy of computerised ECG interpretation. Our aim was to compare the performance of computer-based ECG interpretation with that of a panel of experienced cardiologists.

Methods

All consecutive ECGs performed in a hospital cardiology department over a 1-week period were analysed. Two cohorts were assessed, open access patients from primary care and surgical preoperative assessment patients. Cardiologists were blinded to clinical details and the computerised ECG interpretation. ECGs were analysed by a panel of cardiologists with the consensus view taken as the reference standard. ECGs were interpreted in relation to 'rhythm' and 'other abnormalities' and were classified as normal or abnormal.

Results

Seventy consecutive ECGs were analysed, 47 from open access and 23 from surgical preassessment. The cohort's median age was 60 years (range 27–87 years, male n = 30). Twenty-four ECGs were normal. There was complete disagreement over the computerised ECG interpretation of one ECG, which was deemed of major clinical significance. Partial disagreement occurred in the remainder. The greatest level of disagreement related to the interpretation of left ventricular hypertrophy and ECG evidence of myocardial ischaemia/infarction. Likelihood ratios (LR) were not calculated for negative results as there were no false negative results. LR for abnormal ECG 'rhythm' were 18.3 (6.7–53.4) and for abnormal ECG 'other abnormalities' were 3.15 (2.13–5.11) (Table 1).

Table 1 abstract

Conclusion

Need exists to improve the diagnostic algorithms used by computerised ECG interpretation. It is essential that all automated computerised ECG interpretations be over read by a physician.