Diagnostic accuracy of atrial fibrillation by computerized electrocardiogram analysis versus cardiologist...

Advancements in computerized electrocardiogram (ECG) interpretation have improved efficiency, but concerns remain regarding diagnostic accuracy for atrial fibrillation (AF). Misclassification may result in inappropriate treatment and adverse outcomes.

To compare the diagnostic accuracy of computerized ECG interpretation versus cardiologist over-read for AF and evaluate the clinical consequences of diagnostic errors.

We conducted a retrospective study of 5000 ECGs obtained at the Veterans Affairs Caribbean Healthcare System between September and November 2017. All ECGs were initially interpreted by automated software. A subset of 905 ECGs underwent over-read at the time of care. Separately, two board-certified cardiologists independently reviewed all 5000 ECGs in a blinded fashion and completed standardized interpretation forms. Diagnostic metrics were compared using 2-proportion z-tests and McNemar's test. Logistic regression identified predictors of misclassification. Clinical records were reviewed for treatment consequences.

Computerized interpretation identified 803 (16 %) AF-related ECGs; of these, 70 % were true AF, 11 % false positives, and 19 % false negatives. Cardiologist interpretation showed higher accuracy: sensitivity 87 % vs. 78 %, specificity 99 % vs. 97 %, PPV 98 % vs. 86 %, NPV 99 % vs. 97 % (all p < 0.01). Ventricular pacing (OR 10.2), undetermined rhythm (OR 13.4), and tachycardia (OR 1.8) independently predicted misclassification (p < 0.001). Among misclassified ECGs, 41 % of errors were unrecognized, leading to delayed anticoagulation (18 %), inappropriate initiation (12 %), and three adverse events.

Computerized ECG interpretation of AF is prone to diagnostic error. Cardiologist over-read improves accuracy and may reduce preventable harm.

Computerized ECG interpretation misclassified AF in 30 % of cases, 11 % overcalls and 19 % missed diagnoses. Cardiologist over-reading demonstrated superior accuracy (87 % sensitivity, 99 % specificity), minimizing both false positives and negatives. In 41 % of erroneous computer interpretations, providers failed to recognize the error, leading to delayed anticoagulation, inappropriate therapy, two thromboembolic events, and one bleeding-related hospitalization. Ventricular pacing, undetermined rhythms, and tachycardia were key predictors of misclassification. These findings highlight the limitations of automated ECG analysis and support structured cardiologist over-read to safeguard against diagnostic errors and improve patient outcomes in AF detection.