Diagnostic accuracy of derived versus standard 12-lead electrocardiograms

To compare the diagnostic yield of electrocardiograms (ECGs) recorded by 12 standard leads with that of 12-lead ECGs derived from 3 bipolar EASI leads, we analyzed pertinent ECG data for 290 normal subjects and 497 patients who had had a prior myocardial infarction (MI); the latter group comprised 36 patients with a non-Q MI, 282 patients with a Q-wave MI, and 179 patients with a history of ventricular tachycardia (VT). We first estimated statistically an optimal set of coefficients for deriving the 12 standard leads from EASI leads and assessed this transformation in terms of goodness of fit. To gauge the diagnostic information content of the recorded vs. derived 12-lead ECGs, we performed successively two-group diagnostic classification--based on the Cardiac Infarction Injury Score (CIIS)--separating each of the patient subgroups from the normal group; the classification was repeated for 200 sets of patients selected randomly (with replacement), and the results were plotted as mean receiver operating characteristics. We found that derived 12-lead ECGs correlated well with the recorded ones, and reproduced faithfully the diagnostic features needed for the CIIS. When the CIIS was determined from features of the recorded standard 12 leads, its mean diagnostic performance (assessed in terms of area under the receiver operating characteristics curve) was 0.9004 for detecting non-Q MIs, 0.9546 for Q-wave MIs, and 0.9919 for MIs complicated by a history of VT. When, instead, features of derived 12 leads were used to determine the CIIS, diagnostic performance remained virtually unchanged (at 0.8905, 0.9531, and 0.9906, respectively). We conclude that, in our population, EASI-derived 12-lead ECGs contain nearly the same diagnostic information as standard 12-lead ECGs.