Sensitivity-specificity tradeoff for different alarm strategies in continuous 12-lead ST-segment monitoring

Our objective was to assess both the sensitivity and specificity of various alarm-detection strategies for real-time ST-segment monitoring. The dataset was comprised of ST measurements from 12-lead Mason-Likar electrocardiograms (ML ECG) obtained from 88 patients before and during elective balloon-inflation angioplasty. We compared the ability of the 12-lead ML ECG and its subsets to detect ischemia, based on an index derived from the absolute values of the ST shift in "non-ischemic" and "ischemic" states. The sensitivity and specificity of a classifier using this index for a range of threshold values yielded receiver operating characteristic (ROC) curves. Detection performance was assessed as mean area under 1,000 ROC curves (AUC) generated by a bootstrap method with replacement. Results show that the reduced lead sets with 1 limb lead and 2 precordial leads (e.g. III, V3, V6 or III, V1, V4) attain AUC values that are not significantly different from those of the 12-lead ML ECG.