A method and system for high-throughput prediction of the onset of heart arrhythmias observes trends in abnormal or pathologic morphology of the electrocardiogram (ECG). A first set of ECG signals is monitored from a patient. A baseline measurement is generated from the monitored first set of ECG signals to contain nonpathologic ECG morphologies in each lead. A second set of ECG signals is monitored from the patient and a second baseline measurement is generated from the second set of ECG signals. A residuum signal is generated for each lead based on the baseline measurement and the second baseline measurement. The residuum signals are averaged across the leads. R-wave heterogeneity, T-wave heterogeneity, P-wave heterogeneity, or ST-segment heterogeneity or other indicators of arrhythmia risk or myocardial ischemia are quantified based on the generated residuum signals and the averaged residuum signal.