Improving the predictive ability of the signal-averaged electrocardiogram with a linear logistic model incorporating clinical variables

To improve the predictive accuracy of the signal-averaged electrocardiogram, we created a linear logistic model for predicting ventricular tachycardia during electrophysiologic testing. This signal-averaged electrocardiographic model was created from data obtained from 214 patients undergoing electrophysiologic testing (70 had ventricular tachycardia during electrophysiologic testing) by using stepwise logistic regression to rank eight clinical and nine signal-averaged electrocardiographic variables. The best predictors were ejection fraction, history of infarction, ventricular ectopic pairs or nonsustained ventricular tachycardia on Holter monitoring, QRS duration after 25-Hz filtering, and root mean square voltage of the terminal 40 msec of the QRS complex after 40- and 80-Hz filtering. Cross validation (a statistical technique that can be used to accurately evaluate how a predictive model will perform on a prospective patient population) was used to validate the model. After cross validation, the model's sensitivity was 91% and specificity was 59% for predicting ventricular tachycardia during electrophysiologic testing. This model compared favorably with established 25-Hz late-potential criteria (QRS duration of more than 110 msec and root mean square voltage of less than 25 μV of the terminal 40 msec of the QRS complex; sensitivity, 64%; specificity, 85%) and with established 40-Hz late-potential criteria (QRS duration of more than 114 msec or root mean square voltage of less than 20 μV of the terminal 40 msec of the QRS complex or duration of the low-amplitude signal less than 40 μV at the terminal QRS complex that is greater than 38 msec; sensitivity, 84%; specificity, 54%). Therefore, the predictive ability of the signal-averaged electrocardiogram can be improved by using a linear logistic model that incorporates clinical variables.