Discrimination of ventricular tachycardia from supraventricular tachycardia by a downloaded wavelet-transform morphology algorithm: A paradigm for development of implantable cardioverter defibrillator detection algorithms

Introduction: Present implantable cardioverter defibrillators (ICDs) have algorithms that discriminate supraventricular tachycardia (SVT) from ventricular tachycardia (VT). One type of algorithm is based on differences in morphology of ventricular electrograms during VT and SVT. Prior SVT-VT discrimination algorithms have not undergone real-time evaluation in ambulatory patients until they were incorporated permanently into ICDs. This approach may result in incomplete testing of electrogram morphology algorithms because they are influenced by posture, activity, and electrogram maturation. We downloaded software into implanted ICDs to study a novel algorithm that compares morphologies of baseline and tachycardia electrograms based on differences between corresponding coefficients of their wavelet transforms. This comparison is expressed as a match-percent score. Methods and Results: In 23 patients, we downloaded the wavelet algorithm into implanted ICDs to assess the temporal and postural stability of baseline electrograms as measured by this algorithm and its accuracy for SVT-VT discrimination. Median follow-up was 6 months. Software was downloaded into all ICDs without altering other device functions. With few exceptions, percent template match in baseline rhythm was stable with changes in body position, rest versus walking, isometric exercise, and over time (1 and 3 months). Using the nominal match-percent threshold of 70%, sensitivity for detection of 38 VTs was 100%. Specificity for rejection of 65 SVTs was 78%. SVTs were rejected for a total of 2.7 hours. Inappropriate detections of SVT as VT were caused by electrogram truncation, myopotential interference with low-amplitude electrograms, waveform alignment error, and rate-dependent aberrancy. The first three accounted for 69% of inappropriate detections and could have been prevented by optimal programming. The optimal match-percent threshold was 60% to 70% based on a receiver-operator characteristic curve. After shocks, the median time for baseline electrogram morphology to normalize was 85 seconds. Conclusion: The wavelet morphology algorithm has high sensitivity for VT detection. Inappropriate detections of SVT as VT may be reduced by optimal programming. Downloadable software permits evaluation of new algorithms in implanted ICDs.