Abstract 10300: Performance Evaluation of a Novel Noise Detection Algorithm in Implantable Cardioverter Defibrillators
Introduction: Implantable Cardioverter Defibrillators (ICD) have been proven to reduce sudden cardiac death by termination of arrhythmias with a shock from a RV defibrillation lead. Inappropriate shock due to noise oversensing is an important cause of inappropriate therapy. An algorithm may identify noise due to various causes such as lead conductor/insulation failure, myopotential or external interference. The objective of this study is to evaluate a new algorithm for noise detection that could be implemented in ICD to reduce inappropriate shocks.
Methods: The algorithm utilizes beat-to-beat data derived from stored intracardiac electrograms (EGM) and markers and is based on datasets from previous clinical studies. An event detected on the RV EGM by the ICD is declared as noise if either of the following two conditions is satisfied: (1) there is no corresponding depolarization on the shock EGM or (2) number of deflections on the RV and shock EGMs exceeds a set threshold. For an episode to be classified as noise, 3 of the last 10 beats need to be identified as noise. The algorithm was tested on a dataset of 1,420 patients with dual chamber ICDs who received one or more ICD shocks. The underlying rhythm at the time of the shock was adjudicated by a group of 7 eletrophysiologists from 4 institutions. Depending upon the application, the algorithm can be tuned for sensitivity (SN) or specificity (SP). An algorithm tuned for SP is preferred for implementation in ICD since therapy should not be withheld for arrhythmia.
Results: A total of 3304 shock ICD episodes were analyzed, 129 were classified as inappropriate shocks because of noise and 3,175 as non-noise episodes. The algorithm achieved a SN of 98.4 % (95% CI: 94.5-99.8%) and a SP of 96.7% (95% CI: 96.0-97.3%) at the setting of SN towards noise. When tuned for SP towards noise, the algorithm achieved a SP of 99.7% (95% CI: 99.4-99.8%) and SN of 82.9% (95% CI: 75.3-89.0%), suggesting potential reduction in the number of inappropriate shocks for noise by 83%.
Conclusion: In a large cohort of patients receiving shock therapies for ventricular arrhythmia and noise, a noise detection algorithm discriminates well between noise and ventricular arrhythmia with high SN and SP. The algorithm has potential value to reduce shock incidence.
- © 2011 by American Heart Association, Inc.