ARRHYTHMIAS
AND CONDUCTiON
DISTURBANCES
Spectral Turbulence Analysis of the Signal-Averaged Electrocardiogram and Its Predictive Accuracy for Inducible Sustained Monomorphic Ventricular Tachycardia George J. Kelen, MD, Raphael Henkin, PhD, Ann-Marie Starr, MD, Edward 6. Caref, MA, Dennis Bloomfield, MD, and Nabil El-Sherif, MD
This study was designed to assess the accuracy of a new noninvasive frequency analysis method for predicting patients with inducible sustained monomorphic ventricular tachycardia (VT) at electrophysidogic study and hence the risk of spontaneous ventricular tachyarrhythmias. Signal-averaged electrocardiograms from 3 orthogonal bipolar surface leads were evaluated using a microcomputer-based frequency analysis system that performs analysis of conventional timedomain late potentials as well as incorporating a new technique for spectral analysis of relatively short, overlapping signal segments spanning the whole QRS complex. The spectral analysis technique measured abnormalities anywhere in the entire QRS complex and did so without dependence on any arbitrarily defined frequency, duration or amplitude cutoffs. The hallmark of arrhythmogenic abnormality was hypothesized to be frequent and abrupt changes in the frequency signature of the QRS wave front velocity as it propagates throughout the ventricle around areas of abnormal conduction, resulting in a high degree of spectral turbulence. One-hundred forty-two subjects were studied, including 71 totally normal control subjects (“true negatives”), 33 with both late potentials by time-domain analysis and inducible sustained monomorphic VT (“true positives”), 28 with late potentials but no eviFrom St. Vincent’s Medical Center, Staten Island, New York, the Cardiology Division, Department of Medicine, State University of New York Health Science Center at Brooklyn, and the Veterans Administration Medical Center, Brooklyn, New York. This study was supported in part by Grant HL 31341 from the National Institutes of Health, Bethesda, Maryland, and by the Veterans Health Services and Research Administration Merit Review Funds, Brooklyn, New York. Manuscript received October 4, 1990; revised manuscript received and accepted December 17, 1990. Address for reprints: George J. Kelen, MD, Veterans Administration Medical Center, Cardiology Section (lllA), 800 Poly Place, Brooklyn, New York 11209.
dence of spontaneous or inducible sustained monomorphic VT (“false positives”) and 10 with inducible sustained monomorphic VT but absence of time-domain late potentials (“false negatives”). The frequency analysis technique correctly classified 166% of the true negatives, 97% of the true positives, 66% of the late potentials false positives and 60% of the late potentials false negatives. The total predictive accuracy of frequency analysis for all groups was 94%, compared with 73% for time-domain late potential analysis. The results suggest that a high degree of spectral turbulence of the overall QRS signal during sinus rhythm may provide a more accurate marker for the anatomic-electrophysiologic substrate of reentrant tachyarrhythmias than detection of late potentials in the terminal QRS region by either time- or frequencydomain analysis. Spectral turbulence analysis is applicable to patients irrespective of the QRS duration and the presence or absence of bundle branch block. (Am J Cardiol 1991;67:965-975)
he demonstration of late potentials in the terminal QRS region of the signal-averagedelectrocardiogram (ECG) has been shown in numerous studies to be a clinically useful predictor of vulnerability to ventricular tachyarrhythmias. l -8 Current techniques for time-domain late potential analysis are, however, imperfect. There is a lack of agreement as to optimal filter characteristicsand the best numerical criteria of abnormality.9 The numeric measurementsare very sensitive to the specific algorithm used for determining QRS termination. lo In the presenceof intraventricular conduction defect or bundle branch block, or both, which many patients at potential risk have, interpretation of late potentials may be difficult.” Most published studies have specifically excluded such pa-
T
THE AMERICAN JOURNAL OF CARDIOLOGY MAY 1. 1991
965
tients. In a recent review of previous reportsI the predictive values of the time-domain signal-averagedECG ranged between 20 to 95%, with an average of 64%. Several authors have reported promising results using frequency analysis techniques for ventricular tachycardia (VT) risk assessment. 13-16These studies used Fourier analysis of the terminal portion of the QRS complex and early ST segments,seekingto identify late potentials by unique frequency signaturesin the form of additional high-frequency peaksor abnormal content of high frequencies. All these techniques sought to improve tachyarrhythmia risk detection by better identification of late potentials basedon their putative frequency characteristics. Some of these techniques have been criticized as unreproducible,17-19whereas others await confirmation by independent research. The presentstudy describesand evaluatesa new frequency-domain analysis technique that may overcome someof the disadvantagesof both time-domain late potential analysis and previously advocated methods of frequency analysis, while correctly identifying the arrhythmogenic risk in subjectsmisclassifiedor not suitable for analysis by earlier methods. The frequencydomain analysis technique investigated in the present study is fundamentally different from all others20,21 in that it seeksneither to detect nor to measurelate potentials. Observations, measurementsand calculations are instead performed on the QRS complex as a whole, not on any portion arbitrarily identified by temporal, frequency or amplitude characteristics. Furthermore, no assumptions are made as to whether abnormality is representedby “abnormally, high” or “abnormally low” frequenciesanywhere throughout the whole QRS complex. None of the abnormality criteria invoke any arbitrary value for signal amplitude, duration or frequency. The hallmark of arrhythmogenic abnormality was postulated to be frequent and abrupt changesin the frequency signature of the QRS wave front velocity as it propagatesthroughout the ventricle around areasof abnormal conduction, resulting in a high degree of spectral turbulence. This may provide a more accurate marker for the anatomic-electrophysiologicsubstrate of reentrant tachyarrhythmias. A preliminary report of our study has been published.22
Group A, time-domain “true negatives,” comprised 71 subjects, healthy adult volunteers (mostly hospital staff) with no known history or symptoms of heart disease,absenceof time-domain late potentials and a normal 1Zlead ECG. Subjects with bundle branch block, including incomplete right bundle branch block or a QRS duration 1120 ms, or both, were specifically excluded from this group. Group B, time-domain “true positives,” comprised 33 subjects,all patients with positive time-domain late potentials and inducible sustainedmonomorphic VT at electrophysiologicstudy. All had a history of myocardial infarction and documented spontaneous sustained monomorphic VT (24 patients: > 120 beats/min for 230 seconds) or nonsustained VT (9 patients: >3 beats, 120 ms, or both. Acquisition
of signal-averaged
electrocardiograms:
Recordings of 3-lead orthogonal signal-averagedECGs were obtained with an ART Model 1200 EPX unit (123 subjects)or a prototype Del Mar Avionics Model 183 Cardiac Early Warning System (CEWS) (19 subMETHODS jects), and were stored on floppy diskettes. The 183 Study subjects: The study was performed retrospec- CEWS has been compared with the ART model 1200 tively on the signal-averagedECG of 142 subjects di- EPX and has been found to give comparable results.*O vided into 4 groups basedon the presenceor absenceof X-lead electrodes were placed in the fifth intercostal time-domain late potentials, and the inducibility of sus- spacein the anterior axillary line, Y-lead electrodesat tained monomorphic VT at electrophysiologic study. the top and bottom sternal border, and Z-lead elecPatients with bundle branch blocks and intraventricular trodes anteriorly and posteriorly at the level of the 4th conduction defects were included in the study. intercostal spacejust to the left of the midline. Sam966
THE AMERICAN JOURNAL OF CARDIOLOGY VOLUME 67
pling frequency of the signal was 1,000 Hz for both acquisition devices. All recordings were obtained with subjects lying comfortably in the supine position. lime-domain analysis: Time-domain analysis was performed using bidirectional 4-pole Butterworth filters at 25 to 250 Hz and at 40 to 250 Hz, with calculation of the vector magnitude. The Del Mar 183 CEWS was used for analysis of all recordings. The number of beats averagedranged between 215 and 600. The root-meansquare noise level of the vector magnitude was
AND CONDUCTiON
DISTURBANCES
Spectral Turbulence Analysis of the Signal-Averaged Electrocardiogram and Its Predictive Accuracy for Inducible Sustained Monomorphic Ventricular Tachycardia George J. Kelen, MD, Raphael Henkin, PhD, Ann-Marie Starr, MD, Edward 6. Caref, MA, Dennis Bloomfield, MD, and Nabil El-Sherif, MD
This study was designed to assess the accuracy of a new noninvasive frequency analysis method for predicting patients with inducible sustained monomorphic ventricular tachycardia (VT) at electrophysidogic study and hence the risk of spontaneous ventricular tachyarrhythmias. Signal-averaged electrocardiograms from 3 orthogonal bipolar surface leads were evaluated using a microcomputer-based frequency analysis system that performs analysis of conventional timedomain late potentials as well as incorporating a new technique for spectral analysis of relatively short, overlapping signal segments spanning the whole QRS complex. The spectral analysis technique measured abnormalities anywhere in the entire QRS complex and did so without dependence on any arbitrarily defined frequency, duration or amplitude cutoffs. The hallmark of arrhythmogenic abnormality was hypothesized to be frequent and abrupt changes in the frequency signature of the QRS wave front velocity as it propagates throughout the ventricle around areas of abnormal conduction, resulting in a high degree of spectral turbulence. One-hundred forty-two subjects were studied, including 71 totally normal control subjects (“true negatives”), 33 with both late potentials by time-domain analysis and inducible sustained monomorphic VT (“true positives”), 28 with late potentials but no eviFrom St. Vincent’s Medical Center, Staten Island, New York, the Cardiology Division, Department of Medicine, State University of New York Health Science Center at Brooklyn, and the Veterans Administration Medical Center, Brooklyn, New York. This study was supported in part by Grant HL 31341 from the National Institutes of Health, Bethesda, Maryland, and by the Veterans Health Services and Research Administration Merit Review Funds, Brooklyn, New York. Manuscript received October 4, 1990; revised manuscript received and accepted December 17, 1990. Address for reprints: George J. Kelen, MD, Veterans Administration Medical Center, Cardiology Section (lllA), 800 Poly Place, Brooklyn, New York 11209.
dence of spontaneous or inducible sustained monomorphic VT (“false positives”) and 10 with inducible sustained monomorphic VT but absence of time-domain late potentials (“false negatives”). The frequency analysis technique correctly classified 166% of the true negatives, 97% of the true positives, 66% of the late potentials false positives and 60% of the late potentials false negatives. The total predictive accuracy of frequency analysis for all groups was 94%, compared with 73% for time-domain late potential analysis. The results suggest that a high degree of spectral turbulence of the overall QRS signal during sinus rhythm may provide a more accurate marker for the anatomic-electrophysiologic substrate of reentrant tachyarrhythmias than detection of late potentials in the terminal QRS region by either time- or frequencydomain analysis. Spectral turbulence analysis is applicable to patients irrespective of the QRS duration and the presence or absence of bundle branch block. (Am J Cardiol 1991;67:965-975)
he demonstration of late potentials in the terminal QRS region of the signal-averagedelectrocardiogram (ECG) has been shown in numerous studies to be a clinically useful predictor of vulnerability to ventricular tachyarrhythmias. l -8 Current techniques for time-domain late potential analysis are, however, imperfect. There is a lack of agreement as to optimal filter characteristicsand the best numerical criteria of abnormality.9 The numeric measurementsare very sensitive to the specific algorithm used for determining QRS termination. lo In the presenceof intraventricular conduction defect or bundle branch block, or both, which many patients at potential risk have, interpretation of late potentials may be difficult.” Most published studies have specifically excluded such pa-
T
THE AMERICAN JOURNAL OF CARDIOLOGY MAY 1. 1991
965
tients. In a recent review of previous reportsI the predictive values of the time-domain signal-averagedECG ranged between 20 to 95%, with an average of 64%. Several authors have reported promising results using frequency analysis techniques for ventricular tachycardia (VT) risk assessment. 13-16These studies used Fourier analysis of the terminal portion of the QRS complex and early ST segments,seekingto identify late potentials by unique frequency signaturesin the form of additional high-frequency peaksor abnormal content of high frequencies. All these techniques sought to improve tachyarrhythmia risk detection by better identification of late potentials basedon their putative frequency characteristics. Some of these techniques have been criticized as unreproducible,17-19whereas others await confirmation by independent research. The presentstudy describesand evaluatesa new frequency-domain analysis technique that may overcome someof the disadvantagesof both time-domain late potential analysis and previously advocated methods of frequency analysis, while correctly identifying the arrhythmogenic risk in subjectsmisclassifiedor not suitable for analysis by earlier methods. The frequencydomain analysis technique investigated in the present study is fundamentally different from all others20,21 in that it seeksneither to detect nor to measurelate potentials. Observations, measurementsand calculations are instead performed on the QRS complex as a whole, not on any portion arbitrarily identified by temporal, frequency or amplitude characteristics. Furthermore, no assumptions are made as to whether abnormality is representedby “abnormally, high” or “abnormally low” frequenciesanywhere throughout the whole QRS complex. None of the abnormality criteria invoke any arbitrary value for signal amplitude, duration or frequency. The hallmark of arrhythmogenic abnormality was postulated to be frequent and abrupt changesin the frequency signature of the QRS wave front velocity as it propagatesthroughout the ventricle around areasof abnormal conduction, resulting in a high degree of spectral turbulence. This may provide a more accurate marker for the anatomic-electrophysiologicsubstrate of reentrant tachyarrhythmias. A preliminary report of our study has been published.22
Group A, time-domain “true negatives,” comprised 71 subjects, healthy adult volunteers (mostly hospital staff) with no known history or symptoms of heart disease,absenceof time-domain late potentials and a normal 1Zlead ECG. Subjects with bundle branch block, including incomplete right bundle branch block or a QRS duration 1120 ms, or both, were specifically excluded from this group. Group B, time-domain “true positives,” comprised 33 subjects,all patients with positive time-domain late potentials and inducible sustainedmonomorphic VT at electrophysiologicstudy. All had a history of myocardial infarction and documented spontaneous sustained monomorphic VT (24 patients: > 120 beats/min for 230 seconds) or nonsustained VT (9 patients: >3 beats, 120 ms, or both. Acquisition
of signal-averaged
electrocardiograms:
Recordings of 3-lead orthogonal signal-averagedECGs were obtained with an ART Model 1200 EPX unit (123 subjects)or a prototype Del Mar Avionics Model 183 Cardiac Early Warning System (CEWS) (19 subMETHODS jects), and were stored on floppy diskettes. The 183 Study subjects: The study was performed retrospec- CEWS has been compared with the ART model 1200 tively on the signal-averagedECG of 142 subjects di- EPX and has been found to give comparable results.*O vided into 4 groups basedon the presenceor absenceof X-lead electrodes were placed in the fifth intercostal time-domain late potentials, and the inducibility of sus- spacein the anterior axillary line, Y-lead electrodesat tained monomorphic VT at electrophysiologic study. the top and bottom sternal border, and Z-lead elecPatients with bundle branch blocks and intraventricular trodes anteriorly and posteriorly at the level of the 4th conduction defects were included in the study. intercostal spacejust to the left of the midline. Sam966
THE AMERICAN JOURNAL OF CARDIOLOGY VOLUME 67
pling frequency of the signal was 1,000 Hz for both acquisition devices. All recordings were obtained with subjects lying comfortably in the supine position. lime-domain analysis: Time-domain analysis was performed using bidirectional 4-pole Butterworth filters at 25 to 250 Hz and at 40 to 250 Hz, with calculation of the vector magnitude. The Del Mar 183 CEWS was used for analysis of all recordings. The number of beats averagedranged between 215 and 600. The root-meansquare noise level of the vector magnitude was
