Analysis of Abnormal Intra-QRS Potentials
Improved Predictive Value for Arrhythmic Events With the Signal-Averaged Electrocardiogram
Background Using the signal-averaged ECG (SAECG), this study developed a new electrical index for predicting arrhythmic events: abnormal intra-QRS potentials (AIQP).
Methods and Results We studied 173 patients followed after myocardial infarction for a mean duration of 14±7 months. Sixteen arrhythmic events occurred, defined as sudden cardiac death, documented sustained ventricular tachycardia, or nonfatal cardiac arrest. Noninvasive indices of arrhythmia risk were measured, including AIQP, conventional SAECG, Holter, and left ventricular ejection fraction (LVEF). Abnormal intra-QRS potentials were defined as abnormal signals occurring anywhere within the QRS period. They were estimated with a lead-specific, parametric modeling method that removed the smooth, predictable part of the QRS. AIQPs are characterized by the remaining transient, unpredictable component of the QRS and manifest as low-amplitude notches and slurs. A combined XYZ-lead AIQP index exhibited higher specificity (95%) and predictive value (PV) (+PV, 47%; −PV, 94%) than the conventional SAECG in combination with Holter and LVEF (specificity, 89%; +PV, 25%; −PV, 93%).
Conclusions AIQP improved specificity and predictive value, compared with conventional tests, for prediction of arrhythmic events. AIQP emerged as the best noninvasive univariate predictor of arrhythmic events after myocardial infarction in this study. A review of several other reports shows that AIQP in the present study outperformed the conventional predictive indices reported in those other data sets.
The link between ventricular late potentials recorded in the SAECG and reentrant VT after MI has been well established.1 2 Late potentials represent slow or delayed conduction originating from viable cells within or surrounding myocardial infarct regions. They are defined as abnormal signals that outlast the normal QRS period during normal sinus rhythm. Late potentials result from scarring but are not necessarily a definite marker of reentry. Late potentials may not originate from the actual site of reentry in the case of sustained monomorphic VT. Lack of specificity has led to a low positive PV for arrhythmic events (10% to 25%) in studies of post-MI patients.3 4 5 6 7 8 9 10 Poor sensitivity/PV trade-offs are a common limitation of noninvasive indices of arrhythmia risk, eg, SAECG, Holter monitoring of arrhythmia, and measures of ventricular performance.3 4 5 6 7 8 9 10
This article presents the concept of abnormal intra-QRS potentials, defined as abnormal signals that can occur anywhere within the high-resolution QRS period during normal sinus rhythm. We hypothesize that AIQPs, like conventional late potentials, arise from myocardial infarct regions of scarring and are a potential marker of reentry. AIQPs are low-amplitude notches and slurs that may be invisible in the standard ECG. A computerized method for detection of AIQPs has recently been developed, based on parametric modeling of the high-resolution QRS.11 For many years, notches and slurs visible in the standard ECG have been associated with disruption of ventricular activation in the presence of scarring.12 13 14 15 16 17 However, a link between visible QRS notches and slurs and arrhythmias has not been established. The rationale for a link between AIQPs and arrhythmogenesis is strong, although a formal pathophysiological basis has yet to be established. The premise underlying this link is that abnormal activation related to scarring need not outlast the normal QRS period to participate in or be a marker for a reentry substrate.
Many studies have shown discrepancies between the presence of reentrant VT and the incidence of late potential detection. These include reported lower incidences of late potentials with anterior MI18 and comparative mapping of late potentials from intracardiac and body-surface recordings.19 20 21 In the latter studies, intracardiac recordings of fractionated electrograms at or distal from the site of reentry did not always outlast the normal QRS period in the body-surface SAECG. Most recently, abnormal intra-QRS signals have been implicated in reentrant and focal mechanisms of VT soon after MI.22 23
Other attempts have been made to analyze hypothesized abnormal intra-QRS signals in either the time or frequency domain. Abboud et al24 and Mori-Avi et al25 used frequency-sampled bandpass filtering to assess changes in high-frequency QRS energy in the presence of ischemia. Cain et al26 proposed power spectral analysis as a means of identifying frequency ranges that may be linked to arrhythmogenic signals. In response to the limitations of spectral analysis, we proposed the technique of spectrotemporal mapping,27 28 29 combining information in both the time and frequency domains into a single, joint distribution. Haberl et al30 and Kelen et al31 proposed quantitative approaches to analyzing the SAECG in the time-frequency plane to identify signal features linked to arrhythmia risk. These approaches have technique-dependent limitations,28 29 and their application remains controversial.
In this work, the abnormal intra-QRS signal is considered solely a time-domain phenomenon (ie, a transient, low-amplitude notch or slur), and no assumptions about its frequency characteristics are made. This avoids the pitfalls of bandpass filtering or spectral analysis, which arise because the ECG is poorly characterized in the frequency domain.
This study had two primary objectives: (1) to determine whether an index of abnormal intra-QRS potentials may be predictive of arrhythmic events and (2) to assess whether AIQPs offer significant advantages over conventional measures of late potentials.
A data set of 173 patients presenting to Columbia-Presbyterian Medical Center with an acute MI were studied. Enrollment criteria and characteristics of the data set have been previously reported.5 Patients were followed for a mean duration of 14±7 months. During this period, 16 patients had an arrhythmic end point, defined as sudden cardiac death (n=10), documented sustained VT (n=4), or nonfatal cardiac arrest (n=2). Sudden cardiac death was defined as death within 1 hour of the onset of symptoms in patients without evidence of ongoing ischemia, progressive heart failure, or other serious medical illness or for patients who died in sleep. Patients were divided into two groups: an “arrhythmic-event” group (n=16) and a “no-event” group (n=157). Demographics, medical history, and a 12-lead ECG were obtained on enrollment. An SAECG, 24-hour Holter, and LVEF (preferably by radionuclide methods) were obtained before hospital discharge at a mean of 9.5±5 days after MI. No subjects with ECG evidence of bundle-branch block were included in the data set. Subjects taking β-blockers (45%) or undergoing antiarrhythmic drug therapy (12%) did not have significantly different outcomes from the remaining subjects.5 The SAECG was recorded with orthogonal XYZ leads32 with averaging terminated at a noise end point of 0.3 μV RMS.33 Signal averaging was performed with a Predictor system (Corazonix Corp) using algorithms described previously.34 Data on ventricular arrhythmias were compiled by use of three Holter indices: HPVC, HCPL, and HVT.35 36 37 38
A conventional analysis of the bidirectionally filtered vector magnitude was performed with a bidirectional Butterworth filter with −3 dB cutoff frequencies of 40 Hz (order, 4) and 250 Hz (order, 2).39 The three filtered leads were combined into a vector magnitude waveform.39 Onset and offset of the high-resolution QRS complex were determined automatically34 and subsequently overread to ensure accuracy. Four parameters were measured from the vector magnitude: QRSD, RMS40, LAS, and the noise level present in the ST segment, measured as the RMS value in a 70-ms window.40
Abnormal Intra-QRS Potentials
Abnormal intra-QRS potentials are defined as abnormal signals that occur anywhere within the high-resolution QRS period. This includes the terminal portion, currently defined as late potentials. AIQPs may be thought of as low-amplitude notches and slurs. Langner et al14 15 16 introduced the concept of a high-fidelity ECG in 1952. With an oscilloscope and film recorder, the QRS complex was expanded in time, revealing notches and slurs that were called “high-frequency” components. Flowers et al12 13 showed that the timing of notches, postulated to represent an abrupt change in direction of a ventricular activation wave, correlated with the site of MI. Although notches and slurs have traditionally been considered high-frequency components, we have argued that this terminology is misleading.11 41 Notches and slurs, and hence AIQPs, cannot be isolated by filtering or spectral analysis.11 The frequency-domain description is inadequate since notches and slurs and AIQPs are uniquely characterized in the time domain as transient signals: biphasic or multiphasic spikes, or abrupt changes of gradient.
A new, computerized method has been developed to measure abnormal intra-QRS potentials in the time domain on the basis of parametric modeling.11 The individual-lead SAECG QRS complex is presented unfiltered to be mathematically modeled. Only an approximate knowledge of the QRS onset and offset is needed. The time-domain SAECG signal is preprocessed by decimation to an equivalent sampling rate of 1000 Hz11 and transformed to the frequency domain. This is accomplished via the DCT, defined as where x(k) is the original QRS waveform in the time domain and y(n) is its DCT. The DCT of the QRS, y(n), is then modeled as the impulse response of an ARX parametric model.11 42 This linear model has a rational transfer function. It attempts to fit a smooth waveform to the original signal with a minimum mean squared error. The fit is constrained by the lead-specific model order. Specifically, the ARX model is expressed mathematically as where u(n) is an impulse function, y(n) is the DCT of the original QRS, e(n) is the residual or unpredictable part of y(n), and the set of coefficients ai and bj comprise the model with order (na,nb). The modeled or predictable signal ŷ(n) is computed as a linear regression, expressed conveniently in matrix form as ŷ(n)=φT(n)θ, where the model is θ=(a1a2…anab0b1…bnb) and the transpose of the regression vector is defined as
The modeling process isolates the smooth, predictable part of the QRS. The modeled QRS signal is then restored to the time domain and subtracted from the original QRS signal. The difference is a residual signal that represents the unpredictable, transient part of the QRS, ie, the part that cannot be modeled. The residual is the abnormal intra-QRS signal, or AIQP waveform. It is quantified by computation of the RMS amplitude between the QRS limits.11 The modeling process may be thought of as a form of filtering in which the filter is designed to fit the smooth, predictable part of the QRS.
Selection of the lead-specific ARX model orders is a key step. This was made empirically by use of two test groups of data, the 16 arrhythmic-event patients and 16 randomly chosen no-event patients. A range of models was fitted to each of the 32 subjects. For each lead independently, the model that maximized the ratio of mean AIQP amplitudes between the arrhythmic-event and no-event groups was selected. Low-order models were unable to represent features such as the Q wave adequately. Conversely, high-order models were able to represent some notch and slur features. In both cases, the residual failed to represent AIQPs accurately. The best distinction in mean AIQP amplitudes between the arrhythmic-event and no-event groups was found over a broad range of models of intermediate order. Technical details and full results of this procedure are described in Reference 1111 . As hypothesized, amplitudes of AIQPs were found to be significantly higher statistically in the arrhythmic-event than in the no-event test group. This allowed determination of critical values to discriminate between the two patient test groups. These values are given in Table 1⇓, together with the selected lead-specific model orders. The parameter AIQPXYZ was defined as the presence of an abnormal AIQP value in all leads, ie, AIQPX>9.85 μV, AIQPY>41.6 μV, and AIQPZ>12.0 μV. The model orders in Table 1⇓ are not patient-specific; ie, the same three lead-specific model orders were used for every patient.
Mean values of SAECG parameters between the arrhythmic-event and no-event groups were compared by use of unpaired t tests. A level of P=.05 was considered significant. Similarities between parameters and their tendency to vary independently were assessed with the Spearman rank correlation and concordance (Kendall’s τ b), respectively. We used t tests to determine the significance of each correlation and concordance measurement. A level of P=.05 was considered significant, and P=.01 was considered highly significant. Determination of relative risk and the independence of parameters used as univariate predictors of arrhythmic events was performed with the proportional-hazards (Cox) risk model. Survival curves were computed with the Kaplan-Meier method and assessed for significance with log-rank tests. All statistical analysis was performed with the SAS software package (SAS Institute Inc).
Abnormal Intra-QRS Potentials
Fig 1⇓ is an example of abnormal intra-QRS potential waveforms from the SAECG. Panel a is the X lead from an arrhythmic-event subject with a normal SAECG QRSD (92 ms). The AIQP signal (lower trace) is formed by subtracting the modeled QRS (dashed line) from the original QRS (solid line). A QRS slur is represented in the AIQP waveform. The RMS amplitude of the AIQP signal in the QRS period is 16.7 μV. Panel b, in the same format, is the X lead from a no-event subject with a QRSD also of 92 ms. The modeled complex closely follows the original QRS. The AIQP waveform is uneventful in that there are no significant spikes or transient events. Its RMS amplitude in the QRS period is 4.8 μV.
Table 2⇓ shows the mean RMS amplitudes of AIQPs in the XYZ leads for the arrhythmic-event and no-event groups. Mean AIQP amplitudes are greater in the arrhythmic-event group in all three leads. AIQP values in the Y lead are significantly larger than in the other two leads. The lower, Y lead–specific model order maximized discrimination between arrhythmic-event and no-event subjects but also suggests that because of its ECG morphology, the Y lead is more difficult to model accurately than the other two leads. An F test suggested that variances of the X- and Z-lead data are unequal. Under this assumption, the probability values resulting from a two-tailed t test show only the X-lead AIQP values to be significantly different (P<.05) between the two patient groups. A one-tailed t test testing the initial hypothesis underlying the study—that AIQPs have a greater as opposed to unequal amplitude in the arrhythmic-event group—would halve these probability values and hence would suggest significant differences for both the X and Z leads. Mean QRSDs are also shown in Table 2⇓ and are significantly different between the arrhythmic-event and no-event groups.
Table 3⇓ shows the Spearman rank correlation coefficients between the AIQP parameters and the SAECG QRSD. The correlation was tested for significance (P<.05). The data are presented separately for the arrhythmic-event and no-event groups. In the arrhythmic-event group, AIQP parameters appear to be independent of QRSD. Among the AIQP values, leads X and Z are significantly correlated; however, lead Y appears to be uncorrelated with the other leads. Conversely, in the no-event group, all three XYZ AIQP parameters appear to be correlated with QRSD. Among the AIQP values, all leads appear to be correlated with the exception of lead Y with lead Z. Concordance values, which measure the tendency of the parameters to vary together, follow a pattern very similar to the correlation results for both patient groups. These data suggest that AIQPs and late potentials both tend to be absent in the no-event group. In the arrhythmic-event group, AIQPs and late potentials both tend to be present but may be independent of each other.
Fig 2⇓ shows Kaplan-Meier survival plots for prediction of arrhythmic events by QRSD and by AIQPXYZ, the latter defined above as an abnormal value in all three leads. Abnormal test curves (AIQPXYZ+ and QRSD+) are denoted by solid lines. Normal test curves are denoted by a dotted line (QRSD−) and a dashed line (AIQPXYZ−). Both parameters are significant predictors when tested individually by computing univariate χ2 for the log-rank test (QRSD, P=.046; AIQPXYZ, P<.0001). When the two parameters are considered together, with a stepwise sequence of χ2 for the log-rank test, AIQPXYZ is seen to be the most powerful independent predictor (QRSD, P=.17; AIQPXYZ, P<.0001). The curves are influenced by the loss to follow-up of no-event subjects. From Fig 2⇓, AIQPXYZ outperforms QRSD for both negative and positive test results, suggesting a superior index for prediction of arrhythmic events in this data set.
A Cox proportional-hazards model was set up to assess the risk of developing an arrhythmic event with abnormal QRSD and/or AIQPXYZ parameters. The test of QRSD≥110 ms (n=51) was a borderline significant univariate predictor (P=.055), with an increased risk for arrhythmic events of 2.6 times. AIQPXYZ+ (n=16) was a highly significant univariate predictor (P<.0001), with an increased risk for arrhythmic events of 8.3 times. With a stepwise risk model, AIQPXYZ was the only statistically significant, independent predictor of arrhythmic events when AIQPXYZ and QRSD were considered together (AIQPXYZ: relative risk=7.1, P=.0002; QRSD≥110 ms: relative risk=1.9, P=.20).
The results for QRSD presented in this article differ slightly from those reported by Steinberg et al5 with this data set. The present definition of an abnormal QRSD is QRSD≥110 ms, as opposed to QRSD>110 ms.3 In addition, 173 subjects (with 16 arrhythmic events) were studied as opposed to 182 (also with 16 arrhythmic events) in the original report.3 The omission of 9 no-event subjects by definition had no effect on sensitivity with any of the tests reported and negligible effects on specificity and PVs.
Relative Performances of AIQP and Other Noninvasive Variables for Prediction of Arrhythmic Events
Table 4⇓ lists the clinical performances of the parameters AIQP, QRSD, SAECG (the combination of QRSD, RMS40, and LAS), LVEF, Holter, previous MI, and various logical combinations. Combinations of SAECG, LVEF, and Holter, which were previously reported with this data set by Steinberg et al,5 are shown for comparison with the new AIQP-based data. AIQP parameters combine effectively with a logical and, denoted by AIQPXYZ (AIQPX>9.85 μV RMS andAIQPY>41.6 μV RMS and AIQPZ>12.0 μV RMS). The combination of AIQPXYZ or QRSD≥110 ms yields a modest incremental improvement in the trade-off between sensitivity and +PV over AIQPXYZ alone. However, overall accuracy is reduced, which, taken with the results of the proportional hazards analysis presented above, suggests that AIQPXYZ provides most of the predictive power. A sample of combinations of AIQP and other variables is also presented for the sole purpose of showing the improved efficacy of AIQP compared with the conventional SAECG. AIQPXYZ and H3 (a logical and of the three Holter parameters: HPVC≥10 and HCPL>0 and HVT>0) gave an accuracy of 93%. Specificity and +PV were 100%, with sensitivity at a modest 19%. Conversely, a very high sensitivity of 94% was attained by a logical or combination of AIQPXYZ, H3, QRSD≥120 ms, or LVEF≤30%. Accompanying PVs were 21% (+PV) and 99% (−PV). Specificity (64%) and accuracy (67%) were compromised. Although these sample results cannot be tested for statistical significance (and hence may not be reproducible in other data sets), they compare favorably with those previously reported in this population, significantly outperforming the results of conventional analysis (marked by a dagger [†] in Table 4⇓).5 Table 4⇓ offers a comprehensive view of the best-performing clinical tests with the data set. Other logical combinations of parameters with lower clinical performance were also evaluated but were not included in the table.
Impact of Noise on Clinical Performance of AIQP
None of the lead-specific AIQP parameters correlated with noise by the Spearman rank test. In subjects with low-level terminal QRS activity, ie, RMS40 <30 μV RMS (n=83), QRSD, RMS40, and LAS were significantly correlated both with final averaged noise levels, measured from the vector magnitude, and among each other (P<.0001).
Abnormal Intra-QRS Potentials
The concept of abnormal intra-QRS potentials presupposes that abnormal conduction need not outlast the normal QRS to be linked to arrhythmogenesis. Although a formal pathophysiological basis for AIQP remains to be established, several arguments support this concept. Several authors have reported a lower incidence of late potentials in subjects with VT and anterior MI.18 32 One study documented a normal mean SAECG QRSD in this population.18 In a previous work, we reported similar findings; however, with improved noise reduction, QRSD became identical in anterior and inferior MIs.40 We concluded that late potentials initiate from within the normal QRS period and require a longer relative delay for detection in the case of anterior MI. In one study, late potentials were detected in the SAECG when catheter-mapped late activity from the endocardium coincided with the site of reentrant VT.20 However, when late potentials were not detected, delayed endocardial activity (presumably present within the SAECG QRS) was not coincident with the site of induced VT. This implies that pathophysiological signals, which are a marker for a reentrant substrate, can occur anywhere within the QRS during normal sinus rhythm. Other studies have compared intracardiac and body-surface SAECG recordings.19 21 43 The data presented in these reports suggest that the presence of late potentials is not necessary for the existence of a reentrant substrate. Late potentials also may not be coincident with pathophysiological intracardiac signals.20 21 32 In particular, the report by Simson et al19 demonstrated that endocardial electrograms, with short-lived, fractionated activity that did not outlast the body-surface normal QRS period, could be present in subjects with sustained VT. The concept of abnormal intra-QRS potentials is a logical development of these observations. The study of AIQP extends the search for possible pathophysiological signals to the entire QRS period (including conventional late potentials), as opposed to only the terminal portion.
Comparison of AIQP Results With Previous SAECG Reports
Table 5⇓ reviews the clinical performance of the SAECG and other noninvasive markers of arrhythmia risk for the present study and previous reports. The seven tabulated studies show a wide range in clinical usefulness of the tests reported and in prevalence of arrhythmic events and number of patients studied. In addition, the definition of conventional late potentials—a combination of some or all of QRSD, RMS40, and LAS—varied significantly. Averaged noise levels, which can significantly affect the sensitivity/specificity trade-off,40 also varied significantly and were mostly unreported. The clinical performance data have in many instances been calculated from other data presented in the original reports. The table presents a complete, comparative performance overview of the studies.
AIQP emerged as the best univariate predictor of arrhythmic events, with a higher accuracy and specificity and a better trade-off in sensitivity/+PV than either conventional late potentials (SAECG), heart rate variability, Holter, or left ventricular performance measures. Negative PV was also at least comparable to that of other tests, despite the high specificity of AIQP. Positive PV depends heavily on the prevalence of arrhythmic events, as can be seen from contrasting the studies of Gomes et al6 (prevalence, 14.7% to 18.3%; +PV, 29% to 50%) and Savard et al10 (prevalence, 3.3%; +PV, 7% to 11%). In this case, accuracy is the best overall figure of merit for a test. AIQP has a favorable accuracy (90%) compared with SAECG (mean of six studies, 71%) and has comparable or better accuracy than most combinations of tests.
AIQPs may prove to be a more reliable and comprehensive measure of abnormal activation than conventional late potentials. The incidence of late potentials is expected to be sensitive to averaged noise levels and infarct location.40 In contrast, AIQPs will be detected anywhere within the high-resolution QRS period, including the terminal portion, to which the search for abnormal potentials is currently restricted. Therefore, AIQPs do not depend on infarct location. AIQPs are also little affected by moderate averaged noise levels, as verified by no correlation between AIQPs and noise. The modeling process does not require exact determination of QRS onset and offset, relaxing the most significant technical constraint of SAECG analysis. Finally, AIQPs could probably be estimated accurately from a smaller, noisier ensemble average than the 200 to 600 beats presently needed to reach a 0.3-μV RMS noise end point. Although further work is needed to find the appropriate noise end point for AIQP analysis, a simplified signal averaging procedure might be an advantage for the AIQP method.
In this study, we compare the performance of AIQP with a definition of conventional late potentials based on QRSD alone. This sole use of QRSD as the performance maximum with SAECG requires justification. In comparisons of AIQPs and conventional SAECG indices, Table 5⇑ raises the question of the relative importance of AIQP, QRSD, RMS40, and LAS. In this and previous data sets,40 inclusion of RMS40 or LAS degraded the clinical performance of the SAECG compared with QRSD alone. This was due to the strong correlations among QRSD, RMS40, LAS, and noise. The data in Table 5⇑ support the assertion that QRSD alone is a more accurate index of arrhythmia risk than any combination of the three SAECG parameters. Savard et al10 found QRSD alone to be the single most useful SAECG index in their study of 2461 patients. We also calculated the clinical performance characteristics of QRSD alone from data presented in the article by Kuchar et al.7 In that study, accuracy for QRSD improved by 20 percentage points over SAECG (the best combination of QRSD, RMS40, and LAS). AIQPs gain a potential advantage over QRSD (and hence over the conventional SAECG) in that “late potentials” arising from abnormal conduction in the early part of the QRS may not outlast the normal QRS period and hence will not be detected by conventional means.
The usefulness of the Holter parameters HPVC and HVT (number of PCVs/h≥10 and presence of runs of VT) was not consistent between studies. HVT was measured in 5 studies (Kleiger et al,44 Gomes et al,6 Kuchar et al,7 Farrell et al,9 and the present study) and was found to be a significant index of arrhythmic events in 4 of them (all except Gomes et al6 ). HPVC was measured in 6 studies (all except Savard et al10 ) and was found to be a significant index in 4 (Kleiger et al,44 Cripps et al,8 Farrell et al,9 and the present study). A similar variability in clinical usefulness was found in these studies3 4 5 6 7 8 9 10 for other parameters, such as LVEF, Killip class, and the presence of a previous MI. In the case of Holter, use of the Lown classification scheme may have increased the variability of the results. As suggested by Kuchar et al,7 HPVC may have an association with the likelihood of triggered automaticity. In this context, other factors, such as the relative prevalence of VT, ventricular fibrillation, and sudden death events and time of recording after MI may have affected the measured values and significance of HPVC.
Variability among results of the noninvasive tests is the most prominent feature of Table 5⇑. It is probably accounted for in large part by natural variability (population characteristics, prevalence of arrhythmic events, time of recording, etc) and methodological variability (technical factors such as noise, different measurement algorithms, and different definitions of abnormalcy).
Methodological Considerations for Prospective Studies
AIQP, like late potentials, has a signal processing definition. From this perspective, AIQPs are a logical extension of the late potentials concept. However, a significant advantage of the AIQP concept is that it has a hypothesized pathophysiological basis that may be examined in either experimental or clinical settings. Such studies and future prospective clinical studies are needed to verify whether the increased clinical usefulness of AIQPs over conventional late potentials in this study can be reproduced.
It should be noted that the AIQP modeling procedure has been designed retrospectively with the present data set. The most important step, selecting the lead-specific model order, is fully developed in a technical article by us.11 Prospective studies are needed to evaluate the impact of our predetermined model orders on the clinical performance of AIQP indices in other data sets.
Abnormal intra-QRS potentials extend the concept of late potentials from the terminal portion to the entire QRS complex and provide a new predictive index of arrhythmic events. AIQPs significantly enhanced the clinical efficacy of the SAECG compared with conventional parameters. The present retrospective design yielded a sensitivity, positive PV, and accuracy of 44%, 47%, and 90%, respectively. These results were achieved in a post-MI population of 173 patients, with a prevalence of arrhythmic events of 9%. This performance is a significant improvement over conventional noninvasive, univariate indices of arrhythmia risk. AIQPs may potentially relax the technical requirements of the SAECG by reducing dependence of the measured values on (1) noise reduction by signal averaging and (2) exact determination of QRS limits. Given the present retrospective experimental design, further investigation is needed both in elaboration of a pathophysiological basis of AIQP and in prospective testing with larger clinical populations.
Selected Abbreviations and Acronyms
|AIQP||=||abnormal intra-QRS potential|
|ARX||=||autoregressive moving average|
|DCT||=||discrete cosine transform|
|HCPL||=||Holter No. of couplets/24 h|
|HPVC||=||Holter No. of premature ventricular complexes/h|
|HVT||=||Holter No. of runs of nonsustained VT/24 h (≥3 beats at >100 bpm)|
|LAS||=||duration of terminal QRS signal <40 μV (low-amplitude signal)|
|LVEF||=||left ventricular ejection fraction|
|RMS||=||root mean square|
|RMS40||=||RMS amplitude of terminal 40 ms of QRS|
- Received March 4, 1996.
- Revision received November 12, 1996.
- Accepted November 12, 1996.
- Copyright © 1997 by American Heart Association
Gomes JA, ed. Signal-Averaged Electrocardiography: Concepts, Methods and Applications. Dordrecht, Netherlands: Kluwer Academic Publications; 1993.
El-Sherif N, Turrito G, eds. High Resolution Electrocardiography. Mt Kisco, NY: Futura Publishing Co; 1992.
Kanovsky MS, Falcone RA, Dresden CA, Josephson ME, Simson MB. Identification of patients with ventricular tachycardia after myocardial infarction: signal averaged electrocardiogram, Holter monitoring, and cardiac catheterization. Circulation. 1984;70:264-270.
Buckingham TA, Ghosh S, Homan SM, Thessen CC, Redd DM, Stevens LL, Chaitman BR, Kennedy HL. Independent value of signal-averaged electrocardiography and left ventricular function in identifying patients with sustained ventricular tachycardia with coronary artery disease. Am J Cardiol. 1987;59:568-572.
Gomes JA, Winters SL, Stewart D, Horowitz S, Milner M, Barreca P. A new noninvasive index to predict sustained ventricular tachycardia and sudden death in the first year after myocardial infarction: based on signal-averaged electrocardiogram, radionuclide ejection fraction and Holter monitoring. J Am Coll Cardiol. 1987;10:349-357.
Cripps T, Bennett ED, Camm AJ, Ward DE. High gain signal averaged electrocardiogram combined with 24-hour monitoring in patients early after myocardial infarction for bedside prediction of arrhythmic events. Br Heart J. 1988;60:181-187.
Farrell TG, Bashir Y, Cripps T, Malik M, Poloniecke J, Bennett ED, Ward DE, Camm AJ. Risk stratification for arrhythmic events in postinfarction patients based on heart rate variability, ambulatory electrocardiographic variables and the signal-averaged electrocardiogram. J Am Coll Cardiol. 1991;18:687-697.
Savard P, Rouleau JL, Davies RF, Dupuis R, Gardner M, Lauzon C, Poitras N, Stewart DJ, Sussex B, Talajic M, Warnica WJ. Prediction of arrhythmic events after myocardial infarction using signal-averaged ECG criteria specific for gender, age and myocardial infarction type. Circulation. 1994;90(suppl I):I-389. Abstract.
Gomis P, Jones DL, Caminal P, Berbari EJ, Lander P. Analysis of abnormal signals within the QRS complex of the high resolution electrocardiogram. IEEE Trans Biomed Eng. In press.
Flowers NC, Horan LG, Thomas JR, Tolleson WG. The anatomic basis for high frequency components in the electrocardiogram. Circulation. 1969;39:531-539.
Flowers NC, Horan LG, Tolleson WG, Thomas JR. Localization of the site of myocardial scarring in man by high-frequency components. Circulation. 1969;40:927-934.
Langner PH. The value of high fidelity electrocardiography using the cathode ray oscillograph and an expanded time scale. Am Heart J. 1952;5:249-256.
Langner PH. Further studies in high fidelity electrocardiography: myocardial infarction. Circulation. 1953;8:905-911.
Langner PH, Geselowitz DB, Mansure FT. High-frequency components in the electrocardiograms of normal subjects and of patients with coronary heart disease. Am Heart J. 1961;62:749-755.
Wagner GS, Freye CJ, Palmeri P, Roarke SF, Stack NC, Ideker RE, Harrell FE, Selvester RH. Evaluation of a QRS scoring system for estimating myocardial infarct size, I: specificity and observer agreement. Circulation. 1982;65:342-347.
Simson MB, Untereker WJ, Spielman SR, Horowitz LN, Marcus NH, Falcone RA, Harken AH, Josephson ME. Relation between late potentials on the body surface and directly recorded fragmented electrograms in patients with ventricular tachycardia. Am J Cardiol. 1982;51:105-112.
Schwarzmaier HJ, Karbenn U, Borggrefe M, Ostermeyer J, Breithardt G. Relation between ventricular late endocardial activity during intraoperative endocardial mapping and low-amplitude signals within the terminal QRS complex on the signal-averaged surface electrocardiogram. Am J Cardiol. 1990;66:306-314.
Pogwizd SM, Hoyt RF, Saffitz JE, Corr PB, Cox JL, Cain ME. Reentrant and focal mechanisms underlying ventricular tachycardia in the human heart. Circulation. 1992;86:1872-1887.
Hood MA, Pogwizd SM, Patrick J, Cain ME. Contribution of myocardium responsible for ventricular tachycardia to abnormalities detected by analysis of signal-averaged electrocardiograms. Circulation. 1992;86:1888-1901.
Abboud S, Cohen RJ, Selwyn A, Ganz P, Sadeh D, Friedman PL. Detection of transient myocardial ischemia by computer analysis of standard and signal-averaged high frequency electrocardiograms in patients undergoing percutaneous transluminal coronary angioplasty. Circulation. 1987;76:585-596.
Mori-Avi V, Shargorodsky B, Abboud S, Laniado S, Akselrod S. Effects of coronary occlusion on high frequency content of the epicardial electrogram and body surface electrocardiogram. Circulation. 1987;76:237-243.
Cain ME, Ambos D, Witkowski FX, Sobel BE. Fast-Fourier transform analysis of signal-averaged electrocardiograms for identification of patients prone to sustained ventricular tachycardia. Circulation. 1984;69:711-720.
Lander P, Albert DE, Berbari EJ. Spectro-temporal mapping: the next generation in late potential analysis. Am J Coll Cardiol. 1988;11:119A. Abstract.
Lander P, Albert DE, Berbari EJ. Spectro-temporal analysis of ventricular late potentials. J Electrocardiol. 1990;23:95-108.
Lander P, Albert D, Berbari EJ. Principles of frequency domain analysis. In: El-Sherif N, ed. High Resolution Electrocardiography. Mt Kisco, NY: Futura Publishing; 1988:105-128.
Haberl R, Schels HF, Steinbigler P, Jilge G, Steinbeck G. Top-resolution frequency analysis of electrocardiogram with adaptive frequency determination. Circulation. 1990;82:1183-1192.
Bigger JT, Fleiss JL, Kleiger R, Miller JP, Rolnitzky LM. Multicenter postinfarction research group. The relationships among ventricular arrhythmias, left ventricular dysfunction, and mortality in the 2 years after myocardial infarction. Circulation. 1984;69:250-258.
Armstrong WF, McHenry PL. Ambulatory electrocardiographic monitoring: can we predict sudden death? J Am Coll Cardiol. 1985;5:13B-16B.
Simson MB. Use of signals in the terminal QRS complex to identify patients with ventricular tachycardia after myocardial infarction. Circulation. 1981;64:235-242.
Lander P, Berbari EJ, Rajagopalan CV, Vatterott P, Lazzara R. Critical analysis of the signal-averaged electrocardiogram: improved identification of late potentials. Circulation. 1993;87:105-117.
Lander P, Jones DL, Berbari EJ, Lazzara R. Time-frequency structure of the high-resolution ECG. J Electrocardiol. 1994;27:207-212.
Ljung L. System Identification: Theory for the User. Englewood Cliffs, NJ: Prentice-Hall; 1987:chaps 4 and 10.
Vaitkus PT, Kindwall KE, Marchlinski FE, Miller JM, Buxton AE, Josephson ME. Differences in electrophysiological substrate in patients with coronary artery disease and cardiac arrest or ventricular tachycardia: insights from endocardial mapping and signal-averaged electrocardiography. Circulation. 1991;84:672-678.