Automatic Arrhythmia Identification Using Analysis of the Atrioventricular Association
Application to a New Generation of Implantable Defibrillators
Background Atrioventricular association is a key criterion for arrhythmia diagnosis. Its use in a defibrillator should significantly reduce the incidence of inappropriate shocks. Therefore, we evaluated the diagnostic accuracy of an algorithm that uses dual-chamber sensing and analysis of atrioventricular association to discriminate ventricular from supraventricular arrhythmias in a prototype of an implantable defibrillator.
Methods and Results The algorithm performed a stepwise analysis of arrhythmias. The rhythm was first classified on the basis of cycle lengths. Each episode was then classified as supraventricular or ventricular in origin on the basis of the stability of cycle lengths and atrioventricular association. This algorithm was evaluated in 156 episodes of induced sustained tachycardias. Eighty-nine tachycardias were taken from the Ann Arbor electrogram library; the others were recorded in 50 patients during electrophysiological studies. The atrial and ventricular signals were stored on an external recorder and then injected into an external prototype of a defibrillator system. The algorithm correctly diagnosed 96% of ventricular tachycardia episodes, 100% of ventricular fibrillation episodes, and 92% of double-tachycardia episodes. The mean detection time for ventricular tachycardia was 2.6±0.8 seconds, and for ventricular fibrillation, it was 2.1±0.4 seconds. The positive predictive values for the diagnoses of atrial fibrillation and atrial flutter were 92% and 86%, respectively. For ventricular tachycardia and ventricular fibrillation, the values were 95% and 100%, respectively.
Conclusions Analysis of atrioventricular association promotes reliable differentiation between ventricular and supraventricular tachycardias and should enhance the diagnostic capabilities of implantable defibrillators.
Inappropriate shocks for supraventricular arrhythmias are a common clinical problem with the current generation of implantable cardioverter-defibrillators.1 2 These are known to occur in up to 20% of all patients and in 25% of those with a history of atrial fibrillation.3 4 Various methods, including incorporation of algorithms using onset and stability criteria,5 electrogram configuration,6 and hemodynamic sensing,7 have been tested to overcome this problem while still remaining in a single-chamber ventricular architecture. On the surface ECG, an important criterion for diagnostic differentiation is the analysis of atrioventricular association, and it has been shown that algorithms that incorporate dual-chamber sensing can significantly improve differentiation between supraventricular and ventricular arrhythmias.8 9 10
We report the ability of dual-chamber sensing to differentiate ventricular arrhythmias from supraventricular arrhythmias in a prototype of an implantable defibrillator (Defender 9001, ELA Medical).
The algorithm that was tested in this study uses atrial and ventricular signals and performs a tiered analysis of the arrhythmia in three steps: (1) cycle-length sorting; (2) majority rhythm identification; and (3) VT/SVT/ST sorting. The algorithm uses four criteria to detect and classify tachyarrhythmias: (1) ventricular rate; (2) stability of the ventricular (VV) intervals; (3) type of AV association; and (4) acceleration magnitude and origin.
This step aims to identify the occurrence of an arrhythmia and to analyze its type. Each ventricular cycle is classified according to its coupling time into one of the following categories: SR, VT, or VF. The algorithm measures the VV interval and compares it with two programmable boundaries: the VT cycle length and the VF cycle length. Each cycle is identified as follows: (1) SR cycle if the corresponding VV interval is longer than both the VT and the VF cycle lengths; (2) VT cycle if the VV interval is longer than the VF cycle length and shorter than or equal to the VT cycle length; or (3) VF cycle if the corresponding VV interval is shorter than or equal to the VF cycle length. In the present study, the VT and VF cycle lengths were programmed to 500 and 250 ms, respectively.
Majority Rhythm Identification
This step uses two programmable parameters: (1) Y number of recent cycles taken into account and (2) X percentage of those cycles. The algorithm identifies three categories of majority rhythm: (1) SR majority is detected when X% of the last Y cycles are classified as SR; (2) VT majority is detected when X% of the last Y cycles are classified as VT or VF; and (3) VF majority is detected when X% of the last Y cycles are classified as VF. This rhythm classification is established after each new cycle and always takes into account the last Y cycles. In the present study, the values for X and Y were programmed to 75% and 8 cycles, respectively.
Once a VT majority rhythm is identified, this diagnosis is either confirmed or changed to SVT on the basis of the stability of the VV intervals and on the basis of AV association. SVT as defined herein includes ST, atrial tachycardia, reentrant SVT, atrial fibrillation, and atrial flutter. The VV intervals are classified as stable if the majority of recent VT or VF intervals do not vary by more than a programmed maximum limit (in this study, 63 ms). To assess the AV association, the algorithm measures all the AV intervals within those recent VT or VF cycles. It then identifies whether the AV intervals are stable or unstable. If the intervals are unstable, AV dissociation is diagnosed. In case of stable AV intervals, the next step involves finding the type of AV association, ie, 1:1 or N:1. If there is only one set of stable intervals within the population of analyzed AV intervals, then the AV association is characterized as 1:1. If more than one set of stable AV intervals is found for a given number of VV cycles, the AV association is classified as N:1. The procedure is given below:
To assess this criterion, the algorithm constructs a histogram of VV intervals of cycles classified as VT or VF from the last Y ventricular cycles. The algorithm scans the VV-interval histogram with a window set to a programmed width (63 ms), looking for the position of the window that encompasses a maximum number of intervals in the histogram (Fig 1⇓). When it finds that position, it counts the number of VV intervals in the window. If this number is greater than or equal to X% of Y (in this case, programmed to 75% of 8), the rhythm is classified as stable. In this case, additional analysis is performed to discriminate VT from SVT. All tachycardias with an unstable ventricular rhythm are classified as atrial fibrillation, and no additional criterion is required.
AV association (Figs 2⇓ and 3).⇓ This criterion is evaluated only if the assessment of VV stability leads to the conclusion that VV intervals are stable
To evaluate the AV association criterion, the algorithm uses the VV-interval histogram already constructed for assessing stability and constructs an AV-interval histogram corresponding to the cycles stored within the VV histogram. Up to five AV intervals can be stored for each VV cycle. This corresponds to the memorizing capabilities of the device. For example, if the ratio is 5:1, five AV intervals can be stored for one VV interval. Like the VV histogram, the AV-interval histogram is scanned with a 63-ms window. AV association is defined as established when the ratio of the number of stable AV intervals to the number of stable VV intervals is ≥X% (in the present study, 75%). When AV association is established, it is further classified as 1:1 when the ratio of the number of stable AV intervals to the total number of AV intervals in the histogram is ≥75%; otherwise, it is classified as N:1.
In summary, in the present study, a stable ventricular rhythm without AV association was identified as VT. A stable ventricular rhythm with N:1 AV association was classified as SVT.
All the arrhythmias tested in the present study were induced during electrophysiological studies and injected into the external prototype of the defibrillator after the arrhythmia was established. As a result, the fourth criterion (ie, acceleration magnitude and chamber of origin of the accelerated beat) could not be used for arrhythmia sorting. Therefore, a stable ventricular rhythm with 1:1 AV association was not subjected to additional analysis and was not classified as either VT or SVT.
Analysis of Clinically Induced Tachycardias
The algorithm was prospectively evaluated in 156 episodes of induced sustained (defined as a duration of ≥6 seconds) tachycardias. Among these, 89 episodes from 43 patients were taken from the Ann Arbor, Mich, electrogram libraries. The other 67 arrhythmias were recorded in 50 patients during electrophysiological studies performed for the diagnosis of clinical arrhythmias. The atrial and ventricular signals were recorded on an external analog recorder (TEAC R-71) via bipolar-electrode catheter with 5- or 10-mm interelectrode distance after amplification (gain=×100) and filtering (band pass=0.5 to 1000 Hz) and were then injected (after attenuation by 100) into the external prototype of the implantable-defibrillator system (Defender 9001). Signals from the Ann Arbor database were recorded via bipolar-electrode catheter with 10-mm interelectrode distance on FM magnetic tape at a tape speed of 9.5 cm/s (Hewlett-Packard model 3968A) after amplification at 1 to 500 Hz. Amplifier gain and filter settings were held constant during the entire recording procedure, and a 1-mV calibration signal was entered as a reference at the time of recording.
All values are expressed as mean±SD. The PPV was calculated using the equation PPV=True-Positive/True-Positive+False-Positive×100. Negative predictive value (NPV) was calculated as NPV=True-Negative/True-Negative+False-Negative.
The distribution of arrhythmia types, their cycle lengths, and respective durations are given in Table 1⇓. The ability of the algorithm to characterize the various arrhythmias is summarized in Table 2⇓. Nominal settings for the present study have been chosen as 63 ms for definition of stability and 75% for the value of stable RR because in our previous (unpublished) experience, these were the values for which the sensibility for VT detection was 100%.
Of the 85 episodes of VT, a correct diagnosis was made in 82 cases, giving a sensitivity of 96%. The device failed to diagnose VT in 3 cases. In 1 case, the error was due to a slow VT with a cycle length of 560 ms, whereas the programmed lower limit for VT detection was 500 ms. In another case, VT with a cycle length of 420 ms was erroneously classified because of unstable VV cycles. In the third case, oversensing in the ventricle occurred due to tape artifacts. Three episodes of atrial fibrillation occurring in 3 patients with WPW syndrome were detected as stable and dissociated and therefore diagnosed as VT because the rapidity of the ventricular response led to VV-interval variations that were less than that required for classification as an unstable VV cycle. One case of atrial flutter was wrongly labeled as stable with no AV association because of noise in the atrial channel. The specificity of the algorithm for diagnosing VT was thus 94%. The mean VT detection time was 2.6±0.8 seconds (range, 1.8 to 4.4 seconds). In none of the episodes was the detection time >5 seconds.
All 22 episodes of ventricular flutter or VF were correctly identified, and there was no instance of SVT being misdiagnosed as VF. Thus, the sensitivity and specificity for VF diagnosis were both 100%. The mean VF detection time was 2.1±0.4 seconds (range, 1.5 to 2.9 seconds).
Bitachycardia (Double-Tachycardia Episodes With Simultaneously Occurring Atrial Tachycardia and VT)
Seven VT and 2 VF episodes with atrial fibrillation and 3 VT episodes with associated atrial flutter were classified as bitachycardia. Of 12 episodes of bitachycardia, 11 were correctly identified as having VT. In 1 case, fast VT (VV cycle=360 ms) was accompanied by atrial fibrillation and an N:1 association was detected, leading to erroneous diagnosis as SVT.
Of the 15 episodes of atrial fibrillation, 12 were correctly identified. As mentioned previously, 3 episodes in three patients with WPW syndrome were diagnosed as VT. Their VV cycle lengths were respectively 270, 310, and 330 ms, and because the irregularity criterion best applies to relatively slower rhythms, they were detected as stable and dissociated. The mean detection time for AF was 3.1±1.9 seconds (range, 2.4 to 6.8 seconds).
Of the 12 cases of atrial flutter, 6 were correctly identified. In 5 episodes in four patients, because every second atrial electrogram fell in the postventricular atrial absolute refractory period, the episodes were classified as stable tachycardia having a 1:1 AV association and were not further classified. In another patient, an episode of atrial flutter was detected as stable without AV association because of noise in the atrial channel. The mean detection time for atrial flutter was 2.9±3.1 seconds (range, 2.2 to 12.2 seconds).
ST and Atrial Tachycardias
All 10 cases of these arrhythmias were not identified as such but were classified as being stable, with 1:1 AV association. The mean detection time in this case was 2.4±0.7 seconds (range, 1.7 to 3.5 seconds). Additional sorting of these rhythms would require analysis of acceleration magnitude and chamber of origin, which could not be used in the present study, as mentioned above.
Importance of Nominal Settings
Other values of VV stability or of a different percentage of the number of ventricular cycles have been tested, and specificity for atrial fibrillation detection has always been impaired, a posteriori confirming the appropriate nature of nominal settings in the present study.
The inability to differentiate supraventricular arrhythmias such as atrial fibrillation from ventricular tachyarrhythmias is a frequent clinical problem in the present generation of cardioverter-defibrillators.1 2 The present study is the first report in which an algorithm that uses monitoring of the atrial and ventricular signals and analysis of atrioventricular association was tested. This led to excellent and potentially better discrimination between ventricular and supraventricular arrhythmias.
Several studies have reported a high incidence of inappropriate shocks for atrial fibrillation and for ST, especially during exercise.3 4 In clinical practice, recognition of the problem frequently leads to the prescription of AV node–blocking drugs and the programming of high rates for cutoff. Davies et al11 suggested that a computer algorithm based on changes in atrial and ventricular electrogram morphology might be useful for the differentiation of various arrhythmias. However, the potential for having similar ventricular electrograms during ventricular and supraventricular rhythms was a major limitation to the use of electrogram morphology as the sole means of arrhythmia diagnosis.
An important step toward circumventing the problem of inappropriate shocks was the use of defibrillators with the ability to store electrograms preceding and following shocks. Hook and Marchlinski12 were able to correctly diagnose the cause for inappropriate shocks using such a device. This enabled them to do appropriate reprogramming that prevented recurrence of such discharges. Because this was a postevent analysis, the diagnosis required at least one inappropriate discharge; the clinical usefulness of this facility was thus limited.
Among various diagnostic criteria proposed for distinguishing VT from atrial fibrillation and sinus acceleration has been the stability of RR intervals.13 14 Recently, Swerdlow et al5 6 reported the usefulness of onset and stability criteria for enhancing the diagnostic capabilities of cardioverter-defibrillators. These criteria were seen to decrease the incidence of erroneous diagnosis of atrial fibrillation to <5%. Also, this algorithm resulted in the rejection of 98% of sinus acceleration episodes. However, underdetection of VT was seen to occur in 9.6% of the cases when this algorithm was used.
The values of X and Y were programmed on precedent (and unpublished) experience with the use of computer simulation. The use of eight cycles allowed rapid detection yet was specific enough for fair detection of atrial fibrillation. Seventy-five percent of the VV intervals was chosen to account for potential overdetection in the ventricular channel (such as gain control problems or myopotential detection) and for the AV interval to take into account the potential VA cross-talk. Eight cycles were required in part due to computing constraints (an even number is mandatory) and after a trial-and-error experience during simulation. It appeared that the use of eight cycles was the best compromise between specificity and sensitivity with a bias toward sensitivity of detection. The duration of 500 ms for VT detection was chosen to allow identification of slow VTs. We believed that this was logical for correct, precise identification of the ongoing rhythm when concomitant monitoring of the atrial channel was used. One case of VT was nevertheless slower and therefore not correctly identified. The longest (yet immediately shorter) cycle for VT was 450 ms, and all three cases of this cycle length were correctly identified. SVTs with a cycle length of 480 ms were SVTs with 1:1 AV association (excluded) and one episode of atrial flutter. The VF cycle length was based on the likelihood of shock necessity.
We have reported on the sensitivity and specificity of diagnosis of arrhythmia using the same set of parameters for all arrhythmia episodes. This was done to ensure uniformity of parameters used by the algorithm to test for a variety of clinical ventricular and supraventricular arrhythmias. Thus, in three episodes of VT and in one episode of bitachycardia, the diagnosis was missed. However, all the parameters in this algorithm are programmable, and two of the episodes of VT were particularly slow (hence, probably not immediately life-threatening in the real-life situation) and were correctly identified after reprogramming of cycle length and stability criteria. Therefore, the only missed diagnosis occurred in a case of VT because of defective signals due to tape artifacts and in a case of bitachycardia that was incorrectly labeled as N:1 SVT. In the former case, this should not occur with an implanted device; in the latter case, it is expected that in a real-life situation, the additional criterion of magnitude and chamber of origin of acceleration (not described herein) will help to identify where the arrhythmia started (ie, atria first=SVT). This criterion was not used in the present study because of the protocol design that allowed only analysis of established episodes. It should be emphasized that even though the results of arrhythmia analysis in this study are encouraging, even better results are likely with the use of the latter criterion.
Analysis of atrioventricular association has been an important method for the diagnosis of tachyarrhythmias, both on the surface ECG as well as during electrophysiological studies. Schuger et al8 demonstrated that the simple criterion of differences in atrial and ventricular cycle lengths was capable of identifying 83% of episodes of sustained VTs and accurately differentiating them from SVT. Arrhythmia classification with the use of two intracardiac leads and dual-chamber sensing was reported to have a diagnostic accuracy of >90% in algorithms tested by Leong and Jabri.9 Polikaitis and Arzbaecher10 and Arzbaecher et al15 described the use of an algorithm for tachycardia identification using dual-chamber sensing. Diagnoses made with the use of that algorithm were based on the relative rates of atrial and ventricular electrograms, and the algorithm was able to successfully detect 21 of the 22 arrhythmias tested. Recognition of VT in the presence of atrial flutter or fibrillation continued to be a problem with this algorithm, although the addition of rate stability and multiplicity measures was found to increase its diagnostic accuracy.16 More recently, Militiann et al17 reported the usefulness of an algorithm based on analysis of AV intervals for recognition of VT with 1:1 atrial conduction.
We have previously reported the use of dual-chamber analysis for arrhythmia identification.18 Although differences in methodology and patient populations render comparison between studies difficult, we were able to achieve a sensitivity of 97% and a specificity of 98% for the diagnosis of ventricular arrhythmias with nominal programming parameters using an algorithm that incorporated analysis of AV association in addition to the rate and stability criteria. Also, this algorithm was able to identify atrial fibrillation with a PPV of 92% and atrial flutter episodes with a PPV of 86%. Apart from one instance of erroneous diagnosis due to an episode of VT failing to meet the rate criterion and another due to technically defective signals, the only case of VT in which the diagnosis was erroneous occurred in a patient with fast VT and atrial fibrillation. In that case, N:1 association was detected and the tachycardia was classified as supraventricular. More importantly, except for atrial fibrillation occurring in three patients with WPW syndrome and one case of atrial flutter with noise in the atrial channel, no case of SVT was classified as a ventricular arrhythmia. This should lead to a substantial reduction in the number of inappropriate shocks. The patients with WPW syndrome constitute a rare but special group for whom careful programming would be necessary.
An added advantage of this algorithm over the previously tested dual-chamber analyses has been the ability to classify bitachycardia accurately in the VT zone. Eleven of 12 episodes of bitachycardia were correctly classified in the present study. One of the concerns when complex algorithms are used has been the possibility of longer detection times. The mean VT and VF detection times, however, were quite short in the present study. In none of the episodes of VT or VF was the detection time >5 seconds.
One limitation of this study is the potential occurrence of VTs that are unstable, as is often the case in paroxysmal monomorphic VT. This tachycardia usually occurs in a well-defined setting, which is likely to be known before eventual implant. Because it also is usually relatively irregular, stability can be programmed as high (ie, 125 ms) or even canceled. But if all algorithms have limitations, a rate-only algorithm will either fire (and therefore likely fire inappropriately) or not fire if the cutoff rate is high enough, thereby impairing the sensible use of this algorithm. Our algorithm allows identification of a rhythm that although above the VT rate, may not trigger shock delivery because it is either irregular or dissociated from the atrium.
Another limitation of this study was the inability to use the onset criterion, such that rhythms with long VA Wenckebach sequences or 1:1 AV association were not subclassified. Thus, 5 episodes of atrial flutter and 10 episodes of SVT in which 1:1 AV association was detected could not be classified. The onset criterion was not used in the present study because all the rhythms that were fed to the external prototype were established tachyarrhythmias. This limited the ability to differentiate SVT from VT with 1:1 VA conduction. Although Huagul et al19 found that occurrence of VT with 1:1 VA conduction is rare in patients with implanted defibrillators, the incorporation of the onset criterion would significantly enhance the diagnostic capabilities of this algorithm. Dual-chamber analysis would also significantly reduce the underdetection that has been seen when onset and stability criteria are applied in single-chamber analysis because (1) these criteria are applied at a later stage of arrhythmia sorting and (2) the capability of analyzing the chamber of onset would be available. However, the usefulness of atrial sensing in detecting atrial arrhythmias on a long-term basis in implanted devices needs to be evaluated. Whether incorporation of atrial monitoring is of true additional benefit has not been addressed in depth in the present study. It was addressed in a previous study that used an earlier version of the algorithm. In a smaller patient population, additional atrial channel analysis did not affect VT detection but did increase specificity of SVT identification from 65% to 95%.18 A study comparing single- and dual-chamber analysis using the complete algorithm is currently ongoing in implanted devices. Only then will a fair comparison be available between both types of analysis of the same arrhythmia episodes.
Incorporation of atrial electrogram monitoring and analysis of AV association allow sensitive and specific differentiation between VT and SVT. Analysis of magnitude and origin of acceleration at onset should further increase diagnostic yield, especially in cases of tachycardias with 1:1 AV association. This is particularly important at the beginning of an era during which sophisticated defibrillators will surely need the information obtained from an atrial lead not only for accurate acute tachycardia identification but also for retrospective arrhythmia analysis and possibly for dual-chamber pacing therapy.
Selected Abbreviations and Acronyms
|PPV||=||positive predictive value|
CHG Aix en Provence, France (Dr J.C. Barnay); Cardiological Center of Aalst, Belgium (Prof P. Brugada); Hopital du haut Le´ve`que, Bordeaux, France (Prof J. Clementy); Fort Lauderdale (Fla) Hospital (Dr R. Luceri, Dr P. Ritter); Hopital Nord, Marseille, France (Dr M. Gueunoun, Dr F. Lauribe, Prof S. Levy); Hopital Jean Rostand, Ivry, France (Dr R. Franck, Dr G. Lascault); Hopital Lariboisie`re, Paris, France (Prof A. Leenhardt, Dr B. Cauchemez); and CHU Rouen, France (Dr M. Nair, Prof N. Saoudi, Prof B. Letac).
We would like to thank Re´mi Nitzsche, Christine Henry, and Peter Jacobson for their technical support and help in preparing the manuscript.
- Received January 11, 1996.
- Revision received September 25, 1996.
- Accepted October 7, 1996.
- Copyright © 1997 by American Heart Association
Kelly PA, Cannom DS, Garan H, Mirabal GS, Harthorne JW, Hurvitz RJ, Vlahakes GJ, Jacobs ML, Ilvento JP, Buckley MJ, Ruskin JN. The automatic implantable cardioverter-defibrillator: efficacy, complications and survival in patients with malignant ventricular arrhythmias. J Am Coll Cardiol. 1988;11:1278-1286.
Nunain SO, Roelke M, Trouton T, Osswald S, Kim YH, Sosa Suarez G, Brooks DR, MacGovern B, Guy M, Torchiana DF, Vlahakes GJ, Garan H, Ruskin JN. Limitations and late complications of third-generation cardioverter-defibrillators. Circulation. 1995;91:2204-2213.
Grimm W. Electrocardiographically documented unnecessary, spontaneous shocks in 241 patients with implantable cardioverter defibrillators. PACE. 1992;15:1667-1673.
CEDARS Investigators. Comprehensive Evaluation of Defibrillators And Resuscitative Shocks (CEDARS) study: does atrial fibrillation increase the incidence of inappropriate shock by implanted defibrillators? J Am Coll Cardiol. 1993;21:278A. Abstract.
Cohen TJ, Liem LB. A hemodynamically responsive antitachycardia system: development and basis for design in humans. Circulation. 1990;82:394-406.
Schuger CD, Jackson K, Russel TS, Lehmann MH. Atrial sensing to augment ventricular tachycardia detection by the automatic implantable cardioverter defibrillator: a utility study. PACE. 1988;11:1456-1464.
Leong PHW, Jabri MA. Arrhythmia classification using two intracardiac leads. Proceedings of Computers in Cardiology. Los Alamitos, Calif: IEEE Computer Society Press; 1992:189-192.
Polikaitis A, Arzbaecher R. Sensitivity and specificity of a dual-chamber arrhythmia recognition algorithm for implantable devices. J Electrocardiol. 1995;27:78-83.
Davies DW, Wainwright RJ, Tooley MA, Lloyd D, Nathan AW, Spurrell RA, Camm AJ. Detection of pathological tachycardia by analysis of electrogram morphology. PACE. 1986;9:200-208.
Brachmann J, Sterns LD, Hilbel T, Beyer T, Schoels W, Ruf-Richter J, Karolyi L, Merkely B. Regularity of spontaneously occurring ventricular tachycardias and its influence on stability criteria for detection. Circulation. 1993;88(suppl I):I-353. Abstract.
Swerdlow C, Kass RM, Peter CT. Specificity of interval-stability for rejection of chronic atrial fibrillation. Circulation. 1993;88(suppl I):I-52. Abstract.
Arzbaecher R, Bump T, Jenkings J, Click K, Munkenbeck F, Brown J, Nandhakumar N. Automatic tachycardia recognition. PACE. 1984;7:541-547.
Chiang CMJ, Jenkings JM, DiCarlo LA. The value of rate stability and multiplicity measures to detect ventricular tachycardia in the presence of atrial fibrillation or flutter. PACE. 1994;17:1503-1508.
Militiann A, Yang H, Salacates A, Steinmann R, Meissner M, Mosteller R, Baga J, Pires L, Schuger C, Lehman MH. Computerised recognition of ventricular tachycardia (VT) with 1:1 ventriculo atrial conduction by AV interval analysis. Circulation. 1995;92(suppl I):I-474. Abstract.
Saoudi N, Henri C, Brugada P, Lascaut G, Gueunoun M, Haissaguerre M, Barnay C, Letac B. Automatic ventricular tachycardia recognition during sustained rapid rhythms: superiority of dual chamber over single chamber ventricular monitoring. PACE.. 1994;17:112. Abstract.
Huagul GL, Zardini M, Thakur RJ, Morillo CS, Yee R, Klein GJ. Ventriculo-atrial conduction in patients with implantable cardioverter-defibrillators: implication for tachycardia discrimination by dual chamber sensing. Circulation. 1993;88(suppl I):I-54. Abstract.