- Email: [email protected]
Detection of atrial arrhythmia for cardiac rhythm management by implantable devices
Journal of Electrocardiology Vol. 33 Supplement 2000
Detection of Atrial Arrhythmia for Cardiac Rhythm Management by Implantable Devices
M i l t o n M. M o r r i s , P h D , B r u c e H. K e n K n i g h t , P h D , a n d D o u g l a s J. L a n g , P h D
Abstract: Implantable atrial defibrillators (IAD) should provide pacing therapy whenever appropriate (ie, typical atrial flutter) to minimize shock-related patient discomfort. Additionally, IADs should provide diagnostics regarding atrial arrhythmia type and frequency of occurrence to enable improved physician management of atrial arrhythmia. To achieve this, IADs should accurately classify atria/ arrhythmia such as atrial fibrillation (AF) and atrial flutter (AFL) This article evaluates the performance of an algorithm, atrial rhythm classification (ARC), designed to classify AF and AFL. The ARC algorithm uses maximum rate, standard deviation, and range of the 12 most recent atrial cycle lengths to plot a point in a three-dimensional space. A decision boundary divides the space into 2 regions-faster/unstable atrial cycle lengths (AF) or slower/stable cycle lengths (AFL). Classifications are made on a sliding window of 12 consecutive cydes until the end of the episode is reached. In this way, continuous episode feedback is provided that can be used to help guide device therapy, measure arrhythmia type and frequency of occurrence. Bipolar (1-cm) electrogram episodes of AF (n = 16) and AFL (n = 7) were acquired from 20 patients and retrospectively analyzed using the ARC algorithm. The sensitivity and specificity in this study was 0.993 and 0.982, respectively. The ARC algorithm would have appropriately guided atrial therapy and minimized discomfort associated with defibrillation shocks in this small patient data set warranting further studies. The ARC algorithm may also be beneficial as a diagnostic tool to assist physician management of atrial arrhythmia. Key words: Atrial arrhythmia, detection, atrial defibrillation, implantable devices.
tients in the United States annually (1). Each admission averages 4 days and averages $6,000 for the treatment of AF and its related symptoms and sequelae (2). AF may be idiopathic or a comorbidity of other organic or cardiovascular disease processes (3). The Framingham study, a major source of data regarding the incidence and prevalence of AF (3-6) has reported a stroke risk associated with AF that increases with age (7). Causes of AF include acute alcoholic intake ('hol-
Atrial fibrillation (AF) is the primary hospital admission diagnosis of approximately 225,000 pa-
From Guidant Corporation, St. Paul, MN.
Reprint requests: Mi]ton M. Morris, PhD, Guidant Corporation, 4100 Hamline Ave N, St. Paul, MN 55112; e-mail: [email protected] Copyright 9 2000 by Churchill Livingstone |
0022-0736/00/330S-0020535.00/0 doi: 10.1054/jelc.2000.20305
133
134
Journal of Electrocardiology Vol. 33 Supplement 2000
iday heart syndrome'), electrocution, acute myocardial infarction, acute pericarditis, acute myocarditis, pulmonary embolism, hyperthyroidism, and acute pulmonary disease (3). AF is a common complication with cardiac and thoracic surgery and can also occur in association with mitral valve disease, atrial septal defects, coronary artery disease, and other organic heart diseases (3). Current management strategies for AF include pharmacology, AV node ablation with ventricular pacing for ventricular rate control, surgical/catheter based ablations, and internal/external atrial defibrillation for maintenance of sinus rhythm (SR) (8). Internal atrial defibrillation (lAD) can be achieved through the use of in-hospital elective internal cardioversion or by an implantable cardioverter defibrillator (ICD) with treatment of atrial tachyarrhythmia including atrial flutter (AFL) and AF (9). Treatment of atrial tachyarrhythmias other than AF by implantable devices includes antitachycardia pacing (ATP) and internal atrial defibrillation shocks (10,11). Although pain or discomfort does occur with IAD therapy (12-16), the use of ATE therapy is well tolerated by patients and has been effective for termination of atrial tachyarrhythmias like AFL (17-19).
tachyarrhythmia (ie, ventricular tachycardia and ventricular fibrillation) based on ventricular rate and rate related criteria (20-21). While a similar approach may be used to detect atrial tachyarrhythmia episodes, mean atrial rates during AF and AFL have a region of overlap (Fig. 1) (22). Jung et al. (22) reported that the range of mean atrial cycle lengths for AF and AFL, 111 +_ 237 and 177 _+ 310, respectively, have a region of overlap. Additional criteria may be of benefit w hen rate overlap occurs. Variability in atrial electrogram features (eg, morphology and activation timing) has been called "quintessential" to AF (23). Therefore, techniques for classifying atrial tachyarrhythmia have centered on measures of variability or dispersion of sequential cycle lengths. Such techniques would enable delivery of effective pacing therapy by identifying potentially pace-terminable arrhythmias, and providing meaningful diagnostics for tracking the relative burden of different supraventricular tachycardias. This article describes an algorithm, atrial rhythm classification (ARC), designed to discriminate between AF and AFL. The rationale for, and the description of, the ARC algorithm is provided along with the resuhs of an initial retrospective evaluation.. A discussion of the ARC algorithm performance and study limitations are also provided.
Clinical Rationale for Atrial Rhythm Classification by Implantable Devices Implantable devices should be capable of ATP therapy wh en pace-terminable atrial tachyarrhythmia (ie, typical atrial flutter) are detected to minimize patient discomfort associated with IAD therapy. Additionally, physicians may find diagnostics regarding atrial arrhythmia burden useful. Diagnostics that track arrhythmia type, frequency of occurrence, episode duration, and duration between episodes may be helpful to physicians managing a patient's atrial arrhythmia, particularly if pharmacology is part of the management strategy. Accurate classification of atrial arrhythmia such as AF and other AT may be of additional clinical benefit.
Background Whereas mean rate as a lone classification feature may be insufficient for classification during an episode of AF and AFL, the m axi m um rate within a series of sequential activation intervals may be useful as a biasing tool. Atrial fibrillation frequently presents with paroxysms of short activation intervals. Additionally, variability in atrial electrogram features is typical of AF (23). Figure 2 contains intracardiac electrogram strips recorded using bipolar electrodes placed in the high right atrium and right ventricular apex of human
AFL
Practical Rationale for Atrial Rhythm Classification by Implantable Devices Commercially available implantable cardioverter defibrillators (ICDs) currently classify ventricular
~
AFL & AF ~
AF IP
Atrial Rate (bpm)
Fig. 1. Depiction of rates for AFL relative to AF. Shaded region represents the existence of an overlap zone containing both AF and AFL rates.
Cardiac Rhythm Management .
Fig. 2. Intracardiac electrogram recordings from the high right atrium (HRA) and right ventricular apex (RVA) of (A) AF and (B) AFL, respectively. The AF electrogram of Figure 2A has greater variability in the atrial interval durations and shorter activation intervals when cornpared to the AFL eIectrogram of Figure 2B. Note: The VF markers at the bottom of the electrogram strips are a manifestation of artificial breadboard programming to trigger storage of electrogram and interval data.
subjects. Each episode was played t h r o u g h a breadboard simulation of an implantable device (VENTAK AV, Guidant Corporation, Indianapolis, IN) and telemetered out to strip chart. Figure 2A and 2B contain episodes of AF and AFL, respectively. Note that the AF electrogram of Figure 2A has greater variability in the atrial interval durations and shorter activation intervals w h e n compared to the AFL electrogram of Figure 2B. These features are regularly seen in episodes of AF and AFL and suggest that an algorithm based on m a x i m u m rate ( 1 / m i n i m u m activation interval) and interval variability within a series of sequential atrial activation intervals m a y provide acceptable levels of accuracy with respect to specificity and sensitivity w h e n classifying AF from AFL.
Description of ARC Algorithm for Discriminating AF and AFL The ARC algorithm uses a three-dimensional approach to AF/AFL classification (Fig. 3). Maxi-
Morris et al.
135
m u m rate, range, and standard deviation of 12 sequential atrial activation intervals are used to represent a r h y t h m in a three-dimensional featurespace. A three-dimensional decision b o u n d a r y seen in Figure 3 is used to partition the space into an AF region and an AFL region. A r h y t h m is classified based on the region it occupies. The decision b o u n d a r y is formed by using a mathematical equation describing the interdependence of each parameter: m a x i m u m rate • range • standard deviation = scalar. The 3 dimensional decision b o u n d a r y location and shape can be manipulated by changing the value of the scalar to w h i c h the product of m a x i m u m rate, range and standard deviation are bound. Also depicted in Figure 3 are hypothetical event markers representing typical atrial activations in time for AF and AFL. Note that interval lengths during AF relative to AFL have higher m a x i m u m rates (ie, shorter coupling intervals), a larger range and larger standard deviation. The AF subspace correspond to r h y t h m s with higher rate and larger interval variability.
X of Y Criteria An X of Y criteria can be used with the ARC algorithm to help filter transient errors in beat-tobeat classifications. In this investigation, we set Y at 12 and required at least 75% or 8 of the last 12 windows to be classified as AF to return an AF classification for that m o m e n t in time. An X of 12 criteria is accomplished by tracking the last 12 diagnoses and simply counting h o w m a n y were AF (Fig. 4). If 75% or 8 of the last 12 w i n d o w s were diagnosed as AF, the algorithm returns an AF classification for that m o m e n t in time. Otherwise, an AFL classification is returned for that m o m e n t in time.
Rate Thresholds Rate thresholds m a y also be used with the ARC algorithm. An u p p e r and l o w e r limit on maxim u m rate can be used w h e n a n y series of intervals with a m a x i m u m rate greater t h a n the u p p e r limit is a u t o m a t i c a l l y classified as AF. Conversely, a n y series of intervals with a m a x i m u m rate below a l o w e r limit is a u t o m a t i c a l l y classified as an AFL.
136
Journal of Electrocardiology Vol. 33 Supplement 2000
Fig. 3. Sample three-dimensional decision boundary produced by setting the product of maximum rate, range, and standard deviation of 12 sequential atrial activation intervals to a scalar value of 5. The decision boundary is used to split the space into an AF subspace and an AFL subspace. Event markers are used to depict the timing of typical atrial activations as sensed by an ICE) for AF and AFL.
Materials and Methods Data Acquisition Data for this investigation were acquired from 2 sources: A n n Arbor Electrogram Libraries (AAEL,
8 ofthe past 12 windows diagnosed as AF. Retm'n AF dasslflcatlon
7 ofthe past 12 windows d/aSnosed as AF. Return AFL dasstflcalion
Fig. 4. X of 12 criteria shown using a window of classifications. Two groups of 12 individual window classifications are shown. Each window is used to return a classification by the ARC algorithm for that moment in time. If at least 8 of 12 (75%) of the individual classifications are AF, the ARC algorithm returns a classification of AF for that moment in time.
A n n Arbor, MI, USA) (20 patients, 16 episodes of AF, 7 episodes AFL) and a separate proprietary database collected during routine electrophysiology examinations (12 patients, 10 episodes of AF, 5 episodes of AFL). The AAEL database was generated by recording from unipolar and bipolar (1 cm) distal atrial and ventricular electrograms during elective clinical cardiac electrophysiology (EP) studies. Six French quadrapolar electrode catheters (USCI, Billerica, MA) with an interelectrode distance of 1 cm were inserted and advanced u n d e r fluoroscopic guidance to the high right atrium and right ventricular apex. All recordings were wide-band filtered and sampled at 1,000 Hz with the patient in a supine position. Strip charts of each episode were classified by the electrophysiologist conducting the EP study. The proprietary database was gathered during routine EP examinations. Data were acquired from a bipolar ( - 1 cm spacing) electrode placed in the right ventricular apex and the high right atrium during episodes of supraventricular tacbycardia, including AF and AFL, wide-band filtered and stored at high fidelity (-> 1,000 I-Iz) to TEAC tape. All
Cardiac Rhythm Management
recordings were obtained with the patient in a supine position. Strip charts of each episode w e r e classified by the electrophysiologist conducting the EP study.
Data Analysis To heIp assess the utility of using rate cutoffs a n d an X of Y criterion, 4 discrete receiver operating characteristic (ROC) plots were generated. A discrete operating point is generated by setting a value to each v a r i a b l e - - l o w e r rate cutoff (Ratel), u p p e r rate cutoff (Rate2), and the scalar. An ROC plot was generated for each combination: ARC alone, ARC with rate cutoffs and no X of Y criterion, ARC with the X of Y criterion and no rate cutoffs, and ARC with both rate cutoffs and the X of Y criterion. The ROC plots were produced by using a retrospective analysis of a particular subject population. Therefore, the p e a k p e r f o r m a n c e constitute the best case scenarios for this patient population.
Results Receiver Operating Characteristic Plots The ROC plot (Fig. 5) contains text that provides detailed p e r f o r m a n c e informalion. Sensitivity and specificity information for the operating point closest to the ideal operating point (0,1) is p r o v i d e d along with the corresponding threshold values. R a t e l and Rate2 are the lower and u p p e r rate cutoffs, respectively, and are in beats per minute. The scalar n u m b e r corresponding to the shape and location of the decision b o u n d a r y is also provided.
137
~0.
i
=_oo. -$ ~04 'S
~0
Atrial electrograms w e r e played t h r o u g h a breadboard simulation of the VENTAK AV (Guidant Corporation) implantable cardioverter defibrillator. The b r e a d b o a r d was p r o g r a m m e d to a single zone configuration with the VF threshold artificially set to 40 b p m . The low VF threshold substantially increases the likelihood that an episode w o u l d be declared and data w o u l d be stored during playback. The b r e a d b o a r d declared episodes and atrial interval i n f o r m a t i o n was stored to disk. The atrial interval data w e r e stored as a file accessible by a c o m puter r u n n i n g a MATLAB (MathWorks Inc, Natick, MA) sottware script used to m o d e l the ARC algorithm.
Morris et al.
ROC (ARC): Rate and No X of Y
ROC (ARC): Rate and X of Y
~ 0.2
Experimental Setup
9
(Spec, Seas) - (0.993, 0,982) (Ratet, Rate2, Scalar) = (100, 275,325) (Max Spec, Seas) = (0998,0.837) (Rate1, Rate2, Scalar)= (108,400,4.001
~o.
l
(Spec, Sens) = (0979, 0968) i (Rate1, Rate2, Scalar) = (100,275, fi.00)~ (Max Spec, Seas) = (0.993, 0.811) (Rate1, Rate2, Scalar) = (100,400, 500
02 0.4 0.6 0.8 probability of false alarm (1-specifLcity)
0.2 0.4 0,6 0.8 probability of false alarm (1-specificity)
ROC (ARC): NO Rate and X of Y
ROC (ARC): No Rate and No X of Y
:~0
:•0.B ._~o.e e 0.4 "o # ~o 0.2
~o.
1
1
t
0 (Spec, Sens) = (0.958, 0.925) (Scalar) = (1.50) (Max Spee. Seas) = (0.990, 0.757) (Scalar) : (400)
~0
012 0;4 010 0'8 ~robabirity of false alarm (1 -specificity)
(Spec, Sens) = (0.938, 0.926) (Scalar) = (1.75) (Max Spec. Seas) = (0.093. 0.749) (Scalar) = (5.00)
1 [ ] I
0.2 0,4 0.6 08 1 probaNlity of false alarm (1-specificity)
Fig. 5. Discrete ROC plots used to measure performance for various values for lower rate cutoff (Ratel), upper rate cutoff (Rate2), and the scalar. (A) Performance of the ARC algorithm when using rate cutoffs and an X of Y criterion. (B) Performance of the ARC algorithm when using rate cutoffs and no X of Y criterion. (C) Performance of the ARC algorithm when using X of Y criterion and no rate cutoffs. (D) Performance of the ARC aIgorithm without using X of Y criterion and without using rate cutoffs. The second set of specificity, sensitivity, Ratel, Rate2, and scalar n u m b e r s corresponds to the operating point that m a x i m i z e d specificity. Biasing the p e r f o r m a n c e t o w a r d greater specificity corresponds to being m o r e aggressive with delivering pacing therapy as opposed to missing opportunities to give pacing therapies and increasing the likelihood of going directly to IAD. Based on the sensitivity and specificity values closest to the ideal operating points and those that maximize sensitivity, ARC used with b o t h an X of Y criterion and rate cutoffs m a x i m i z e d p e r f o r m a n c e . The sensitivity and specificity for ARC with both rate cutoff and X of Y criteria that came closest to the ideal were 0.993 and 0.982, respectively. This corresponded to rate cutoffs of i00 b p m and 275 b p m and a scalar value of 3.25. The sensitivity and m a x i m u m specificity for ARC with both rate cutoff and X of Y criteria w e r e 0.998 and 0.837, respectively. This corresponded to rate cutoffs of t 0 0 b p m and 400 b p m and a scalar value of 4.00.
Discussion In this investigation, the ARC algorithm accurately classified AF f r o m AFL. The use of rate cutoffs
138
Journal of Electrocardiology Vol. 33 Supplement 2000
and an X of Y criteria w e r e incrementally beneficial with respect to p e r f o r m a n c e . This could be seen by the i m p r o v e m e n t s in operating points on the ROC plots.
VENTAK
*
......
AV I)R
~ i'='T i= ~3i=
i I i
I ==
l
r l
I HRAbp
Database : :.
The AAEL and p r o p r i e t a r y databases were important to this investigation for several reasons. First, the AAEL data contained episodes that w e r e classified by an i n d e p e n d e n t physician expert. Second, the lead configuration of the electrodes closely mimics w h a t is currently used in commercial devices. Third, the data w e r e acquired f r o m h u m a n s in a controlled m a n n e r . The p e r f o r m a n c e m e a s u r e s of this investigation should be robust for those reasons.
.
.
.
.
.
.
.
.
,
,,
::[:
Fig. 6. Depiction of an episode that alternates between a more organized and a less organized activation pattern as sensed from a bipoIe in the high right atrium. Note: The VF markers at the bottom of the electrogram strips are a manifestation of artificial breadboard programming to trigger storage of electrogram and interval data.
Limitations AF and flutter are global p h e n o m e n a and are not optimally sensed w i t h closely spaced (1 cm) bipolar eIectrodes that are d o m i n a t e d by proximal {near field) electrical activity. Additionally, as AF has m a n y different causes, it m a y also h a v e m a n y different eleCtrophysiologic m e c h a n i s m s that m a y affect the electrocardiographic recordings. No att e m p t to control AF "type" was m a d e in this study.
m a y contain this AF/AFL behavior. A small subset of episodes in this study contained r h y t h m s that m o v e d b e t w e e n regular and irregular intervals. W h e t h e r these episodes were truly e x a m p l e s of AF/AFL hybrids was not explored in this study. However, because the ARC algorithm can track r h y t h m s continuously, a shift from AF to AFL detected by the ARC algorithm w o u l d enable pacing t h e r a p y to be delivered during transient periods of flutter-like r h y t h m s .
Retrospective Evaluation of Algorithm ROC curves w e r e used to determine the ideal operating point and the utility of additional criteria (X of Y and rate cutoffs) by using carefully collected data as a training set. Future w o r k for this algorithm will include a prospective evaluation of the ARC algorithm with similarly collected data as a test set. It was d e t e r m i n e d that the X of Y and Rate Cutoff criteria produced a m o r e accurate algorithm with respect to p e r f o r m a n c e as captured on the ROC plots. The lower cutoff rate (Ratel) that m a x i m i z e d p e r f o r m a n c e was 100 bpm. This was the lowest value of Ratel evaluated in the study. This means, increasing the value for R a t e l did not result in improved p e r f o r m a n c e . This m a y indicate that a Ratel cutoff is not necessary to achieve o p t i m u m performance.
Tracking Behavior The p h e n o m e n o n of atrial a r r h y t h m i a that oscillate b e t w e e n fibrillation and flutter has b e e n discussed (24). Figure 6 contains an SVT episode that
Conclusion The use of rate alone does not suffice for discriminating AF from AFL. The ARC algorithm is a m o r e robust analysis t h a n rate by utilizing the variability of the atrial electrogram. A retrospective analysis of the ARC algorithm on this data set indicates that high levels of specificity and sensitivity m a y be achieved w h e n classifying AF and AFL. The rationale for the use of the ARC algorithm contains clinical and practical c o m p o n e n t s . The potential clinical utility of the ARC algorithm is: 1) enabling preferential use of well tolerated ATP pacing over internal defibrillation for treating selected atrial a r r h y t h m i a , and 2) providing diagnostics to physicians at patient follow-ups. Although i m p o r t a n t limitations of this study should not be ignored, the rationale for, and the p e r f o r m a n c e of, the ARC algorithm warrants consideration for use in implantable cardioverter defibrillators.
Cardiac Rhythm Management
References 1. Cairns JA, Connolly SJ: Nonrhcmnatic atrial fibrilla lion. Risk of stroke at~d role of antithrombotic therapy. Circulation 84:469, 1991 2. Dell'Orfano JT, Patcl 1I, Wolbrctte DL, et al: Acute treatment of atrial fibrillation: Spontaneous conversion rates and cost of care. Am J Cardiol 83:788, 1999 3. Levy S, Breithardt G, Campbell RW, et al: Atrial fibrillation: Current knowledge and recommendations for management. Eur Heart J 19:1294, 1998 4. Kannel WB, Abbot RD, Savage DD, et al: Epidemiologic features of chronic atrial fibrillation; The Franfingham Study. N Engl J Med 306:1018, 1982 5. Brand FN, Abort RD, Kannel WB, et aI: Characteristics and prognosis of lone atrial fibrillation: 30-years follow up in the Framingharn Study. JAMA 254: 3449, 1985 6. Wolf PA, Abbott RD, Kannel WB: Atrial fibrillation as an independent risk factor for stroke; The Framingham Study. Stroke 22:983, 1991 7. Wolf PA, Abbott RD, Kannel WB: Atrial fibrillation: A major contributor to stroke in the elderly. Arch Intern Med 147:1561, 1987 8. Murgatroyd FD, Canlm AJ: Nonpharmoacological M a n a g e m e n t of Atrial Fibrillation. Futura Publishing Company, Inc. Armonk, NY, 1997 9. Jung W, Pfeiffer D, Wolpert C, et al: Which patients do benefit from an implantablc atrial defibrillator? J Am ColI CardioI 27:302A, 1996 (abstr) 10. Reddy RK, Bardy GH: Implantable Dual Chamber Defibrillators for Atrial Defibrillation. In: Murgatroyd FD, Carom AJ (eds): Nonpharmacological Managemerit of Atrial Fibrillation. Futura Publishing Company, Inc, Armonk, NY, p 439, 1997 11. KenKnight BH, Lang D J, Scheiner A, et ah Atrial Defibrillation for Implamable Cardioverter-Deflbrillators: Lead Systems, Waveforms, Detection Algorithms, and Results. In: Singer I, Barold SS, Carom AJ. (eds): NonpharmacologicaI Therapy of Arrhythmias for the 21st Century: The State of the Art. Futura Publishing Co, Inc, Armonk, NY, p 457, 1998 12. A m m e r R, Alt E, Ayers G, et ah Pain threshold for low energy intracardiac cardioversion of atrial fibrillation with low or no sedation [published erratum appears in Pacing Clin ElectrophysioI 1997 Apr;20(4 Pt 1):viii]. PACE 20(1 Pt 2):230, 1997 13. Jung J, Heisel A, Fries R, et al: Tolerability of internal
14.
15.
i6.
17.
18.
19.
20.
21.
22.
23.
24.
9
Morris et al.
t39
low-energy shock strengths currently needed for endocardial atrial cardioversion. Am J Cardiol 80: 1489, I997 Kidwai BJ, Harbtson MT, Allen JD, et ah Waveform optimization [or internal atrial defibrillation: A ne w rounded biphasic trapezoid shock. J Am Coll Cardiol 33:159A, I999 (abstr) Harbinson MT, Alien JD, I m a m Z, et ah Rounded biphasic waveform reduces energy requirements for transvcnous catheter cardioversion of atrial fibrillation and flutter. PACE 20:226, 1997 Ayers GM: How can atrial defibrillation be made more tolerable? In: Murgatroyd FD, Carom AJ (eds): Nonpharmacological m a n a g e m e n t of atrial fibrillation. Futura Publishing Co Inc, Armonk, NY, 1997, p 475 Hii JT, Mitchell LB, Duff HJ, et al: Comparison of atrial overdrive pacing with and without extrastimuli for termination of atrial flutter. Am J Cardiol 70:463, 1992 Waksman R, Pollack A, Berkovits BV, et ah Amodecremental pacing for the interruption of ventricuIar tachycardia and atrial flutter. J Electrocardiol 25:339, 1992 Watson RM, Josephson ME: Atrial flutter. I. EIectrophysiologic substrates and modes of initiation and termination. Am J CardioI 45:732, 1980 Morris MM, Marcovecchio AF, KenKnight BH, et al: Retrospective evaluation of detection enhancements in a dual-chamber implantable cardioverter defibrillator: Implications for device programming. PACE 22:849, 1999 Jenkins JM, CaswelI SA: Detection algorithms in implantable cardioverter defibrillators. Proc IEEE 84: 428, 1996 Jung J, Hohenberg G, Heisel A, e{ ah Discrimination of sinus rhythm, atrial flutter, and atrial fibrillation using bipolar endocardial signals. J Cardiovasc ElectrophysioI 9:689, 1998 Swiryn S, Schoenwald A, Sahakian A: Detection of atrial fibrillation by pacemakers and antiarrhythmic devices. In: Nonpharmacological Management ot Atrial Fibrillation Futura Publishing Company, Inc, Armonk, NY, 1997, p 309 Wietholt D, Hart J, Chen L, e~ ah On the mechanisms of alternation b et w een atrial flutter and atrial fibrillation in man. PACE 23:735, 2000
Detection of Atrial Arrhythmia for Cardiac Rhythm Management by Implantable Devices
M i l t o n M. M o r r i s , P h D , B r u c e H. K e n K n i g h t , P h D , a n d D o u g l a s J. L a n g , P h D
Abstract: Implantable atrial defibrillators (IAD) should provide pacing therapy whenever appropriate (ie, typical atrial flutter) to minimize shock-related patient discomfort. Additionally, IADs should provide diagnostics regarding atrial arrhythmia type and frequency of occurrence to enable improved physician management of atrial arrhythmia. To achieve this, IADs should accurately classify atria/ arrhythmia such as atrial fibrillation (AF) and atrial flutter (AFL) This article evaluates the performance of an algorithm, atrial rhythm classification (ARC), designed to classify AF and AFL. The ARC algorithm uses maximum rate, standard deviation, and range of the 12 most recent atrial cycle lengths to plot a point in a three-dimensional space. A decision boundary divides the space into 2 regions-faster/unstable atrial cycle lengths (AF) or slower/stable cycle lengths (AFL). Classifications are made on a sliding window of 12 consecutive cydes until the end of the episode is reached. In this way, continuous episode feedback is provided that can be used to help guide device therapy, measure arrhythmia type and frequency of occurrence. Bipolar (1-cm) electrogram episodes of AF (n = 16) and AFL (n = 7) were acquired from 20 patients and retrospectively analyzed using the ARC algorithm. The sensitivity and specificity in this study was 0.993 and 0.982, respectively. The ARC algorithm would have appropriately guided atrial therapy and minimized discomfort associated with defibrillation shocks in this small patient data set warranting further studies. The ARC algorithm may also be beneficial as a diagnostic tool to assist physician management of atrial arrhythmia. Key words: Atrial arrhythmia, detection, atrial defibrillation, implantable devices.
tients in the United States annually (1). Each admission averages 4 days and averages $6,000 for the treatment of AF and its related symptoms and sequelae (2). AF may be idiopathic or a comorbidity of other organic or cardiovascular disease processes (3). The Framingham study, a major source of data regarding the incidence and prevalence of AF (3-6) has reported a stroke risk associated with AF that increases with age (7). Causes of AF include acute alcoholic intake ('hol-
Atrial fibrillation (AF) is the primary hospital admission diagnosis of approximately 225,000 pa-
From Guidant Corporation, St. Paul, MN.
Reprint requests: Mi]ton M. Morris, PhD, Guidant Corporation, 4100 Hamline Ave N, St. Paul, MN 55112; e-mail: [email protected] Copyright 9 2000 by Churchill Livingstone |
0022-0736/00/330S-0020535.00/0 doi: 10.1054/jelc.2000.20305
133
134
Journal of Electrocardiology Vol. 33 Supplement 2000
iday heart syndrome'), electrocution, acute myocardial infarction, acute pericarditis, acute myocarditis, pulmonary embolism, hyperthyroidism, and acute pulmonary disease (3). AF is a common complication with cardiac and thoracic surgery and can also occur in association with mitral valve disease, atrial septal defects, coronary artery disease, and other organic heart diseases (3). Current management strategies for AF include pharmacology, AV node ablation with ventricular pacing for ventricular rate control, surgical/catheter based ablations, and internal/external atrial defibrillation for maintenance of sinus rhythm (SR) (8). Internal atrial defibrillation (lAD) can be achieved through the use of in-hospital elective internal cardioversion or by an implantable cardioverter defibrillator (ICD) with treatment of atrial tachyarrhythmia including atrial flutter (AFL) and AF (9). Treatment of atrial tachyarrhythmias other than AF by implantable devices includes antitachycardia pacing (ATP) and internal atrial defibrillation shocks (10,11). Although pain or discomfort does occur with IAD therapy (12-16), the use of ATE therapy is well tolerated by patients and has been effective for termination of atrial tachyarrhythmias like AFL (17-19).
tachyarrhythmia (ie, ventricular tachycardia and ventricular fibrillation) based on ventricular rate and rate related criteria (20-21). While a similar approach may be used to detect atrial tachyarrhythmia episodes, mean atrial rates during AF and AFL have a region of overlap (Fig. 1) (22). Jung et al. (22) reported that the range of mean atrial cycle lengths for AF and AFL, 111 +_ 237 and 177 _+ 310, respectively, have a region of overlap. Additional criteria may be of benefit w hen rate overlap occurs. Variability in atrial electrogram features (eg, morphology and activation timing) has been called "quintessential" to AF (23). Therefore, techniques for classifying atrial tachyarrhythmia have centered on measures of variability or dispersion of sequential cycle lengths. Such techniques would enable delivery of effective pacing therapy by identifying potentially pace-terminable arrhythmias, and providing meaningful diagnostics for tracking the relative burden of different supraventricular tachycardias. This article describes an algorithm, atrial rhythm classification (ARC), designed to discriminate between AF and AFL. The rationale for, and the description of, the ARC algorithm is provided along with the resuhs of an initial retrospective evaluation.. A discussion of the ARC algorithm performance and study limitations are also provided.
Clinical Rationale for Atrial Rhythm Classification by Implantable Devices Implantable devices should be capable of ATP therapy wh en pace-terminable atrial tachyarrhythmia (ie, typical atrial flutter) are detected to minimize patient discomfort associated with IAD therapy. Additionally, physicians may find diagnostics regarding atrial arrhythmia burden useful. Diagnostics that track arrhythmia type, frequency of occurrence, episode duration, and duration between episodes may be helpful to physicians managing a patient's atrial arrhythmia, particularly if pharmacology is part of the management strategy. Accurate classification of atrial arrhythmia such as AF and other AT may be of additional clinical benefit.
Background Whereas mean rate as a lone classification feature may be insufficient for classification during an episode of AF and AFL, the m axi m um rate within a series of sequential activation intervals may be useful as a biasing tool. Atrial fibrillation frequently presents with paroxysms of short activation intervals. Additionally, variability in atrial electrogram features is typical of AF (23). Figure 2 contains intracardiac electrogram strips recorded using bipolar electrodes placed in the high right atrium and right ventricular apex of human
AFL
Practical Rationale for Atrial Rhythm Classification by Implantable Devices Commercially available implantable cardioverter defibrillators (ICDs) currently classify ventricular
~
AFL & AF ~
AF IP
Atrial Rate (bpm)
Fig. 1. Depiction of rates for AFL relative to AF. Shaded region represents the existence of an overlap zone containing both AF and AFL rates.
Cardiac Rhythm Management .
Fig. 2. Intracardiac electrogram recordings from the high right atrium (HRA) and right ventricular apex (RVA) of (A) AF and (B) AFL, respectively. The AF electrogram of Figure 2A has greater variability in the atrial interval durations and shorter activation intervals when cornpared to the AFL eIectrogram of Figure 2B. Note: The VF markers at the bottom of the electrogram strips are a manifestation of artificial breadboard programming to trigger storage of electrogram and interval data.
subjects. Each episode was played t h r o u g h a breadboard simulation of an implantable device (VENTAK AV, Guidant Corporation, Indianapolis, IN) and telemetered out to strip chart. Figure 2A and 2B contain episodes of AF and AFL, respectively. Note that the AF electrogram of Figure 2A has greater variability in the atrial interval durations and shorter activation intervals w h e n compared to the AFL electrogram of Figure 2B. These features are regularly seen in episodes of AF and AFL and suggest that an algorithm based on m a x i m u m rate ( 1 / m i n i m u m activation interval) and interval variability within a series of sequential atrial activation intervals m a y provide acceptable levels of accuracy with respect to specificity and sensitivity w h e n classifying AF from AFL.
Description of ARC Algorithm for Discriminating AF and AFL The ARC algorithm uses a three-dimensional approach to AF/AFL classification (Fig. 3). Maxi-
Morris et al.
135
m u m rate, range, and standard deviation of 12 sequential atrial activation intervals are used to represent a r h y t h m in a three-dimensional featurespace. A three-dimensional decision b o u n d a r y seen in Figure 3 is used to partition the space into an AF region and an AFL region. A r h y t h m is classified based on the region it occupies. The decision b o u n d a r y is formed by using a mathematical equation describing the interdependence of each parameter: m a x i m u m rate • range • standard deviation = scalar. The 3 dimensional decision b o u n d a r y location and shape can be manipulated by changing the value of the scalar to w h i c h the product of m a x i m u m rate, range and standard deviation are bound. Also depicted in Figure 3 are hypothetical event markers representing typical atrial activations in time for AF and AFL. Note that interval lengths during AF relative to AFL have higher m a x i m u m rates (ie, shorter coupling intervals), a larger range and larger standard deviation. The AF subspace correspond to r h y t h m s with higher rate and larger interval variability.
X of Y Criteria An X of Y criteria can be used with the ARC algorithm to help filter transient errors in beat-tobeat classifications. In this investigation, we set Y at 12 and required at least 75% or 8 of the last 12 windows to be classified as AF to return an AF classification for that m o m e n t in time. An X of 12 criteria is accomplished by tracking the last 12 diagnoses and simply counting h o w m a n y were AF (Fig. 4). If 75% or 8 of the last 12 w i n d o w s were diagnosed as AF, the algorithm returns an AF classification for that m o m e n t in time. Otherwise, an AFL classification is returned for that m o m e n t in time.
Rate Thresholds Rate thresholds m a y also be used with the ARC algorithm. An u p p e r and l o w e r limit on maxim u m rate can be used w h e n a n y series of intervals with a m a x i m u m rate greater t h a n the u p p e r limit is a u t o m a t i c a l l y classified as AF. Conversely, a n y series of intervals with a m a x i m u m rate below a l o w e r limit is a u t o m a t i c a l l y classified as an AFL.
136
Journal of Electrocardiology Vol. 33 Supplement 2000
Fig. 3. Sample three-dimensional decision boundary produced by setting the product of maximum rate, range, and standard deviation of 12 sequential atrial activation intervals to a scalar value of 5. The decision boundary is used to split the space into an AF subspace and an AFL subspace. Event markers are used to depict the timing of typical atrial activations as sensed by an ICE) for AF and AFL.
Materials and Methods Data Acquisition Data for this investigation were acquired from 2 sources: A n n Arbor Electrogram Libraries (AAEL,
8 ofthe past 12 windows diagnosed as AF. Retm'n AF dasslflcatlon
7 ofthe past 12 windows d/aSnosed as AF. Return AFL dasstflcalion
Fig. 4. X of 12 criteria shown using a window of classifications. Two groups of 12 individual window classifications are shown. Each window is used to return a classification by the ARC algorithm for that moment in time. If at least 8 of 12 (75%) of the individual classifications are AF, the ARC algorithm returns a classification of AF for that moment in time.
A n n Arbor, MI, USA) (20 patients, 16 episodes of AF, 7 episodes AFL) and a separate proprietary database collected during routine electrophysiology examinations (12 patients, 10 episodes of AF, 5 episodes of AFL). The AAEL database was generated by recording from unipolar and bipolar (1 cm) distal atrial and ventricular electrograms during elective clinical cardiac electrophysiology (EP) studies. Six French quadrapolar electrode catheters (USCI, Billerica, MA) with an interelectrode distance of 1 cm were inserted and advanced u n d e r fluoroscopic guidance to the high right atrium and right ventricular apex. All recordings were wide-band filtered and sampled at 1,000 Hz with the patient in a supine position. Strip charts of each episode were classified by the electrophysiologist conducting the EP study. The proprietary database was gathered during routine EP examinations. Data were acquired from a bipolar ( - 1 cm spacing) electrode placed in the right ventricular apex and the high right atrium during episodes of supraventricular tacbycardia, including AF and AFL, wide-band filtered and stored at high fidelity (-> 1,000 I-Iz) to TEAC tape. All
Cardiac Rhythm Management
recordings were obtained with the patient in a supine position. Strip charts of each episode w e r e classified by the electrophysiologist conducting the EP study.
Data Analysis To heIp assess the utility of using rate cutoffs a n d an X of Y criterion, 4 discrete receiver operating characteristic (ROC) plots were generated. A discrete operating point is generated by setting a value to each v a r i a b l e - - l o w e r rate cutoff (Ratel), u p p e r rate cutoff (Rate2), and the scalar. An ROC plot was generated for each combination: ARC alone, ARC with rate cutoffs and no X of Y criterion, ARC with the X of Y criterion and no rate cutoffs, and ARC with both rate cutoffs and the X of Y criterion. The ROC plots were produced by using a retrospective analysis of a particular subject population. Therefore, the p e a k p e r f o r m a n c e constitute the best case scenarios for this patient population.
Results Receiver Operating Characteristic Plots The ROC plot (Fig. 5) contains text that provides detailed p e r f o r m a n c e informalion. Sensitivity and specificity information for the operating point closest to the ideal operating point (0,1) is p r o v i d e d along with the corresponding threshold values. R a t e l and Rate2 are the lower and u p p e r rate cutoffs, respectively, and are in beats per minute. The scalar n u m b e r corresponding to the shape and location of the decision b o u n d a r y is also provided.
137
~0.
i
=_oo. -$ ~04 'S
~0
Atrial electrograms w e r e played t h r o u g h a breadboard simulation of the VENTAK AV (Guidant Corporation) implantable cardioverter defibrillator. The b r e a d b o a r d was p r o g r a m m e d to a single zone configuration with the VF threshold artificially set to 40 b p m . The low VF threshold substantially increases the likelihood that an episode w o u l d be declared and data w o u l d be stored during playback. The b r e a d b o a r d declared episodes and atrial interval i n f o r m a t i o n was stored to disk. The atrial interval data w e r e stored as a file accessible by a c o m puter r u n n i n g a MATLAB (MathWorks Inc, Natick, MA) sottware script used to m o d e l the ARC algorithm.
Morris et al.
ROC (ARC): Rate and No X of Y
ROC (ARC): Rate and X of Y
~ 0.2
Experimental Setup
9
(Spec, Seas) - (0.993, 0,982) (Ratet, Rate2, Scalar) = (100, 275,325) (Max Spec, Seas) = (0998,0.837) (Rate1, Rate2, Scalar)= (108,400,4.001
~o.
l
(Spec, Sens) = (0979, 0968) i (Rate1, Rate2, Scalar) = (100,275, fi.00)~ (Max Spec, Seas) = (0.993, 0.811) (Rate1, Rate2, Scalar) = (100,400, 500
02 0.4 0.6 0.8 probability of false alarm (1-specifLcity)
0.2 0.4 0,6 0.8 probability of false alarm (1-specificity)
ROC (ARC): NO Rate and X of Y
ROC (ARC): No Rate and No X of Y
:~0
:•0.B ._~o.e e 0.4 "o # ~o 0.2
~o.
1
1
t
0 (Spec, Sens) = (0.958, 0.925) (Scalar) = (1.50) (Max Spee. Seas) = (0.990, 0.757) (Scalar) : (400)
~0
012 0;4 010 0'8 ~robabirity of false alarm (1 -specificity)
(Spec, Sens) = (0.938, 0.926) (Scalar) = (1.75) (Max Spec. Seas) = (0.093. 0.749) (Scalar) = (5.00)
1 [ ] I
0.2 0,4 0.6 08 1 probaNlity of false alarm (1-specificity)
Fig. 5. Discrete ROC plots used to measure performance for various values for lower rate cutoff (Ratel), upper rate cutoff (Rate2), and the scalar. (A) Performance of the ARC algorithm when using rate cutoffs and an X of Y criterion. (B) Performance of the ARC algorithm when using rate cutoffs and no X of Y criterion. (C) Performance of the ARC algorithm when using X of Y criterion and no rate cutoffs. (D) Performance of the ARC aIgorithm without using X of Y criterion and without using rate cutoffs. The second set of specificity, sensitivity, Ratel, Rate2, and scalar n u m b e r s corresponds to the operating point that m a x i m i z e d specificity. Biasing the p e r f o r m a n c e t o w a r d greater specificity corresponds to being m o r e aggressive with delivering pacing therapy as opposed to missing opportunities to give pacing therapies and increasing the likelihood of going directly to IAD. Based on the sensitivity and specificity values closest to the ideal operating points and those that maximize sensitivity, ARC used with b o t h an X of Y criterion and rate cutoffs m a x i m i z e d p e r f o r m a n c e . The sensitivity and specificity for ARC with both rate cutoff and X of Y criteria that came closest to the ideal were 0.993 and 0.982, respectively. This corresponded to rate cutoffs of i00 b p m and 275 b p m and a scalar value of 3.25. The sensitivity and m a x i m u m specificity for ARC with both rate cutoff and X of Y criteria w e r e 0.998 and 0.837, respectively. This corresponded to rate cutoffs of t 0 0 b p m and 400 b p m and a scalar value of 4.00.
Discussion In this investigation, the ARC algorithm accurately classified AF f r o m AFL. The use of rate cutoffs
138
Journal of Electrocardiology Vol. 33 Supplement 2000
and an X of Y criteria w e r e incrementally beneficial with respect to p e r f o r m a n c e . This could be seen by the i m p r o v e m e n t s in operating points on the ROC plots.
VENTAK
*
......
AV I)R
~ i'='T i= ~3i=
i I i
I ==
l
r l
I HRAbp
Database : :.
The AAEL and p r o p r i e t a r y databases were important to this investigation for several reasons. First, the AAEL data contained episodes that w e r e classified by an i n d e p e n d e n t physician expert. Second, the lead configuration of the electrodes closely mimics w h a t is currently used in commercial devices. Third, the data w e r e acquired f r o m h u m a n s in a controlled m a n n e r . The p e r f o r m a n c e m e a s u r e s of this investigation should be robust for those reasons.
.
.
.
.
.
.
.
.
,
,,
::[:
Fig. 6. Depiction of an episode that alternates between a more organized and a less organized activation pattern as sensed from a bipoIe in the high right atrium. Note: The VF markers at the bottom of the electrogram strips are a manifestation of artificial breadboard programming to trigger storage of electrogram and interval data.
Limitations AF and flutter are global p h e n o m e n a and are not optimally sensed w i t h closely spaced (1 cm) bipolar eIectrodes that are d o m i n a t e d by proximal {near field) electrical activity. Additionally, as AF has m a n y different causes, it m a y also h a v e m a n y different eleCtrophysiologic m e c h a n i s m s that m a y affect the electrocardiographic recordings. No att e m p t to control AF "type" was m a d e in this study.
m a y contain this AF/AFL behavior. A small subset of episodes in this study contained r h y t h m s that m o v e d b e t w e e n regular and irregular intervals. W h e t h e r these episodes were truly e x a m p l e s of AF/AFL hybrids was not explored in this study. However, because the ARC algorithm can track r h y t h m s continuously, a shift from AF to AFL detected by the ARC algorithm w o u l d enable pacing t h e r a p y to be delivered during transient periods of flutter-like r h y t h m s .
Retrospective Evaluation of Algorithm ROC curves w e r e used to determine the ideal operating point and the utility of additional criteria (X of Y and rate cutoffs) by using carefully collected data as a training set. Future w o r k for this algorithm will include a prospective evaluation of the ARC algorithm with similarly collected data as a test set. It was d e t e r m i n e d that the X of Y and Rate Cutoff criteria produced a m o r e accurate algorithm with respect to p e r f o r m a n c e as captured on the ROC plots. The lower cutoff rate (Ratel) that m a x i m i z e d p e r f o r m a n c e was 100 bpm. This was the lowest value of Ratel evaluated in the study. This means, increasing the value for R a t e l did not result in improved p e r f o r m a n c e . This m a y indicate that a Ratel cutoff is not necessary to achieve o p t i m u m performance.
Tracking Behavior The p h e n o m e n o n of atrial a r r h y t h m i a that oscillate b e t w e e n fibrillation and flutter has b e e n discussed (24). Figure 6 contains an SVT episode that
Conclusion The use of rate alone does not suffice for discriminating AF from AFL. The ARC algorithm is a m o r e robust analysis t h a n rate by utilizing the variability of the atrial electrogram. A retrospective analysis of the ARC algorithm on this data set indicates that high levels of specificity and sensitivity m a y be achieved w h e n classifying AF and AFL. The rationale for the use of the ARC algorithm contains clinical and practical c o m p o n e n t s . The potential clinical utility of the ARC algorithm is: 1) enabling preferential use of well tolerated ATP pacing over internal defibrillation for treating selected atrial a r r h y t h m i a , and 2) providing diagnostics to physicians at patient follow-ups. Although i m p o r t a n t limitations of this study should not be ignored, the rationale for, and the p e r f o r m a n c e of, the ARC algorithm warrants consideration for use in implantable cardioverter defibrillators.
Cardiac Rhythm Management
References 1. Cairns JA, Connolly SJ: Nonrhcmnatic atrial fibrilla lion. Risk of stroke at~d role of antithrombotic therapy. Circulation 84:469, 1991 2. Dell'Orfano JT, Patcl 1I, Wolbrctte DL, et al: Acute treatment of atrial fibrillation: Spontaneous conversion rates and cost of care. Am J Cardiol 83:788, 1999 3. Levy S, Breithardt G, Campbell RW, et al: Atrial fibrillation: Current knowledge and recommendations for management. Eur Heart J 19:1294, 1998 4. Kannel WB, Abbot RD, Savage DD, et al: Epidemiologic features of chronic atrial fibrillation; The Franfingham Study. N Engl J Med 306:1018, 1982 5. Brand FN, Abort RD, Kannel WB, et aI: Characteristics and prognosis of lone atrial fibrillation: 30-years follow up in the Framingharn Study. JAMA 254: 3449, 1985 6. Wolf PA, Abbott RD, Kannel WB: Atrial fibrillation as an independent risk factor for stroke; The Framingham Study. Stroke 22:983, 1991 7. Wolf PA, Abbott RD, Kannel WB: Atrial fibrillation: A major contributor to stroke in the elderly. Arch Intern Med 147:1561, 1987 8. Murgatroyd FD, Canlm AJ: Nonpharmoacological M a n a g e m e n t of Atrial Fibrillation. Futura Publishing Company, Inc. Armonk, NY, 1997 9. Jung W, Pfeiffer D, Wolpert C, et al: Which patients do benefit from an implantablc atrial defibrillator? J Am ColI CardioI 27:302A, 1996 (abstr) 10. Reddy RK, Bardy GH: Implantable Dual Chamber Defibrillators for Atrial Defibrillation. In: Murgatroyd FD, Carom AJ (eds): Nonpharmacological Managemerit of Atrial Fibrillation. Futura Publishing Company, Inc, Armonk, NY, p 439, 1997 11. KenKnight BH, Lang D J, Scheiner A, et ah Atrial Defibrillation for Implamable Cardioverter-Deflbrillators: Lead Systems, Waveforms, Detection Algorithms, and Results. In: Singer I, Barold SS, Carom AJ. (eds): NonpharmacologicaI Therapy of Arrhythmias for the 21st Century: The State of the Art. Futura Publishing Co, Inc, Armonk, NY, p 457, 1998 12. A m m e r R, Alt E, Ayers G, et ah Pain threshold for low energy intracardiac cardioversion of atrial fibrillation with low or no sedation [published erratum appears in Pacing Clin ElectrophysioI 1997 Apr;20(4 Pt 1):viii]. PACE 20(1 Pt 2):230, 1997 13. Jung J, Heisel A, Fries R, et al: Tolerability of internal
14.
15.
i6.
17.
18.
19.
20.
21.
22.
23.
24.
9
Morris et al.
t39
low-energy shock strengths currently needed for endocardial atrial cardioversion. Am J Cardiol 80: 1489, I997 Kidwai BJ, Harbtson MT, Allen JD, et ah Waveform optimization [or internal atrial defibrillation: A ne w rounded biphasic trapezoid shock. J Am Coll Cardiol 33:159A, I999 (abstr) Harbinson MT, Alien JD, I m a m Z, et ah Rounded biphasic waveform reduces energy requirements for transvcnous catheter cardioversion of atrial fibrillation and flutter. PACE 20:226, 1997 Ayers GM: How can atrial defibrillation be made more tolerable? In: Murgatroyd FD, Carom AJ (eds): Nonpharmacological m a n a g e m e n t of atrial fibrillation. Futura Publishing Co Inc, Armonk, NY, 1997, p 475 Hii JT, Mitchell LB, Duff HJ, et al: Comparison of atrial overdrive pacing with and without extrastimuli for termination of atrial flutter. Am J Cardiol 70:463, 1992 Waksman R, Pollack A, Berkovits BV, et ah Amodecremental pacing for the interruption of ventricuIar tachycardia and atrial flutter. J Electrocardiol 25:339, 1992 Watson RM, Josephson ME: Atrial flutter. I. EIectrophysiologic substrates and modes of initiation and termination. Am J CardioI 45:732, 1980 Morris MM, Marcovecchio AF, KenKnight BH, et al: Retrospective evaluation of detection enhancements in a dual-chamber implantable cardioverter defibrillator: Implications for device programming. PACE 22:849, 1999 Jenkins JM, CaswelI SA: Detection algorithms in implantable cardioverter defibrillators. Proc IEEE 84: 428, 1996 Jung J, Hohenberg G, Heisel A, e{ ah Discrimination of sinus rhythm, atrial flutter, and atrial fibrillation using bipolar endocardial signals. J Cardiovasc ElectrophysioI 9:689, 1998 Swiryn S, Schoenwald A, Sahakian A: Detection of atrial fibrillation by pacemakers and antiarrhythmic devices. In: Nonpharmacological Management ot Atrial Fibrillation Futura Publishing Company, Inc, Armonk, NY, 1997, p 309 Wietholt D, Hart J, Chen L, e~ ah On the mechanisms of alternation b et w een atrial flutter and atrial fibrillation in man. PACE 23:735, 2000