Morphology discrimination in implantable cardioverter-defibrillators: consistency of template match percentage during atrial tachyarrhythmias at different heart rates

Abstract

Background

Morphology discrimination (MD) in implantable cardioverter-defibrillators (ICDs) is based on the comparison of the ventricular electrogram during tachycardia with a stored reference template obtained during baseline rhythm. However, the effect of heart rate on the template match percentage during supraventricular tachyarrhythmias (SVT) is not known. The purpose of this study was to evaluate the performance of the template match percentage during SVT at different heart rates.

Methods and results

Stored electrograms of 868 tachyarrhythmias from 88 patients with a dual-chamber ICD (St Jude Medical, USA) were analysed by the investigators. The effect of heart rate on template match percentage was estimated by regression analysis. For performance measures, data were corrected for multiple episodes in a patient by using the generalized estimating equation method. The mean template match percentage was 86.6 ± 22.2% (median 100%) for SVT episodes. No significant differences in template match percentage between fast [ventricular cycle length (CL) 300–350 ms] and slow (ventricular CL >400 ms) SVTs were observed (85.4 ± 27.0 vs. 87.1 ± 19.7%). Using nominal settings, MD alone provided sensitivity and specificity of 70.2% and 89.4% overall, respectively. Morphology discrimination in conjunction with rate branch analysis, sudden onset, and stability yielded sensitivity and specificity of 98.5% and 91.2%, respectively.

Conclusion

Morphology discrimination has a consistently high template match percentage during SVTs, which is independent of ventricular CL. The consistent high match percentage results in high specificity for arrhythmia discrimination.

Implantable cardioverter-defibrillator, Arrhythmias, Atrial fibrillation, Morphology, Electrogram

Introduction

The implantable cardioverter-defibrillator (ICD) has become standard therapy in patients who have survived life-threatening ventricular tachyarrhythmias or who are at risk for ventricular arrhythmias. ^1–3 However, delivery of inappropriate therapy due to misclassification of supraventricular tachycardia (SVT) as ventricular tachycardia (VT) is the most reported adverse event in defibrillator patients. ^4–7 Even with advancing ICD technology, accurate rhythm diagnosis to avoid inappropriate device therapy remains a clinical challenge. Recently, morphology-based discrimination (MD) algorithms have been implemented in devices for further improvement in arrhythmia discrimination. ^8–10 These discrimination algorithms focus on the premise that the ventricular electrogram changes during VT compared with the stored template obtained during supraventricular baseline rhythm. A concern of this methodology is that the performance of MD may vary at different ventricular cycle lengths (CLs). Any decrease in the template match percentage during atrial tachyarrhythmias may put the patient at risk for inappropriate device therapy. Therefore, the purpose of this study was to evaluate the performance of the morphology template match percentage during SVTs at different heart rates.

Methods

Study population

The study population consisted of 88 patients who underwent a first ICD implantation for established indications. All patients received a dual-chamber ICD incorporating an MD algorithm (St Jude Medical, Sylmar, CA, USA). During implantation, efforts were made to eliminate the presence of far-field R wave sensing. Defibrillation leads providing true bipolar ventricular sensing were implanted in 98% of the patients. Two patients (2%) had a defibrillation lead with integrated ventricular sensing.

Arrhythmia discrimination and device programming

The initial step in dual-chamber arrhythmia discrimination by the Atlas DR system is rate branch analysis based on the relative rates in the atria and ventricles, which results in three rate branches. These branches are ventricular rate faster than the atrial rate (V > A), ventricular rate slower than the atrial rate (V < A), and ventricular rate equals the atrial rate (V = A). If tachyarrhythmias are detected in the V > A rate branch, the device proceeds to deliver therapy without any arrhythmia discrimination. Based on the rate branch classification V < A and V = A, further arrhythmia discrimination occurs by the application of ‘sudden onset’, ‘stability’, and ‘morphology discrimination’. Ventricular tachycardia diagnosis is based on the combination of discriminators using two types of logic: ‘Any’ logic (deliver therapy if one discriminator indicates VT) and ‘All’ logic (deliver therapy if all discriminators indicate VT).

The MD algorithm compares each ventricular complex during tachycardia with a stored reference template obtained during the patient's baseline rhythm. After alignment of peaks of the tachycardia and the reference template complex, a percent template match score for each ventricular complex is calculated. This template match percentage represents the morphologic similarity of the tachycardia complex to the stored reference template. In all devices, the MD algorithm was activated in nominal settings: number of matches five of eight beats and template match score 60%. On the basis of the MD, an arrhythmia is classified as VT when at least five of eight beats have a template match percentage <60%. Table 1 summarizes the programming of selective discriminators. The morphology template had to be updated at least once every day, except for patients who had a poor match percentage during a temporary atrial pacing test.

Table 1

Programming of arrhythmia discrimination

Parameter	Value
Sudden onset (%)	16
Stability (ms)	30–40
Morphology discrimination
Template match threshold	60%
Number of beats	Five of eight matches
Supraventricular tachycardia upper limit	Ventricular fibrillation
Algorithm logic setting	ALL
Safety timers	OFF

Morphology discrimination and atrial pacing

For the verification of template match percentages, each patient underwent testing by atrial pacing at least 2 weeks after ICD implantation. The device was temporarily programmed to atrial pacing (AAI mode) at rates of 90, 120, and 150 bpm. During atrial pacing, real-time electrograms were recorded to analyse beat-to-beat template percentages. In the case of poor match percentage (<80%), a new template was acquired during atrial pacing and stored as a reference template. In addition, the automatic template update was disabled.

Data collection

Follow-up started at the time of ICD implantation. All patients were regularly followed at 3-month intervals and were advised to contact our outpatient clinic after a symptomatic event as soon as possible. At every follow-up visit or visit prompted by ICD therapy, all stored data of tachyarrhythmia episodes were collected. Two experienced electrophysiologists reviewed the stored electrograms. In the case of disagreement between the two reviewers about the stored electrograms, a third one was consulted and decided. The date of each episode, type, morphology (monomorphic or polymorphic), mean CL of the tachyarrhythmia, type, duration, and outcome of delivered ICD therapy were recorded. The stored arrhythmias were classified as ventricular tachyarrhythmia or atrial tachyarrhythmia without a co-existent ventricular tachyarrhythmia. As the atrial electrogram was present, the presence of atrioventricular dissociation was used to classify ventricular tachyarrhythmia. Otherwise, ventricular tachyarrhythmia was defined as an event with a sudden increase in the rate combined with a change in the electrogram morphology when compared with the baseline rhythm. Stored diagnostic data obtained with each episode provided information on the arrhythmia classification by the tachyarrhythmia discrimination algorithms of the device.

Data analysis

Data are presented as mean ± SD if normally distributed, otherwise by median. The set of tachyarrhythmia episodes cannot be considered as independent as patients can contribute one or more tachyarrhythmia episodes to the data set. To correct for these factors, the performance of MD was analysed by using the generalized estimating equation (GEE) statistical method with an exchangeable correlation structure to correct the varying number of episodes from each patient. ^{11 , 12} Sensitivity was defined as the number of VT episodes classified by MD divided by the total number of VT episodes classified by the investigator. Specificity was defined as the number of SVT classified by the MD algorithm divided by the total number of SVT classified by the investigator.

Mean template match percentages for all the episodes of VT and SVT were calculated. The effect of heart rate on template match percentage was estimated by the regression analysis. Statistical analysis was performed with SPSS (release 11.5, SPSS Inc., Chicago, IL, USA) and SAS (release 8.2, SAS Institute, Cary, NC, USA) for Windows.

Results

Study population

The study population consisted of 88 patients. The clinical characteristics of the patients are summarized in Table 2 . The ICD was indicated as primary prophylaxis in 38 patients (43%) and as secondary prophylaxis in 50 patients (57%). At the time of implantation, all patients had sinus rhythm. The mean QRS duration was 127 ± 33 ms, and the QRS duration of at least 120 ms was present in 50% of the patients. A history of atrial tachyarrhythmias was documented in 19 patients (22%), paroxysmal atrial fibrillation in 12 patients, and paroxysmal atrial flutter in 7 patients. The programmed fibrillation and tachycardia zones were 283 ± 13 and 366 ± 45 ms, respectively. Eighteen patients (21%) had two tachycardia zones activated.

Table 2

Patients' clinical characteristics ( n = 88)

Patient characteristics	n (%)
Age (years)	60 ± 14
Male	68 (77)
Arrhythmic indication for ICD
Ventricular fibrillation	19 (22)
Ventricular tachycardia	31 (35)
Non-sustained ventricular tachycardia	38 (43)
Prior history of atrial tachyarrhythmias	19 (22)
Ejection fraction	30 ± 9
Underlying cardiac disease
Coronary artery disease	52 (59)
Dilated cardiomyopathy	12 (13)
Hypertrophic cardiomyopathy	7 (8)
NYHA classification
I	21 (24)
II	57 (65)
III	10 (11)
Pharmacological treatment
Amiodarone	19 (22)
Beta-blockade	58 (66)
Digoxin	11 (13)
ACE-inhibitor	57 (65)
Diuretics	47 (53)
Lipid-lowering drug	43 (49)

Continuous data are presented as mean ± SD.

ACE, angiotensin-converting enzyme.

Morphology discrimination and atrial pacing

During atrial testing, the overall mean template match percentage was 84.6 ± 27.7%. None of the patients developed atrial fibrillation due to atrial pacing. Nine patients (10%) demonstrated atrial pacing-induced changes in the template match percentage ( Figure 1 ). The mean template match score was 50.0 ± 24.1%. In these patients, a new template was aquired during atrial pacing at 150 bpm, and the automatic template update was disabled.

Figure 1

Atrial pacing-induced changes in the template match percentage. From top to bottom: surface ECG, marker channel (template match percentage, atrial intervals, and ventricular intervals), near-field ventricular electrogram, and far-field ventricular electrogram. A, atrial-paced event; R, ventricular-sensed event; √, match with stored reference template; X, non-match with stored reference template.

Spontaneous tachyarrhythmias and morphology discrimination

The mean follow-up was 34 months, with a cumulative follow-up of 2959 patient-months. During this follow-up, a total of 1029 episodes were recorded in 47 (53%) patients. Of these episodes, 150 were excluded because of detection in the fibrillation zone, where discrimination algorithms did not apply. Another 11 episodes were excluded because of the feature bigeminal avoidance. Thus, 868 episodes were eligible for analysis. On the basis of the physician classification, there were a total of 486 episodes of true SVT and 382 episodes of true VT ( Figure 2 ). In Figure 3 , the number of episodes by CL is presented.

Figure 2

Number of spontaneous episodes per patient for ventricular tachycardia (VT) and supraventricular tachycardia (SVT). The error bars extend down to the minimum value and up to the maximum value. The box extends from the 25th percentile to the 75th percentile, with a black box at the median (50th percentile).

Figure 3

Number of supraventricular tachycardia (SVT) and ventricular tachycardia (VT) episodes by cycle length.

In 30 patients, a total of 486 SVT episodes (mean ventricular rate 405 ± 78 ms) occurred. The mean template match percentage was 86.6 ± 22.2% (median 100%). At high heart rates (CL 300–350 ms), the mean template match percentage was 85.4 ± 27.0%. The mean template match percentage was 87.1 ± 19.7% at low heart rates (CL >400 ms). With nominal programming, 453 of the 486 SVT episodes were correctly classified by MD, which results in an absolute specificity of 92.6% (GEE-corrected 89.4%). The ventricular electrogram was significantly changed in 33 SVT episodes. During these episodes, the mean template match percentage was 29.5 ± 29.3% (median 23.0%).

A total of 382 VT episodes (mean ventricular rate 415 ± 94 ms) were recorded in 30 patients. The device correctly classified 377 of the 382 episodes as VT, yielding a sensitivity of 98.6% (GEE-corrected 98.5%). Of these 377 episodes, 374 (99.2%) were detected in the V > A rate branch. The sensitivity for VT detection in this rate branch was 100%, which is based on the comparison between atrial and ventricular rates. The mean template match percentage was 37.9 ± 40.9% (median 10.5%). Based on MD alone, 236 of the 382 VT episodes were correctly classified, resulting in an absolute sensitivity of 61.8% (GEE-corrected 70.2%).

Consistency of morphology discrimination

For all SVT episodes, the mean template match score was 86.6 ± 22.2%. In order to analyse the effect of heart rate on percent template match score, the linear regression analysis was performed ( Figure 4 ). Comparing the mean template match percentage at different heart rates during SVT, the template match percentage seems to be very consistent, even at increasing heart rates. Sensitivity and specificity values were calculated for MD as a single discriminator at different heart rates ( Table 3 ).

Figure 4

Scatterplot of template match percentage by cycle length for supraventricular tachycardia (open square, SVT) episodes. The lines represent the mean template match percentage including the 95% confidence intervals of a linear regression analysis for supraventricular tachycardia (dashed line).

Table 3

Sensitivity and specificity for morphology discrimination at different heart rates (cycle length)

Cycle length (ms)	Episodes ( n )	Patients ( n )	Sensitivity		Specificity
			Absolute (%)	GEE-corrected (%)	Absolute (%)	GEE-corrected (%)
300–400	471	43	70.6	72.9	91.2	85.1
300–350	257	36	80.8	72.5	92.0	84.4
355–400	214	22	55.1	89.7	88.2	91.9
405–450	106	9	77.4	77.8	97.3	91.6

GEE, generalized estimating equation.

Morphology discrimination provided a high specificity to distinguish SVT from VT, independent of ventricular CL. The loss in sensitivity by MD is compensated by the rate branch analysis in dual-chamber devices. The combination of rate branch analysis and MD resulted in a sensitivity of 98.6% (GEE-corrected 98.5%) and a specificity of 94.2% (GEE-corrected 91.2%). For fast tachyarrhythmias (CL 300–350 ms), sensitivity was 98.3% (GEE-corrected 98.8%) and specificity was 88.3% (GEE-corrected 84.3%). For tachyarrhythmias with CL >350 ms, sensitivity was 98.8% (GEE-corrected 98.1%) and specificity was 96.8% (GEE-corrected 96.4%).

Discussion

The present study evaluated the template match percentage of the MD algorithm during SVTs at different heart rates. As a single discriminator, the MD algorithm demonstrated a consistently high template match percentage during SVT, independent of ventricular CL. Specifically, MD with nominal settings resulted in an overall specificity of 92.6% (GEE-corrected 89.4%). In contrast, MD as a single discriminator has limited sensitivity for appropriate diagnosis of VT (GEE-corrected sensitivity of 70.2%). The loss in sensitivity is compensated by the rate branch analysis in dual-chamber devices.

The delivery of inappropriate ICD therapy due to misclassification of atrial tachyarrhythmias is a substantial complication in ICD recipients. ^4–7 Accurate discrimination of tachyarrhythmias is critically important to prevent inappropriate device therapy. Arrhythmia discrimination based on measurements of arrhythmia onset, stability, and AV relationship has provided mixed results. Challenges for these interval-based discrimination algorithms are SVTs with 1:1 AV conduction ^{13 , 14} and atrial fibrillation with rapid ventricular response. ¹⁵ A strategy to improve arrhythmia discrimination further is to apply morphology-based algorithms, which are based on the morphological comparison of the ventricular electrogram in sinus rhythm and those during tachycardias. ^16–18 Previous studies evaluated the performance of the MD algorithm. ^{9 , 19–21} In a prospective study, MD detected 77% of the VTs and appropriately classified 71% of the SVTs when applied to ventricular CLs ranging from 288 to 401 ms. ¹⁹ Using a per-episode analysis, Glikson et al . ²¹ reported a specificity value of 82% (GEE-corrected 85%) for SVTs with cycle lengths ranging from 315 to 590 ms. Using the same method of analysis, the present study observed high specificity values for MD of SVTs, independent of ventricular CL (88.2–97.3%; GEE-corrected 85.1–91.1%). Despite the high specificity, the MD algorithm has limited sensitivity to detect VTs (GEE-corrected 70.2%). The limited sensitivity was also found in previous studies. ^{19 , 22} The MD algorithm uses a near-field electrogram from the bipolar sensing electrodes as an electrogram source for morphology analysis. The analysis of near-field electrograms has the potential to reduce sensitivity for the detection of VTs, as this electrogram source provides only information on local activation. In clinical practice, any loss in sensitivity exposes the patient to potential risk. In dual-chamber devices, the loss in sensitivity is compensated by the rate branch analysis (V > A), in which MD does not apply. ^20–22 Recently, the MD algorithm was updated to improve the alignment of complexes by the tallest peak. The effect of this improved algorithm on sensitivity is not known.

Apart from the MD algorithm, two other MD algorithms are implemented in current ICDs: rhythm ID and wavelet. Rhythm ID uses both electrogram sources for morphology analysis: a near-field electrogram from the bipolar sensing electrodes and a far-field electrogram constructed between the defibrillation coils and can. For MD, this algorithm uses a vector time and correlation analysis, which is based on the measurement of voltage differences over time. In the dual-chamber mode, the rhythm ID algorithm incorporates interval-based discriminators. Arrhythmia discrimination in the single-chamber mode is based solely on MD. Gold et al . ⁸ reported on the performance of the rhythm ID algorithm applied to induced arrhythmias and showed 99% sensitivity and 97% specificity in the single-chamber mode. In a recent study, the rhythm ID algorithm in the single-chamber mode achieved an overall specificity of 91.7% (GEE-corrected 88.4%) for spontaneous SVT episodes without compromising sensitivity. ²³ This reported specificity is nearly identical to the 92.6% (GEE-corrected 89.4%) overall specificity observed in our study for MD.

The wavelet algorithm uses a far-field electrogram source (distal coil to can) for morphology analysis. Swerdlow et al . ²⁴ reported on the performance of the wavelet algorithm applied to induced and spontaneous arrhythmias and showed a sensitivity of 100% and a specificity of 78%. In a recent study evaluating the wavelet algorithm under clinical circumstances, the algorithm achieved a specificity of 75.8% for SVT and a sensitivity of 98.6% for sustained VT episodes. ¹⁰ The algorithm failed to discriminate 175 of the 724 SVT episodes (24.2%) with CLs, where the algorithm was programmed to apply (311–401 ms). For SVT episodes with a similar range of CLs, the MD algorithm in our study demonstrated a specificity of 91.2% (GEE-corrected 85.1%). The major reason reported for inappropriate classification of SVTs by the wavelet algorithm was a change in the electrogram morphology during tachycardia when compared with the stored reference template. Possible explanations for changes in the electrogram morphology are rate-related aberrancy, influence of anti-arrhythmic therapy, progression of underlying cardiac disease, and postural changes in the far-field electrogram. ²⁵

The evaluated MD algorithm proved to be accurate and stable during atrial tachyarrhythmias at different heart rates. Combined with interval-based discriminators, e.g. stability and ‘adaptive’ onset settings with the logic ‘All’, the MD resulted in high specificity. In addition, a method to prevent inappropriate classifications of SVTs is atrial testing to assess the template match percentages. ²⁶ In our study, only a minority of patients (10%) demonstrated poor template matching during atrial testing. In these patients, a reference template was loaded during fast atrial pacing, and the automatic update of templates was deactivated. In order to optimize arrhythmia discrimination and to prevent inappropriate ICD therapy, we recommend to perform template match analysis by temporary atrial testing and to optimize arrhythmia discrimination settings accordingly.

Limitations

The results of this study apply only to this particular MD algorithm and cannot be generalized to MD algorithms of other manufacturers. In addition, it should be noted that the results of this study apply only to dual-chamber devices and should not be generalized to single-chamber devices. The rate branch analysis and more important temporary atrial testing are only available in dual-chamber devices. Morphology discrimination can achieve high specificity values, but at the expense of sensitivity to detect VTs. The loss in sensitivity caused by MD is compensated by the rate branch analysis in dual-chamber devices. Further, the MD algorithm was tested at their nominal settings, although other settings may have been better. In a previous study, we demonstrated by the ROC analysis that any improvement in sensitivity was accompanied by a loss in specificity and vice versa. ²²

Conclusion

In the present study, the MD algorithm demonstrated a consistently high template match percentage during SVT, which was independent of ventricular CL. This consistent high match percentage resulted in a high specificity for MD, which prevents inappropriate classification and subsequently inappropriate therapy. In contrast, MD has a limited sensitivity for detection of VT. In dual-chamber devices, the loss of sensitivity was compensated by the rate-branch analysis, as the majority of VT episodes are classified in the V > A rate branch.

Conflict of interest: D.A.M.J.T. received research grants from Biotronik (Netherlands), Boston Scientific (Netherlands), and St Jude Medical (Netherlands) and is a consultant for Cameron Health (USA). L.J.J. received research grants and speaker fees from Biotronik, Boston Scientific, Medtronic, Sorin, and St Jude Medical.