Beat-to-Beat Repolarization Variability in LQTS Patients with the SCN5A Sodium Channel Gene Mutation

Beat-to-Beat Repolarization Variability in LQTSPatients with the SCN5A Sodium Channel GeneMutation

JEAN-PHILIPPE COUDERC, WOJCIECH ZAREBA. LAURA BURATTINI,and ARTHUR J. MOSS

From the Cardiology Unit, Department of Medicine University of Rochester Medical Center,Rochester, New York

COUDERC, J., ET AL.: Beat-to-beat Repolarization Variability in LQTS Patients with the SCN5A SodiumChannel Gene Mutation. Current techniques evaluating beat-to-beat variability of repolarization rely onaccurate determination of T wave endpoints. This studvproposes a Twave endpomt-independent methodto quantify repolarization variability in a standard 12-iead ECC using a wavelet transformation. Ourmethod was used to identify repolarization variability in long QT syndrome patients (LQTS) with theSCN5A sodium channel gene mutation. Using wavelet transformations based on the second Caussianderivative, we evaluated repolorization variability in 11 LQTS patients wiih the mutation, 13 noncarrierfamily members, and 28 unrelated healthy subjects. Time-domain repolarization variability parameters(SDRTo. SDBTm) and wavelet parameters describing temporal (beat-to-beat) variability of repolarizationin time (TVT) and in amplitude (TVA) were analyzed. Reproducibility of wavelet parameters and rela-tionship of wavelet-based variability with heart rate and preceding RR interval were investigated. Thewavelet-based method quantified beat-to-beat variability of the entire repolarization segment (regardlessofQT interval identification) providing insight into variability in repolarization morphology. Our methodshowed that SCN5A carriers have significantly increased repolarization variability in amplitude (23% ±14% vs S + 4%. P < 0.001) and in time (14 ± 17 ms vs 3 ± 2 ms. P < 0.004) compared to noncarriers.Variability of repolarization amplitude was found to be heart rate dependent with variability decreasingwith increasing heart rate. Relative error describing reproducibility of TVA and TVT was ^5% and ^10%, respectively. Our method quantifies repolarization variability in amplitude and in time without theneed to identify T or U wave endpoints. Wavelet-detected repolarization variability contributes to pheno-typic identification of SCN5A carriers, with more pronounced beat-to-beat variability in repolarizationamplitude than in time. (PACE 1999; 22:1581-1592)

long QT syndrome, ventricular repolarization, wavelet, beat-to-beat variability, SCN5A genemutation

Introduction

Recent experimental and clinicalhave shown that beat-to-beat variability of repo-larization morphology is associated with an in-

This work was supported in part by grants from NIH (ROl-HL-33843) and Marquette Electronics, Inc., Milwaukee, Wiscon-sin, USA.

Address for reprints: Jeiin-Philippe Couderc, Ph.D., Heart Re-stiarnh, 601 Eimwood Avenue, Box 653. Rochestor, New York14642. Fax: (716) 473-2751; e-mail: heartjpcfoiheart.rochester.edii

Received ]uly 29, 1998: revised December 21, 1998; acceptedFebruary 17.'l999.

creased risk of arrhythmic events. ECG analysis ofbeat-to-beat variability of repolarization is fre-quently performed by automatic computation ofQT or QT apex intervals/"'^ However, these algo-rithms have limited performance and repro-ducibility for measuring repolarization durationon a beat-to-beat basis due to the influence of therandom noise, baseline wandering, and respira-tory modulation on T wave shape. Another limita-tion of the QT duration-based techniques used forbeat-to-beat repolarization variability is that theydo not provide insight into beat-to-beat variabilityin the shape of repolarization waves. Therefore,new methods evaluating beat-to-beat changes in

PACE, Vol. 22 November 1999 1581

COUDERC, ET AL.

repolarization morpbology are needed to enhancethe clinical usefulness of repolarization variabil-ity in the identification of patients at risk for ma-lignant ventricular arrhythmias. The primary oh-jective of this study was to develop and test a nevi'signal-processing approach for the quantificationof heat-to-heat repolarization variahility in a stan-dard 12-lead ECG using a wavelet decompositiontechnique. Our prior experience with wavelettransformation technique used for the late poten-tial detection in the high resolution signal-aver-aged ECG indicated potential usefulness of thistechnique for repolarization analysis.^ Indeed, thewavelet transformation was chosen hecause it pro-vides insight into time frequency components ofthe entire repolarization interval optimizing thetrade-off between time and frequency resolution.The secondary objective of this study was to de-termine whether beat-to-heat variability of repo-larization can he useful in phenotypic characteri-zation of long QT syndrome patients with theSCN5A sodium channel gene mutation.

Methods

Study Population

The study population consisted of threegroups: 11 LQTS patients with the SCN5A sodiumchannel gene mutation (SCN5A carriers; 26 ± 15years, 5 males and 6 females, QT^ = 607 ± 23 ms,heart rate: 53 ± 12 heats/min), 13 unaffected fam-ily members (SCN5A noncarriers; 25 ± 20 years, 6males and 7 females, QT^ = 390 ± 26 ms, heartrate: 70 ± 15 beats/min) from 2 LQTS families,and 28 healthy subjects [32 ± 13 years, 16 malesand 12 females, QT^ = 390 ± 26 ms, heart rate: 66± 1 1 beats/min). An additional group consistingof 10 healthy subjects (23 ± 11 years, 6 males and4 females) in whom ECGs were repeated for 5 min-utes and 24 hours after first recording (heart rate:67 ± lObeats/min, 65 ± 11 beats/min and 71 ± 11beats/min, respectively) was used to evaluate thereproducibility of the wavelet transformationmethod. All subjects gave informed consent forparticipating in the study.

Data Acquisition and Time-domainQuantification of Beat-to-Beat Variability

The standard 12-lead ECGs were recorded us-ing the MAC12 digital 12-lead ECG machine (Mar-

quette Electronics, Inc., Milwaukee, WI, USA)providing a 250-Hz sampling rate of signal ECGacquired continuously for 10 seconds. The Rpeaks were detected using an adaptative thresholdfor the amplitude of the first derivative of theECG.̂ Only ECGs without premature beats wereconsidered. The time-domain repolarization pa-rameters were computed after preprocessingphases involving baseline removal and the attenu-ation of respiratory modulation. ̂ "'"̂ ^ The durationof the RTm interval segment (the time from the Rpeak to the maximum amplitude [apex] of the Twave) and the RTo interval (the time from the Rpeak to the T wave offset) were automaticallycomputed for each beat. Detection of the T waveapex and offset were based on automatic algo-rithms previously described.'^ Standard deviationof RTm intervals (SDRTm) and RTo intervals(SDRTo) in consecutive beats of the same leadwere computed, and their median values in 12leads were determined.

Beat-to-Beat Variability Quantified Using theWavelet Transformation Method

The 10-second digital ECG signal was pro-cessed following several steps including: repolar-ization windowing, wavelet transformations, syn-chronization of wavelet transformations using across-correlation technique (measure of beat-to-beat variability in time), calculation of the map-ping of time-frequency variability, and detectionof maxima for quantifying beat-to-beat variability.

Mathematical details ofthe wavelet transformprocessing are provided in the Appendix. Briefly,the calculation of the ECG interval used for theanalysis of ventricular repolarization is processed.This interval was defined independently of Twave endpoints identification and encompassedthe interval between the points located 100 ms af-ter the R peak (to reduce potential influence oftheQRS complex) and 220 ms before the next R peak(to avoid potential influence of P wave). The lastheat ofthe 10-second recordings was not analyzedhecause it usually does not include the entire re-polarization segment. Because the duration of thisinterval may vary for each beat, the shortest inter-val in a 10-second series was used for each con-secutive beat. Then, the ECG was filtered with awavelet function based on the second Gaussian

1582 November 1999 PACE, Vol. 22

BEAT-TO-BEAT REPOLARIZATION VARIABILITY IN LQTS

derivative.^^ This wavelet transformation corre-sponds to a 10 redundant frequency-band filteringencompassing frequencies between 4 and 78 Hz.Figure 1 gives an example of the wavelet filteringof the repolarization segments from two consecu-tive beats from a representative ECG. Wavelet fil-tering was applied to raw ECG signals from whichthe components of the QRS were partially re-moved using a Hanning window. Removal of mainQRS components is necessary to avoid introduc-ing undesirable components in the wavelet trans-formation ofthe ECGs. Indeed, the low frequencywavelet has long duration (see wavelet durationon Fig.l), and the onset ofthe repolarization seg-ment can be slightly affected by the QRS compo-nents.

ECG. Lead V5

Figure 1. Examples of wavelet transformations of twoconsecutive beats from the lead Vs of a LQTS patientwith the SCN5A sodium channel gene mutation. Theupper panel contains an ECG including two consecutiveHR intervals. The lower panel is the three-dimensionalrepresentation ofthe wavelet transformation (front view)of the corresponding ECG repolarization segment. Thethree dimensions are time (x axis in ms), amplitude ofwavelet transformation (y axis, eq. mV), and frequencybands represented by the ten curves (z axis in Hertz). Thewavelet transformation is applied to the raw ECGsignals, and it corresponds to filtering the ECGcomponents following the 10 frequency bands with atime precision given by the wavelet duration. Tbe beat-to-beat variability is based on the quantification ofamplitude and time shift changes between the 10 filteredrepolarization segments constituting the wavelettransformations.

Two quantifiers of temporal (or heat-to-beat)variability in amplitude (TVA) and in time (TVT)were designed. These parameters quantify mor-phological changes of the entire ventricular repo-larization segment. TVA and TVT have a value foreach pair of consecutive beats of the 10-secondECG recording including all 12 leads.

These parameters are computed for each con-secutive pair of beats. The TVT parameter is basedon a synchronization process that is adjusted forbeat-to-beat changes in T wave amplitude distri-bution in the prespecified window. The synchro-nization is based on a cross-correlation technique.The TVT calculation is illustrated in Figure 2.Wavelet transformations are first synchronizedusing the R-peak location. Then, cross-correlationfunctions are applied individually to each fre-quency band of the two consecutive beats. Theshift needed to align each filtered segment is oh-tained from the maximal value of the cross-corre-lation sequence (Fig. 2E and F). TVT is the medianvalue of time shifts calculated from the first fivelow frequency hands (including frequencies be-tween 4—23 Hz). Higher frequencies were not usedbecause they did not contribute significantly tobeat-to-beat variability. This approach has heenpreviously validated using simulated ECGs.̂ ^Based on the median values from redundant fre-quency components, TVT provides stable and ac-curate estimation of the shift between amplitudedistribution of two noisy signals.

TVA quantifies temporal variability in ampli-tude. As shown in Figure 3, the wavelet transfor-mations are resynchronized (using the TVT value)and normalized in amplitude (using the maximumamplitude from the wavelet transformation of thetwo consecutive heats). Normalized amplitude isused to estimate beat-to-beat variability with sim-ilar reference in each patient. A mapping of thetemporal time-frequency variability was com-puted corresponding to the subtraction ofthe syn-chronized wavelet transformations ofthe two con-secutive heats. The computation of TVA consistedof tracking the evolution of maxima in the map-ping. A maximum was defined as the point wherethe wavelet transformation magnitude was greaterthan the amplitude of its two nearest samples.These maxima represented singularities of the re-polarization segment due to variability of T wavemorphology and noise (Fig. 3C and D). All max-


COUDERC, ET AL.

Healthy subject, l̂ cad VS SCN5A Carrier. Lead VS

ECGs

—IJent b

— BealbM

>E,

<

0.7

oa0-5

0-1

03

0-2

01

0

-0.1

B A

k111H IH 11

jlj

— Beat b— Beaih*-!

1soo

Tinwimsl

Wavelet Inmslbnniaions

1000Tiiiie (ms)

10000

8000-

6000

dOOO

2000

0

/

/

/ /

0 TVT=Oms

\

0 \

\ \

0 \ \

Cross-coirdati on sequences(10*

- 4 0 0 -200 200 -100 600Shift |ms)

-1000 -500 -250

Figure 2. This figure illustrates the processes involved for obtaining the TVT value estimatingbeut-io-beat variability between a pair of consecutive beats. (A, B) superimposed repolarizationintervals from two consecutive beats for a healthy subject and for a SCN5A gene carrier.respectively. (G. D] corresponding superimposed wavelet transformations of the ECGs portionsfor all frequency hands. (E. F) cross-correlation sequences between the wavelet transformation areplotted for frequency bands between 4-23 Hz (5 frequency bands). The solid vertical line locatesthe zero shift ahscissafor the cross-correlation sequence. Time location of maximum of each cross-correlation sequence are dotted and laheied. For the healthy subject, no shift is needed toinorphologically synchronize the different frequency components. However, for the SGN5A carrierpatient, a shift of 20 ms is detected between the wavelet transformations. This shift is markedusing the dashed verticale line. The scale of the upper panels were adjusted to provide bettervisualization of recorded signals.


•E.

<


Healthy subjecr. l*ad VS SCN5A Cmier. [.eod V5

Synchronized wavekt t n n s forma (ions

— Beatb— Beat h+l

— Beaib•—Beaib+I

800Timelmsl

200 400 BOO 800 1000

; (ms)

0.2

015

Mapping of temporal time-rrcqycncy variability• 25

0-25

0.2

=E 015

<

.0.1

fl:05

300Timclmsi

Accepted maxima as repolarizalion variability0 35

TVA=3 %

J 0.15 •

Figure 3. This figure illustrates how TVA is computed. (A, B) contain the same curves as Figure2 C and D. but after the synchronization using the TVT value (20 ms for the carrier ofLQT3, noshift for the healthy subject) and amplitude normalization. Wavelet transforms of the twoconsecutive beats are subtracted leading to a representation corresponding to a mapping of thetemporal time-frequency variability (G. D) ofthe repolariz.ation segment. This mapping shows thatthe SGN5A gene carrier (right column, lower panel) has higher amplitude than the healthy subjectwhere mapping is pratically flat (left column, lower panel). On these same panels, maximasatisfying the amplitude criterion (see details in text) are marked by dots. The last lower panelscontain the same mappings in which only maxima satisfying all criteria are displayed (frequencyextent and location criteria). For the healthy subject, one set of aligned maxima is detected withan amplitude extent of 3%. For the SCN5A carrier, two sets of maxima are accepted as T wavebeatto-beat variability with an amplitude of 2% and 21 % leading to a TVA value of 23%.

PACE. Vol. 22 November 1999 1585

COUDERC, ET AL.

ima were connected to maxima from adjacent fre-quency bands if tbe deviation between their re-spective time localization was < 16 ms (4 sam-ples). The following set of criteria have beendesigned:

1. Amplitude: only maxima below the upperamplitude of 90% of the detected local maximaare used to estimate the variability in amplitude.

2. Frequency extent: only maxima connectedwith at least two other maxima from adjacent fre-quency bands are retained.

3. Frequency location: lines of connectedmaxima must have at least one of the maxima lo-cated in the first three frequency bands corre-sponding to frequencies between 4 and 14 Hz.

This set of empirical criteria makes it possibleto separate informative maxima from maxima dueto noise. The TVA value corresponds to the differ-ence in amplitude between the highest and lowestconnected maxima values in all accepted lines ofconnected maxima (Fig. 3 E and F). It is importantto note that more than one line of connected max-ima can be detected in the same mapping. For ex-ample, in Figure 3, changes of the T-apex ampli-tude and T wave up-slope contribute to increasingthe TVA value.

The reported TVT and TVA values were com-puted between each pair of consecutive beats inall 12 precordial leads, and median values from allsets of consecutive pairs of beats were averaged forthe 12-lead ECG.

Statistical Analysis

The results are presented as mean ± standarddeviation and median. Comparisons ofthe valuesof each parameter between groups were per-formed using the nonparametric Mann-Whitneytest and Chi-square test where appropriate. Re-gression analysis was used to determine the rela-tionship between repolarization variability pa-rameters and mean RR interval and RR variabilityof preceding beats. The heart-rate adjusted TVAthreshold was determined by identifying the 97.5percentile regression line for healthy subjects. Re-producibility of repolarization variability parame-ters was evaluated using the relative error compu-tation [[Measurement A—Measurement B) /(Measurement A + Measurement B) / 2)]. A Pvalue < 0.05 was considered significant.

ResultsT Wave Variability in SCN5A Carriers

Table 1 shows time-domain and wavelet-based parameters of repolarization variability inthe three studied groups of healthy subjects,SCN5A carriers, and their noncarrier family mem-bers. LQTS patients with the SCN5A sodiumchannel gene mutation had significantly higher re-polarization variability as measured by time-do-main (P < 0.02) and wavelet (P < 0.005) parame-ters. Repolarization variability in amplitude andSDRTm were the most significant parameters dif-ferentiating carriers and noncarriers. No signifi-cant differences were found in any of the repolar-ization variability measures between healthysubjects and the SCN5A noncarriers group. Thebottom panel of Figure 3 shows an example of in-creased amplitude of time-frequency componentsin the SCN5A gene carriers (panel F) compared tohealthy subject (panel E). The discriminant powerof wavelet in comparison to time-domain parame-ters of repolarization variability was assessed us-ing cutoffs defined as the value > 97.5 percentilefor healthy subjects (Table I). Using these criteria,increased repolarization variability in time wasidentified in 36% of SCN5A carriers and 15% ofnoncarriers when tested with the time-domainmethod (SDRTo) and in 55% of carriers and noneof noncarriers when using TVT. SDRTm could beconsidered as a time-domain surrogate of variabil-ity in repolarization morphology and was found tobe elevated in 55% of carriers. TVA measuringbeat-to-beat changes in repolarization segmentamplitude was increased in 64% of carriers and8% of noncarriers.

The combination of TVA and TVT parametersdid not significantly improve the diagnostic per-formance of the wavelet method, indicating that Twave amplitude variability is the dominating fea-ture of repolarization variability. These character-istics of the repolarization segment are illustratedin Figure 4, which shows the plotted, superim-posed consecutive T waves for a healthy subjectand a SCN5A patient (with an R-peak synchro-nization). The healthy subject has very small vari-ability ofthe T wave shape and duration, whereasthe SCN5A carrier patient has increased variabil-ity ofthe T wave shape and amplitude with negli-gible variability in the T wave duration. In this ex-



Table I.

Repolarization Variability by Time-Domain and Wavelet Methods in the SCN5A Carriers, SCN5A Noncarriers,and Healthy Subjects

RepolarizationVariability

Parameters

Time DomainSDRTo (ms)

SDRTm (ms)

WaveletsTVT {ms)

TVA (%)

mean ± sdmedian

> 14 ms*mean ± sd

median> 7 m s *

mean ± sdmedian> 6 ms*

mean ± sdmedian

> 15%'t

HealthySubjects(N - 28)

7.7 ± 3.26.3

2 (7%)3.6 ± 1.7

3.11 (4%)

2.0 ± 4.0170

7.2 ± 4.06.6

1 (4%)

SCN5ANoncarriers

(N - 13)

8.6 ± 5.56.6

2 (15%)3.2 ± 1.0

3.00

2.8 ± 1.74,00

8.0 ± 3.96.6

1 (8%)

SCN5ACarriers(N-11)

30.6 ± 39.011.0

4 (36%)28.6 ±41.8

7.36 (55%)

13.7 ± 16.78.0

6 (55%)22,7 ± 13.8

18.67 (64%)

P Valuest

0.0150.23

<0.0010,006

0.0040.006

<0.0010,004

tP value from the Mann-Withney nonparametric test for comparison of the SCN5A carriers versus noncarriers. Comparison of SCN5Anon-carriers and healthy subjects did not show significant differences. 'Criteria based on a cutoff defined as the value > 97,5% percentilefor healthy subjects, tThe value of threshold is heart-rate adjusted following analysis shown in Figure 5, SDRTo = standard deviationfrom the R peak to the T wave offset; SDRTm = standard deviation from the R peak to the T wave apex; TVA = temporal variability ofrepolarization in amplitude; TVT = temporal variabiiity of repolarization in time.

ample, the wavelet parameters are able to detectand quantify T wave variability, whereas the time-domain parameter measuring total repolarizationduration (SDRTo) cannot because the beat-to-beatvariability appears in the early part ofthe repolar-ization segment.

Relationship Between Wavelet-based Parametersof T Wave Variability and Heart Rate

Investigating the relationship betweenwavelet measures of repolarization variability andthe mean heart rate values (Fig. 5), we found thatin healthy subjects higher heart rate was associ-ated with lower variability of repolarization am-plitude (P = 0.001, r = 0.6). In the LQTS patients,the small number of patients did not allow us todraw a definitive conclusion regarding the TVAand heart rate relationship. However, slope of re-gression line was similar to the slope for healthysubjects. No significant relationship betweenheart rate and TVT values was observed in healthy

subjects and in LQTS patients. To evaluate repo-larization variability after adjustment for heartrate, we compared TVA values in SCN5A genecarriers to TVA values predicted for relevant heartrate. The TVA values were > 97.5 percentile forhealthy subjects in 7 of 11 (64%) LQTS patients.

We also analyzed the influence of the preced-ing RR value on quantitative estimation of the re-polarization variahility. This was obtained by cor-relating the TVA and TVT values with thedifference between the RR values of the studiedand the preceding beats [ARR). As shown in thelower panels of Figure 5, there is a significant, al-though weak, positive linear relation (P — 0.001, r= 0.24) hetween TVA and \RRin the healthy sub-jects. This relation was not confirmed in theSCN5A patients. The ^RR values in SCN5A carri-ers with and without elevated TVA value weresimilar (202 ms vs 164 ms, respectively; P = 0.35).TVT was not correlated with ARR in healthy sub-jects or in SCN5A patients.

PACE, Vol. 22 November 1999

COUDERC, ET AL.

Healthy subject

SDRTm = 0 msSDRTo = 4 ms -

TVA-3 %TVT=lms -

0 200

SCN5A Carrier

800 1000Time (ms)

SDRTm = 4 msSDRTo = 4 ms

TVA=71TVT=10ms

1000Time (ms)

Figure 4. Example of time-domain presentation of beat-to-beat variability and the values of time-domain andwavelet parameters in healthy subject (upper panel) andin the SCN5A gene carrier (lowerpanel). All consecutivebeats in a fO-second ECC recording of V5 are super-imposed.

Reproducibility ofthe Wavelet-basedParameters

Reproducibility of wavelet repolarizationvariability parameters was assessed in 10 healthysubjects by computing the values of time-domain(SDRTo, SDRTm) and wavelet parameters [TVT,TVA) in three sets of ECG recordings: baselineand ECG repeated after 5 minutes and 24 hours(Table II). The reproducihility of these parametersat 5-minute and 24-hour intervals was quantifiedusing relative error. TVA parameter was highlyreproducible with a relative error < 6%. TVT andSDRTm showed a comparable reproducibilitywith a relative error of 10%. SDRTo appeared as a

nonreproducible parameter with a relative errorof 34%.

Number of Beats and T wave Variability

Numher of beats recorded in 10-second 12-lead ECG vary depending on heart rate. Therefore,the question remains whether the magnitude of Twave variability is related to the numher ofrecorded heats. In our analysis, the SCN5A genemutation carriers had a lower numher of heats inthe 10-second ECG than noncarriers and healthysubjects (8 ± 2, 11 ± 3, 10 ± 2. respectively; P =0.005). An arbitrary number of 5 consecutive beats(instead of 10-second ECG) was chosen to deter-mine and compare the TVA and TVT valuesamong three groups. TVA values derived from 5-beat ECG series were significantly higher in carri-ers than in noncarriers, and healthy subjects (32 ±25 %, 10 ± 8 % and 7 ± 7 %. respectively, P <0.05). Respective TVT values were 20 ± 8 ms forSGN5A carriers and 3 ± 2 ms for noncarriers andhealthy suhjects (P < 0.05).

Discussion

We describe a wavelet-based signal-process-ing technique detecting and quantifying beat-to-beat repolarization variability in a digital standard12-lead ECG. Our objective was not to demon-strate that wavelet-based parameters have supe-rior prognostic or diagnostic values than classicaltime-domain parameters, hut rather to describethe benefit of quantifying beat-to-beat variabilityofthe T wave morphology. There are several novelaspects and benefits of this new method worthemphasizing. The most important methodologicaladvantage of the wavelet approach is that it doesnot require exact identification of T wave onset,peak, and offset, and it allows quantification of re-polarization variability independently of QT mea-surement accuracy. This methodological featureof the wavelet-based technique overcomes prob-lems related to poor performance of T wave end-point detection algorithms. The wavelet-hasedmethod quantifies beat-to-heat changes in repolar-ization morphology in the entire segment of inter-est, i.e.. between QRS complex and following Pwave, therefore taking into account the variahilityof ST segment, T waves, and U waves [if present).This approach avoids distinction between the T



Healthy Subjects SCN5A gene mutation

90%

«D%

30%

10%

10%

0%

r=0.60p=0.001

•

•

r=0.2ep=0.43

r=0,00p=0.99

100 150

ARR (ms)

Figure 5. Analysis ofthe relationship between mean heart rate values (HB) and beat-to-beat RBvariability values (ARB) and wavelet measures of beat-to-beat repolarization (TVA) variability inhealthy subjects and SCN5A gene carriers.

Table 11.

Reproducibility of Time-Domain and Wavelet Parameters at 5-Minute and 24-hour Intervais

N - 10 First ECG 5-Minute interval 24-Hour Interval

Time DomainSDRTo (ms)Relative errorSDRTm (ms)Relative error

WaveletsTVT (ms)Relative errorTVA (%)Relative error

6.2

2.9

2.1

6,1

± 1.9

+ 0.6

± 1.8

± 1.6

6.9 ± 1.918%

3.1 ± 0.95%

2.1 ± 1.90%

5.8 ± 2.95%

4.9 ± 1.834%

2.6 ± 0.610%

1.9 ± 1.410%

5.9 ± 2.01%

Relative error was computed using the formula; [(measurement A - measurement B)/(measurement A +measurement B)/2].

PACE. Vol. 22 November 1999 1589

COUDERC, ET AL.

wave and the U wave and is in agreement with theciirrent understanding that the TU complex repre-sents the overall repolarization process.

Currently used techniques measuring themagnitude of the standard deviation of QT (RT)interval duration provide quantification of theoverall changes in repolarization duration in aprespecified set of beats or time intervals withoutinsight into actual variahility of repolarizationmorphology and duration in consecutive heats.For example, the new technique recently pub-lished hy Berger and co-authors,^"* relies on themanual determination of a template heat, andbeat-to-heat variahility is measured using this ar-hitrary reference. In our approach, the beat-to-heatvariability is estimated considering each heat as areference for the following beat. Indeed, the heat-to-beat variability should involve temporal associ-ation between consecutive beats, an approach thathas not been addressed. Beat-to-beat recordings ofionic current kinetics and action potential mor-phology demonstrate that the repolarization pat-tern of an individual beat is influenced by the re-polarization processes and morphology of thepreceding beat.̂ '̂"̂ '̂ * Thus, it seems particularlyimportant to evaluate electrocardiographic repo-larization variability in consecutive pairs of beats.

Beat-to-beat repolarization variability usingthe wavelet transformation technique allowsquantitative comparison of repolarization mor-phology [amplitude and time-duration) in eachpair of beats, thus providing new insight into thedynamic repolarization process. We demonstratedthat the LQTS carriers ofthe SCN5A sodium genechannel mutation have increased time-domainand wavelet-based measures of repolarizationvariability. However, the wavelet-based techniqueprovided a more distinctive phenotypic descrip-tion of LQTS carriers in comparison to noncarri-ers, mainly due to the ability of the waveletmethod to quantify beat-to-beat variahility in am-plitude. As shown in the ECG tracing for anSCN5A gene carrier (Fig. 4), markedly increasedbeat-to-beat variability in repolarization morphol-ogy (expressed by high values of TVA) may occurdespite subtle beat-to-beat variability in repolar-ization duration (expressed hy small values ofSDRTo or TVT). Beat-to-heat variation in the num-ber of ionic sodium channels showing impairedactivation and in the amplitude of leaking current

seem to precipitate beat-to-beat variability in ac-tion potential and T wave morphology.^ •̂ ''•̂ ^ Ourclinical observation on the dominant role of am-plitude variability in the SCN5A carriers empha-sizes the advantage of the wavelet-hased methodover traditional time-domain techniques in de-tecting alteration in repolarization morphology inthis disease entity. However, it should be empha-sized that TVT can he different from zero even forconsecutive T waves with the same duration iftheir morphologies are strongly different. Funda-mentally, changes in the amplitude distributionbetween two consecutive T waves with the sameduration have an effect on amplitude (TVA) andtime-shift (TVT) parameters.

Heart rate influenced T wave variability inamplitude, with higher heart rate contributing tolower TVA values. However (in the limited rangeof heart rates), we did not observe significant asso-ciation hetween heart rate and TVT values. Basedon these observations, the interpretation of theTVA values was heart-rate adjusted. Because ouranalysis was based on resting 12-lead ECC record-ings, we had limited ability to determine heart ratedependency of TVA and TVT parameters in thewide range of heart rate values. Further studies us-ing long-term Holter-recorded ECGs will allow de-termination of full heart rate dependency ofwavelet-hased T wave variahility parameters.

Resting heart rate also affects the numher ofbeats available in a 10-second ECG. We recalcu-lated TVA and TVT values using just 5 arbitrarilychosen consecutive beats from a 10-second seriesto determine whether our wavelet-hased methodis able to identify increased T wave variabilityeven in cases of profound bradycardia. Despite us-ing just four pairs of consecutive heats, wo foundthat SCN5A carriers still show significantly higherTVA and TVT values than noncarriers and controlhealthy subjects. This observation further empha-sizes that our method can identify very transientchanges in T wave morphology.

A clinically important feature of our waveletrepolarization variahility method is that thismethod can be successfully and reproduciblyused for the analysis of repolarization variabilityin standard 12-lead ECG recordings as short as 10seconds. This implies that our algorithm could heincorporated in commercially available digitalECC machines. Longer ECC recordings including



Holter monitoring, exercise testing, or bedsideECG monitoring will be even more suitable foranalysis of repolarization variability and its dy-namic changes over time.

This wavelet-based technique for analysis ofrepolarization variability provides a new approachfor clinical analysis of patients at risk for arrhyth-mic events. In comparison to traditional time-do-main techniques, the wavelet-based method hasthe following advantages: it does not require iden-tification of T wave onset and offset, it encom-passes the entire repolarization segment (betweenQRS complex and following P wave) allowing re-polarization assessment of T and U wave variabil-ity, and it provides insight into variability in therepolarization amplitude. Ability to assess beat-to-beat variability based on pairs of consecutive beatsis another clinically relevant feature of thewavelet-based method. Our study documented theclinical performance of this new technique in longQT syndrome patients with the SGN5A sodiumchannel gene mutation. These patients had signif-icantly greater beat-to-beat variability of repolar-ization amplitude than their unaffected relatives.This clinical observation indicates that wavelet-based parameters of repolarization variabilitymight improve phenotypic characterization ofLQTS patients with different gene mutations.

Appendix

The wavelet transformation of the ECC isbased on a so-called 'mother wavelet'. The secondderivative of a Caussian function, also called'Mexican Hat* has been used in this study. Thediscrete definition of this function is given inequation 1:

<p[n) = (n^ - l]e-""' (1)

The set of 10 wavelets is designed using equa-tion 2 where 10 logarithmically progressing valuesof the parameter a called 'scale parameter' areused. This parameter allows the wavelet to be con-tracted or dilated.

1 , iM (2)

The values of a are obtained using the formulawith a =2" with a varying between 0 and 3.21with a step of 0.35.

The wavelet transformation ofthe ECC calledW|cG of the digitized ECC signal Ecg(n) relative tothe basic wavelet ip''(n) at scale a is defined inequation 3:

= TF-'[E^ X Ip^'] (3)

Acknowledgments: We are grateful to A. S. Manalan, M.D.and I. Harbeck, R,N. far their assistance with the data collection.

Ecg and ip"' are the Fourier transformations ofEcg(n) and of the complex conjugate of thewavelet (f>°(n). TF'^ is the inverse discrete-timeFourier transformation. Time-duration and fre-quency bands of each wavelet have been calcu-lated using the formula described in reference 12.

References

1. Shimizu W, Antzelevitch C. Sodium channelblock with mexilitine is effective in reducingdispersion of repolarization and preventingtorsade de polntes in LQT2 and LQT3 models ofthe long QT syndrome. Circulation 1997; 96:2038-2047.

2. Viskin S, Heller K, Barron HV, et al. Postextrasys-tolic U wave augmentation, a new marker of in-creased arrhythmic risk in patients without thelong QT syndrome. J Am Coll Cardiol 1996;28:1746-1752.

3. Cilmour RF, Riccio ML. Locati EH, et al. Time andrate-dependent alterations of the QT interval pre-cede the onset of torsade de pointes in patientswith acquired QT prolongation. J Am Coll Cardiol1997; 30:209-217.

4. Merri M, Moss AI, Benhorin I, et al. Relation be-tween ventricular repolarization duration and car-

diac length during 24-honr Holter recording. Find-ings in normal patients and patients with long QTsyndrome. Circulation 1992; 85:1816-1821.

5. Merri M, Alberti M, Moss AJ. Dynamic analysis ofventricular repolarization dnration from 24-hourHolter recordings. IEEE Trans Biomed Engineer1993; 40:1219-1225.

6. Lombardi F, Sandrone C, Porta A, et al. Spectralanalysis of short term R-Tapex interval variabilityduring sinus rhythm and fixed atrial rate. EurHeart J 1996; 17:769-778.

7. Bruggemann T, Eisenreich S, Behrens S. et al. Con-tinuous QT interval measurements from the 24-hour electrocardiography and risk after myocar-dial infarction. ANE 1997; 2:264-273.

8. Conderc JP, Fareh S, Chevalier PH, et al Stratifica-tion of time-frequency abnormalities in the signalaveraged high-resolution ECC in post-infarction


COUDERC, ET AL.

patients and congenital long QT syndrome. ] Elec-trocardio! 1996; 29 (Suppl.):18()-188,

9. Merri M, Benhorin J, Alberli M, et al. Electrocar- 14.diographic quantification of ventricular repolar-ization. Circulation 1989; 80:1301-1308,

10, Badilini F. Time and Frequency Analysis of STSegment Displacement Signal in Ambulatory ECCRecordings | Doctoral Thesis]. Rochester, NY, Col- 15.lege of Engineering and Applied Sciences; 1991.

11, Badilini F. Moss AJ, Titlebaum L. Cnbic: splinebaseline estimation in ambulatory ECC recordings.Proceedings IEEE/EMBS Annual Conference. NewYork: IEEE/EMBS, 1991, pp. 584-585. 16.

12. Conderc JP. Zareba W. Bnrattini L, et al. Detectionof abnormal time-frequency components ofthe QTinterval using wavelet transformation technique.In A. Murray, S. Swiryn (eds.): Computers in Car-diology 1997, Lund, Sweden, IEEE Computer So- 17.ciety Press, 1997, pp. 661-664.

13. Couderc JP. Zareba W. Moss AJ. New-method forthe quantification of beat-to-beat T wave temporal 18,variability based on interscale changes in wavelettransform of tbe ECC: Application to simulated ECCsignals. Proceedings LASTED International Confer-

ence, Spain. Anaheim, lASTED/Acta Press 1998,pp. 213-216,Berger RD, Kasper EK, Baughman KL, et al. Beat-to-beat QT interval variability. Novel evidence forrepolarization liability in ischemic and nonis-chemic dilated cardiomyopathy. Circulation 1997;96:1557-1565.Lukas A, Antzelevitch C. Differences in the elec-trophysiological response of canine ventricularepicardium and endocardium to iscliemia. Role ofthe transient outward current. Circulation 1993;88:2903-2915.El-Sherif N, Zeiler RH. Craelius W. et al. QTU pro-longation and polymorphic ventricular tach-yarrhythmias due to bradycardia-dependent earlyafler depolarization. Afterdepolarizations and ven-tricular arrhythmias. Circ Res 1988; 63: 286-305.Bennett PB, Yazawa K, Makita N. et al. Molecularmechanism for an inherited cardiac arrhythmia.Nature 1995; 376:683-685.Shimizu W, Antzelevitch. Characteristics of spon-taneous as well as stimulation-induced torsade depointes in LQT2 and LQT3 models of the long QTsyndrome, (abstract] Circulation 1997; 96:1-554.

1592 November 1999 PACE. Vol. 22

Documents

Beat-to-Beat Repolarization Variability in LQTS Patients with the SCN5A Sodium Channel Gene Mutation