RELATIONSHIP BETWEEN QUANTITATIVE T-WAVE

ALTERNANS ESTIMATES AND PARAMETERS DESCRIBING

CLINICAL STATUS OF PATIENT IN ACCUTE PHASE OF

MYOCARDIAL INFARCTION AND OUTCOME RESULTS

Algimantas Krisciukaitis, Renata Simoliuniene, Andrius Macas, Robertas Petrolis,

Eimante Kamile Puodziunaite, Zita Bertasiene and Viktoras Saferis

Lithuanian University of Health Sciences, Eiveniu Str.4, Kaunas, Lithuania

Keywords: T-wave alternans, Principal component analysis, Myocardial infarction.

Abstract: Relationship between parameters describing clinical status of the patient in acute phase of myocardial

infarction, outcome results and quantitative estimates of T-wave alternans, known prognostic factor of

severe cardiac arrhythmias or sudden cardiac death, was investigated in aim to reveal their usefulness and

incremental diagnostic utility. Integrated estimate, reflecting differences between S-T,T complexes of odd

and even cardiocycles, obtained by means of Principal Component Analysis showed significant correlation

with Left Ventricular Ejection Fraction and rehospitalization of patient within 6 months.

1 INTRODUCTION

ECG T-wave alternans (TWA), a beat-to-beat

alternation in the morphology and amplitude of the

ECG ST segment or T wave, reflects temporal-

spatial heterogeneity of repolarization (Nearing et al.

2003). The evidence linking TWA with arrhythmias

spans more than a century, dating from the

pioneering observations reported in (Hering in

1909). Macroscopic levels of TWA have been

detected under diverse clinical conditions in

association with life-threatening arrhythmias,

including acute myocardial ischemia and infarction,

Prinzmetal’s angina, heart failure, and

channelopathies such as the Brugada and long QT

syndromes (Nieminen et al. 2010). Numerous

studies have demonstrated that screening with

microvolt T-wave alternans (MTWA) testing in

patients with ischemic cardiomyopathy is effective

in identifying patients at high and low risk for

sudden cardiac death (Chow et al. 2006);

(Bloomfeld et al. 2004). Gehi with co-authors

reporting results of meta-analysis about TWA as

predictor of ventricular tachyarrhythmic events

(Gehi et al. 2005) acknowledge that incremental

prognostic value of MTWA when used with other

known risk factors for cardiac arrhythmia, such as

Left Ventricular Ejection Fraction (LVEF) in

patients with ischemic cardiomyopathy, remains

unclear. Replying to this publication Chan with

colleagues (Chan et al. 2006) rises the question

whether TWA is simply a surrogate marker of

patients with greater disease burden and severity, or

it is an independent predictor of cardiac arrhythmias.

To answer this question multivariable modelling that

adjusts for demographics, LVEF, clinical

comorbidities, medication treatment, as well as

electrophysiologic variables (e.g., Holter

monitoring, ECG QRS duration, electrophysiologic

study) should be done.

On the other hand evaluation of TWA over the

last two decades has evolved from visual inspection

of the ECG to the use of computerized analytical

methods for detection of non-visible TWA in the

microvolt range. However clinical performance of

the methods differs and there is no “Golden

Standard” method for detection and evaluation of

TWA so far. Results of “Physionet Challenge 2008”

(Moody 2008) revealed big variety in results

between the methods for detection and evaluation of

TWA even on simulated data. Analysis of clinical

recordings in critical situations is another challenge

for the methods.

Known methods of detection of TWA are giving

more or less comparable results, however estimate

indicating simply presence or absence of the

448

Krisciukaitis A., Simoliuniene R., Macas A., Petrolis R., Puodziunaite E., Bertasiene Z. and Saferis V..

RELATIONSHIP BETWEEN QUANTITATIVE T-WAVE ALTERNANS ESTIMATES AND PARAMETERS DESCRIBING CLINICAL STATUS OF PATIENT

IN ACCUTE PHASE OF MYOCARDIAL INFARCTION AND OUTCOME RESULTS.

DOI: 10.5220/0003856404480452

In Proceedings of the International Conference on Bio-inspired Systems and Signal Processing (BIOSIGNALS-2012), pages 448-452

ISBN: 978-989-8425-89-8

 2012 SCITEPRESS (Science and Technology Publications, Lda.)

phenomena is never used. There is big variety of

methods of quantitative evaluation of TWA giving

sometimes mutually incomparable results. Review

of methods is given in (Martínez et al., 2005). Some

methods give estimates in mV as for alternans only

in amplitude, others give non linear energy estimates

obtained from spectral analysis. There are also

methods based on multivariate analysis (Principal

Component Analysis (PCA) or Karhunen Loeve

Transform) giving linear estimates, but

incomparable with amplitude evaluation, because

they depend on basis functions used for truncated

expansion of the signals. Most of methods are

elaborated using simulated signals (e.g. Clifford et

al. 2008) and tested with clinical recordings where

TWA was evoked by atrial pacing or comparing

recordings of healthy and ichaemic patients. Such

rough evaluation usually does not reveal any

incremental prognostic usefulness and could be the

main reason for hesitations expressed by Gehi (Gehi

et al., 2005) and Chan (Chan et al., 2006).

The aim of this study was to evaluate usefulness

of several quantitative estimates obtained by means

of method based on Principal Component Analysis

analyzing recordings in patients in acute phase of

myocardial infarction. We expect that comparison of

them to routinely obtained standard estimates of

patient status and outcome results can reveal their

incremental diagnostic utility.

2 METHODS

2.1 Signal Registration and

Pre-processing

Clinical recordings of the signals for investigation

we performed during 24h follow up of the patients

hospitalized in the acute phase of myocardial

infarction in Cardiology Clinics of Lithuanian

University of Health Sciences in Kaunas

(Permission of Kaunas Region Ethics Committee for

Biomedical Research Nr. 169/2004). One lead chest

ECG signal was recorded by means of Heartlab™

system (Dregunas, 1999) (certificate No. LS.

08.02.1957) using 12 bit resolution A/D conversion

at 1000 Hz sampling rate. The age of patients (64

females and 114 males) was ranging from 35 till 84.

Their clinical status according Killip-Kimball

classification was following: 50 patients were of

class I, 109 – of class II, 10 – of class III and 9 – of

class IV.

Signal pre-processing we started with detection

of fiducial point of each cardiocycle – peak of ECG

R-wave. Advanced two steps R-wave peak time

point determination method was necessary to

achieve sufficient accuracy according to our

experience reported in (Simoliuniene et al., 2008).

After preliminary detection using filtered derivative

of the ECG signal we maximized cross-correlation

of the sliding in time R-wave template with the ECG

signal. R-wave template was constructed from first

10 cardiocycles of the recording and updated after

every processed cardiocycle. The values of samples

of just found R-wave of current cardiocycle was

added as 10% of values of updated template shape.

Standardized arrays of 428 time points of R-waves

corresponding to about 6 minutes duration

(depending on actual heart rate) were prepared from

every recording for further analysis.

A mean value of 10 consequent samples in

interval between the end of T-wave of preceding

cardiocycle and beginning of P-wave of current

cardiocycle, was considered as a baseline reference

point of each cardiocycle. Bicubic spline

interpolation using these reference points was used

to calculate baseline wander component, which was

subtracted from the original ECG signal.

Excerpts of ECG signal samples representing S-

T,T complexes of each cardiocycle were taken for

further analysis. Number of samples corresponding

to 2/3 of mean length of RR intervals in the

recording was considered as a length of interval of

samples representing S-T, T complex. This interval

was starting at 100

sample after fiducial point of

cardiocycle. The length of QT interval is varying in

regard to heart rate. This variation is tolerable if we

want to analyze only amplitude alternans, however it

hampering detail shape analysis of S-T, T complex.

We applied time stretching of the ordinary S-T,T

interval to align it with the others using bicubic

spline interpolation, maximizing cross-correlation

with the template constructed from the first 10

cardiocycles. Estimated coefficients for QT interval

time stretching were close to the values reported by

(Sagie et al., 1992), proposed as substitution of

classical Bazett formula. Corrected (stretched)

arrays of samples from each cardiocycle formed

matrix of samples X, which was giving a redundant

but comprehensive representation of variety of the

shape of S-T,T complexes from the recording

considered for analysis:















nppp

xxx

,2,1,

,21,21,2

,12,11,1

...

.........

...

(1)

where x

i,j

is the i

sample of the j

cardiocycle.

RELATIONSHIP BETWEEN QUANTITATIVE T-WAVE ALTERNANS ESTIMATES AND PARAMETERS

DESCRIBING CLINICAL STATUS OF PATIENT IN ACCUTE PHASE OF MYOCARDIAL INFARCTION AND

OUTCOME RESULTS

449

2.2 Principal Component Analysis of

the S-T,T Complexes

Principal Component Analysis was already

successfully used in detection and evaluation of

TWA (Simoliuniene et al., 2008) in test recordings

of “Physionet Challenge 2008” dataset (Moody,

2008).

PCA is used to reduce dimensionality of

redundant representation of S-T,T complexes. The

PCA transforms the original data set into a new set

of vectors (the principal components) which are

uncorrelated and each of them involves information

represented by several interrelated variables in the

original set. The calculated principal components are

ordered so that the very first of them retain most of

the variation present in all the original variables.

Thus it is possible to perform a truncated expansion

of S-T,T complexes representing vectors by using

only the first several principal components. Every

vector x

representing ordinary S-T,T complex is

then represented by linear combination of the

principal components



multiplied by coefficients

i,k





kkii



(2)

Variation of coefficients w

i,k

represents variation in

shape of S-T,T complexes. It was shown in our

previous works that TWA is represented by beat-to-

beat variation of one or mostly few coefficients w

(Simoliuniene et al., 2008).

Minimal yet sufficient number of principal

components to be used for truncated representation

of the signals was determined by means of Wold’s

criteria (Wold, 1978):

)1(

)(





mPRESS

mPR

(3)

where PRESS(m) is calculated as following:







ijij

xxmPRESS

)

()(

(4)

ijm

here is the estimate of the original data set

based not on all but the first m principal

components;

x - the original data set. Final

determination of number was done according to our

experience (Krisciukaitis et al. 2006).

2.3 Detection of T-wave Alternans

Detection of TWA was performed step by step in

consequent intervals of the recordings. TWA was

detected by two methods. First method uses

normalized estimate of power spectral density of the

coefficients w at the highest frequency. Episode of

TWA was registered in case when this estimate at

the highest frequency in 128 coefficients interval

was at least two times bigger then mean of 10

neighbouring lower frequency estimates

(Simoliuniene et al., 2008). Second method is based

on the idea used by (Nearing et al., 2002), that shape

of S-T,T complexes in odd and even cardiocycles

should be similar between each other in the groups

and different between these groups. Unlike moving

average complexes used by (Nearing et al., 2002) we

used derivative quantitative estimates of shape of the

S-T,T complexes – coefficients of principal

components. The shape of ordinary S-T,T complex

is optimally represented by several coefficients of

principal components as a point in multidimensional

orthogonal space. Performing a t-test for means

between two 32 coefficient sets of such coefficients

formed from odd and even cardiocycles (64 is a total

number of cardiocycles in the tested interval) we can

reveal even fine differences in their shape. TWA

was detected when means of two sets were

statistically different at significance level p<0.05.

Final detection of TWA was performed

consolidating results of these two methods, taking

into account only episodes where results of methods

coincided.

2.4 Evaluation of T-wave Alternans

Four quantitative estimates of TWA were selected

for testing:

nTWA – Number of TWA episodes over tested

interval;

siTWA – sum of absolute value of integrated

differences between S-T,T complexes restored using

first five principal components and mean

coefficients of odd and even cardiocycles;

smTWA – maximal amplitude of difference between

restored cardiocycles;

scTWA – sum of absolute values of highest

frequency component of coefficient sequences

extracted by means of hi-pass filter.

All estimates are normalized due to the fixed length

of processed recording interval and give

comprehensive representation of phenomena in

terms of presence/absence, frequency and amplitude.

siTWA and smTWA could be comparable with

estimates used in (Nearing et al. 2002), however

giving better representation being more robust to

BIOSIGNALS 2012 - International Conference on Bio-inspired Systems and Signal Processing

450

noise.

2.5 Estimates of Patient Status and

Outcome Results

Following standard estimates of patient status and

clinical outcome were used:

 Killip-Kimball class index;

 Left Ventricular Ejection Fraction (LVEF);

 Localization of injury by infarction;

 Rehospitalization within 6 months;

 Rehospitalization within 12 months;

In addition mean heart rate and triangular heart rate

variability index (HRVi) were estimated according

(Eur Heart J 1996).

3 RESULTS

PCA of several recordings was performed as a pilot

study to establish minimal yet sufficient number of

principal components to be used for further analysis.

Values of Wold’s criteria PR and percentage of

variation represented by corresponding number of

first principal components is presented in Figure 1.

First five principal components representing more

then 90% of variation in S-T,T complexes were

selected according our experience (Krisciukaitis et

al. 2006) as minimal yet sufficient set of basis

functions for truncated representation of the signals.

0.2

0.4

0.6

0.8

12345678

Nr. Principal Components

Var

Figure 1: Wold’s criteria PR and part of represented

variation (Var) in signal for determination of minimal yet

sufficient number of Principal components for truncated

representation of S-T,T complexes. Specific point in

dependency of PR criteria marked by arrow.

We selected 178 recordings for processing,

where signals were of acceptable quality and free of

power network or other noises and movement

artefacts.

Episodes of TWA were found in 41 out of 178

recordings. Visually was possible to observe

episodes of alternans in T-wave amplitude and/or

shape in some of these 41 recordings and no one

such episode was found in the rest of recordings.

Interestingly TWA was found not only in the

recordings showing low HRVi and high heart rate.

Mean values of RR intervals in TWA episodes

containing recordings were ranging from 750 ms till

1100 ms with HRVi spanning from 3.27 till 25.

Moreover we did not find any episodes of TWA in

some recordings with mean RR intervals below 750

ms. We also did not find any significant relationship

between heart rhythm parameters and any of TWA

parameters. No significant relationship was found

also between TWA and Killip-Kimball class

estimates.

We found statistically significant correlation

between siTWA and LVEF, which was r = -0.456

(p=0.017). Correlation was estimated using

Spearman rank correlation because distribution of

data was non Gaussian.

Values of siTWA showed also relationship with

clinical outcome results: 0.22±0.037 (mean ± SE) in

7 patients who were rehospitalized within 6 months

versus 0.19±0.011 for the rest of 32 patients in TWA

positive group. There was no significant difference

when compared in regard to rehospitalization within

12 months.

TWA estimates obtained using separately only

one of the methods, either spectral (Simoliuniene et

al., 2008), either t-test for means between odd and

even PC coefficient sets, did not show any

significant relationship with LVEF, clinical

outcome, or other parameters describing status of the

patient.

4 DISCUSSION

Importance of aspiration to increase reliability of

TWA detection is supported by the fact that only

TWA estimates obtained using combined analysis

methods involving as much as possible

comprehensive information showed significant

relationship to LVEF, parameter describing clinical

status of the patient in acute phase of myocardial

infarction. LVEF was obtained by means of

independent from ECG method. Increase in siTWA

was related with decrease in LVEF, it complies with

data published by other authors (Chan et al., 2008).

Another fact, that TWA detection method

(Simoliuniene et al., 2008) comparatively good

scored up by results on test recordings from

PhysioNet database shows in some cases disputable

results on clinical recordings, demonstrates that

RELATIONSHIP BETWEEN QUANTITATIVE T-WAVE ALTERNANS ESTIMATES AND PARAMETERS

DESCRIBING CLINICAL STATUS OF PATIENT IN ACCUTE PHASE OF MYOCARDIAL INFARCTION AND

OUTCOME RESULTS

451

further studies on clinical recordings are of great

importance.

We found TWA episodes not only in cases of

high mean heart rate and low heart rate variability

unlikely results reported in earlier publications

concerning phenomena. This fact shows need for

further clinical studies.

Prognostic value of siTWA was demonstrated by

statistically significant relationship with

rehospitalization of patients within 6 months. It

complies with the results of survival studies reported

in (Nieminen et al., 2010). However, our results at

the moment do not provide any significant evidence

of incremental prognostic value of estimates of

TWA.

Further investigations need database of clinical

recordings including as big as possible variety of

standardized clinical recordings. A network based

databank of such recordings based on international

cooperation would be a solution for future

investigations.

5 CONCLUSIONS

Elaboration of quantitative estimates of TWA is

playing major role in development of diagnostic

methods and acquiring by them of wider acceptance

as risk stratification tool.

ACKNOWLEDGEMENTS

This research was funded by a grant (No. MIP-

68/2010) from the Research Council of Lithuania.

REFERENCES

Bloomfield D. M., Steinman R. C., Namerow P. B.,

Parides M., Davidenko J., Kaufman E. S., Shinn T.,

Curtis A., Fontaine J., Holmes D., et al., Microvolt T-

wave alternans distinguishes between patients likely

and patients not likely to benefit from implanted

cardiac defibrillator therapy: a solution to the

Multicenter Automatic Defibrillator Implantation Trial

(MADIT) II conundrum. Circulation 2004;110:1885–

1889.

Chan P. S., Kereiakes D. J., Bartone C., Chow T.,

Usefulness of microvolt T-wave alternans to predict

outcomes in patients with ischemic cardiomyopathy

beyond one year. Am J. Cardiol. 2008 Aug

1;102(3):280-4.

Chan P. S., Nallamothu B. K., Chow T., Microvolt T-

wave alternans: where do we go from here? [letter to

the editor]. J. Am Coll Cardiol 2006;47:1736.

Chow T., Kereiakes D. J., Bartone C., Booth T., Schloss

E. J., Waller T., Chung E., Menon S., Nallamothu B.

K., Chan P. S., Prognostic utility of microvolt T-wave

alternans in risk stratifying patients with ischemic

cardiomyopathy. J. Am Coll Cardiol 2006;47:1820 –

1827.

Clifford G. D., Nemati S., Sameni R., An Artificial Multi-

Channel Model for Generating Abnormal

Electrocardiographic Rhythms. Computers in

Cardiology 2008;35:773−776

Dregunas K., Povilonis E., Cardiac output and

hemodynamic monitoring system “Heartlab”.

"Biomedical engineering" (Proc.Int.Conf.), Kaunas

1999, p.100-105.

Gehi A. K., Stein R. H., Metz L. D., Gomes JA. Microvolt

T-wave alternans for the risk stratification of

ventricular tachyarrhythmic events: a meta-analysis. J.

Am Coll Cardiol 2005;46:75– 82.

Hering H. E., Experimentelle studien an Saugetieren uber

das Elektrocardiogram. Zeitschrift für experimentelle

Pathologie und Therapie 1909;7:363–378.

Krisciukaitis A., Tamosiunas M., Jakuska P., Veteikis R.,

Lekas R., Saferis V., Benetis R., Evaluation of

ischemic injury of the cardiac tissue by using the

principal component analysis of an epicardial

electrogram. Comput Methods Programs Biomed.

2006 May; 82(2):121-129.

Martínez J. P. and Olmos S., Methodological Principles Of

Twa Analysis: A Unified Framework. IEEE

Transactions On Biomedical Engineering, Vol. 52,

No. 4, April 2005.

Moody G. B., The PhysioNet / Computers in Cardiology

Challenge 2008: T-Wave Alternans. Computers in

Cardiology 2008;35:505−508.

Nearing B. D. and Verrier R. L., Modified moving average

analysis of T-wave alternans to predict ventricular

fibrillation with high accuracy. J. Appl Physiol 92: 41–

549, 2002;

Nearing B. D., Verrier R. L., Tracking cardiac electrical

instability by computing interlead heterogeneity of T-

wave morphology. J. Appl Physiol 2003;95:2265–

2272.

Nieminen T., Verrier R. L., Usefulness of T-wave

alternans in sudden death risk stratification and

guiding medical therapy.

Ann Noninvasive

Electrocardiol. 2010 Jul;15(3):276-88.

Simoliuniene R., Krisciukaitis A., Macas A., Baksyte G.,

Saferis V., Zaliunas R., Principal Component Analysis

Based Method for Detection and Evaluation of ECG

T-Wave Alternans. Computers in Cardiology 2008,

vol. 1-2, p. 757-760.

Task Force of the European Society of Cardiology and the

North American Society of Pacing and

Electrophysiology. Heart rate variability. Standards of

measurement, physiological interpretation, and clinical

use. Eur Heart J. 1996;17:354-81.

Wold S, Cross-validatory estimation of the number of

components in factor and pricipal component models.

Technometrics, (1978) 20 pp. 397-405.

BIOSIGNALS 2012 - International Conference on Bio-inspired Systems and Signal Processing

452